BIOGEOCHEMICAL CONSEQUENCES OF LAND COVER AND LAND USE CHANGES IN THE AGRICULTURAL FRONTIER OF THE BRAZILIAN AMAZON BY GILLIAN LAURA GALFORD B.A., WASHINGTON UNIVERSITY IN ST. LOUIS, 2004 M.S., BROWN UNIVERSITY, 2006 A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY IN THE DEPARTMENT OF GEOLOGICAL SCIENCES AT BROWN UNIVERSITY PROVIDENCE, RHODE ISLAND MAY 2010         © Copyright 2010 by Gillian Laura Galford         This dissertation by Gillian L. Galford is accepted in its present form by the Department of Geological Sciences as satisfying the dissertation requirement for the degree of Doctor of Philosophy. Date __________ ______________________________________ John F. Mustard, Advisor Date __________ ______________________________________ Jerry M. Melillo, Advisor Recommended to the Graduate Council Date __________ ______________________________________ Dr. Timothy Herbert, Reader Date __________ ______________________________________ Dr. Leah Vanwey, Reader Date __________ ______________________________________ Dr. Pedro Sanchez, Reader Approved by the Graduate Council Date __________ ______________________________________ Dr. Sheila Bonde, Dean of the Graduate School             iii  iv    CURRICULUM VITAE GILLIAN GALFORD Brown University‐ MBL Joint Graduate Program  7 MBL St. | Woods Hole, MA 02543 | 508.289.7489 | ggalford@mbl.edu  EDUCATION Brown- MBL Joint Graduate Program in Biological & Environmental Sciences    Ph.D. ‘10  Department of Geological Sciences (Brown University) and the Ecosystems Center (MBL)  Thesis title: Biogeochemical Consequences of Land Cover and Land Use Changes in the  Agricultural Frontier of the Brazilian Amazon, Advisors: John Mustard and Jerry Melillo    M.Sc. ‘06  Thesis title: Growing Agricultural Frontiers: Spread and Intensification of Soybean and  Other Mechanized Crops in the Southwestern Brazilian Amazon   Washington University in St. Louis  B.A. ’04, cum laude  Double majors: Earth & Planetary Sciences and Environmental Studies (Social Studies)   Senior thesis title: Geochemical coating & mineral distribution mapping, Kilauea lava  flow, Advisor: Dr. Raymond Arvidson  School of International Training, Arusha, Tanzania Semester program: Tanzania Wildlife Ecology & Conservation, fall 2002    Independent study project: Elephant Migration Corridors of the Manyara‐ Tarangire  Ecosystem, Advisor: Lara Foley (Tarangire Elephant Project)  PUBLICATIONS Galford,  G.L.,  J.  Melillo,  D.  Kicklighter,  J.F.  Mustard,  T.  Cronin,  C.E.P.  Cerri  (In  Review).  Carbon  emissions and uptake from 105 years of land‐cover and land‐use change at the agricultural  frontier of the Brazilian Amazon. Ecological Applications.   Galford, G.L., J. Melillo, J.F. Mustard, C.E. Cerri (In Review). Global Frontier of Land‐Use Change: The  Expansion  and  Intensification  of  Croplands  in  the  Southwestern  Amazon.  Global  Change  Biology.   Agricultural Transitions Research Alliance  (In Prep).  The Amazon: Agricultural  Transitions and their  Unintended Environmental Consequences. Ambio.   Galford, G.L., J.F. Mustard, J. Melillo, A. Gendrin, C.C. Cerri, C.E.P. Cerri (2008). Wavelet analysis of  MODIS time series to detect expansion and intensification of row‐crop agriculture in Brazil.  Remote Sensing of Environment, 112: 576‐587  Neibur, C.S., R.E. Arvidson, E.A. Guinness, and G.L. Galford (2003). Lower Missouri River Flood  Plain At Arrow Rock Before and After the Great Floods of 1993, in At The Confluence:  Rivers, Floods, and Water Quality in the St. Louis Region, ed. R. E. Criss, D. A. Wilson,  Missouri Botanical Gardens Press.     v    RESEARCH EXPERIENCE Nitrogen in African Ecosystems Working Group, The Millennium Villages Project, The Earth Institute at Columbia University and affiliated institutions, 2008-present; Participating in  this interdisciplinary collaboration between US and African scientists to understand how to get  nitrogen into depleted agricultural soils while maximizing nitrogen use efficiency to increase soil  fertility and thus food production while minimizing unintended consequences or disturbances to  ecosystem goods and services. My role is in extrapolations from plot‐level studies of nitrogen cycling  to landscape level analysis through remote sensing and biogeochemical modeling. Funded for one  NSF workshop in Segou, Mali (July 2008) and submitting additional proposals currently. Brown University- MBL, Joint Graduate Program in Biological and Environmental Sciences Ph.D. Candidate, Department of Geological Sciences and The Ecosystems Center, September 2004-present; Generally investigating the impacts of such Land‐Cover and Land‐Use  Change (LCLUC) on biogeochemical cycling, particularly the emission of greenhouse gases CO2 and N2O.  Specific products include estimates of the rates and magnitude of LCLUC in the southern Amazon states Mato  Grosso and Rondônia, Brazil by deriving new methods of processing and interpreting MODIS remote sensing  products. Incorporated LULUC products with in situ biogeochemical data in a regional scale processed‐based  terrestrial ecosystem models to understand how temporal changes in spatial distributions of land cover and  land use affect carbon and nitrogen fluxes to the atmosphere.     Brown University, Research Mentor and Collaborator, Center for Environmental Studies and Spatial Structures in Social Sciences Initiative, December 2005- June 2006; Designed and  facilitated staff research in remote sensing and GIS on the ecological resistance and resilience to Hurricane  Katrina for integration with demographic and socioeconomic data to understand the social and ecological  vulnerabilities of impacted communities and identify the differential impacts of the disaster. Assisted in a field  campaign to collect field validation data on resistance and resilience of natural and human‐dominated  ecosystems.  Washington University/Earth & Planetary Sciences Remote Sensing Laboratory, Undergraduate Honors Research, Department of Earth & Planetary Sciences, January 2003-May 2004; Utilized and tested new algorithms with MASTER and ASTER data sets to detect  geochemical coatings and mineral distributions in the Kilauea lava flows. Determined the algorithm efficacies  before they were used in the then upcoming Mars CRISM mission. Organized and led 15 undergraduate  students in a field trip to the Ka’u Desert to collect ASD spectral data on the lava flows.  Washington University/Earth & Planetary Science Remote Sensing Laboratory, NASA Missouri Space Grant Consortium Intern, Department of Earth & Planetary Sciences, 2000-2004; • Completed satellite and field data fusion through spectral library classifications to map the lithology of the  Southern Soda Mountains, Mojave Desert. Conducted fieldwork for verification, calibration and additional  data coverage of the site using an ASD spectrometer.    • Monitored the Lower Missouri River Floodplain biogeomorphology in response to the Great Floods of 1993.  Demonstrated  the  increasing  homogeneity  and  dominance  of  riparian  forests  in  agricultural  floodplain  abandoned  after  the  floods  due  to  remaining  flood  control  structures,  disproving  the  hypothesis  that  areas  would revert to natural vegetation conditions without management interventions. Data sets included various  resolution satellite data, field validation spectral libraries, and historical land surveys.   Washington University/Earth & Planetary Science Remote Sensing Laboratory, Research Assistant to Mars Exploration Rover (MER) Mission, Department of Earth & Planetary Sciences, 2003-2004; Developed, tested and maintained data archives before and during the  active mission.     vi    African Wildlife Foundation and Tarangire Elephant Project, Independent Study Research Project, School for International Training, Arusha, Tanzania, November- December 2002;  Designed  and implemented an Independent Study Project. Collaborated with researchers  at the Tarangire National Park Elephant Research to compliment their on‐going research program and goals.  Supervised  a  local  Maasai  villager  as  field  assistant,  working  in  Swahili.  Tracked  and  mapped  previously  unknown elephant migration corridors between national parks, conservation areas, game controlled areas and  mountain forest reserves using GPS and GIS applications. Produced maps of migration routes for intervention  points to scare the elephants away in areas where human‐wildlife conflicts had resulted in several elephant  shootings.   ORAL PRESENTATIONS Galford, G.L., J.M.Melillo, D.W. Kicklighter, J. Mustard, T. Cronin, C.E.P. Cerri (2009). Modeling  Greenhouse Gas Emissions from 100 Years of Land‐Cover and Land‐Use Change on the  Amazon Agricultural Frontier. Eos 90(52), Fall Meet. Suppl., Abstract B24B‐02.  Galford, G.L., J. Melillo, J. Mustard, C.E. Cerri (2008). Understanding the Environmental Impacts of  Crop Expansion and Intensification. At: Amazon In Perspective: Integrated Science for a  Sustainable Future (LBA/GEOMA/PPBio), Manaus, Brazil. 20 Nov.  Galford, G.L. (2008, Invited talk). Land‐use transitions along Brazil’s Amazonian Agricultural Frontier.  Bay Paul Center at the Marine Biological Laboratory. Woods Hole, MA. 11 April.   Galford, G.L. (2008, Invited talk). Land‐use transitions along Brazil’s Amazonian Agricultural Frontier.  American Society for Photogrammetry and Remote Sensing, Providence, RI. 08 April.   Galford, G.L. (2007, Invited talk). Land use and global change. Global Change Speaking Series. Center  for Environmental Studies, Brown University. Providence, RI. 07 March.  Galford, G.L. (2007, Invited talk). Remote Sensing for Land Use Studies. Biology Dept. Brown Bag,  Tufts University. Boston, MA, 13 April.  CONFERENCE ABSTRACTS   Galford, G.L., J.M.Melillo, J. Mustard, C.E.P. Cerri (2008). Global Frontier of Land‐Use Change:  Recent Explosion of Croplands in the Southwestern Amazon. Eos Trans. AGU 89(53), Fall  Meet. Suppl., Abstract GC51A‐0649.  Galford, G.L., J. Mustard, J.M.Melillo, C.E.P. Cerri, J.C. Brown, J. Kastens, W. Jepson (2008). Rapid  land‐use changes in the agricultural frontier of Brazil: Remote sensing, ecosystems modeling  and socioeconomic analysis. NASA Joint Carbon Cycle and Ecosystems Science Team  Meeting. Adelphi, MD.   Galford, G. L. (2008). Rapid land‐use changes for large‐scale industrialized agriculture in the Brazilian  Amazon. American Association of Geographers Annual Meeting. Boston, MA.  Galford, G.L., J.F. Mustard, J.M. Melillo, D. Kicklighter, C.E. Cerri, C.C. Cerri (2007). Rapid Land‐Use  Change for Large‐Scale Industrialized Agriculture and Associated Estimates of Greenhouse  Gas Emissions in The Brazilian Amazon. Eos Trans. AGU 88(52), Fall Meet. Suppl., Abstract  B42A‐0895.  Galford, G.L., J.F. Mustard, J.M. Melillo, C.E. Cerri, C.C. Cerri, S.M. Pelkey (2007). MODIS‐based  estimates of row‐crop agricultural expansion in Rondonia and Mato Grosso. LBA‐ECO 11th  Science Team Meeting. Salvador, Brazil. Abstract ID: 27.   Galford, G.L. J.Mustard, J. Melillo, C.C. Cerri, C.E.P. Cerri, D. Kicklighter, B. Felzer (2007).  Biogeochemical consequences of land‐use transitions along Brazil’s agricultural frontier.  NASA Land‐Cover and Land‐Use Change Science Team Meeting. Adelphi, MD.       vii    Galford, G.L., J.F. Mustard, J.M. Melillo, A. Gendrin, C.C. Cerri, C.E. Cerri (2006). Growing Agricultural  Frontiers: Spread and Intensification of Mechanized Agriculture in the Southwestern  Brazilian Amazon. Eos Trans. AGU 87(52), Fall Meet. Suppl., Abstract B31C‐1124.  Galford, G., R.E. Arvidson, F.P. Seelos, B. Jolliff(2001). Eos Trans. AGU, 82(47), Fall Meet. Suppl.,  Abstract P52B‐0586.   T EACHING E XPERIENCE Guest Instructor, Remote  Sensing  module, Watson  International  Scholars  of  the  Environment,  Watson  Institute,  Brown  University,  fall  2009,  spring  2007.  Designed  and  taught  introductory  lectures  on  remote  sensing theory and applications for land‐use and environmental studies.   Teaching Consultant, Harriet W. Sheridan Center for Teaching and Learning, Brown University 2007‐2008.  Provided critical verbal and written feedback to instructors, upon their request, by observing their classroom  teaching style and meeting in a private consultation. Teaching Fellow, Land  Use  and  the  Environment,  Watson  International  Scholars  of  the  Environment,  Watson Institute, Brown University, Spring 2007, Instructor: Dr. Steven Hamburg.   Developed  course  syllabus  as  co‐instructor.  Facilitated  seminar  classes,  delivered  subject  lectures,  lead  field  trips, designed and graded critical analysis writing projects, organized guest lectures. Teaching Assistant, Geographical  Information  Systems  for  Environmental  Applications,  Department  of  Geological  Sciences,  Brown  University,  fall  2006,  Instructor:  Dr.  Wilfrid  Rodriguez.  Lectured  on  different  subjects, created participatory class activities, lead laboratory sections, graded labs and reports. Volunteer Science Teacher, Vartagan Gregorian Elementary School, Providence, RI, 2005‐2007. Designed  and taught elementary school lessons after science funding was cut in the local school district.  Teaching Assistant, Land  Dynamics  &  the  Environment,  The  Pathfinder  Program  in  Environmental  Sustainability,  Washington  University  in  St.  Louis,  fall  2003,  Instructor:  Dr.  Raymond  Arvidson.  Conducted  office hours and help sessions, assisted with lectures, prepared and taught labs, graded lab assignments, co‐ lead field trips. Naturalist/Teacher, Portland  Parks  and  Recreation  Summer  Nature  Camp,  summer  2002,  2003.  Taught  students ages 5‐12 environmental science, nature crafts, and outdoor skills at Hoyt Arboretum, Powell Butte  and Kelly Point Parks. Soils Science Instructor, Multnomah  County  Education  Service  District  Outdoor  School,  1997‐2000.  Volunteered  one  week  a  semester  to  teach  soils  science  to  6th  grade  students  in  this  residential  education  program. Inspired my path into geological sciences.  Group Leader and Mentor, Advocates  for  Women  in  Science,  Engineering  and  Mathematics,  1999‐2000.  Created curriculum for after‐school science projects for 10 middle school girls. Organized field trips to OHSU and other  local science facilities to learn from professional women in science.     P ROFESSIONAL A FFILIATIONS American Geophysical Union 2003-present American Association of Geographers 2007-present Ecological Society of America 2004-present     viii    AWARDS AND HONORS Earth Institute Fellow, Columbia University               2010‐2012  NASA Earth and Space Science Fellowship                      2006‐ present  Brown‐MBL Dissertation Improvement Grant              2009‐2010  Graduate Fellowship Award, Brown University              2004‐2006  Harriet W. Sheridan Center Teaching Certificate I, Brown University                  2006  Women’s Leadership Award, Washington University Women’s Society                  2004  C. Werner Memorial Scholarship for Academic Achievement                     2004  M.E. Bewig Memorial Award, Department of Earth & Planetary Sciences                  2004  The Pathfinder Program in Environmental Sustainability            2000‐2004  Astronaut Scholars Honors Society                       2003‐ present  Astronaut Scholarship Foundation Scholarship                        2003  Portland Area Community Employees (PACE) Scholarship           2001‐2003  Women’s Leadership Training Institute                          2002  Washington University Peer Advisor of the Year             2001‐ 2002  Target All‐Around Scholarship                  2001‐2002  Washington University Dean’s List                          2000  Washington University Eliot Scholar                2000‐2004  Grant High School Valedictorian                           2000  Multnomah County Outdoor School Outstanding Student Leader                   2000  Portland State University Academic Achievement Outstanding Student         1998‐2000  L EADERSHIP AND S ERVICE E XPERIENCES Journal Reviewer, Remote Sensing of Environment, 2008‐present. Co-Founder, MBL Community Garden, 2008. Coordinated and lead development and construction. Chair and co-founder, Geodyssey (Earth & Planetary Sciences Geology Club) , 2001‐2004. Founded  club,  organized  field  trips  and  collaborated  with  other  campus  groups  for  education  outreach  and  geology events.  President, Washington  University  Outing  Club,  2003‐2004.  Drafted  the  mission  statement,  “…Dedicated to providing outdoor recreation and experiential education to all University community  members  interested  regardless  of  financial  means  or  past  experience.”  Organized  and  lead  club  meetings trips. Served as a liaison to school administration and the greater St. Louis community.  Board of Directors and Leadership Training Director,  2003‐2004, The  Wilderness  Project  at  Washington  University,  Created  an  experiential  environment  for  facilitation  transition  to  college,  promoting personal growth and discussing pressing social issues. Volunteer, Elizabeth  Danforth  Butterfly  Garden,  2001‐2004.  Maintained  this  native  garden  dedicated to butterfly habitat restoration with the W.U. Women’s Club. Field Supervisor, Amigos  de  las  Américas,  Paraguay,  2000.  Coordinated  public  health  and  environmental  education  projects  in  rural  communities.  Organized  supply  deliveries  with  Paraguayan  sponsoring  agencies.  Created  and  fostered  relationships  with  local  supporters  to  aid  project sustainability. Supervised 10 volunteers for 8 weeks, arranged volunteer housing, food, and      ix    emergency  transportation.  Conducted  weekly  assessments  of  volunteer  safety,  health  and  project  progress. Completed performance evaluations for each volunteer upon project completion.  Board Member, Portland  Training  Chapter  of  Amigos  de  las  Américas,  1999‐2000.  Recruited  and  selected volunteers, planned fundraisers, served as a trainer at retreats.  Volunteer, Amigos  de  las  Américas,  Dominican  Republic,  Summer  1998.  Promoted  public  health,  community  sanitation  and  environmental  education.  Worked  on  community‐identified  goals:  community  sanitation,  nutrition,  maternal  health,  dental  hygiene,  smoking  prevention,  and  reforestation.  Constructed  10  latrines,  planted  2  community  gardens  and  425  trees,  distributed  200+ toothbrushes.  A DDITIONAL S KILLS Portuguese: conversational intermediate, literary comprehension beginning Spanish: conversational fluency, literary comprehension advanced Sw ahili: conversational fluency, literary comprehension advanced Wilderness First Responder, certified March, 2002         x    PREFACE Introduction Changes in land cover and land use caused by human activities are major anthropogenic components of global environmental change [Foley et al., 2005; Forester et al., 2007]. The land-cover and land-use change (LCLUC) science community began with investigations into the location, timing and magnitude of LCLUCs (e.g., Skole and Tucker 1993). Today, many studies also try to estimate the impacts of LCLUC on agricultural and natural ecosystem function. One major concern is how to balance human development needs for food, fiber, fodder and fuel with the potentially negative impacts on ecosystem functions, such as regulation of greenhouse gas emissions, water cycling, or habitat provisioning. Agricultural development in the tropics has become a large focus of the LCLUC community, as we become increasingly aware of the impacts of tropical deforestation on carbon emissions and biodiversity [Hansen et al., 2008; Myers et al., 2000]. The Brazilian Amazon is one of the most rapidly developing agricultural frontiers, at the expense of natural Amazon rainforest and cerrado (savanna). The Amazon rainforest has long been recognized for its role in fostering biodiversity, in the global hydrologic cycle and in mitigation of greenhouse gases through carbon cycling, earning it the nickname “lungs of the world.” The cerrado, although less well- known than the Amazon forest, is a global biodiversity hotspot [Myers et al., 2000]. Although it is home to more endemic species than the rainforest, less than 2% of the cerrado is in protection, and there is rapid conversion of natural cerrado to pasture and cropland agriculture. Changes to the natural forest and cerrado systems have     xi    global teleconnections through the climate system [Ramos da Silva et al., 2008], in addition to the regional and global consequences for greenhouse gas emissions, threats to biodiversity, and changes in water cycling. Expansion of cropland in the Brazilian Amazon is a relatively new phenomenon that has recently accelerated, as I present in Chapter 1. For many decades, land-use transitions driven by smallholder agriculture, usually for pasture, were the largest source of deforestation in the Amazon [Browder et al., 2004]. Initial clearings were government-supported for colonization and border security, giving way to pasture development through various government aid programs to supply beef. Today, pasture is the largest land-use in the Amazon, but its rate of growth is outpaced by the recent, rapid growth of row-crop agriculture, particularly in the southern Amazon state of Mato Grosso [Barona, 2008; Brown et al., 2005; Nepstad et al., 2006]. The rapid growth in row-crop agriculture, in response to global product demand and national development goals, has been possible because of advances in mechanized farm technology, crop breeding and crop engineering [Brown et al., 2007; Fearnside, 2001; Nepstad et al., 2006]. These croplands are very large, with over 75 percent of the large-scale, mechanized farms cultivating areas more than 500 hectares and ranging to over 5,000 hectares [Fundação Agrisus, 2006] with crops that include soybeans, maize, dry-land rice, cotton and millet. The result has been large-scale conversions of natural ecosystems or lower-production agricultural lands, such as pasture, to expansive row-crop agriculture [Jepson et al., 2008]. Further expansion of agriculture in the region is expected as Brazil and other developing countries grow their economies. The increasing world demand for beef,     xii    soybeans and biofuels will reshape the landscape of Amazon region in the near term, with the expansion of new pastures and croplands and the intensification of extant agriculture. In the last decade, the LCLUC community has focused on a “hot spot” approach of targeting areas where deforestation and agriculture place natural ecosystems in great peril (e.g., Myers et al. 2000). Remote sensing has a long history for detecting LCLUC at local, regional and global scales [DeFries et al., 1998; Fisher and Mustard, 2007]. LCLUC studies have come to rely on repeat observations over time at resolutions appropriate to resolve the processes under investigation. Most recently, the MODIS sensor has provided moderate resolution data (250m to 1 km), with repeat overpasses every few days that allows for detailed analysis of changes in land cover, land use, and phenology [Bradley and Mustard, 2007; Fisher et al., 2006; Galford et al., 2008]. A new front of LCLUC research is incorporating ecosystems modeling with LCLUC data produced from remote sensing analyses to improve estimates of carbon cycling. Many global ecosystems models are being downscaled for regional studies with the LCLUC emphasis on deforestation and agricultural hotspots and with the aid of regionally tuned LCLUC. Remote sensing analysis provides regionally tuned information on spatial and temporal processes such as land clearing and fires, current land uses, and changes in phenology [Brown et al., 2007; Galford et al., 2008; Morton et al., 2008]. These analyses are essential for the regional modeling efforts as they overcome limitations of global data sets in depicting regionally appropriate or accurate land covers and land uses [Bradley and Mustard, 2007]. In     xiii    the Amazon, the drivers of LCLUC are well known [Fearnside, 2001; Jasinski et al., 2005; Morton et al., 2006; Nepstad et al., 2006] and there is a rich history of field research in pastures and forests (e.g., Steudler et al. 1996, Neill et al. 1997, Melillo et al. 2001), but the regional ecosystems impacts are just starting to be understood through ecosystems modeling [DeFries et al., 2008; Potter et al., 2009; Van der werf et al., 2006]. The biogeochemical impacts of LCLUC in the cerrado and croplands have only recently begun to be studied [Carvalho et al., 2009; Jantalia et al., 2007; Miranda et al., 2008]. A few recent carbon budget studies have used ecosystems modeling coupled with new LCLUC products derived from MODIS observations, but they focused only on carbon emissions from land-clearing [DeFries et al., 2008] or carbon dynamics in natural ecosystems [Potter et al., 2009]. The work presented in Chapters 2 and 3 is the first integrated study of both carbon and nitrogen dynamics considering net budgets that account for natural land cover, land clearing, and post- clearing land uses. Today, policy considerations must weigh conservation with decisions of where, how, and how much new land to bring into agricultural use to meet production goals. Deforestation rates significantly increased in the first half of this decade [INPE, 2008a]. Since 2005, lower rates have been observed, but it is unclear if they can be maintained. Deforestation scenarios for the Amazon, based on planned infrastructure development (e.g., paving roads, building hydroelectric plants) to ensure economic growth for Brazil, show large impacts on the southwestern Amazon [Soares-Filho et al., 2006]. In addition, a new biofuels national program announced in 2006 is expected to have a major effect on the Amazon and other     xiv    parts of Brazil, perhaps greater than the impact of the Pro-alcohol Program in 1973, which made Brazil a world leader in ethanol production. Conversely, there are increasing global pressures and incentives for conservation of natural ecosystems, including private sector projects for offsetting carbon emissions elsewhere in the world and government initiatives, such as the UNCCC REDD+ Programme. Overview of Chapters The work presented here incorporates advances in remote sensing and ecosystems modeling to address impacts of LCLUC on landscape composition and the resulting consequences for carbon cycling and greenhouse gas emissions using historical and future scenarios of LCLUC. In Chapter 1, I track contemporary (2000 to 2006) changes in land cover and land use with a new remote sensing methodology (Galford et al. [2008], Appendix A), applied to the entire state of Mato Grosso. I apply a simple bookkeeping model to estimate the impacts of land-cover and land-use change on the regional carbon budget. In Chapter 2, I expand the land-use data set to cover historical changes in land use (1901-2006), reconstructed with government census and survey data that I disaggregate for a spatial reconstruction. The biogeochemical impacts of 105 years of land-use change are evaluated using a process-based ecosystems model (Terrestrial Ecosystems Model, TEM). I find a land-use legacy effect not included in previous land-use modeling efforts for this region. Additionally, the long-term historic modeling effort allows me to assess the CO2 fertilization effect on intact vegetation     xv    that shorter-term studies have dismissed, despite recent field studies documenting this effect in the Amazon [Laurance et al., 2009]. Finally, in Chapter 3, I combine deforestation scenarios and agricultural land use scenarios in TEM to estimate the range of future greenhouse gas emissions, including carbon dioxide, methane and nitrous oxide. The results from this chapter show that nitrous oxide emissions from croplands will become an increasing source of greenhouse gas emissions under all scenarios. I discuss the trade-offs for agricultural development and potential conservation in terms of carbon emissions or uptake. References Barona, E. 2008. Indentifying the role of crop production in land cover change in Brazil, 1990-2006. McGill University, Montreal. Bradley, B. A., and J. F. Mustard. 2007. Comparison of phenology trends by land cover class: a case study in the Great Basin, USA. Global Change Biology 14:334-346. Browder, J. O., M. A. Pedlowski, and P. M. Summers. 2004. Land use patterns in the Brazilian Amazon: Comparative farm-level evidence from Rondônia. Human Ecology 32:197-224. Brown, C. J., M. Koeppe, B. Coles, et al. 2005. Soybean production and conversion of tropical forest in the Brazilian Amazon: The case of Vilhena, Rondônia. Ambio 34:462-469.     xvi    Brown, J. C., W. Jepson, J. H. Kastens, et al. 2007. Multitemporal, moderate-spatial- resolution remote sensing of modern agricultural production and land modification in the Brazilian Amazon. GIScience & Remote Sensing 44:117- 148. Carvalho, J. L. N., C. E. P. Cerri, B. J. Feigel, et al. 2009. Carbon sequestration in agricultural soils in the Cerrado region of the Brazilian Amazon. Soil and Tillage Research 103:342-349. DeFries, R., D. Morton, G. van der werf, et al. 2008. Fire-related carbon emissions from land-use transitions in southern Amazonia. Geophysical Research Letters 35. DeFries, R. S., M. Hansen, J. R. G. Townshend, et al. 1998. Global land cover classifications at 8 km spatial resolution: The use of training data derived from Landsat imagery in decision tree classifiers. International Journal of Remote Sensing 19:3141-3168. Fearnside, P. M. 2001. Soybean cultivation as a threat to the environment in Brazil. Environmental Conservation 28:23-28. Fisher, J. I., and J. F. Mustard. 2007. Cross-scalar satellite phenology from ground, Landsat, and MODIS data. Remote Sensing of Environment 109:261-273. Fisher, J. I., J. F. Mustard, and M. A. Vadeboncoeur. 2006. Green leaf phenology at Landsat resolution: Scaling from the field to the satellite. Remote Sensing of Environment 100:265-279. Foley, J. A., R. DeFries, G. P. Asner, et al. 2005. Global consequences of land use. Science 309:570-574.     xvii    Forester, P., V. Ramaswamy, P. Artaxo, et al. 2007. Changes in Atmospheric Constituents and in Radiative Forcing. In S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H. L. Miller, editors. Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Fundação Agrisus. 2006. Rally da Safra 2006: Situação do plantio direto e da integração lavoura-pecuária no Brasil. Florianópolis. Galford, G. L., J. F. Mustard, J. Melillo, et al. 2008. Wavelet analysis of MODIS time series to detect expansion and intensification of row-crop agriculture in Brazil. Remote Sensing of Environment 112:576-587. Hansen, M. C., S. V. Stehman, P. V. Potapov, et al. 2008. Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data. PNAS 105:9439-9444. INPE. 2008. Taxa de desmatamento annual: Estimativas Anuais desde 1988 até 2008 Projecto PRODES: Monitoramento da floresta Amazônica Brasilera por satélite. Instituto Nacional de Pesquisas Espaciais. Jantalia, C. P., D. V. S. Resck, B. J. R. Alves, et al. 2007. Tillage effect on C stocks of a clayey Oxisol under a soybean-based crop rotation in the Brazilian Cerrado region. Soil and Tillage Research 95:97-109.     xviii    Jasinski, E., D. Morton, and R. DeFries. 2005. Physical landscape correlates of the expansion of mechanized agriculture in Mato Grosso, Brazil. Earth Interactions 9:1-18. Jepson, W., C. J. Brown, and M. Koeppe. 2008. Agricultural intensification on Brazil's Amazonian soybean frontier. Pages 73-92 in A. Millington and W. Jepson, editors. Land Change Science in the Tropics. Springer Publications, Boston. Laurance, S. G. W., W. F. Laurance, H. E. M. Nasciemento, et al. 2009. Long-term variation in Amazon forest dynamics. Journal of Vegetation Science 20:323- 333. Melillo, J. M., P. Steudler, B. J. Feigl, et al. 2001. Nitrous oxide emissions from forests and pastures of various ages in the Brazilian Amazon. Journal of Geophysical Research 106:34,179-134,188. Miranda, A., H. S. Miranda, J. Lloyd, et al. 2008. Fluxes of carbon, water and energy over Brazilian cerrado: an analysis using eddy covariance and stable isotopes. Plant, Cell & Environment 20:315-328. Morton, D. C., R. DeFries, J. Randerson, et al. 2008. Agricultural intensification increases deforestation fire activity in Amazonia. Global Change Biology 14:2262-2275. Morton, D., R. DeFries, Y. E. Shimbukuro, et al. 2006. Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. PNAS 103. Myers, N., R. A. Mittermeier, C. G. Mittermeier, et al. 2000. Biodiversity hotspots for conservation priorities. Nature 403.     xix    Neill, C., J. M. Melillo, P. A. Steudler, et al. 1997. Soil Carbon and Nitrogen Stocks Following Forest Clearing for Pasture in the Southwestern Brazilian Amazon. Ecological Applications 7:1216-1225. Nepstad, D. C., C. M. Stickler, and O. T. Almeida. 2006. Globalization of the Amazon soy and beef industries: Opportunities for conservation. Conservation Biology 20:1595-1603. Potter, C., S. Klooster, A. Huete, V. Genovese, et al. 2009. Terrestrial carbon sinks in the Brazilian Amazon and Cerrado region predicted from MODIS satellite data and ecosystem modeling. Biogeosciences Discussions 6:947-969. Ramos da Silva, R., D. Werth, and R. Avissar. 2008. Regional impacts of future land-cover changes on the Amazon Basin wet-season climate. Journal of Climate 21:1153-1170. Skole, D., and C. Tucker. 1993. Tropical deforestation and habitat fragmentation in the Amazon: Satellite data from 1978 to 1988. Science 260:1905-1910. Soares-Filho, B. S., D. C. Nepstad, L. M. Curran, et al. 2006. Modeling conservation in the Amazon basin. Nature 440:520-523. Steudler, P. A., J. M. Melillo, B. J. Feigel, et al. 1996. Consequence of forest-to- pasture conversion on CH4 fluxes in the Brazilian Amazon Basin. Journal of Geophysical Research 101:18,547-518,554. Van der werf, G., J. Randerson, L. Giglio, et al. 2006. Interannual variability of global biomass burning emissions from 1997 to 2004. Atmospheric Chemistry and Physics Discussions 6:3175-3226.     xx    ACKNOWLEDGEMENTS   To committee members Tim Herbert, Leah VanWey and Pedro Sanchez-- your time and efforts are appreciated. Jack Mustard counseled me to apply to the Brown-MBL program while I was doing fieldwork, five time zones away, two days before the application was due. Since then, he has provided continued commitment and inspiration, given me an intellectual home, and played a crucial role in helping me and his co-mentor, Jerry Melillo, navigate Brown. At our first meeting, Jerry said we do “science for society’s sake” and this has been a theme of our partnership ever since. Thanks to Jerry and Jack for being a team but also letting me explore my own path. Steven Hamburg and Laura Sadovnikoff deserve recognition for bringing me into the Watson International Scholars of the Environment (WISE) program, a unique and exceptional experience. Lynn Carlson has been a constant lifesaver. Jeremy Fisher and Bethany Bradley’s willingness and ability to give sound advice early on was so appreciated that I still ask for it today. The administrative support at Ecosystems eased my transition and helped me navigate since day one-- thank you especially to Mary Ann Seifert. The Brown-MBL program has been a very unique path and the students have been an essential support network despite our far-flung interests. Thank you, Erica Lasek-Nesselquist, Susie Theroux, Alex Valm, Yawei Luo, Yuko Hasegawa, Priya Dutta and Shelby Hayhoe. At Ecosystems, David Kicklighter has had perpetual patience as I have learned TEM and more about ecosystems modeling. This work would not be     xxi    possible without his participation and commitment, for which I am so grateful. Much thanks also to Tim Cronin. The Ecosystems RAs, particularly Clara Funk and Christina Maki who befriended me even before I moved, made Woods Hole feel like home. Thanks to Elissa Schuett, Kate Morkeski and Aaron Strong for warming our first home and loving my dog, Kailua. Thanks to the other community members who have made Woods Hole such a beautiful home—my roommates Connor Ahearn, Peter Rowell and Robin Littlefield, and my friends, Alyssa Hoffert, Courtney Fadness, Valerie Walbek, Kate Stebinger, Anna Stevens, Ellie Tripp and Howard J. Roche III. My work in Brazil is made possible through a partnership between MBL and Universidade de São Paulo. Carlos, Ado, Ana and Alexandra Cerri welcomed me as a foreign researcher, as well as part of their family, and became livelong friends. Carlos and Ado provided data, logistical support, and undying enthusiasm. João Carvalho, Francisco Mello, and Cindy Moreira provided lots of data, translations, logistical support and friendship. Obrigada a todos. At MBL, I have been honored to share an office with Sarah Butler, a brilliant scientist and an amazing friend. She has helped me through quandaries big and small, in science and beyond, in the office and out on our many runs. I am sure the future holds more road races for our team, but I hope also future collaborations. Ray Arvidson is responsible for my interests in remote sensing, dedication to teaching, and for that phone call to Jack from Hawaii. Without him, I never would have embarked on this path. Thanks for always being there.     xxii    Heather Schoeck and Monica Sievertson were my first female mentors and role models in science. They instilled the confidence to pursue my passions while finding ways to give back to the greater good. I look back on these early mentors with great joy as my mind skips ahead to my post-doctoral co-advisors, Ruth DeFries and Cheryl Palm. They have long been an inspiration as women in science and leaders in their field. The last few months have been bright knowing I would work with them, and on quite an amazing project. Thank you to the friends who are my family, offering love and support over the years and who, though distant, have always been there even when they do not understand what I do or why I do the things I do-- Kendra Manton, Kaye Jones, Caryn Corwin, Nigel Davies, and Jen Griffes. The Rajaratnams, the Flemmings, and Jessica Knoll—open doors, open arms and the best relatives I could ask for. On the worst of days, and often despite myself, Andrew Schroth made me smile. On the other days, he brought new adventures, helped me relax and provided a smoked turkey or two. To Griffin and Gleneden Galford, for the best hugs a sister could ever get. To the ones that combed the beaches for pebbles with their toddler and who knew I would be an earth scientist long before I did, I am so grateful for those moments in the sand. To all four of my parents, for always believing in me. For always giving me the freedom to explore as far as my heart could carry me and at the same time, a soft place to land. For making me laugh like no one else can. For being just exactly what we need each other to be—a loving family.         xxiii    Table of Contents CURRICULUM VITAE ......................................................................................................iv PREFACE ..................................................................................................................... x ACKNOWLEDGEMENTS .................................................................................................xx LIST OF TABLES........................................................................................................ xxiv LIST OF FIGURES ....................................................................................................... xxv CHAPTER 1 .................................................................. Error! Bookmark not defined. Tables ................................................................................................................... 36 Figures.................................................................................................................. 42 CHAPTER 2 ................................................................................................................ 52 Tables ................................................................................................................... 89 Figures.................................................................................................................. 93 Appendix 2.A. ..................................................................................................... 101 CHAPTER 3 .............................................................................................................. 105 Tables ................................................................................................................. 145 Figures................................................................................................................ 150 CONCLUSIONS ......................................................................................................... 154 APPENDIX A ............................................................................................................. 171 Tables ................................................................................................................. 205 Figures................................................................................................................ 212     xxiv      LIST OF TABLES Table 1.1 Field data sets.......................................................................................... 36  Table 1.2 Biomass estimates ................................................................................... 37  Table 1.3 Validation results ...................................................................................... 38  Table 1.4 Pasture-to-cropland transitions by biome ................................................. 39  Table 1.5 Major components of Brazil’s carbon budget ........................................... 40  Table 1.6 Greenhouse gas budget (Mato Grosso) ................................................... 41  Table 2.1 Potential vegetation types ........................................................................ 89  Table 2.2 Land cover compositions ......................................................................... 90  Table 2.3 TEM simulations ...................................................................................... 91  Table 2.4 Comparison of studies ............................................................................. 92  Table 3.1 Land covers and land uses by scenario ................................................. 145  Table 3.2 Carbon losses from land clearing ........................................................... 146  Table 3.3 Carbon dynamics in agricultural lands ................................................... 147  Table 3.4 Net greenhouse gas budgets ................................................................. 148  Table 3.5 Carbon in intact natural ecosystems ...................................................... 149  Table A.1 Wavelet powers and accuracy ............................................................... 205  Table A.2 Area in croplands by year ...................................................................... 206  Table A.3 Analysis of unclassified pixels ............................................................... 207  Table A.4 Accuracy assessments .......................................................................... 208  Table A.5 Overall accuracy .................................................................................... 209  Table A.6 Error matrix for classifications at Fazenda Santa Lordes....................... 210  Table A.7 Cropping patterns by year ..................................................................... 211      xxv    LIST OF FIGURES Figure 1.1 Study area .............................................................................................. 44  Figure 1.2 Land-use transitions ............................................................................... 45  Figure 1.3 Cropland Extensification ......................................................................... 46  Figure 1.4 Areas of land-use change ....................................................................... 47  Figure 1.5 Map of cropland extensification .............................................................. 48  Figure 1.6 Cropland area by biome.......................................................................... 49  Figure 1.7 Carbon emissions from cropland extensification ..................................... 50  Figure 1.8 Map of cropland intensification ............................................................... 51  Figure 1.9 Changes in cropping patterns by biome ................................................. 52  Figure 2.1 Study area .............................................................................................. 94  Figure 2.2 Historical reconstruction of agricultural uses ........................................... 95  Figure 2.3 Carbon losses from land-use conversions .............................................. 96  Figure 2.4 Carbon emissions from land-cover and land-use change in Mato Grosso ................................................................................................................................. 97  Figure 2.5 Carbon emissions by land-use transition ................................................ 98  Figure 2.6 Cumulative carbon losses ....................................................................... 99  Figure 2.7 Carbon dynamics in natural vegetation ................................................. 100  Figure 2.A.8 Cropland and pasture reconstructions ............................................... 104  Figure 3.1 Extent of natural land covers, land use in 2006 .................................... 151  Figure 3.2 Climate data sets .................................................................................. 152  Figure 3.3 Changes in agricultural area by scenario .............................................. 153  Figure A.1 Overview of methodology ..................................................................... 212      xxvi    Figure A.2 Location map ........................................................................................ 213  Figure A. 3 False-color infrared MODIS image ...................................................... 214  Figure A.4 EVI time series ..................................................................................... 215  Figure A.5 Histogram of standard deviations ......................................................... 216  Figure A.6 Input data and residuals ....................................................................... 217  Figure A.7 Wavelets............................................................................................... 218  Figure A.8 Examples of wavelet-smoothed time series ......................................... 219  Figure A.9 Example wavelet-smoothed time series ............................................... 220  Figure A.10 Extent of croplands by year ................................................................ 221  Figure A.11 Area detected by cropping pattern ..................................................... 222          CHAPTER 1   The Amazon frontier of land-use change: Croplands and consequences for greenhouse gas emissions Gillian L. Galford a,b, Jerry Melillob, John F. Mustard a, Carlos E.P. Cerri c, Carlos C. Cerri d a) Geological Sciences, Brown University, United States b) The Ecosystems Center, MBL, United States c) Escola Superior de Agricultura Luiz de Queiroz, Universidade de São Paulo, Brazil d) Centro de Energia Nuclear na Agricultura, Universidade de São Paulo, Brazil                   1  2    KEYWORDS: Amazon, deforestation, cropland expansion, remote sensing, MODIS, Mato Grosso, greenhouse gas emissions, land-cover and land-use change ABSTRACT The Brazilian Amazon is one of the most rapidly developing agricultural frontiers in the world. We assess changes in cropland area in the Brazilian agricultural frontier state of Mato Grosso using remote sensing and develop a greenhouse gas emissions budget. The most common type of intensification (“double-cropping intensification”) in this region is a shift from single to double cropping patterns and associated changes in management, including increased fertilization. With remote sensing techniques, we document cropland extensification and double-cropping intensification validated with field data for 85 percent accuracy for detecting croplands and 64 percent and 89 percent accuracy for detecting single and double cropping patterns, respectively. Our results show that croplands more than doubled from 2001-2006 to cover about 100,000 square kilometers, and that new double- cropping intensification occurred on over 20 percent of croplands. We also see variation in the rates of extensification and double-cropping intensification interannually and both are proceeding rapidly. We estimate the carbon loss for the period 2001-2006 due to conversion of natural vegetation to row-crop agriculture in Mato Grosso averages 169 Tg CO2 equivalents per year, over half the typical fossil fuel emissions for the country in recent years. INTRODUCTION     3    Today, agricultural land uses utilize almost 40 percent of earth’s land surface, with rapid growth in tropical regions [Foley et al., 2005; Hansen et al., 2008]. Economic development and global markets have driven large-scale conversions of tropical ecosystems to agricultural use in recent decades [Nepstad et al., 2006]. Agricultural changes in tropical regions may be linked, through global markets, to demands on the other side of the world. Meeting these demands can have large environmental impacts including the release of greenhouse gases through biomass burning and biogeochemical processes affected by clearing activities [Forester et al., 2007]. Localizing land clearing activities in time and geography is essential for improving estimates of impacts on carbon emissions, biogeochemical cycles, climate, and biodiversity [Hansen et al., 2008]. These transformations are particularly evident in the Brazilian Amazon, where clearing rates reached as high as 27,772 km2 in 2004 [INPE, 2008b]. Mato Grosso state is a global hotspot of tropical deforestation and is a major contributor to the Brazilian Amazon “arc of deforestation,” as the state accounts for 38% of Amazon deforestation since 2000 (Figure 1.1, [Hansen et al., 2008; INPE, 2008b]. Land-use transitions driven by smallholder agriculture, usually for pasture, have been the largest source of deforestation in this region for decades. Initial clearings were government-supported for colonization and border security, giving way to pasture development through various government aid programs. Today, pasture is the largest land-use in the Amazon, but its rate of growth is outpaced by the recent, rapid growth of row-crop agriculture, particularly in the southern Amazon state of Mato Grosso [Barona, 2008;     4    Brown et al., 2005; Nepstad et al., 2006]. These cropland operations are very large, with over 75 percent of the large-scale, mechanized soybean farms cultivating areas of more than 500 hectares and ranging to over 5,000 hectares [Fundação Agrisus, 2006]. Recent advances in mechanized farm technology, crop breeding and crop engineering enable the rapid growth in row-crop agriculture, in response to global product demand and national development activities [Brown et al., 2007; Fearnside, 2001; Nepstad et al., 2006]. The result has been large-scale conversions of natural ecosystems or lower-production agricultural lands (e.g., pasture) to expansive row- crop agriculture [Jepson et al., 2008]. Large-scale deforestation continues with little transition back to forest [Rudel et al., 2005]. In the Brazilian Amazon agricultural frontier, the major expansion of croplands with mechanized agriculture is emerging as a significant new dynamic that needs to be further examined [Morton et al., 2006]. The dynamics within agricultural uses in Mato Grosso are complex and represent a variety of pathways leading to cropland extensification and double- cropping intensification (Figure 1.2). Desired increases in crop production may be met through new lands put into production (extensification) or through increased production on existing agricultural lands (intensification) [Boserup, 1965]. Extensification is a horizontal expansion of agriculture, converting natural systems into productive land use. In Mato Grosso, cropland extensification involves land clearing through slash and burn, and leaves no residual slash or woody debris that could hinder the mechanized tilling and planting. Intensification, or the increased production per unit area or per unit time by switching patterns of agricultural management, can be implemented in many different ways, including shifting     5    cropping patterns, selecting new cultivars, controlling weedy growth and pests, using irrigation, and adding nutrients [Gregory et al., 2002; Keys and McConnell, 2005]. A change in cropping-pattern intensity, such as increased plantings of the same crop or a shift in cultivars, are a common type of intensification well documented in other parts of the world [Turner II and Ali, 1996]. In Mato Grosso, double-cropping intensification increases production per area by changing cropping patterns from single cropping (typically soybean) to double cropping (typically soybean followed by corn) in one growing season, enabled by the use of fertilizers in the double-cropping system [Fundação Agrisus, 2006]. Quantifying the space-time dynamics of developing croplands and double-cropping patterns is critical for many areas of ecological assessment and agricultural sustainability, including improved estimates of greenhouse gas emissions, regional climate modeling, hydrological cycling, biodiversity monitoring and agricultural soil fertility. Understanding and documenting land-cover and land-use change over broad spatial and temporal scales has come to rely on remote sensing, particularly in large areas such as the Amazon [Adams et al., 1995; Alves, 2002; Skole and Tucker, 1993]. Since 1988, the Brazilian government has used remote-sensing analyses to track deforestation for enforcement of forest protection laws, often with great success, but also with some uncertainty due to relatively infrequent repeated measurements, the limitations of validation over such a large and remote area as the Amazon, and conservative detection techniques used to minimize false positives [INPE, 2008b]. Today, we can reduce the error in these estimates using higher frequency observations provided by the Moderate Resolution Imaging     6    Spectroradiometer (MODIS; [Justice, 1998] and even distinguish pasture, crops and natural land cover [Anderson et al., 2005; Brown et al., 2005; Galford et al., 2008; Morton et al., 2005]. Recent studies using remote sensing data have moved beyond deforestation to study complex land-use phenomena, such as fire frequency and transitions of natural areas and pastures to croplands [Galford et al., 2008; Morton et al., 2008]. Biogeochemical processes, including nutrient cycling, carbon storage and greenhouse gas regulation, are ecosystem services affected by cropland extensification and intensification in the Amazon. Land clearing, or extensification, releases large amounts of carbon into the atmosphere through slash and burn. Intensification may lead to the depletion of soil nutrients, depending on management, and can contribute to greenhouse gas emissions through increased use of nitrogen fertilizers. Soybeans (used in single cropping) are nitrogen (N) fixing and require no additional nitrogen, although some farms apply modest amounts (up to 10 kg/ha) [Galford, unpublished]. In double-cropping, soybean is followed by corn or other non-N fixing crops that require the application of nitrogen fertilizer as part of the intensification process. Roughly 2 to 5 percent of nitrogen fertilizer applied is converted to nitrous oxide (N2O), a greenhouse gas 300 times more potent than CO2 [Crutzen et al., 2008; Forester et al., 2007; Scanlon and Kiely, 2003]. The subsequent emissions from land-use management may be small compared to carbon emissions from deforestation, but they are important when aggregated over longer time scales, when considering their implications for best management     7    practices for agricultural sustainability, and as sources of local emissions with feedbacks to the global climate system. These two dynamics of cropland land use and cover change, extensification and intensification, are proceeding at a rapid pace in Mato Grosso driven by the emergence of large-scale mechanized agriculture. The evolution of croplands from new extensification to intensified double-cropping patterns is a land-use change process that is well-known for this region [Fundação Agrisus, 2006] but remains unquantified by previous remote sensing studies or by government records (surveys and census data). In this analysis, we quantitatively assess the rates of change associated with these processes using new observations from remotely sensed data, coupled with validation work, to estimate the consequences on regional greenhouse gas emissions.   MATERIALS AND METHODS Study area A complex set of land-use transitions occur in Mato Grosso (Figure 1.2). Pasture and croplands are created from natural ecosystems of forest, cerradão (savanna woodland), and cerrado (scrub savanna). Subsequently, pastures may be converted to cropland and croplands shift from single to double cropping patterns after a few years of cultivation. The Amazon forest has long been recognized for its high biodiversity and role in the global climate as well as for threats to it from deforestation and logging [Dale et al., 1994; Shukla et al., 1990; Skole and Tucker, 1993]. The cerrado is less well-known, but is one of the world’s biodiversity hot spots threatened by agricultural development, as two-thirds of the Brazilian cerrado’s     8    native extent has been converted to agriculture [Klink and Machado, 2005; Myers et al., 2000]. Much of the remaining intact cerrado is in Mato Grosso, but the threat of agricultural conversion creates a challenging juxtaposition of agricultural development and conservation priorities [Conservation International, 2008; Klink and Machado, 2005; Mello, 2007]. Remote sensing of croplands We used newly developed remote sensing techniques [Galford et al., 2008] to analyze MODIS data to detect croplands in Mato Grosso. We created a MODIS Enhanced Vegetation Index (EVI) [Huete et al., 2002] time series from 8-day composite surface reflectance data at a moderately coarse resolution of 500 meters, appropriate for the size of croplands in Mato Grosso (the majority are over 200 ha, many are over 1,000 ha) [Alves, 2002].For this MODIS product, the reflectance value of each pixel represents the best observation over the 8-day period, so there may be several different observation dates within one image. To create a more accurate time series, the actual observation date of each pixel was extracted from the MODIS data product’s ‘date of observation flag’ instead of using a single date of observation for all pixels. We stacked the EVI observations into a time series of remotely sensed green-leaf phenology. Due to the frequent anomalous observations (noise) in these data from clouds and other instrumental and observational effects, we temporally smoothed the time series with a wavelet transform in order to distinguish true phenological peaks from noise. Croplands were detected from the wavelet-smoothed time series by their characteristically high annual standard     9    deviation of green-leaf phenology, since crops have large annual phenological changes as they go from essentially bare soil to extremely uniform green cover. After croplands were identified, we distinguished single- and double-cropping patterns from the number of peaks in the wavelet-smoothed time series over the growing year. We defined a “growing year” that begins in August of one calendar year and ends in July of the following year, which allows us to track the wet growing season peak in January. Each year is analyzed independently of its class from the previous year. We have named the growing year for the year of harvest (e.g. August 2000-July 2001 is referred to as “2001”). The results are classes of cropland, non- cropland and cropping pattern classes (single and double crops). Galford et al. [2008], discuss more details on this methodology. To create a spatially coherent product, we smoothed the cropping pattern classes using a three-by-three pixel window to sieve outlying classes, designating them as unclassed [ITT, 2008]. Unclassed pixels were then clumped [ITT, 2008] into the dominate class within the three-by-three pixel window. Cropping patterns classes from the sieving and clumping procedure were then validated against field data (see “Remote sensing validation” for further discussion). Further phenological analyses of single and double cropping patterns were used to determine areas of extensification (new croplands) and double-cropping intensification (a shift from single to double cropping). We used the 2000-2001 (“2001”) growing season as the baseline year for quantifying change. We classed “extensification” as any area that moved from “not cropland” in 2001 to cropland in any subsequent year and remained in cropland through 2006. We identified the     10    source ecosystem cleared for the cropland using the potential natural vegetation map of Mello [2007], in order to calculate greenhouse gas emissions. Mello [2007] collected and compiled natural vegetation data from various state-level offices in Mato Grosso using a Geographic Information System (G.I.S.). We combined our cropland classification scheme, the natural vegetation map [Mello, 2007] and pasture maps [Morton et al., 2009; Morton et al., 2006] to determine the land-use transitions. We know where and when new croplands are developed directly from each natural ecosystem or from pastures and their former natural ecosystem. The pasture data set is likely an overestimate of planted pasture in the cerrado region as it accounts for managed pastures and natural (unmanaged) grasslands that may or may not be used as natural pastures. Despite this caveat, this is the best spatially explicit estimate of pasture lands available for the study period. Pasture-to-cropland transitions were identified where new croplands were detected in areas mapped as pasture in the previous growing season. Remote sensing validation We used three new cropland data sets for validation (Table 1.1). Each data set documents observed land uses at the time of field visits and/or reconstructed land uses from farm records and G.I.S. maps generated by the researchers or provided by the farms. We included validation points from farms on the border of Mato Grosso, in the neighboring state of Rondônia, to increase the validation data set. This is a soybean-growing region similar to the croplands in Mato Grosso and the data from this area were treated the same as the other field data. Each data set     11    had a slightly different spatial format, such as point observations or polygons of field areas. We standardized the representation of the ground truth observations to facilitate validation. First, all data sets were converted to point data (Figure 1.1) and matched to the corresponding pixels of the remote sensing classified images. Errant field points were identified in a cloud-free 2005 MODIS reflectance image and removed by hand if they met one of the following criteria: 1) points occur on mixed pixels (e.g. forest edges), 2) two points recording the same land use occur in one pixel that would cause double counting , or 3) two points reporting different land uses occur in one pixel, as we cannot say which use makes up the majority of the pixel. The points in these data sets are not randomly stratified across the classes, as they were collected for purposes other than this validation exercise. They may over- represent forested areas and double cropping patterns, but these are the best available field data at the scale and detail required for validation. After validation, the cropland areas detected were spatially subset for the state of Mato Grosso. For further validation, we used Instituto Brasileiro de Geografia e Estatística [IGBE, 2009] annual estimates of croplands to compare with the remote sensing results. We assessed the accuracy of our aggregate statewide cropland area detection by direct comparison to the statewide numbers reported by IBGE [2009]. Annual production records are aggregated by IGBE from monthly data under the direction of the Coordenador Estadual de Pesquisas Agropecuárias (state coordinator of agricultural research) with the aid of the IBGE data collection network, other local government offices, and the producers at município, regional and state     12    levels [IGBE, 2002]. We used the state level data because of uncertainties in the finer-scale records related to how data are collected and reported. For example, data are collected for each município (equivalent to a U.S. county), but the total cropland area in one município may be larger than the total area in the município because a single farm may straddle two municípios and the cropland area will be reported in the município housing the farm headquarters. Comparing remote sensing results to the government estimates on a larger, aggregated level removed some of these smaller-scale artifacts. Estimating greenhouse gas emissions We used a bookkeeping model to calculate carbon losses associated with land-cover and land-use change. We estimated the biomass lost during land-use conversions to croplands by accounting for the area affected from remote sensing inputs. We used published biomass estimates for each biome and for pastures (Table 1.2) and assumed constant values across each biome. In Mato Grosso, abandonment of croplands is rare so vegetation regrowth is not considered [Rudel et al., 2005]. All aboveground biomass was assumed to be completely lost by burning because cultivation practices, such as plowing and harvesting with large machinery, require fields be free of roots, stumps and other forest remnants that could damage farm equipment. Unlike clearing for pastures where some trees are left standing and many stumps and logs persist for decades, croplands are devoid of any signs of the former land-cover.     13    Carbon emissions, as CO2 equivalents (CO2-e), due to the loss of aboveground biomass in land-use transitions from pasture and the major land- covers were given as: CO2-e = (CBIOMASS ) (Area) (CO2 conversion) (1) where CBIOMASS is the carbon content in aboveground biomass of pasture or natural land cover (Table 1.2), Area is the land area converted to crops, and the CO2 conversion scales to CO2 equivalents. Greenhouse gas emissions from intensification We estimated total fertilizer use as a function of area by cropping pattern. We assumed that single crops are soybean, as is typical in the region [Fundação Agrisus, 2006; Galford, unpublished], with a fertilizer input of 10 kg N/ha. Soybean agriculture typically does not require nitrogen additions, but most farmers in the region add this modest amount as “insurance.” Double cropping systems may require fertilizer applications when the second crop (corn) is planted. To estimate the average applied N fertilizer for corn planted in a double cropping pattern, we combined farm records of fertilizer use, field trial fertilization rates that produced crop productivities approximately the Mato Grosso average corn productivity, and recommended N fertilization rates. Studies show that corn may require up to 120 kg N/ha to reach optimal productivity [Mar et al., 2003; Rezende, 2007], but the total N fertilizer added may be somewhat reduced by the N fixed by the soybean grown as     14    the first crop. Data from government surveys show that Mato Grosso corn crops have an average productivity yield of 3.71 metric tons/ha (IBGE 2009). Field trials in the Amazon forest and cerrado regions achieve these productivity levels using fertilizer doses ranging from 0-77 kg N/ ha [Cruz et al., 2005; Embrapa, 2008; Mar et al., 2003; Souza and Sorrato, 2006]. Recommended fertilizer doses range from 34- 120 kg N /ha [Broch and Pedroso, 2008; 2009; Embrapa, 2008]. Farm records from the Amazon show a comparatively low dose of N fertilizer is typically used, with a mean of 23 kg N/ ha (range from 0-70 kg N/ ha) [Cerri et al., 2007; Edgar, 2007; Galford, unpublished]. From these sources, the average N fertilization for corn as a secondary crop is 34.3 kg N/ha [Broch and Pedroso, 2008; 2009; Cerri et al., 2007; Cruz et al., 2005; Edgar, 2007; Embrapa, 2008; Galford, unpublished; Mar et al., 2003; Souza and Sorrato, 2006]. For this work, we used the average fertilizer rate (34 kg N/ha) for second crops and include a high (75 kg N/ha) and low range (0 kg N/ ha) in our sensitivity analysis. We then used the area in single and double crops to estimate N2O losses associated with the statewide fertilizer application: N2O emissions (units CO2-e) = (Fertilizer) (N2O production rate) (GWPN2O) (Area) (2) where Fertilizer is the N fertilizer dose in kg N/ha, the N2O production is a 3 percent conversion rate of N fertilizer to N2O emissions [Crutzen et al., 2008; Scanlon and Kiely, 2003], the Global Warming Potential of N2O (GWPN2O) is 300 to convert to CO2 equivalent emissions [Forester et al., 2007] and Area is the land area being fertilized as a double crop     15    RESULTS Remote Sensing Accuracy We assessed our accuracy in remote sensing analyses with the comparison of field data using overall accuracy, producer’s and user’s accuracy, and KAPPA Khat metrics (Table 1.3). We find our overall accuracy is 86 percent. Further accuracy assessments were conducted to better represent the nature of the data set. The “producer’s” accuracy is a measure of omission by using column totals and “user’s” accuracy is a measure of commission using row totals. The KAPPA Khat statistic assesses accuracy while accounting for the off-trace elements from the error matrix. A Khat value close to one is much better than a random classification, zero is random and negative numbers are worse than random [Jenson, 2005]. Croplands were detected with producer’s accuracy of 98 percent and user’s accuracy of 82 percent. The Khat for the cropland and non-cropland classes was 0.82. For the detection of mechanized agriculture single and double cropping land use, producer’s and user’s accuracies were 64 percent and 42 percent for single crops and 89 percent and 84 percent for double crops. The Khat statistic for non-croplands, single and double crop classes was 0.74. The field data under-represent single crops (Table 1.3) as many of the field sites had no recorded single crops and were not representative of the region, so this may be a bias that causes the validation to suggest that the remote sensing may systematically underestimate the presence of single crops.     16    Our statewide estimates of cropland area generally agree with government estimates [IGBE, 2009], with a tendency to under-represent the cropland area. On average across the study period, the remote sensing data detects 65 percent (standard deviation = 17) of the cropland estimated in IBGE 2009. Looking at individual years, the lowest agreement between these data sets is 40 percent and the highest is 82 percent. Changes in croplands Cropland extensification Total cropland area in Mato Grosso more than doubled from 2001-2006, increasing from 45,497 km2 to 99,488 km2 (Figures 1.3, 1.4, 1.5, and 1.6). We document an average annual rate of increase for agricultural extensification of 0.47 percent. By 2006, Mato Grosso croplands covered 11 percent of the state (Figure 1.6). We see different cropland extensification rates for the different natural biomes (Figures 1.4 and 1.5). When weighted by area of potential natural vegetation type, the cerrado supplied the largest relative amount of land for croplands, with croplands accounting for 18 percent of the total cerrado potential vegetation area. Over 22,000 km2 of new croplands emerged in areas of cerrado potential vegetation, a 10 percent increase in cropland area for Mato Grosso’s cerrado region (Figure 1.6). Rates of conversion to cropland vary by land-cover and land-use sources (Figure 1.4). On average, over 60 percent of conversions in the cerrado region were cerrado to pasture to cropland transitions (Table 1.4). In the cerradão regions,     17    transitions from cerradão to pasture to cropland were the transition pathway for 70% of new croplands (Figure 1.4, Table 1.4). Forest to pasture to cropland conversions accounted for over 60 percent of the forest land-use transitions, and total cropland area in this region increased 24,200 km2 over the study period (Figure 1.4, Table 1.4). Double-cropping intensification Double-cropping intensification increased significantly over the period 2001- 2006, occurring on less than 7 percent (63,962 km2) of Mato Grosso in 2001 and 17 percent (157,436 km2) of the state by 2006. Averaged over the study period, double- cropping intensification accounts for 40% of all croplands, but is not uniformly distributed across the natural ecosystems (Figure 1.8). By biome, cerrado has the largest overall level of double-cropping intensification, with double cropping patterns at almost 50 percent of all croplands (Figure 1.9). Cerradão and forest both have double cropping patterns in less than half the cropland area. Cropland intensification to double cropping in areas with a pasture-to-cropland land-use trajectory account for over half of all the area under double-cropping intensification occurs largely in areas of former cerrado. Rates of conversion to double-cropping intensification vary by source land cover and land use (Figure 1.4). Double-cropping patterns account for an annual average of 47 percent of cerrado croplands, increasing their share of the cerrado land cover from 4 to 8 percent over the study period. In the cerradão system, double cropping rates as a percentage of all cropland area in cerradão ranged from a low of     18    18 percent in 2003 to a high of 30 percent in 2004. Double cropping increased over three-fold during the study period in the forested region, ending with 3 percent double-cropping land use. Estimated greenhouse gas emissions We estimate an annual average of 169 Tg CO2 equivalent per year emissions from cropland extensification in Mato Grosso (Figure 1.7), with forest-to-cropland transitions having the highest carbon emissions at 115 Tg per year CO2 equivalent. Cerrado-to-crop, cerradão-to-crop, and pasture-to-crop transitions have modest emissions (16, 16 and 21 Tg per year CO2 equivalent). Emissions from fertilizer additions with single crops are comparatively low at 0.2 Tg per year CO2 equivalent, equally divided between single and double crops. By adding existing and non-redundant estimates, we developed a relatively complete greenhouse gas budget for Mato Grosso (Table 1.6). The combined estimates are 396 Tg CO2 equivalents per year. It includes estimates of carbon losses from forest conversion to pasture, natural ecosystem and pasture conversions to cropland, pasture maintenance fires and cattle emissions of methane (Table 1.6) [DeFries et al., 2008]; after Steudler et al. [1996]. DISCUSSION Land-use change A growing human population along with dietary changes is expanding the global area of agriculture, with the southern Amazon being one of the most rapidly     19    growing agricultural regions in the world. This and previous studies in the Amazon frontier have documented rapid land-use change for pastures and croplands and point out the differing land-use trajectories within the region [Brown et al., 2007; Galford et al., 2008; Morton et al., 2006]. Here, we show that remote sensing techniques track landscape level processes of cropland extensification and double- cropping intensification with temporal and spatial detail not provided by census or agricultural surveys. We document the average annual rate of agricultural extensification (0.47 percent increase each year) doubled from the 1990-1996 annual rate (0.24 percent) reported in the agricultural census data [IGBE, 2009]. We estimate slightly higher rates of forest conversion to cropland (average >1,500 km2 per year) as compared to previous studies (average 1,350 km2 per year) [Morton et al., 2006]. This could be for several reasons; our remote sensing algorithm is more specifically tuned for crop dectection, we use different land cover data sets to define the forest biomes, and the years included in the study are slightly different. This data set confirms that most lands moved into cropland do not revert to pasture or natural vegetation; in fact, just as suggested by Rudel et al. [2005], croplands show little reversal to natural vegetation at least over the period studied and this region. We observe high interannual variability in intensification and extensification rates within each biome. Extensification was particularly high between the years 2002 and 2003 and occurred mainly in cerrado areas. The cerrado is favoriable for extensification for two reasons. The cerrado region has some of the oldest croplands in the state and existing croplands serve as nucleation for new croplands, as infrastructure is already in place and farm equipment is easily moved between fields.     20    Another incentive to choose extensification in cerrado instead of forest is the ease of clearing cerrado, as lower aboveground biomass and smaller root structures make it easier to clear cerrado for croplands. Extensification rates for cerrado to cropland transitions slow in 2005 and 2006 while double-cropping intensification continues to increase. The continued double-cropping intensification in cerrado may be due to a shift in the relative profits from Brazilian soybeans dropping with falling Brazilian currency in 2004 [Nepstad et al., 2006], which would make intensification more cost- effective than extensification at that time. Futher, double-cropping intensification is more widespread in the cerrado than in other biomes, because double-cropping intensification generally occurs a few years after extensification and there is a longer history of extensification in the cerrado (primarily southern Mato Grosso). For croplands from areas of forest and cerradão, extensification continues steadily while intensification rates remain low. Existing croplands from the forest and cerradão biomes represent a potential target for intensification that follows environmentally- sound production guidelines. The process of cropland development is dynamic and the land-cover and land-use change story does not stop at land clearing. Suitable areas for mechanized croplands [Jasinski et al., 2005] need to be large and amenable to large-scale mechanized agriculture, e.g., fairly level topography. Soils may be amended with lime to reduce the aluminotoxicity. Lands are cleared through slash and burn until there are no roots or slash to foul mechinery. The large-scale nature of these croplands is coupled both to scale-dependence of profitable farming and the investors who will encourage that scale. From our observations, extensification of     21    new single crops largely proceeds by expanding into adjacent lands, rather than leaping to new development distrant from the previous cropland activity centers. This spatial clumping is highly evident in the cerrado region, where most croplands have been concentrated. We see that double cropping is commonly carried out in the densest regions of mechanized croplands, not outlying regions that are newly established. Double-cropping patterns (soy-corn) typically emerge after several years of single cropping (soy). The time-lag of extensification explains why the highest rates of double cropping are found in the cerrado, as this is the oldest cropland region in the state and hold the greatest amount of cropland by bioregion. Greenhouse gas consequences Land-use changes across the globe have contributed 35 percent of anthropogenic CO2 emissions over the last 150 years [Foley et al., 2005]. In the global context, Mato Grosso accounts for over 2 percent of the contemporary global greenhouse gas emissions from land-cover and land-use changes [Denman et al., 2007b], while comprising less than 1 percent of the earth’s land surface. Annual greenhouse gas emissions from land-use extensification in Mato Grosso are equivalent to over half the Brazilian carbon emissions from fossil fuel burning and over 25 percent of all carbon emissions from Amazonian deforestation and from cerrado clearing (Table 1.5). Mato Grosso cropland extensification alone contributes at least half of the estimated CO2 losses from all biomass burning throughout southern-hemisphere South America [Van der werf et al., 2006]. Pasture is still the dominant land use in the Amazon and Mato Grosso, but this work shows new clearing for cropland extensification accounts for almost half of Mato Grosso’s total     22    greenhouse gas budget (Table 1.6). If current trends in land-use change continue, cropland extensification may become the largest regional source of greenhouse gas emissions. Greenhouse gas emissions in Mato Grosso related to cropland intensification from N fertilizer, totaling 0.1 Tg CO2 equivalent per year, are quite small today. Depending on shifting crop types, fertilizer prices, and management strategies that affect soil fertility and/or nitrogen use efficiency, there is potential for these emissions to grow substanially. Regional field trials show diminishing returns on increasing fertilizer application [Mar et al., 2003; Souza and Sorrato, 2006], suggesting increasing N fertilization in this region may lead to increased N2O emissions without a proportional increase in crop productivity. We lack information on N2O production from soybean fields and how this changes over time (e.g. years planted in soy), which requires further field studies. Remaining uncertainties in the total greenhouse gas budget could be clarified through total cost accounting that includes emissions from termite mounds, other animal sources, and post-clearing soil biogeochemistry dynamics of based on management, such as the impacts of tillage practices on soil carbon storage [Cerri et al., 2004]. Ecosystems Consequences This work provides information on the mosaic of landscape heterogeneity, land-cover types, and length of crop cover as it changes with single and double cropping that may be useful in future climate modeling efforts. These large-scale     23    changes in land-cover in the Amazon have global teleconnections in the global climate system (e.g., increased precipitation in the the U.S. Midwest) [Avissar and Werth, 2005]. Climate research in the Amazon shows that, locally, cleared lands adjacent to natural ecosystems can alter convection patterns, leading to increased dry season cloud cover, increased high energy storms, and stronger storm clouds [Ramos da Silva et al., 2008; Roy and Avissar, 2002]. Croplands may have negative feedbacks to their own microclimate, as models show low stature vegetation becomes drier and warmer than the original natural ecosystems [Costa et al., 2007]. Land-use change affects the bidirectional relationship between biodiversity and ecosystem function. Here, we document the rate of disappearance of natural ecosystems and provide information on changes in cropland area that may be used to estimate impacts on biodiversity, while adding richness in understanding trade- offs between agricultural development and natural ecosystem conservation. Species gains and losses related to changing habitats may alter ecosystem function. There are several ecosystems characteristics related to biodiversity could be assessed with remote sensing methods. It is imperative to note that conservation of these uniquely diverse ecosystems and their associated species biodiversity is important in its own right and that ecosystem function and climate regulation are only a few of many additional reasons for conservation. The agricultural mosaic of land-use and the dynamic nature of cropland uses impacts environmental and agricultural sustainability. The landscape configuration can have cascading affects on surrounding intact ecosystems. Information on land use configurations in the matrix of natural land covers can be applied in conservation     24    work to reduce environmental impacts, such as in protecting waterways and wildlife corridors [Green et al., 2005]. The agricultural dynamics we observe suggest that cropland extensification and double-cropping intensification will likely continue in this region through the coming decades. Thus, conservation priorities must focus on 1) increasing double-cropping intensification over new extensification to reduce the regional carbon emissions and preserve other ecosystem services while responding to global product demand, and 2) understanding and communicating best management practices for croplands that synergistically reduce the environmental impact of croplands and increase farm production. Conclusions As suggested by Morton et al. [2006], new land clearing for croplands is an emerging force that is rapidly increasing in the Brazilian agricultural frontier. We have verified this trend using a different set of detection algorithms, while enhancing the story of cropland dynamics, by tracking single crops maturing to double cropping patterns. Double-cropping intensification is emerging as a new and major component of regional land uses, with rapid increases along with extensification. Information from remote sensing on the location and extent of cropping patterns is a useful addition to other data sets, such as deforestation detections and crop surveys conducted by the Brazilian government. There are a wide range of applications and implications from these cropland dynamics. Here, we quantify the greenhouse gas implications of cropland extensification and intensification but to assess the true environmental impact, this information on land-cover and land-use changes needs to     25    be evaluated for its interactions with local and regional climate, hydrology, and biodiversity. ACKNOWLEDGEMENTS This work was supported by NASA’s Earth and Space Science Fellowship (G.L. Galford) and Large-Scale Biosphere Atmosphere Experiment in Amazonia (Grant no. NNG06GE20A), and the Environmental Change Initiative at Brown University. We kindly thank colleagues Cindy Moreira, Marcos Neto and CENA drivers for their assistance in the field research. A special thanks to J. Christopher Brown, Claudia Stickler and their colleagues for the use of their field data. We are grateful to Kim Cahill and three anonymous reviewers for their comments on an earlier version of this manuscript.    REFERENCES  Adams, J. B., D. E. Sabol, V. Kapos, R. A. Filho, D. Roberts, M. O. Smith, and A. R. Gillespie. 1995. Classification of multispectral images based on fractions of endmembers: Application to land-cover change in the Brazilian Amazon. Remote Sensing of Environment 52:137-154. Alves, D. S. 2002. Space-time dynamics of deforestation in Brazilian Amazonia. International Journal of Remote Sensing 23:2903-2908. Anderson, L. O., Y. E. Shimabukuro, R. DeFries, and D. Morton. 2005. Assessment of land cover and land use changes in the Brazilian Amazon using multitemporal fraction images derived from Terra MODIS: Examples from the     26    state of Mato Grosso. IEEE Transactions on Geosciences and Remote Sensing 2:315-318. Avissar, R., and D. Werth. 2005. Global hydroclimatological teleconnections resulting from tropical deforestation. Journal of Hydrometeorology 6:134-145. Barona, E. 2008. Indentifying the role of crop production in land cover change in Brazil, 1990-2006. McGill University, Montreal. Boserup, E. 1965. The conditions of agricultural growth: The economics of agrarian change under population pressure. George Allen and Unwin Limited, London, United Kingdom. Broch, D. L., and R. S. Pedroso. 2008. Custo de Produção da Cultura do Milho. Page 1 in F. MS, editor. Fundação MS, Maracaju, MS. Brown, C. J., M. Koeppe, B. Coles, and K. P. Price. 2005. Soybean production and conversion of tropical forest in the Brazilian Amazon: The case of Vilhena, Rondônia. Ambio 34:462-469. Brown, J. C., W. Jepson, J. H. Kastens, B. D. Wardlow, J. M. Lomas, and K. P. Price. 2007. Multitemporal, moderate-spatial-resolution remote sensing of modern agricultural production and land modification in the Brazilian Amazon. GIScience & Remote Sensing 44:117-148. Buschbacher, R., C. Uhl, and A. S. Serrao. 1988. Abandoned pastures in eastern Amazonia II: Nutrient stocks in the soil and vegetation. Journal of Ecology 76:682-699.     27    Castro, E. A., and J. B. Kauffman. 1998. Ecosystem structure in the Brazilian Cerrado: A vegetation gradient of aboveground biomass, root mass and consumption by fire. Journal of Tropical Ecology 14:263-283. Cerri, C. C., M. Bernoux, C. E. P. Cerri, and C. Feller. 2004. Carbon cycling and sequestration opportunities in South America: the case of Brazil. Soil Use and Management 20:248-254. Cerri, C. E. P., M. Easter, K. Paustian, K. Killian, K. Coleman, M. Bernoux, P. Falloon, D. Powlson, N. Batjes, E. Milne, and C. C. Cerri. 2007. Simulating SOC changes in 11 land use change chronosequences from the Brazilian Amazon with the RothC and Century models. Agriculture, Ecosystems & Environment 122:46-57. Conservation International. 2008. Cerrado. Biodiversity hotspots. www.biodiversityhotspots.org 20 January 2008. Costa, M. H., S. N. M. Yanagi, P. J. O. P. Souza, A. Ribeiro, and E. J. P. Rocha. 2007. Climate change in Amazonia caused by soybean cropland expansion, as compared to caused by pastureland expansion. Geophysical Research Letters 34. Crutzen, P. J., A. R. Mosier, K. A. Smith, and W. Winiwarter. 2008. N2O release from agro-biofuels production negates global warming reduction by replacing fossil fuels. Atmospheric Chemistry and Physics 8:389-395. Cruz, J. C., F. T. F. Pereira, I. A. P. Filho, and A. M. Coelho. 2005. Resposta de Cultivares de Milho à Adubação Nitrogenada em Cobertura in P. e. A.     28    Ministério de Agricultura, editor. Embrapa Milho e Sorgo, MG, Sete Lagoas, MG, Brazil. Dale, V. H., S. M. Pearson, H. Offerman, and R. V. O'Neill. 1994. Relating patterns of land-use change to faunal biodiversity in the Central Amazon. Conservation Biology 8:1027-1036. DeFries, R., D. Morton, G. van der werf, L. Giglio, G. J. Collatz, J. Randerson, R. A. Houghton, P. Kasibhatla, and Y. E. Shimabukuro. 2008. Fire-related carbon emissions from land-use transitions in southern Amazonia. Geophysical Research Letters 35. Denman, K. L., G. Brasseur, A. Chidthaisong, P. Ciais, P. M. Cox, R. E. Dickinson, D. Hauglustaine, C. Heinze, E. Holland, D. Jacob, U. Lohmann, S. Ramachandran, P. L. d. S. Dias, S. C. Wofsy, and X. Zhang. 2007. Couplings between changes in the climate system and biogeochemistry. In S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H. L. Miller, editors. Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Edgar, J. 2007. IFB CropWatchers: FarmWeek 11/30/2007. Embrapa. 2008. Cultivares de milho, feijão, girassol, e mandioca para o Agrest Sergipano com foco na Agricultura Familiar e no Agronegócio. Page 31 in H. W. L. Carvalho, I. R. Olivera, E. R. C. Donald, S. R. R. Ramos, C. A. P. Pacheco, W. M. G. Fukuda, M. J. Peloso, C. G. P. Carvalho, K. E. O. Melo, L.     29    F. Feitosa, and A. F. Menezes, editors. Documentos / Embrapa Tabulerios Costeiros. Embrapa Tabuleiros Costeiros, Aracaju, SE. Fearnside, P. M. 2001. Soybean cultivation as a threat to the environment in Brazil. Environmental Conservation 28:23-28. Foley, J. A., R. DeFries, G. P. Asner, C. Barford, G. Bonan, S. R. Carpenter, F. S. Chapin, M. T. Coe, G. C. Daily, H. K. Gibbs, J. H. Helkowski, T. Holloway, E. A. Howard, C. J. Kucharik, C. Monfreda, J. A. Patz, I. C. Prentice, N. Ramankutty, and P. K. Snyder. 2005. Global consequences of land use. Science 309:570-574. Forester, P., V. Ramaswamy, P. Artaxo, T. Berntsen, R. Betts, D. W. Fahey, J. Haywood, J. Lean, D. C. Lowe, G. Myhre, J. Nganga, R. Prinn, G. Raga, M. Schulz, and R. Van Dorland. 2007. Changes in Atmospheric Constituents and in Radiative Forcing. In S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H. L. Miller, editors. Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Fundação Agrisus. 2006. Rally da Safra 2006: Situação do plantio direto e da integração lavoura-pecuária no Brasil. Florianópolis. Galford, G. L. unpublished. Field observations, GIS maps and farm records. Collected 2005, 2007.     30    Galford, G. L., J. F. Mustard, J. Melillo, A. Gendrin, C. C. Cerri, and C. E. P. Cerri. 2008. Wavelet analysis of MODIS time series to detect expansion and intensification of row-crop agriculture in Brazil. Remote Sensing of Environment 112:576-587. Green, R. E., S. J. Cornell, J. P. W. Scharlemann, and A. Balmford. 2005. Farming and the fate of wild nature. Science 307:550-555. Gregory, P. J., J. S. I. Ingram, R. Andersson, R. A. Betts, V. Brovkin, T. N. Chase, P. R. Grace, A. J. Gray, N. Hamilton, T. B. Hardy, S. M. Howden, A. Jenkins, M. Meybeck, M. Olsson, I. Ortiz-Monasterio, C. A. Palm, T. W. Payn, M. Rummukainen, R. E. Schulze, M. Theirn, C. Valentin, and M. J. Wilkinson. 2002. Environmental consequences of alternative practices for intensifying crop production. Agriculture, Ecosystems & Environment 88:279-290. Hansen, M. C., S. V. Stehman, P. V. Potapov, T. R. Loveland, J. R. G. Townshend, R. DeFries, K. W. Pittman, B. Arunarwati, F. Stolle, M. K. Steininger, M. Carroll, and C. Dimiceli. 2008. Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data. PNAS 105:9439-9444. Houghton, R. A., K. T. Lawrence, J. L. Hackler, and S. Brown. 2001. The spatial distribution of forest biomass in the Brazilian Amazon: a comparison of estimates. Global Change Biology 7:731-746. Houghton, R. A., D. Skole, C. A. Nobre, J. L. Hackler, K. T. Lawrence, and W. H. Chomentowski. 2000. Annual fluxes of carbon from deforestation and regrowth in the Brazilian Amazon. Nature 403:301-304.     31    Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83:195-213. IGBE. 2002. Pesquisas Agropecuárias.in M. d. Planejamento, editor. Ministério do Planejamento, Orçamento e Gestão, Rio de Janeiro. IGBE. 2009. Banco de Dados Agregados: Sistem IBGE de Recuperação Automática (SIDRA). INPE. 2008. Sistema DETER. Instituto Nacional de Pesquisas Espaciais. ITT. 2008. ENVI Reference Guide, version 4.6. ITT Visual Information Solutions. Jasinski, E., D. Morton, and R. DeFries. 2005. Physical landscape correlates of the expansion of mechanized agriculture in Mato Grosso, Brazil. Earth Interactions 9:1-18. Jenson, J. R. 2005. Thematic map accuracy assessment. Pages 495-513 Introductory Digital Image Processing: A Remote Sensing Perspective. Pearson Pretice Hall, Upper Saddle River. Jepson, W., C. J. Brown, and M. Koeppe. 2008. Agricultural intensification on Brazil's Amazonian soybean frontier. Pages 73-92 in A. Millington and W. Jepson, editors. Land Change Science in the Tropics. Springer Publications, Boston. Justice, C. O. 1998. The moderate resolution imaging spectroradiometer (MODIS): Land remote sensing for global change research. IEEE Transactions on Geosciences and Remote Sensing 36:1228-1249.     32    Keys, E., and W. H. McConnell. 2005. Global change and the intensification of agriculture in the tropics. Global Environmental Change 15:320-337. Klink, C. A., and R. B. Machado. 2005. Conservation of the Brazilian Cerrado. Conservation Biology 19:707-713. Mar, G. D., M. E. Marchetti, L. C. F. Souza, M. C. Goncalves, and J. O. Novelino. 2003. Fertilidade do solo e nutricao de plantas: Producao do milho safrinha em funcao de doses e epocas de aplicacao de nitrogenio. Bragantia 62:267- 274. Marland, G., T. A. Boden, and R. J. Andres. 2008. Global, regional, and national CO2 emissions. Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy, Oak Ridge, Tenn., USA. Mello, F. 2007. Estimativas dos estoques de carbono dos solos nos Estados de Rondônia e Mato Grosso anteriores à intervenção antrópica. Universidade de São Paulo, Piracicaba, Brazil. Ministry of Science and Technology. 2004. Carbon Dioxide Emissions and Removals from Forest Conversion and Abandonment of Managed Lands. Brazilian Ministry of Science and Technology, Brasilia. Ministry of Science and Technology. 2006. Carbon dioxide emissions and removals from forest conversion and abandonment of managed lands First Brazilian Inventory of Anthropogenic Greenhouse Gas Emissions. Brazilian Ministry of Science and Technology, Brasilia. Morton, D., R. DeFries, and Y. E. Shimabukuro. 2009. Cropland expansion in cerrado and transition forest ecosystems: Quantifying habitat loss from     33    satellite-based vegetation phenology. In C. A. Klink, editor. Cerrado Land Use and Conservation, Advanced Applied Biodiversity Science. Conservation International, Washington, D.C. Morton, D. C., R. DeFries, J. Randerson, L. Giglio, W. Schroeder, and G. van der werf. 2008. Agricultural intensification increases deforestation fire activity in Amazonia. Global Change Biology 14:2262-2275. Morton, D. C., R. S. DeFries, Y. E. Shimabukuro, L. O. Anderson, F. D. B. Espirito- Santo, M. Hansen, and M. Carroll. 2005. Rapid assessment of annual deforestation in the Brazilian Amazon using MODIS data. Earth Interactions 9:1-22. Morton, D. C., R. S. DeFries, Y. E. Shimbukuro, L. O. Anderson, E. Arai, F. d. B. Espirito-Santo, R. Freitas, and J. Morisette. 2006. Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. PNAS 103. Myers, N., R. A. Mittermeier, C. G. Mittermeier, G. A. B. d. Fonseca, and J. Kent. 2000. Biodiversity hotspots for conservation priorities. Nature 403. Nepstad, D. C., C. M. Stickler, and O. T. Almeida. 2006. Globalization of the Amazon soy and beef industries: Opportunities for conservation. Conservation Biology 20:1595-1603. Nogueira, E. M., P. M. Fearnside, B. W. Nelson, R. I. Barbosa, and E. W. H. Kiezer. 2008. Estimates of forest biomass in the Brazilian Amazon: New allometric equations and adjustments to biomass from wood-volume inventories. Forest Ecology and Management 256:1853-1867.     34    Potter, C., S. Klooster, A. Huete, V. Genovese, M. Bustamante, L. G. Ferreira, R. Cosme de Oliveira Junior, and R. Zepp. 2009. Terrestrial carbon sinks in the Brazilian Amazon and Cerrado region predicted from MODIS satellite data and ecosystem modeling. Biogeosciences Discussions 6:947-969. Ramos da Silva, R., D. Werth, and R. Avissar. 2008. Regional impacts of future land-cover changes on the Amazon Basin wet-season climate. Journal of Climate 21:1153-1170. Rezende, P. J. L. A. 2007. Cultivares, doses de fertilizantes e densidades de semeadura no cultivo de milho safrinha. UFLA, Lavras. Roy, S. B., and R. Avissar. 2002. Impact of land use/land cover change on regional hydrometeorology in Amazonia. Journal of Geophysical Research 107. Rudel, T. K., O. Coomes, E. Moran, A. Angelsen, F. Achard, E. Lambin, and J. Xu. 2005. Forest transitions: towards an understanding of global land use change. Global Environmental Change 14:23-31. Saatchi, S. S., R. A. Houghton, R. D. Alvala, J. V. Soares, and Y. Yu. 2007. Distribution of aboveground live biomass in the Amazon basin. Global Change Biology 13:816-837. Saatchi, S. S., R. A. Houghton, D. S. Alves, and B. W. Nelson. 2009. LBA-ECO LC- 15 Amazon Basin Aboveground Live Biomass Distribution Map: 1990-2000. Oak Ridge National Laboratory Distributed Active Archive Center. Scanlon, T. M., and G. Kiely. 2003. Ecosystem-scale measurements of nitrous oxide fluxes for intensely grazed, fertilized grassland. Geophysical Research Letters 30:1852-1856.     35    Shukla, J., C. Nobre, and P. Sellers. 1990. Amazon Deforestation and Climate Change. Science 247:1322-1325. Skole, D., and C. Tucker. 1993. Tropical deforestation and habitat fragmentation in the Amazon: Satellite data from 1978 to 1988. Science 260:1905-1910. Souza, E. F. C., and R. P. Sorrato. 2006. Efeito de fonts e doses de nitrogênio em cobertura, no milho safrinha, em plantio direto. Revista Brasileira de Milho e Sorgo 5:395-405. Steudler, P. A., J. M. Melillo, B. J. Feigel, C. Neill, M. C. Piccolo, and C. C. Cerri. 1996. Consequence of forest-to-pasture conversion on CH4 fluxes in the Brazilian Amazon Basin. Stickler, C. M. unpublished. Field validation data for the Xingu Basin. Collected 2006. Turner II, B. L., and A. M. A. Ali. 1996. Induced intensification: Agricultural change in Bangladesh with implications for Mathus and Boserup. PNAS 93:14984- 14991. Van der werf, G., J. Randerson, G. J. Collatz, and L. Giglio. 2003. Carbon emissions from fires in tropical and subtropical ecosystems. Global Change Biology 9:547-562. Van der werf, G., J. Randerson, L. Giglio, G. J. Collatz, P. Kasibhatla, and A. F. J. Arellano. 2006. Interannual variability of global biomass burning emissions from 1997 to 2004. Atmospheric Chermistry and Physics Discussions 6:3175- 3226.     36      Tables   Table 1.1 Field data sets   Description of field data sets used for accuracy assessment for the classes of cropland and cropping patterns detected with remote sensing. DATA SET  METHODS   SPATIAL DATA  YEAR  [Brown et al., 2007]  Farm interviews & mapping for land‐ Field mapping  2005  use history  as polygons    [Galford, unpublished]  Farm records & mapping for land‐use  Field mapping  2006  history  as polygons    [Stickler, unpublished]  Regional transect  GPS point data  2006  Observations of land‐cover and land‐   use changes                          37                                        Table 1.2 Biomass estimates Biomass estimates (aboveground + belowground) used to calculate carbon savings and losses. A (*) indicates the estimate used here, (∆) represents the low estimate and (○) represents the high estimate used. Land cover or land use  Carbon (106 g/ha)  Reference(s)  Forest  130∆‐442○  [Saatchi et al., 2007; Saatchi et al.,  130‐220 (200*)  2009]     [Houghton et al., 2001]  Cerradão   98∆  [Nogueira et al., 2008; Saatchi et  113*  al., 2007]  195○  [Saatchi et al., 2007]  Cerrado  13∆  [Castro and Kauffman, 1998]  33*  [Ministry of Science and  45○  Technology, 2004; Saatchi et al.,    2009]    Pasture (well managed)  10  [Buschbacher et al., 1988]                    38                                            Table 1.3 Validation results Validation results (pixels counts) for remotely-sensed classes of cropping patterns compared to the collective pool of all field data sets.   Remotely‐sensed    Field data        classes  Single cropping  Double cropping  Non‐cropland  Row total  User's accuracy Single cropping  47  16  50  113  41.6 percent  Double cropping  26  163  4  193  84.5 percent  Not row‐crop  0  4  383  387  99.0 percent  Column total  73  183  437  693    Producer's  accuracy  64.4 percent  89.1 percent  87.7 percent        Overall accuracy 85.6 percent                          39                                        Table 1.4 Pasture-to-cropland transitions by biome   The percentage of natural vegetation to pasture to cropland transitions out of all transitions to cropland by natural vegetation type and year. “Year of conversion” is the year prior to the first crop harvest. For example, an area converted in 2005 would first be harvested in 2006, so only conversions between 2001 and 2006 are reported here. Also, see Figure 1.4 for net (area) transitions along each land-use trajectory.   Year of conversion   Pasture‐to‐cropland transitions  (percent of all cropland transitions)  NATURAL ECOSYSTEM   CERRADO  CERRADÃO  FOREST  OF ORIGIN  2002  60  67  64  2003  66  65  60  2004  82  78  68  2005  42  67  53  Annual average  63  69  61  Standard deviation  17  6  6      40                                          Table 1.5 Major components of Brazil’s carbon budget Comparison of select major components of the carbon cycle in Brazil. Tg C per  Region  Greenhouse gas source  year  Year of estimate  Reference  Brazil  Fossil fuels  89  2005          [Marland et al., 2008]  Amazon  Forest clearing  116  2006   [Ministry of Science and          Technology, 2006]                      Cerrado clearing  52  2006        [Ministry of Science and        Technology, 2006]  Ecosystem dynamics        driven by remotely  70, 130  2000, 2002      sensed phenology      [Potter et al., 2009]    Amazon fire emissions,  200‐500  2003  including deforestation              [Van der werf et al., 2003]  Deforestation  100  1989‐1998          [Houghton et al., 2000]            Mato  Cropland extensification  46  2000‐2006  Grosso                                  41                                                                  Table 1.6 Greenhouse gas budget (Mato Grosso)   Total annual greenhouse gas budget (CO2 equivalent emissions per year) for Mato Grosso accounting for conversions to pasture and cropland, pasture maintenance, and methane emissions from cattle. This budget accounts for cattle emissions using a mean emission rate of 55 kg methane/head [Steudler et al., 1996] and a cattle herd of 19,600,000 for Mato Grosso (IBGE 2009). Tg CO2 equivalent per  EMISSIONS ESTIMATE    year [DeFries et al., 2008]  Conversion of forest to pasture  183.3   Galford et al., this paper  Conversion of pasture and natural ecosystems  169   to cropland  [DeFries et al., 2008]  Pasture maintenance (fire)  22.0 After Steudler et al. [1996]  Methane from the cattle herd  21.5 Galford et al., this paper  Cropland fertilization  0.2 TOTAL    396.0         42                                      Figures   Figure 1.1. The study area of Mato Grosso state, the frontier of agricultural development in the Brazilian Amazon, is shown here in the context of the Brazilian legal Amazon and with potential natural vegetation [Mello, 2007]. Field data points used for validation are shown with orange dots.   Figure 1.2. Predominating land-use trajectories for Mato Grosso. The foci of this paper are transitions to row-crop agriculture and intensification within existing croplands. Figure 1.3. Total cropland area (km2) in Mato Grosso by natural ecosystem of origin.   Figure 1.4. Transitions to croplands detected with remote sensing from 2001-2006. The two primary cropland development paths analyzed in this paper (natural- ecosystem-directly-to-cropland or pasture-to-cropland transitions) are illustrated for     43    each of the natural ecosystems of origin. The “net” cropland is total increase in cropland area from either land-use trajectory. The “total area” is the cropland area in 2006 that came from either trajectory. Also, see Table 1.4 for annual transition rates. Figure 1.5. Agricultural cropland extensification (new areas of cropland from 2001- 2006) mapped by natural ecosystem of origin, largely nucleating from existing croplands. Figure 1.6. The fraction of cropland land-use occupying lands in each originating biome. The large solid bar represents the proportion of all lands in Mato Grosso in cropland use. Figure 1.7. CO2 equivalent emissions from biomass loss and fertilizer additions. “Pasture” represents cropland extensification into areas of pasture that were previously created from areas of native vegetation. The range of emissions comes from a low, mid and high range of biomass estimates (Table 1.4). Figure 1.8. Agricultural cropland intensification is mapped by natural ecosystem of origin for the cropland, where intensification is the transition from non-croplands or single cropping patterns to double cropping patterns over the study period.     44    Figure 1.9. Total cropland area from 2001-2006 and dynamics between area of single and double cropping patterns for Mato Grosso (a) and by areal percent of each natural ecosystem of origin, cerrado (b), cerradão (c), and forest (d).                       Figure 1.1 Study area           45    Figure 1.2 Land-use transitions                           46    Figure 1.3 Cropland Extensification     47    Figure 1.4 Areas of land-use change Areas of land-use change from 2001-2006       48    Figure 1.5 Map of cropland extensification                   49    Figure 1.6 Cropland area by biome     50    Figure 1.7 Carbon emissions from cropland extensification           51    Figure 1.8 Map of cropland intensification       52      Figure 1.9 Changes in cropping patterns by biome                       CHAPTER 2   Carbon emissions and uptake from 105 years of land-cover and land-use change at the agricultural frontier of the Brazilian Amazon Gillian L. Galford a,b, Jerry Melillob, David Kicklighterb, John F. Mustard a, Timothy Cronin b, Carlos E.P. Cerri c, Carlos C. Cerri d a) Geological Sciences, Brown University, United States b) The Ecosystems Center, MBL, United States c) Escola Superior de Agricultura Luiz de Queiroz, Universidade de São Paulo, Brazil d) Centro de Energia Nuclear na Agricultura, Universidade de São Paulo, Brazil Ecological Applications (In Review) 52  53    Abstract Tropical ecosystems play a large and complex role in the global carbon cycle. Clearing of natural ecosystems for agriculture leads to large pulses of CO2 to the atmosphere from terrestrial biomass. Concurrently, the remaining intact ecosystems, especially tropical forests, may be sequestering a large amount of carbon from the atmosphere in response to global environmental changes including climate changes and an increase in atmospheric CO2. Here we use an approach that integrates census-based historical land-use reconstructions, remote-sensing based contemporary land-use change analyses and simulation modeling of terrestrial biogeochemistry to estimate the net carbon balance over the period 1901-2006 for the state of Mato Grosso, Brazil, which is one of the most rapidly changing agricultural frontiers in the world. By the end of this period, we estimate that of the state’s 925,225 km2, 221,092 km2 had been converted to pastures and 89,533 km2 have been converted to croplands, with forest-to-pasture conversions being the dominant land-use trajectory but with recent transitions to croplands increasing rapidly in the last decade. These conversions have led to a cumulative release of 4.8 Pg C to the atmosphere, with about 80% from forest clearing and 20% from the clearing of cerrado. Over the same period, we estimate that the residual undisturbed ecosystems accumulated 0.3 Pg C in response to CO2 fertilization. Therefore, the net emissions of carbon from Mato Grosso over this period were 4.5 Pg C. Net carbon emissions from Mato Grosso since 2000 averaged 146 Tg C yr-1, on the order of Brazil’s fossil fuel emissions during this period. These emissions were associated with the expansion of croplands to grow soybeans. While alternative     54    management regimes in croplands, including tillage, fertilization and cropping patterns promote carbon storage in ecosystems, they remain a small portion of the net carbon balance for the region. This detailed accounting of a region’s carbon balance is the type of foundation analysis needed by the new United Nations Collaborative Programme for Reducing Emissions from Deforestation and Forest Degradation (REDD). Key words: Land-cover and land-use change; Amazon; Cerrado; carbon emissions; Terrestrial Ecosystems Model; pasture; crops; soils; global warming; REDD; CO2 fertilization 1. Introduction The role of the tropics in the global carbon cycle is still an unsettled question (Schimel 2007). Two recent analyses based on inverse models have led to different conclusions about how the tropics are functioning in the global carbon cycle. The first study, by Jacobsen et al. (2007), has found a strong tropical source that implies there are large emissions from land-use change, primarily due to deforestation, and suggest that mechanisms, such as CO2 fertilization, that stimulate carbon uptake in the remaining tropical forests are relatively unimportant. The second study, by Stephens et al. (2007), suggests that the uptake and storage of carbon in the tropics balances deforestation emissions. Research on carbon emissions from land-use change in the tropics indicates that the amount is large, but uncertain. Canadell et al. (2007) estimate tropical land-     55    use emissions for this decade to be 1.5 PgC +/- 0.5 PgC. Typical methods for estimating carbon emissions from tropical deforestation track changes in the fate of carbon pools after cutting, often using simple response functions to describe carbon losses similar to those initially developed by Houghton et al. (1983). The areas deforested are derived from either national scale assessments (FAO 2006) or satellite analyses (INPE 2009a). The resulting estimates are highly sensitive to vegetation biomass at the time of cutting as well as the accuracy of the estimates of the area cleared. The Amazon region has long been recognized for its role in global ecosystems services, such as water cycling and carbon storage, but rapid development of agriculture now threatens carbon sinks and its net impact on carbon cycling is rapidly changing. The history of intensive research on the role of Amazon ecosystems in the carbon cycle spans several decades and much of what we understand about the net uptake or release of carbon by intact tropical forests and cerrado areas is based on site-level studies using a range of methods including the eddy flux technique and repeated measures of aboveground biomass on inventory plots. For forests and cerrado of the Amazon Basin, eddy flux methods generally show net carbon uptake (e.g., Grace et al. 1995; Malhi et al. 1998; Miranda et al. 2008), although one study has shown net carbon release from an old growth forest in the central Amazon (Saleska et al. 2003). A synthesis of Amazon forest survey data by Phillips et al. (1998) has shown forests functioning as net carbon sinks. This has been confirmed by a recent analysis of forest monitoring data from permanent plots in the Amazon by Laurance et al. (2009), who found an acceleration of tree     56    growth over several decades that robustly supports a hypothesis of carbon accumulation from CO2 fertilization. In addition to plot-level studies of ecosystems processes, accurate satellite- based regional information on the location and temporal dynamics of land-cover and land-use changes are needed for regional carbon budget estimates to reduce uncertainty on the role of the tropics in the global carbon budget and, in particular, to elucidate where land-use change emissions are greater than natural carbon sinks. One powerful approach for estimating regional carbon budgets involves the coupling of state-of-the-art ecosystem models with land-cover and land-use information derived from remote sensing products. Because of its frequent overpasses and the availability of the data it generates, the Moderate Resolution Imaging Spectrometer (MODIS) is an important source of information for these studies (Justice 1998). This has been especially true for the Brazilian Amazon during the Large-Scale Biosphere- Atmosphere (LBA) study, a long-term, internationally supported research program led by the Brazilian government to understand the regional biological, chemical and physical functions of the Amazon and alterations due to land-use and climate changes in the context of sustainable development and global climate (LBA-ECO 2009). For example, DeFries et al. (2008) and Morton et al. (2008) have used a process-based biogeochemistry model, DECAF, coupled with fire data derived from MODIS to estimate carbon emissions from burning of biomass in the state of Mato Grosso in the southern Amazon. Potter et al. (2009) used MODIS observations of vegetation together with the CASA model, another biogeochemistry model, to simulate regional patterns of productivity and net carbon fluxes in intact forests of     57    the Amazon in response to climate variability. These studies, and others like them in the Amazon, separately explain patterns in carbon dynamics from natural vegetation and carbon losses from land-use transitions. An important next step in understanding changes in the carbon cycle in the Brazilian Amazon is to study combined changes in land cover, land use, land management and changing environmental conditions. While remote sensing imagery is useful for studying recent carbon dynamics in this region, the influence of land-use history beyond the period of remote sensing observations on these carbon dynamics also needs to be examined. Here, we explore the combined role of this complex set of controls on regional-scale carbon emissions from the state of Mato Grosso, home of Brazil’s fastest growing agricultural frontier (Hansen et al. 2008, Galford et al. In Review; INPE 2009b). We quantify land-use emissions in the context of changes in the carbon budget of the state’s remaining intact natural ecosystems, the forests and the cerrado. Beyond previous studies focusing on natural systems or emissions from conversions, we integrate the landscape dynamics of natural vegetation, land-use change, and management using explicit land-use data derived from remote sensing in a process-based ecosystems model. The objectives of this study are: 1) to understand the role of historical land-use change on carbon emissions in the frontier of the Brazilian Amazon and 2) to understand the role of intact natural systems in the regional greenhouse gas budget including the loss of carbon sequestration potential due to land-use change.     58    2.0 Methods We develop a 105-year integrated land-use data set and use it with the process- based Terrestrial Ecosystem Model (TEM) to calculate the impacts of land-use change on regional-scale carbon emissions from the rapidly developing agricultural frontier of the state of Mato Grosso, Brazil within the context of other changing environmental factors. We also assess how land management of croplands (fertilizer application, tillage, and cropping practices) may influence storage of soil organic matter. The simulations occur at a moderate resolution (1 km2) that is relevant to managing resources for carbon credits. Below, we describe the study area, the TEM along with model modifications implemented for this study, and the development of land-use and other data sets used by TEM. We also describe the simulation experiments we carried out to examine the relative importance of land-use changes, different land management options and CO2 fertilization on carbon dynamics of Mato Grosso. 2.1 Study area Mato Grosso is a large state (925,225 km2) in the southern Brazilian Amazon. The natural vegetation of Mato Grosso is a mix of cerrado (savanna) and tropical forest (Figure 2.1; Castro and Kauffman 1998) that experiences a humid tropical climate with a short dry season (June-September). For the purposes of this study, we identify six natural vegetation types (Table 2.1) simplified from the data presented by Mello (2007). The cerrado, considered a global biodiversity hot spot (Myers et al. 2000), varies in community structure from tree-rich areas (cerradão), to grass (campo limpo) and shrub-dominated areas (cerrado stricto senso). Across     59    Brazil, two-thirds of the cerrado’s native extent has been converted to agriculture and much of the remaining intact cerrado is in Mato Grosso (Klink and Machado 2005; Conservation International 2008). Beginning in the 1940s, throughout Mato Grosso, large-scale clearing of natural vegetation for pasture and cropland continues with little transition back to secondary growth (Rudel et al. 2004). Today a complex set of transitions is occurring in the region, including the clearing of natural vegetation for both pastures and croplands and the replacement of pastures to grow crops, primarily soybeans. Pastures are currently the dominant land-use type in this region, but cropland area is rapidly increasing and brings with it a number of important management decisions related to cropping patterns, fertilizer use, and tillage. In an effort to increase production, some farmers in the region are changing from single (typically soybean) to double cropping (typically soybean followed by corn) in a growing season, enabled by the use of fertilizers in the double cropping system (Keys and McConnell 2005, Fundação Agrisus 2006). In this region, there are three types of tillage patterns used: no tillage, conservation tillage and conventional tillage. Conservation tillage is a rotational program of roughly 3 years with no-tillage and conventional tillage using deep disking plows in the fourth year. 2.2 Model description For this study, we simulate monthly terrestrial carbon and nitrogen dynamics using a version of the process-based Terrestrial Ecosystem Model (TEM, Felzer et al. 2004) that has been modified to better incorporate the effects of land management observed in Mato Grosso. As described in Felzer et al. (2004), the     60    exchange of carbon dioxide between terrestrial ecosystems and the atmosphere depends on both ecosystem metabolism, as influenced by local environmental conditions, and land management practices. The uptake of atmospheric carbon dioxide during photosynthesis by plants is simulated as gross primary production (GPP) and is influenced by atmospheric carbon dioxide and ozone concentrations, photosynthetically active radiation, air temperature, evapotranspiration, soil available nitrogen, and canopy stature. Carbon dioxide is returned to the atmosphere from respiration of both autotrophs and heterotrophs. Autotrophic respiration by plants is dependent upon the amount of vegetation biomass, air temperature and GPP. Heterotrophic respiration is dominated by microbial respiration, which is associated with the decomposition of organic matter and is influenced by the amount and C:N ratio of soil organic matter, air temperature and soil moisture. During the conversion of natural lands to agriculture, some of the carbon in vegetation biomass is assumed to be lost as emissions to the atmosphere from human-induced fires and the rest of the vegetation biomass is partitioned among slash, 10-year and 100-year wood product pools. The slash is added to the soil organic pool, where it is assumed to decompose at the same rate as soil organic matter, whereas the woody product pools are assumed to decompose linearly. For land areas converted to row-crop agriculture, we assume that all slash that could hinder the mechanized tilling and planting of fields is removed through continued windrowing and burning to improve the accessibility and use of farm machinery. As a result, no slash is left behind in these areas to decompose. For tropical forests converted to pasture, we assume that 33% of the vegetation biomass is left as slash. In cerrado and cerradão areas,     61    we assume 50% of the vegetation biomass is left as slash during conversion to pasture. These assumptions are consistent with the 100% combustion completeness for cropland and 50-90% combustion completeness for pastures reported by Morton et al. (2008). The harvest of crops converts 40% of the crop biomass to agricultural product pools with the remaining biomass added to soil organic matter. The agricultural pools are assumed to be consumed or decomposed linearly within a year of harvest. Additional details of these land-use change dynamics may be found elsewhere (McGuire et al. 2001; Tian et al. 2003; Felzer et al. 2004). In Felzer et al. (2004), the timing of the planting and harvesting of crops is simulated based on growing-degree-days. For this study, we prescribe the dates of crop planting and harvests based on whether a field is classified in the land-use dataset as a single or double crop. For areas that grow a single crop, we assume that the crop is planted in November and harvested in March. For areas of double cropping, we assume that the first crop is planted in October and harvested in January and that the second crop is planted in February and harvested in June. We further assume that the second crop of double-cropped areas is optimally fertilized. No fertilization occurs in single-cropped areas or of the first crop of double-cropped areas and we assume that there is there is no net loss of nitrogen, so the single crop or first crop of double-cropped areas acts as a nitrogen-fixing crop, such as soy. In addition to prescribed planting and harvest dates, we also modified TEM to account for the effects of tillage on soil organic carbon stocks. In previous model applications, no-till practices have been assumed to occur in all croplands. To better account for the effects of tillage on carbon dynamics, we assume in this study that     62    tillage doubles the decomposition rate of soil organic carbon because mixing of the soil profile and the breakup of soil aggregates exposes more carbon to oxidation than would happen in undisturbed soils (Balesdent et al. 2000). Tillage occurs every 4th year in a conservation-tillage regime, so we simulate the tillage effect on decomposition once every four years and assume no-till conditions in the intervening years. Since large areas of Mato Grosso have historically been converted for pastures rather than row-crop agriculture, we added algorithms to TEM to consider the influence of grazers on pasture biogeochemistry. In a pasture, grazers are assumed by TEM to consume 5% of the standing vegetation biomass every month as forage. Of the forage consumed, 83% of the carbon is respired back to the atmosphere and 17% is transferred to the soil organic carbon pool as manure. For the corresponding nitrogen in forage, 50% is transferred to the soil organic nitrogen pool as manure and 50% is transferred to the soil available nitrogen pool as urine. No additional fertilization of pastures is assumed to occur.    Thus, the calculation of net carbon exchange (NCE) between a terrestrial ecosystem and the atmosphere as follows: NCE = GPP – RA – RH – EF –EP (1a) RH = RM + RG (1b) EP = EA + E10 + E100 (1c)     63    where the variables are: autotrophic respiration (RA), heterotrophic respiration (RH), microbial respiration (RM), emissions from fires associated with land-clearing (EF), 10-year and 100-year woody product pools (E10 and E100, respectively), and agricultural product pools (EA). A positive NCE indicates a net carbon sink, while a negative NCE indicates a carbon source. The TEM is calibrated to data on carbon and nitrogen stocks and fluxes at intensively studied field sites (Raich et al. 1991; McGuire et al. 1992). For this study, we assume that the natural land cover of Mato Grosso can be represented by six dominant vegetation types: Tropical Evergreen Forests, Tropical Deciduous Forests, Riparian Forests, Cerradão, Cerrado Stricto Senso, and Campo Limpo. For Tropical Evergreen Forests, Tropical Deciduous Forests and Riparian Forests, we use the TEM parameterizations (McGuire et al. 1992) appropriate for tropical evergreen forests, but we assume the other vegetation types represent a mixture of mesic trees, xeromorphic trees and shrubs, and grasses (Table 2.2). For mesic trees, we also use the TEM parameterizations associated with tropical evergreen forests, but we use the TEM parameterizations associated with xeromorphic forests and woodlands for xeromorphic trees and shrubs and the grassland parameterizations for grasses. 2.3 Input Data To develop regional estimates of carbon and nitrogen stocks and fluxes, the TEM needs spatially explicit data for elevation, soil texture, land cover, climate and atmospheric chemistry. To meet these data input needs, we developed a series of     64    data sets at a spatial resolution of 1km x 1km. For elevation, we use the TerrainBase v1.1 data set from the National Geophysical Data Center (NGDC 1994). For soil texture, we use data provided by Batjes et al. (2004). To capture the effects of land- use change and climate change on terrestrial carbon dynamics, we developed spatially explicit time-series data sets at the 1km x 1 km spatial resolution to prescribe changes in land cover, climate and atmospheric chemistry data, as described below. 2.3.1 Land-use data sets We use two approaches over different time periods to develop a time-series data set of historical land-use change from 1901 to 2006 (See Appendix 2.A). For the period 1901 to 2001, we combine information from census estimates (SIDRA 2009a) and annual crop statistics (SIDRA 2009b) with the distribution patterns of pasture, single crops and double crops determined from remote sensing imagery for the year 2001 (Galford et al. Chapter 1, Morton et al. 2006, 2009). For the period between 2001 and 2006, we use the changes in distribution patterns of pastures, single crops and double crops as determined from remote sensing imagery (Galford et al. Chapter 1, Morton et al. 2006, 2009). 2.3.2 Climate & atmospheric chemistry data sets The version of TEM used in this study requires spatially explicit data for three climate variables (air temperature, precipitation, cloudiness) and one atmospheric- chemistry variable (ozone) at a monthly time step. In addition, the model uses a time series of annual atmospheric CO2 concentrations based on the ice core record of Etheridge et al. (1996) and flask measurements at the Mauna Loa Observatory in     65    Hawaii (Keeling et al. 2005). Mean surface air temperature, precipitation, and cloudiness are obtained for the period 1901-2000 from datasets produced by the Climate Research Unit (CRU) of the University of East Anglia (New et al. 2002, Mitchell et al. 2004). The global ozone dataset is described in Felzer et al. (2005). The datasets normally exist at a spatial resolution of 0.5 x 0.5 degrees and are thus down-sampled to the 1-km grid used in this study such that all 1-km grid cells contained within a much larger 0.5 x 0.5 degree grid cell simply use the value of a climate driver from the larger grid cell. Because sharp climate gradients generally do not occur in the state of Mato Grosso, and continuous long-term observations are not readily available, we believe this to be an adequate treatment. For the period 2001-2006, which is not covered by the CRU dataset, we use baseline climatological averages (baseline period 1948-2000) for cloudiness, and use a delta/ratio approach to blend temperature and precipitation from the CRU and NCAR/NCEP reanalysis (1948-2006) products. Within the NCAR/NCEP reanalysis dataset, temperature differences and precipitation ratios are calculated for each month in the period 2001- 2006 relative to the 1948-2000 average value for the corresponding calendar month. These differences (temperature) and ratios (precipitation) are then added to or multiplied by, respectively, the 1948-2000 average value of temperature (precipitation) from the CRU dataset to obtain the blended value. 2.4 Design of simulation experiments Besides developing regional estimates of carbon stocks and fluxes, our simulation approach also allows us to explore the relative importance of different environmental factors influencing terrestrial carbon dynamics in Mato Grosso and     66    the potential impacts of different land management options. After conducting a simulation that best represents historical conditions in Mato Grosso (S1 in Table 2.3) to estimate historical carbon dynamics, we then conducted six additional simulations to examine the effects on carbon dynamics of land-use change, CO2 fertilization, and several land management options including tillage, cropping patterns and nitrogen fertilizer applications (Table 2.3). To determine land-use change effects, we conducted a simulation (S2 in Table 2.3) in which climate and atmospheric chemistry are allowed to change, but the land cover is the same as that found in the year 1901. As our historical land-use change datasets indicated that the region was covered by natural vegetation until 1934, the difference in carbon fluxes and stocks between simulations S1 and S2 captures the effects of all land-use change in Mato Grosso including interactive effects with other environmental factors and the loss of carbon sequestration potential by replacing natural vegetation with agriculture. To better quantify the effects of land conversions on carbon dynamics, we stratified the net carbon flux based on annual transitions in land cover (33 total transitions, e.g., tropical evergreen forest to pasture, cerrado to single crop, cerradão to cerradão, pasture to single crop, single crop to double crop, double crop to double crop). To evaluate the effects of CO2 fertilization on carbon dynamics in Mato Grosso, we conducted a simulation (S3 in Table 2.3) in which climate, ozone and land-use are allowed to change over the historical period, but atmospheric CO2 concentrations remain at the same level found in 1901. The CO2 fertilization effect is then determined by the difference in carbon stocks and fluxes between the S3 and     67    S1 simulations. With this approach, we capture both the direct effects of enhanced atmospheric CO2 on carbon dynamics along with any interactive effects of CO2 with other environmental factors. Running several land-use scenarios that include different types of management allows us to evaluate the impacts of each type of management. The conservation tillage used in our control simulation (S1) is the most representative of practices in the region, as it balances the agronomic benefits and drawbacks to no- tillage and conventional tillage. To more fully explore the impacts of tillage on the amount of soil organic carbon found in croplands, we conducted one simulation (S4 in Table 2.3) in which no tillage is assumed to occur on all croplands and another simulation (S5 in Table 2.3) in which conventional tillage (soils are tilled every year) is assumed to occur on all croplands. We do not implement a tillage effect on decomposition under no-till conditions and we implement the tillage effect on decomposition rates every year for conventional tillage rather than every 4th year as for conservation tillage. To assess the potential impacts of cropping patterns on soil carbon dynamics, we conducted a simulation (S6 in Table 2.3) in which all croplands were assumed to be single cropped. We used single cropping with and without fertilization (S7 and S6, respectively, in Table 2.3) to evaluate the potential impacts of nitrogen fertilizer application. 3. Results 3.1 Historical land-use reconstruction Our reconstruction of land use in Mato Grosso indicates that pastures have historically been the dominant land use in this region (Figure 2.2). Pastures first     68    appear in 1934 and there was modest pasture development through 1960. From 1960 to 2000, there were accelerated pasture developments. Croplands emerged in the region in 1971, but their area increased rapidly starting in the late 1990s as soybean varieties became available for the Amazon and dramatically increased from 2000. By 2006, over 221,092 km2 were in pasture use and 89,533 km2 were in cropland use. We distinguished double-cropping patterns, which appear at large- scales in the early 1980s. Of the cropland area in 2006, 54,159 km2 were single cropped and 35,374 km2 were double cropped. The cerrado region hosts more of the croplands than the other natural ecosystems, and by 2006, almost 20% of the cerrado region has been converted to croplands (Galford et al. Chapter 1). Pastures tend to be more spatially disparate, while croplands nucleate around existing croplands (Figure 2.3). We did not consider secondary growth as one of the regional land-cover and land-use trajectories because it has been identified as of minor importance in Mato Grosso (Rudel et al. 2005), although Fearnside et al. (2009) suggest regrowth may have a large impacts on post-clearing carbon stocks. 3.2 Historical carbon dynamics in Mato Grosso Our simulations show that over the 105 years between 1901 and 2006, Mato Grosso lost 4.8 Pg carbon, largely due to land clearing for pasture and croplands. Over the same period, these losses were slightly offset by the uptake of 0.3 Pg carbon by intact natural ecosystems. The net loss of carbon from land ecosystems in Mato Grosso over the study period was therefore 4.5 Pg C (Figure 2.4).     69    Carbon losses from land-use changes were minimal from the 1930s through the 1950s, but steadily increased from the 1960s through 1990s with the expansion of pasture areas and the beginning of the expansion of cropland areas (Figure 2.5). The most rapid carbon losses from land-use change have occurred since 2000. Between 2000 and 2006, annual emissions from land-use were on average 146 Tg C yr-1. For the entire study period, carbon losses from forest clearing accounted for about 80% or 3.8 Pg C of the 4.8 Pg C lost from Mato Grosso, with most of the rest, 19% or 0.9 Pg C, lost from the clearing of cerrado. The remaining carbon loss (about 1%) results from pasture to cropland transitions. The clearing of forests and cerrado for pasture were the dominant land-use transitions (Figure 2.5). Forest clearing for pasture was responsible for about 75% of the total carbon loss, and the clearing of cerrado for pasture was responsible for about another 15%. The direct clearing of forests and cerrado for croplands, a relatively recent phenomenon in Mato Grosso, accounted for 10%. There is a strong spatial relationship evident between the highest rates of carbon loss from forested regions, as well as, with areas of agriculture (Figure 2.3). Most of the lost carbon, 4.5 Pg C, came from vegetation cut and burned during clearing or the decay of the resulting wood products. Some vegetation carbon was rapidly oxidized to CO2 by burning and some was more slowly oxidized to CO2 in microbially mediated decay. A much smaller amount came from the decay of soil organic matter following clearing (0.3 Pg C). For both vegetation and soils, most of this carbon loss comes from forested areas (Figure 2.6). Forests have a     70    greater proportion of carbon loss from soils relative to vegetation, as compared to cerrado. 3.3 Carbon balance in natural ecosystems Starting in 1901 with changing climate and CO2, we find an increase of 0.28 Pg C in intact natural vegetation. In our simulations, if atmospheric CO2 is held constant (S3 in Table 2.3) over this study period, we find that these natural ecosystems would have lost 0.47 Pg C instead of gaining carbon. This indicates that CO2 fertilization enhances carbon storage in these intact natural ecosystems by 0.75 Pg C over the study period. Most of this additional carbon storage occurs in tropical deciduous forests and cerrado stricto senso, together account for 75% of the carbon sink from CO2 fertilization (Figure 2.7). 3.4 Carbon dynamics related to pasture and cropland management After clearing, we estimate that soils in agricultural systems can either gain or lose carbon. In forest-to-pasture and cerrado-to-pasture transitions, we simulate a long-term net gain in soil carbon, but this is a minor carbon increase to the large net carbon losses from clearing. For all croplands, we simulate carbon losses from the soils, with the smallest losses associated with conservation tillage and nitrogen fertilization application in double cropping patterns. Conventional tillage increased carbon losses by 20%. Our estimate of the amount of fertilizer required for croplands, 48 kg N ha-1 yr-1 (standard deviation 11 kg N ha-1 yr-1) over the study period is within the range of fertilizer doses observed on farms and suggested by field trials (Galford et al. Chapter 1).     71    4. Discussion Carbon losses from land-use changes in Mato Grosso have been the subject of two recent studies in addition to ours. DeFries et al. (2008) estimated carbon emissions using remote sensing to detect fires and process-based biogeochemical modeling of combustion and decay along different land-use trajectories (Table 2.4). For the period 2001-2005, DeFries et al. (2008) estimate 89 Tg C year-1 is lost. Another study by Fearnside et al. (2009) uses a bookkeeping model to account for combustion, decay, release of graphite particles, carbon storage by replacement biomass, emissions from cattle and changes in forest methane sources and sinks. That study uses annual deforestation statistics reported by the Brazilian government (INPE 2009b) for 2007, the lowest deforestation year on record (2,040 km2 yr-1 cleared). To compare to DeFries et al. (2008), we used the Fearnside et al. (2009) emissions rates per unit area with the average annual deforestation rates reported by INPE (2009b) for the period 2001 to 2005 (Table 2.4). This adjusted estimate, based on Fearnside et al. (2009), becomes 113 Tg C year-1. For the 2001 to 2005 period, our results show emissions of 163 Tg C year-1. Considered together, these studies highlight the importance of differences between assumptions and considerations in estimates of carbon emissions. We understand many of the differences that have caused these variations in carbon emission estimates for Mato Grosso, including biomass estimates, input land-use change data set accuracy, biome boundary definitions, and land-use change trajectories. Initial biomass estimates account for much of the difference in carbon emission estimates between our study and Fearnside et al. (2009). Our     72    biomass estimates are higher for tropical evergreen forests (217 ±75 Mg C ha-1) than the 169 Mg C ha-1 reported by Fearnside et al. (2009), but are quite similar for all other natural vegetation classes between the two studies. There may be differences in biome boundaries and land-use transitions between our study and DeFries et al. (2008), and, in particular, we included carbon losses from the cerrado-to-pasture transition that was not addressed by their study. Accuracy from the input land-use change remote sensing products is another potential source of error. For example, our work uses land-use data on croplands that were detected with 85% accuracy through remote sensing methods that introduces a source of potential error in the results. These factors highlight the need for better quantification of biomasses across climate gradients that can also be used to define biome boundaries and the importance of field-validation for all data sets. Mato Grosso is a hotspot for carbon emissions from land-use changes in the Legal Amazon, which includes both forest and cerrado. Our annual carbon emissions estimate for Mato Grosso is larger than Brazilian government estimates (Ministry of Science and Technology 2006) for forest clearing in the Legal Amazon (117 Tg C year-1), but smaller than the corresponding estimates for land clearing in both forest and cerrado (167 Tg C year-1). The government estimates used PRODES deforestation detections and estimated forest regrowth through visual interpretation of 44 Landsat satellite images that was then extrapolated to the rest of the biome. In a bookkeeping approach, the government estimates combined the net deforestation with allometric estimates of biomass through field survey data from the RADAMBRASIL project to estimate carbon emissions, a method which Fearnside et     73    al. (2009) points out will over emphasize lower biomass ecosystems and under- report carbon emissions. Agriculture makes it difficult to accumulate carbon, but managing agricultural systems to minimize soil C losses or increase carbon sequestration in soils is important for sustainable development. Our findings suggest that best practices from management of both pastures and crops can minimize further carbon losses or actually promote a small carbon sink, and are supported by field studies (Bayer et al. 2006, Carvalho et al. 2009, Jantalia et al. 2007, Neill et al. 1997, Moraes et al. 1996). We simulate that, over time, pasture soils have the potential to sequester more carbon relative soil carbon stocks in the pre-disturbance natural ecosystem. We find that conservation tillage with nitrogen fertilizer for the second crop in a double cropping system promotes the largest amount of carbon storage in soils. The application of N fertilizer increases soil carbon stocks through increased crop productivity and returns of crop residue to soils. Conservation tillage decreases some of the negative impacts of tillage; decomposition rates increase with tillage, as tillage introduces residues into the soil where there are often more favorable conditions for decomposition (Karlen and Cambardella 1996, Six et al. 1999), promotes soil microbial activity through aeration (Kladivko 2001), and disrupts soil aggregates that otherwise protect carbon from microorganisms (Karlen and Cambardella 1996). There are other benefits of conservation tillage or no-tillage, such as reduced run off and soil erosion, that are not considered in this paper but are important for both agricultural and environmental sustainability (Lal and Kimble 1997).     74    Recent studies (Laurance et al. 2009) suggest that increasing atmospheric CO2 is increasing forest aboveground biomass in the Amazon. Laurance et al. (2009) observed a 4% growth in forest plots that they suggest is due to a CO2 fertilization effect. The TEM model predicts a 1% increase in vegetation carbon storage from CO2 effects over the 1901 to 2006 period, suggesting the future role of CO2 fertilization in carbon sequestration for remaining intact forests may be important. Previous biogeochemical modeling studies in the Amazon (e.g. Potter et al. 2009) have focused on short time scales of a few years, over which CO2 fertilization is assumed to have no effect but this work shows that increasing atmospheric CO2 levels may have long-term impacts on carbon storage in natural vegetation Going beyond contemporary and historical carbon budgets, we estimate foregone carbon sequestration from the original land cover that has been lost due to land-use changes. In the absence of any land use from 1901 to 2006, our analysis (S2 in Table 2.3) suggests carbon stocks in Mato Grosso would have increased 2.7%, representing a carbon sink of 0.5 Pg C. Thus, land-use change has reduced the capacity of Mato Grosso ecosystems to sequester carbon by 0.2 Pg C (i.e., 0.5 Pg C described above minus the 0.3 Pg C sequestered in residual natural lands, see Figure 2.7) in addition to causing a net loss of 4.5 Pg C over the historical period from 1901 to 2006. For the 2000 to 2006 period, Mato Grosso carbon emissions as estimated by us account for between 1 and 2% of recent estimates of land-use change carbon emissions, which are on the order of 1.5 Pg C year-1 (Canadell et al. 2007). The     75    uncertainty in pan-tropical carbon emission estimates could be reduced by applying the coupled remote sensing and modeling approach presented here in a wide survey across the tropics. Critical steps in applying this linked approach involve reducing uncertainty surrounding these estimates. This could be achieved by increasing field data sets for training and validating remote sensing detections of land use and by ecological field studies across environmental gradients in major biomes that could provide useful information in model parameterization and calibration while reducing uncertainties in the spatial variation of biomass estimates (Fearnside et al. 2009). Our analysis suggests that emissions from Mato Grosso are larger than Brazil’s fossil fuel emissions. We estimate that land clearing in Mato Grosso releases an average of 146 Tg C year-1 since 2000, which is greater than Brazil’s current fossil fuel emissions (89 Tg C year-1, Marland et al. 2008) and suggests the actual net emissions from Brazil are much higher than solely the fossil fuel and cement emissions typically considered. In fact, several studies have suggested that total emissions from a country should be the sum of fossil fuel, cement and deforestation emissions, which might raise Brazil’s global rank for top emitting countries (Boden 2009, Cerri et al. 2004, Cerri et al. 2007). In addition, we estimate an average contemporary (since 2000) carbon sink in intact natural land cover of 28.4 Tg C year -1 in Mato Grosso. The net carbon budget for this deforestation hotspot in the Amazon is a strong argument for including both land-use change emissions (carbon loss) and uptake by intact natural vegetation (carbon sink) along with fossil fuel emissions when calculating national-scale carbon emissions, not just for Brazil but for all countries. This is the type of detailed information necessary to     76    support policy decisions, such as the UNFCCC Programme on Reducing Emissions from Deforestation and Degradation (REDD). 5. Conclusions The analyses we present here represent a significant step forward in integrating historical land use with modern land use change quantified from remotely sensed data, and using these as critical boundary conditions on biogeochemical modeling. Through this process, we quantify changes in carbon stocks, including the effects of increased atmospheric CO2. Thus, this analysis moves beyond previous studies that focus on natural systems or emissions from conversions and accommodate the full dynamic processes of land cover change and land management practices. This work shows that the historical role of land-use change has made serious contributions to the net carbon emissions from this region, but that recent, rapid changes in land use primarily through clearing of forests has increased the annual emissions greatly. We find that intact natural ecosystems have been a continuing net sink for carbon, but if historical land-use change is any indicator of the future, future deforestation could seriously reduce this carbon sink, as well as, release carbon currently stored in natural vegetation. It is clear that some agricultural expansion will continue (Soares-Filho et al. 2006). Our evaluation of greenhouse gas emissions from land-use change in this region shows a way forward on future mitigation of global warming. The role of this region in mitigation depends on reducing emissions from land-cover and land-use     77    change, conserving of intact natural vegetation that is both a carbon reservoir and an active carbon sink, and following best management practices for both pasture and croplands that minimizes further carbon losses or actually promotes a small carbon sink.     78    Citations Alves, D. S. 2002. Space-time dynamics of deforestation in Brazilian Amazonia. International Journal of Remote Sensing 23:2903-2908. Balesdent, J., C. Chenu, and M. Balabane. 2000. Relationship of soil organic matter dynamics to physical protection and tillage. Soil and Tillage Research 53 (3- 4):215-230. Batjes, N., M. Bernoux, and C. Cerri. 2004. Soil data derived from SOTER for studies of carbon stocks and change in Brazil (ver. 1.0). ISRIC - World Soil Information, Wageningen. Bayer, C., L. Martin-Neto, J. Mielniczuk, A. Pavinato, and J. Dieckow. 2006. Carbon sequestration in two Brazilian Cerrado soils under no-till. Soil and Tillage Reserach 86:237-245. Boden, T. A., G. Marland, and R.J. Andrew. 2009. Global, regional and national fossil-fuel CO2 emissions Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy, Oak Ridge, Tenn., U.S.A. Canadell, J. G., C. Le Quere, M. R. Raupach, C. B. Field, E. T. Buitenhuis, P. Ciais, T. J. Conway, N. P. Gillett, R. A. Houghton, and G. Marland. 2007. Contributions to accelerating atmospheric CO2 growth from economic activity, carbon intensity and efficiency of natural sinks. PNAS 104:18866-18870. Carvalho, J. L. N., C. E. P. Cerri, B. J. Feigel, M. Piccolo, V. P. Godinho, and C. C. Cerri. 2009. Carbon sequestration in agricultural soils in the Cerrado region of the Brazilian Amazon. Soil and Tillage Reserach 103:342-349.     79    Cerri, C. C., M. Bernoux, C. E. P. Cerri and C. Feller. 2004. Carbon cycling and sequestration opportunities in South America: the case of Brazil. Soil Use and Management 20: 248-254. Cerri, C. E. P., G. Sparovek, M. Bernoux, W. E. Easterling, J. M. Melillo, C. C. Cerri. 2007. Tropical agriculture and global warming: impacts and mitigation options. Scientia Agricola 64: 83-99. Conservation International. 2008. Cerrado. Biodiversity hotspots. www.biodiversityhotspots.org 20 January 2008. DeFries, R., D. Morton, G. van der Werf, L. Giglio, G. J. Collatz, J. Randerson, R. A. Houghton, P. Kasibhatla, and Y. E. Shimabukuro. 2008. Fire-related carbon emissions from land-use transitions in southern Amazonia. Geophysical Research Letters 35. Etheridge, D. M., L. P. Steele, R. L. Langenfelds, R. J. Francey, J. M. Barnola, and V. I. Morgan, 1996. Natural and Anthropogenic Changes in Atmospheric CO2 over the Last 1000 Years From Air in Antarctic Ice and Firn. Journal of Geophysical Research-Atmospheres 101(D2):4115-4128 FAO (United Nations). 2006. Global Forest Resources Assessment 2005: Progress Towards Sustainable Forest Management. Food and Agricultural Organization of the United Nations, Rome. Fearnside, P.M., C.A. Righi, P.M.L.A. Graça, E.W.H. Keizer, C.C. Cerri, E.M. Nogueira, and R.I. Barbosa. 2009. Biomass and greenhouse-gas emissions form land-use change in Brazil’s Amazonian “arc of deforestation”: The states of Mato Grosso and Rondônia. Forest Ecology and Management, In Press.     80    Felzer, B., D. W. Kicklighter, J. M. Melillo, C. Wang, Q. Zhuang, and R. Prinn. 2004. Effects of ozone on net primary production and carbon sequestration on the conterminous United States using a biogeochemistry model. Tellus 56B:230- 248. Felzer, B. J., J. Reilly, J. M. Melillo, D. W. Kicklighter, C. Wang, R. Prinn, and Q. Zhuang. 2005. Future effects of ozone on carbon sequestration and climate change policy using a global biogeochemical model. Climatic Change 73:345- 373. Fundação Agrisus. 2006. Rally da Safra 2006: Situação do plantio direto e da integração lavoura-pecuária no Brasil. Agroconsult, Florianópolis. Galford, G. L., J. M. Melillo, J. F. Mustard, C. E. P. Cerri, L. VanWey, and C. C. Cerri. In Review. Global frontier of land-use change: The extensification and intensification of croplands in the Amazon and consequences for greenhouse gas emissions. Galford, G. L., J. F. Mustard, J. Melillo, A. Gendrin, C. C. Cerri, and C. E. P. Cerri. 2008. Wavelet analysis of MODIS time series to detect expansion and intensification of row-crop agriculture in Brazil. Remote Sensing of Environment 112:576-587. Grace, J., J. Lloyd, J. McIntyre, A. Miranda, P. Meri, H. Miranda, C. Nobre, J. Monteith, J. Massheder, I. Wright, and J. Gash. 1995. Carbon dioxide uptake by an undisturbed tropical rain forest in southwest Amazonia, 1992 to 1993. Science 270:778-780.     81    Hansen, M. C., S. V. Stehman, P. V. Potapov, T. R. Loveland, J. R. G. Townshend, R. DeFries, K. W. Pittman, B. Arunarwati, F. Stolle, M. K. Steininger, M. Carroll, and C. Dimiceli. 2008. Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data. PNAS 105:9439-9444. Houghton, R. A., J. E. Hobbie, J. M. Melillo, B. Moore, B. J. Peterson, G. R. Shaver, and G. M. Woodwell. 1983. Changes in the carbon content of terrestrial biota and soils between 1860 and 1980: A net release of CO2 to the atmosphere. Ecological Monographs 53:235-262. Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83:195-213. INPE. 2009a. Sistema DETER. Accessed online 09 September 2009. http://www.obt.inpe.br/deter/ INPE. 2009b. Projecto PRODES: Monitoramento da floresta Amazônica Brasileira por satélite. Accessed online 09 September 2009. http://www.obt.inpe.br/prodes/ Jacobsen, A. R., S. E. M. Fletcher, N. Gruber, J. L. Sarmiento, and M. Gloor. 2007. A joint atmosphere-ocean inversion for surface fluxes of carbon dioxide: 1. Methods and global-scale fluxes. Global Biogeochemical Cycles 21. Jantalia, C. P., D. V. S. Resck, B. J. R. Alves, L. Zotarelli, S. Urquiaga, and R. M. Boddey. 2007. Tillage effect on C stocks of a clayey Oxisol under a soybean-     82    based crop rotation in the Brazilian Cerrado region. Soil and Tillage Research 95:97-109. Justice, C. O. 1998. The moderate resolution imaging spectroradiometer (MODIS): Land remote sensing for global change research. IEEE Transactions on Geosciences and Remote Sensing 36:1228-1249. Karlen, D. L. and C. A. Cambardella. 1996. Conservation strategies for improving soil quality and organic matter storage. IN: Structure and Organic Matter Storage in Agricultural Soils, Advances in Soil Science, Carter, M.R., B.A. Stewart (eds), CRC Press, Boca Raton, pp. 395-420. Keeling, C. D., S. C. Piper, R. B. Bacastow, M. Wahlen, T. P. Whorf, M. Heimann, and H. A. Meijer. 2005. Atmospheric CO2 and 13CO2 exchange with the terrestrial biosphere and oceans from 1978 to 2000: observations and carbon cycle implications, pp. 83-113 IN: A History of Atmospheric CO2 and its effects on Plants, Animals, and Ecosystems, J. R. Ehleringer, J.R., T. E. Cerling, and M. D. Dearing (eds), Springer Verlag, New York. Data continually updated. Keys, E., and W. H. McConnell. 2005. Global change and the intensification of agriculture in the tropics. Global Environmental Change 15:320-337. Kladivko, E. J. 2001. Tillage systems and soil ecology. Soil & Tillage Research 61:61-76. Klink, C. A., and R. B. Machado. 2005. Conservation of the Brazilian Cerrado. Conservation Biology 19:707-713.     83    Lal, R., and J. M. Kimble. 1997. Conservation tillage for carbon sequestration. Nutrient Cycling in Agroecosystems 49:243-253. Laurance, S. G. W., W. F. Laurance, H. E. M. Nasciemento, A. Andrade, P. M. Fearnside, E. R. G. Rebello, and R. Condit. 2009. Long-term variation in Amazon forest dynamics. Journal of Vegetation Science 20:323-333. LBA-ECO. 2009. LBA-ECO. Accessed online 09 September 2009. http://www.lbaeco.org/lbaeco/index.html. Malhi, Y., A. D. Nobre, G. J., K. B., P. M.G., G. A., and S. Saleska. 1998. Carbon dioxide transfer over a Central Amazonian rain forest. Journal of Geophysical Research 103:31,593-531,612. Marland, G., T. A. Boden, and R. J. Andres. 2008. Global, regional, and national CO2 emissions. Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy, Oak Ridge, Tenn., USA. McGuire, A. D., J. M. Melillo, L. A. Joyce, D. W. Kicklighter, A. L. Grace, B. Moore III, and C. J. Vorosmarty. 1992. Interactions between carbon and nitrogen dynamics in estimating net primary productivity for potential vegetation in North America. Global Biogeochemical Cycles 6:101-124. McGuire, A. D., J. M. Melillo, and D. Kicklighter. 2001. Carbon balance of the terrestrial biosphere in the twentieth century: Analyses of CO2, climate and land-use effects with four process-based ecosystem models. Global Biogeochemical Cycles 15:183-206.     84    Mello, F. 2007. Estimativas dos estoques de carbono dos solos nos Estados de Rondônia e Mato Grosso anteriores à intervenção antrópica. Universidade de São Paulo, Piracicaba, Brazil. Ministry of Science and Technology. 2006. Carbon dioxide emissions and removals from forest conversion and abandonment of managed lands. IN: First Brazilian Inventory of Anthropogenic Greenhouse Gas Emissions. Brazilian Ministry of Science and Technology, Brasilia. Miranda, A., H. S. Miranda, J. Lloyd, J. Grace, R. J. Francey, J. A. McIntyre, P. Meir, P. Riggan, R. Lockwood, and J. Brass. 2008. Fluxes of carbon, water and energy over Brazilian cerrado: an analysis using eddy covariance and stable isotopes. Plant, Cell & Environment 20:315-328. Mitchell, T., T.R. Carter, P.D. Jones, M. Hulme, and M. New. 2004. A comprehensive set of high-resolution grids of monthly climate for Europe and the Globe: The observed record (1901-2000) and 16 scenarios (2001-2100). Working Paper 55. Tyndall Center for Climate Change. University of East Anglia, Norwich, UK. 25 p. Moraes, J.F.L., B. Volkoff, C.C. Cerri, and M. Bernoux. 1996. Soil properties under Amazon forest and changes due to pasture installation in Rondônia, Brazil. Geoderma 70:63-81. Morton, D., R. DeFries, and Y. E. Shimabukuro. 2009. Cropland expansion in cerrado and transition forest ecosystems: Quantifying habitat loss from satellite-based vegetation phenology. IN: C. A. Klink, editor. Cerrado Land     85    Use and Conservation, Advanced Applied Biodiversity Science. Conservation International, Washington, D.C. Morton, D. C., R. DeFries, J. Randerson, L. Giglio, W. Schroeder, and G. van der Werf. 2008. Agricultural intensification increases deforestation fire activity in Amazonia. Global Change Biology 14:2262-2275. Morton, D. C., R. S. DeFries, Y. E. Shimbukuro, L. O. Anderson, E. Arai, F. d. B. Espirito-Santo, R. Freitas, and J. Morisette. 2006. Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. PNAS 103. Myers, N., R. A. Mittermeier, C. G. Mittermeier, G. A. B. d. Fonseca, and J. Kent. 2000. Biodiversity hotspots for conservation priorities. Nature 403. National Geophysical Data Center (NGDC). 1994. TerrainBase v. 1.1, 5-minute digital terrain model data. Boulder, Colorado. Neill, C., B. Fry, J.M. Melillo, P. A. Stuedler, C. C. Cerri, J. F. L Moraes, M. C. Piccolo, and M. Brito. 1997. Soil organic matter stocks following deforestation for pasture in the southwestern Brazilian Amazon. Ecological Applications 7:1216-1225. New, M., D. Lister, M. Hulme, and I. Makin. 2002. A high-resolution data set of surface climate over global land areas. Climate Research 21. Phillips, O., Y. Malhi, N. Higuchi, W. F. Laurance, P. V. Núñez, R. M. Vásquez, S. G. Laurance, L. V. Ferreira, M. Stern, S. Brown, and J. Grace. 1998. Changes in the carbon balance of tropical forests: Evidence from long-term plots. Science 282:439-442.     86    Potter, C., S. Klooster, A. Huete, V. Genovese, M. Bustamante, L. G. Ferreira, R. Cosme de Oliveira Junior, and R. Zepp. 2009. Terrestrial carbon sinks in the Brazilian Amazon and Cerrado region predicted from MODIS satellite data and ecosystem modeling. Biogeosciences Discussions 6:947-969. Raich, J. W., E. B. Rastetter, J. M. Melillo, D. W. Kicklighter, P. A. Steudler, B. J. Peterson, A. L. Grace, B. Moore III and C. J. Vorosmarty.1991. Potential net primary productivity in South America: application of a global model. Ecological Applications 1, 399-429. Rudel, T.K., O. Coomes, E. Moran, A. Angelsen, F. Achard, E. Lambin, and J. Xu. 2005. Forest transitions: towards an understanding of global land use change. Global Environmental Change 14(4):23-31. Saleska, S., S. D. Miller, D. M. Matross, M. L. Goulden, S. C. Wofsy, H. R. d. Rocha, P. B. d. Camargo, P. Crill, B. C. Daube, H. C. d. Freitas, L. Hutyra, M. Keller, V. Kirchhoff, M. Menton, J. W. Munger, E. H. Pyle, A. H. Rice, and H. Silva. 2003. Carbon in Amazon Forests: Unexpected seasonal fluxes and disturbance losses. Science 302:1554-1557. Schimel, D. 2007. Carbon cycle conundrums. PNAS 104:18353-18354. SIDRA. 2009a. Censo Agropecuário. Sistema IGBE de Recuperação Automática. SIDRA. 2009b. Produção Agrícola Municipal. Sistema IGBE de Recuperação Automática. Six, J., E.T. Elliott, and K. Paustian. 1999. Aggregate and soil organic matter dynamics under conventional and no-tillage systems. Soil Science Society of America Journal 63:1350-1358.     87    Soares-Filho, B. S., D. C. Nepstad, L. M. Curran, G. C. Cerqueira, R. A. Garcia, C. A. Ramos, E. Voll, A. McDonald, P. Lefebvre, and P. Schlesinger. 2006. Modelling conservation in the Amazon basin. Nature 440:520-523. Stephens, B. B., K. R. Gurney, P. P. Tans, C. Sweeney, W. Peters, L. Bruhwiler, P. Ciais, M. Ramonet, P. Bousquet, T. Nakazawa, S. Aoki, T. Machida, G. Inoue, N. Vinnichenko, J. Lloyd, A. Jordan, M. Heimann, O. Shibistova, R. L. Langenfelds, L. P. Steele, R. J. Francey, and A. S. Denning. 2007. Weak Northern and Strong Tropical Land Carbon Uptake from Vertical Profiles of Atmospheric CO2. Science 316:1732-1735. Tian, H., J. M. Melillo, D. W. Kicklighter, S. Pan, J. Liu, A. D. McGuire, B. Moore III. 2003. Regional carbon dynamics in monsoon Asia and its implications for the global carbon cycle. Global and Planetary Change 37, 201-217, doi:10.1016/S0921-8181(02)00205-9.     88    Acknowledgements This work was supported by NASA’s Earth and Space Science Fellowship (G.L. Galford) and NASA’s Large-Scale Biosphere Atmosphere Experiment in Amazonia (Grant no. NNG06GE20A).     89    Tables   Table 2.1 Potential vegetation types   Potential vegetation types, areal extent in Mato Grosso and main species represented in each land cover (Mello 2007). Potential natural Area, Main species vegetation km2 Bertholletia excelsa, Parkia pendula, Swietenia macrophylla, Virola sp., Astronium nelson-rosae, Hymenaea sp., 137,670 Caryocar villosum, Bertholletia excelsa, Euterpe oleracea, Tropical Evergreen Forests Theobroma cf. subincanum, Theobroma sp., Iriartea sp., Attalea maripa, Euterpe precatoria, Oenocarpus sp. Cedrela fissilis, Aspidosperma sp., Myroxylon peruiferum, Chorisia speciosa, Zanthoxylum riedelianum, Apeiba tibourbou, Tabebuia roseo-alba, T. serratifolia, T. Tropical Deciduous 371,876 impetiginosa, Pseudobombax longiflorum e P. tomentosum, Forests Astronium fraxinifolium, Anadenanthera macrocarpa e A falcata, Myracrodruon urundeuva, Acrocomia sclerocarpa, Orbignia speciosa 38,602 Inga sp., Ficus sp., Talauma ovata, Xylopia emarginata Riparian Forests Hymenaea stigonocarpa, Tabebuia caraiba, Annona Cerrado stricto senso coriacea, Caryocar brasiliensis, Hancornia speciosa, Davilla 264,572 (savanna; 20-50% canopy elliptica, Salvertia convallariaeodora, Curatella americana, cover) Kielmeyera coriacea, Qualea sp., Byrsonima sp. Pterodon pubescens, Bowdichia virgilioides, Hymenaea Cerradão 41,191 courbaril, Magonia pubescens, Qualea sp., Kielmeyera (dense savanna) coriacea, Emmotum nitens, Machaerium sp., Dalbergia sp. Anacardium humile, Annona dioica, Dimorphandra mollis, Campo Limpo 65,738 Alibertia sp., Solanum lycocarpum, Salvertia (savanna grassland) convallariaeodora 5,576 Other (water, unclassified)     90    Table 2.2 Land cover compositions   Modeled compositions of land cover types in Mato Grosso Land Cover Type Mesic Trees Xeromorphic Trees and Grass (Percent) Shrubs (Percent) (Percent) Tropical Evergreen 100 0 0 Forests Tropical Deciduous 100 0 0 Forests Riparian Forests 100 0 0 Cerrado Stricto Senso 10 60 30 Cerradão 70 30 0 Campo Limpo 0 20 80     91    Table 2.3 TEM simulations Descriptions of simulations for TEM according to transient data sets used for the land-cover scenarios. Land-Cover/ Atmospheric Simulation Effects Examined Management Scenario CO2 Both single (no fertilization) and double S1 Control Variable crops with conservation tillage Land-use change All natural vegetation (no S2 Variable (S1 – S2) management) CO2 Fertilization Both single (no fertilization) and double S3 Constant (S1 – S3) crops with conservation tillage Tillage Both single (no fertilization) and double S4 Variable (S1 – S4) crops with no tillage Tillage Both single (no fertilization) and double S5 Variable (S1 – S4) crops with conventional tillage Cropping pattern Cropland – all single cropped (no S6 Variable (S4 – S6) fertilization, no tillage) N fertilizer application Cropland – all single cropped (optimal S7 Variable (S7 – S6) fertilization, no tillage)     92    Table 2.4 Comparison of studies Comparison of carbon emission estimates for Mato Grosso. Net average area cleared (2001-2005, km2 year-1) INPE (2009b)/ DeFries et al. Fearnside et al. This paper (2008) (2009) approach* Forest transitions 7371 8991 6,223 Cerrado → - - 7,599 Pasture Cerrado → Crop 4,284 - 2,338 Pasture → Crop 2,862 - 6,577 Model overview Process-based cohort Process-based cohort Bookkeeping model model model Combustion (pasture and cropland clearings), decay (process-based Combustion, decay, model distinguishes graphitic particle heterotrophic and release, carbon storage autotrophic respiration, from replacement Combustion, decay including the long-term biomass, emissions decay of slash and from cattle, and loss of wood products), carbon forest methane sources dynamics for and sinks agricultural vegetation and soils simulated with process-based model Net annual average emissions (2001-2005) 84 Tg C year-1 113 Tg C year-1 163 Tg C year-1 *Assumes the same net emissions on a per area basis as Fearnside et al. 2009 calculate for 2007-2008, here adjusted for higher deforestation rates from 2001- 2005.     93    Figures Figure 2.1. Mato Grosso study area shown with potential natural vegetation classes (after Mello 2007). Figure 2.2. Reconstruction of croplands (a) and pasture (b) in Mato Grosso by area, plotted with estimates from government surveys, census and remote sensing used to constrain the reconstruction. Figure 2.3. Potential vegetation, land use in 2006 and total carbon lost from land-use conversions over the study period. Figure 2.4. Carbon emissions from land-use and land-use change and carbon uptake by intact natural vegetation, including emissions from conversions from natural vegetation to agriculture, transitions between pasture and crop agriculture, and management following - clearing. Figure 2.5. Land-use change emissions by land-use transition. Figure 2.6. Cumulative losses in carbon stocks from land-use change, grouped by forest- and cerrado-type biomes (a). Cumulative carbon losses from vegetation carbon (b) and soil carbon (c). Figure 2.7.Carbon uptake by intact natural vegetation by land cover class.       94    Figure 2.1 Study area       95    Figure 2.2 Historical reconstruction of agricultural uses     96    Figure 2.3 Carbon losses from land-use conversions 1 [Galford et al., In Review-b], 2Morton et al. 2006, 2009       97    Figure 2.4 Carbon emissions from land-cover and land-use change in Mato Grosso       98    Figure 2.5 Carbon emissions by land-use transition       99    Figure 2.6 Cumulative carbon losses A.   B.   C.       100    Figure 2.7 Carbon dynamics in natural vegetation       101    Appendix 2.A.   We use the 2001 remote sensing data sets for pasture and cropland as the starting point for creating a historical land-use data sets and use the annual historical total areas of pasture and crops reported in state-level census estimates (Censo Agropecuário; SIDRA 2009a) and annual crop statistics (Produção Agrícola Municipal, PAM; SIDRA 2009b) to extend land-use change backwards through time. We use an iterative process to randomly select pixels for removal from the current year’s cropland and pasture extents, separately, until the respective total areas for cropland and pasture for the previous year equaled the area prescribed by our interpolated census constraints. While the spatial reconstruction does not attempt to be historically and geographically precise, it accurately depicts the statewide trends in land-use transitions in pasture and croplands that we can use to estimate impacts on carbon and nitrogen cycling. In constraining total pasture and cropland area from the census and annual statistics, we had to address the effects of changes in political boundaries, interpolation between records from different sources and interpolation between decadal estimates on our annual estimates. In 1978, Mato Grosso split almost in half to form two states; Mato Grosso and Mato Grosso do Sul. Prior to 1978, the Mato Grosso census records include areas that today are part of Mato Grosso do Sul, which are not part of our study area. For the late 1970s, the overlap between PAM (Mato Grosso and Mato Grosso do Sul reported together) and the Censo Agropecuário (Mato Grosso do Sul separately) data sets for crops were used to normalize the pre-1978 Mato Grosso records to reflect only the area of Mato Grosso     102    as it exists today (Figure 2.A.1). From 1940-1975, we relied on decadal census estimates and used linear interpolation to make an annual time series, after normalizing for the land area in Mato Grosso do Sul. The pasture area was reconstructed through annual linear interpolation of decadal census data from 1935 to 1996, after normalization to remove records accounting for Mato Grosso do Sul (Figure 2.A.1). From 1996 to 2001, the pasture area was reconstructed through linear interpolation between the 1996 census data and the 2001 remote sensing pasture information. Between 2001 and 2006, remote sensing imagery is used to determine both the distribution and extent of pastures and croplands. For pastures, remote sensing estimates in 2001 and 2004 are used directly and estimates for the other years are determined through linear interpolation. For single cropped and double cropped areas, we use newly developed remote sensing data sets for cropland estimates and cropping patterns that are well-validated (Galford et al. 2008, Galford et al. In Review). These estimates are derived from phenological analysis of a wavelet- smoothed time series of MODIS Enhanced Vegetation Index (EVI; Huete et al. 2002) at a moderately coarse resolution of 500 meters, appropriate for the size of croplands in Mato Grosso (the majority over 200 ha, many over 1,000 ha; Alves 2002, Galford et al. 2008). This data set shows that croplands more than doubled from 2001-2006 to cover about 100,000 square kilometers, with double-cropping patterns accounting for roughly 20% of croplands. We spatially combined the two estimates of pasture for 2001 and 2004; the Morton et al. (2006) data is a better     103    estimate of pasture in forested areas, whereas the Morton et al. (2009) data is a better estimate for pasture in areas of cerrado. We randomly fill in pasture each year at a rate suggested by the census and annual crop statistics within the spatial constraints from the remote sensing data for 2001 and 2004.     104    Figure 2.A.8 Cropland and pasture reconstructions Reconstruction of croplands (a) and pasture (b) in Mato Grosso by area, plotted with estimates from government surveys, census and remote sensing used to constrain the reconstruction. a. b.         CHAPTER 3 Greenhouse gas emissions from land-cover and land-use change: scenarios of deforestation and agricultural management to 2050 Gillian L. Galford 1, 2, Jerry M. Melillo1, David W. Kicklighter1, Timothy W. Cronin1, 3, Carlos E.P. Cerri4, John F. Mustard2, Carlos C. Cerri5 1 The Ecosystems Center, MBL, Woods Hole, MA 2 Geological Sciences, Brown University, Providence, RI 3 Now at: Earth, Atmospheric and Planetary Sciences, MIT, Boston, MA 4 Escola Superior de Agricultura “Luiz de Queiroz”, Universidade de São Paulo, Piracicaba, Brazil 5 Centro de Energia Nuclear na Agricultura, Universidade de São Paulo, Piracicaba, Brazil 105  106    0. Abstract The Brazilian Amazon is one of the most rapidly developing agricultural areas in the world, and represents a potentially large future source of greenhouse gas emissions from land clearing and subsequent agricultural management. We examine scenarios of deforestation and post-clearing land use to estimate the future (2006 to 2050) impacts on carbon dioxide (CO2), methane (CH4) and nitrous oxide (N2O) emissions for the agricultural frontier state of Mato Grosso using a process-based ecosystems model, the Terrestrial Ecosystems Model (TEM). We estimate a net emission of greenhouse gases from Mato Grosso ranging from 4.19 to 15.63 Pg CO2- equivalents (CO2-e) from 2006 to 2050. Deforestation is the largest source of greenhouse gas emissions but land uses following clearing account for a large portion (20-40%) of the net greenhouse gas budget. Due to land-cover and land-use change, there is a foregone carbon sequestration 0.2-0.5 Pg CO2-e by intact forests and cerrado. Clearing natural areas for croplands releases large amounts of carbon immediately on conversion, whereas the creation of pastures transfers a large amount of slash to the soil organic matter pool that releases carbon slowly over time through decomposition. We find that both deforestation and future land-use management play important roles in the net greenhouse gas emissions of this frontier, suggesting that the both should be considered in emissions policy. Key words: Amazon, land-cover and land-use change, scenarios, pasture, crops, Terrestrial Ecosystems Model (TEM), greenhouse gases, carbon dioxide, methane, nitrous oxide, carbon sink     107    1. Introduction Today, just a few frontiers of tropical land-use changes are responsible for 20% of anthropogenic greenhouse gas emissions [Denman et al., 2007a], including the Brazilian Amazon. The Amazon region has long been recognized for its role in regulating global carbon and hydrologic cycles, but today the natural landscape is being impacted by climate change and rapid agricultural development. Pastures have been the dominant land use for decades but, recently, the rate of formation of new cropland areas has surpassed the rates of pasture formation [Barona, 2008; Nepstad et al., 2006]. In the last decade, soybean agriculture has boomed in an arc running along the southern extent of the Brazilian Legal Amazon due to advances in crop breeding, global market demand, and national demands for food, fiber and fuel [Nepstad et al., 2008]. New conservation incentives from government and the private sector may change the patterns of development in the future. If these incentives take hold, there may be increased economic viability for preserving natural ecosystems in the future. One government carbon trading project already underway is the Amazon Fund established by Brazil in response to REDD, to which Norway has already pledged $1 billion dollars by 2025 for forest conservation. In the private sector, The Marriott Corporation gives US$1/night/room to the Foundation for a Sustainable Amazon in Amazonas state. The Brazilian Amazon is at a crossroads for management and policy decisions regarding conservation and development that will affect its future carbon     108    emissions [Fearnside et al., 2009]. Rapid shifts in drivers of land use, such as a national biofuels program or emerging international markets, may shift the cropland portfolio from food and fodder crops to fuel crops, or increase conversion of natural ecosystems or pastures to croplands in response to public and private initiatives. The Amazon Scenarios Project sought to understand the responses of land use, forests, climate, biodiversity and watersheds to policy interventions. This group released a set of scenarios on potential future land-use changes for the period 2000 to 2050 based on biophysical features, socioeconomic factors, infrastructure development, and different development scenarios related to conservation laws and enforcement [Soares-Filho et al., 2006]. While these scenarios project the extent and timing of deforestation in the future, they do not account for post-clearing land management. There has yet to be an in-depth study of the post-clearing and intact vegetation carbon and nitrogen budgets to balance, or in addition to, carbon losses from deforestation. Carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O) are the main greenhouse gases associated with the land-cover and land-use changes in Mato Grosso. Biomass burning associated with land clearing for agriculture releases large amounts of CO2 [DeFries et al., 2008]. Carbon uptake by intact natural vegetation is the primary regional carbon sink and may be enhanced by future increases in atmospheric CO2 [Laurance et al., 2009]. Pastures and croplands can gain or lose carbon, depending on management [Cerri et al., 2007; Galford et al., In Review-a]. Cattle grazing on pastures are the largest source of methane, with each animal emitting roughly 55 kg CH4 year-1 [Steudler et al., 1996]. Large nitrous oxide     109    emissions come from intact natural forests [Garcia-Montiel et al., 2004], croplands following N-fertilization, and from recently established pastures [Melillo et al., 2001]. Here we examine several questions to help us understand the future of the Amazon frontier: 1) What will regional greenhouse gas emissions be, given future scenarios of deforestation and land use? 2) How does post-clearing land management influence emissions, such as the cases of traditional dominance of pastures or the rapidly emerging croplands where fertilizer is used? 3) What role will forests place in mitigating greenhouse gases? To address these questions, we employ a process-based ecosystems model, the Terrestrial Ecosystems Model (TEM), with a set of deforestation and land-use scenarios and a conservative climate scenario. 2. Background and Methods 2.1 Study Area Mato Grosso is a large state (925,225 km2) in the southern Brazilian Amazon, covered by a mix of cerrado and forests, croplands and pastures (Table 3.1; Figure 3.1). Today, the state is mostly natural vegetation with tropical forests accounting for 41% and cerrado accounting for 23% of the current land cover. The cerrado, considered a global biodiversity hot spot [Myers et al., 2000], varies in community structure from tree-rich areas (cerradão), to grass- (campo limpo) and shrub- dominated areas (cerrado stricto senso). Beginning in the 1940s and continuing today, there is large-scale clearing of natural vegetation for pasture and cropland with little transition back to secondary     110    growth and replacement of pastures to grow crops [Barona, 2008; Nepstad et al., 2006]. Croplands bring with them important management decisions related to cropping patterns, fertilizer use, and tillage. In Mato Grosso, it is common to shift from single (typically soybean) to double cropping (typically soybean followed by corn) in an effort to increase production on the same amount of farmland, enabled by the use of fertilizers [Fundação Agrisus, 2006; Sanchez et al., 1982]. A modified no-tillage regime is most commonly used in this region with a rotational program of three years with no-tillage followed with conventional tillage using deep disking plows in the fourth year [Brown et al., 2005; Galford, unpublished; Stickler, unpublished]; we refer to this as conservation tillage. 2.2 Future land-cover and land-use change in Mato Grosso Soares-Filho et al. [2006] made publicly available a set of temporally- and spatially-explicit Amazon deforestation scenarios for 2000 to 2050. To model changes in land-cover from deforestation, we used the deforestation extent from two scenarios, Business-As-Usual (BAU) and Governance (GOV) [Soares-Filho et al., 2006], modified to consider only new deforestation relative to the land-use footprint of 2006 (Figure 3.1) [Galford et al., In Review-b]. For 2000 through 2006, we used detailed land-use information from remote sensing [Galford et al., In Review-b] on cropland and pasture areas (Figure 3.1). This was our starting footprint for land use for all scenarios. The BAU deforestation scenario assumed that: 1) contemporary trends in deforestation rates will continue, 2) all planned road development will be carried out, 3) compliance to conservation laws on private land will remain low, and     111    4) no new protected areas will be created. The GOV deforestation scenario assumed: 1) implementation of environmental legislation, 2) enforcement of legislation, including conservation of forest areas on private lands, land-use zoning and expansion of protected areas [Soares-Filho et al., 2006]. Post-clearing land use will affect biogeochemical cycling of carbon and nitrogen and determines the magnitude of greenhouse gas emissions. To address the impacts of post-clearing land use on biogeochemistry, we considered two types of future land use associated with the BAU and GOV deforestation scenarios: 1) all new land use as pasture (Pasture), and 2) all new land use as cropland (Crop). In these scenarios, croplands shifted from single to double cropping at random, providing they had at least 3 years in single cropping. Changing from single to double cropping after roughly three years is the common practice in this region [Fundação Agrisus, 2006; Galford et al., In Review-b]. For the Pasture and Crop scenarios, new lands were cleared each year (dictated by the deforestation scenario discussed above) and put into the appropriate land use for that scenario. The Crop and Pasture land-use scenarios represent the extremes of land use development— the actual portfolio of land uses may be somewhere in between. In addition to these deforestation and land-use scenarios, we conducted an additional simulation (CONST) where no change in land cover and land use is assumed to occur after 2006. This simulation allows us to understand the legacy carbon consequences of land-use changes that occurred prior to the study period. The CONST scenario shows the impacts of changing climate and CO2 on natural vegetation and agricultural systems in the absence of future land-use change. The     112    BAUCrop, BAUPasture, GOVCrop and GOVPasture scenarios examined the impacts of different deforestation patterns with the range of possible land uses. 2.3 Terrestrial Ecosystems Model We simulated monthly terrestrial carbon and nitrogen dynamics using a version of the process-based Terrestrial Ecosystem Model (TEM) that incorporates the effects of land management observed in Mato Grosso [Galford et al., In Review- a]. The TEM is calibrated to data on carbon and nitrogen stocks and fluxes at intensively studied field sites. Additional details of these ecosystem dynamics may be found elsewhere [Galford et al., In Review-a]. In the TEM, the net exchange of carbon dioxide between terrestrial ecosystems and the atmosphere (NCE) depends on both ecosystem metabolism, as influenced by local environmental conditions, and land management practices: NCE = GPP – R – EF – EP (1) where GPP is gross primary production (i.e. the uptake of atmospheric carbon dioxide by plants during photosynthesis), which is influenced by atmospheric carbon dioxide and ozone concentrations, photosynthetically active radiation, air temperature, evapotranspiration, soil available nitrogen, and canopy stature; R is respiration of both autotrophs (RA) and heterotrophs (RH); EF is the carbon released from burning during land clearing; and EP is the decay of woody and agricultural products [Galford et al., In Review-a]. A positive NCE indicates a net carbon sink, while a negative NCE indicates a carbon source.     113    We estimate the impacts of cropland management in TEM using nitrogen fertilizer and soil tillage factors. We assume that no N fertilizer applications occur in single-cropped areas or for the first crop (assumed to be soy) of double-cropped areas, but the second crop (assumed to be corn) is assumed to be optimally fertilized. All croplands were assumed to use conservation tillage, where tillage occurs only once every fourth year [Galford et al., In Review-a].   2.4 Nitrous oxide emissions We estimate the net N2O emissions from Mato Grosso by estimating the contributions from forests, pastures, and fertilized cropland separately: N2O emissions = N2OFORESTS + N2OCROPS + N2OPASTURES (2) In cerrado regions, N2O emissions are small and typically below the detection limits of field measurements [Bustamante et al., 2006] so we assume cerrado emissions were zero except for areas of fertilized cropland. The N2O emission estimates are converted to carbon dioxide equivalents (CO2-e) by multiplying the emissions by 300, the global warming potential of N2O relative to CO2 at a 100-year time horizon [Forester et al., 2007]. For intact forests, Garcia-Montiel et al. [2004] found a strong linear relationship (p < 0.0001) of N2O emissions to soil respiration (Rs). We make use of this relationship using the TEM estimates of respiration to relate Rs to N2O emissions, as follows:     114    N2OFORESTS = - 4.78 + 0.20Rs (4a) Rs = RH + α (RA) (4b) where α (0.35, Garcia-Montiel et al. [2004]) is the fraction of autotrophic respiration (RA) of plants assumed to be root respiration. In TEM, RA is dependent upon the amount of vegetation biomass, air temperature and GPP. Heterotrophic respiration (RH) is dominated by microbial respiration, which is associated with the decomposition of organic matter and is influenced by the amount and C:N ratio of soil organic matter, air temperature and soil moisture. As described earlier, we assume fertilizer was used only for second crops in a double cropping pattern [Fundação Agrisus, 2006]. We assume that 3% of the applied fertilizer is lost as N2O [Crutzen et al., 2008; Scanlon and Kiely, 2003]. The N fertilization rate (NFERT) is determined by TEM from which we estimate N2O emissions from crops (N2OCROPS) to be: N2OCROPS = 0.03NFERT (5) In pastures of the southwestern Amazon, extensive field measurements show that N2O fluxes are quite high in the first 3 years after clearing forest (3.1 to 5.1 kg N ha-1 year-1). By the sixth year in pasture and beyond, N2O fluxes from pastures are less than emissions from the forest, measuring just 0.1 to 0.4 kg N ha-1 year-1 [Melillo et al., 2001]. We used these emission rates to estimate the range of N2O emissions     115    from pastures dependent on their age, assuming that pastures of 4 to 5 years age emitted 0.4 to 3.1 kg N ha-1 year-1. The TEM cohort structure allows us to track the annual changes in area of pasture within each age category. To account for the potential range of N2O emissions from pastures, we use a low and a high emissions rate for age. We then calculate net pasture N2O emissions (N2OPASTURE) as the sum of the N2O emissions from each age category, as follows: N2OPASTURE = 3.1 Area YOUNG + 0.4 Area MID + 0.1 AreaOLD (6a) For low N2O emissions: N2OPASTURE = 3.1 AreaYOUNG + 0.4 AreaMID + 0.1 AreaOLD (6b) For high N2O emissions: N2OPASTURE = 5.1 AreaYOUNG + 3.1 AreaMID + 0.4 AreaOLD (6c) where AreaYOUNG is the area of pastures ages 0-3 years, AreaMID is the area of 4-5 year-old pastures, and AreaOLD is the area of pastures 6 years or older. 2.5 Methane emissions For methane (CH4) emissions, we consider only those emissions resulting from cattle. A comprehensive methane study in the southern Amazon by Steudler et al. [1996] shows that CH4 emissions from cattle are six times greater than emissions from soil and that termite emissions are a minor source. Based on the work of     116    Steudler et al. [1996], we use an average emission rate of 55 kg CH4 year-1 per cattle head and converted to CO2 equivalence (CO2-e) using a global warming potential for methane of 25 [Forester et al., 2007]. We estimate emission rates based on both a high stocking rate (1.5 au/ha) associated with well-managed pastures and a low stocking rate (0.48 au/ha) currently observed for Mato Grosso [IGBE, 2009]. 2.6 TEM data sets To develop regional estimates of carbon and nitrogen stocks and fluxes, the TEM needs spatially explicit data for elevation, soil texture, land cover, climate and atmospheric chemistry variables at a spatial resolution of 1km x 1km. The elevation and soil texture are from Galford et al. [In Review-a] and the future land cover data sets have been described earlier (Section 2.2). For climate, we use spatially explicit data sets recently developed by Melillo et al. [2009] to represent future global climate as influenced by a policy to control greenhouse gas emissions from industrial and fossil fuel sources with an atmospheric stabilization of 550 ppmv CO2 concentration by 2100. Atmospheric CO2 concentrations are global annual averages and reach 473 ppmv by 2050 (Figure 3.2.a). The monthly mean air temperature and precipitation data are downscaled from a 0.5o latitude x 0.5o longitude spatial resolution to 1 km2 such that all 1-km grid cells contained within a much larger 0.5ogrid cell simply use the value of a climate driver from the larger grid cell. Because sharp climate gradients generally do not occur in the state of Mato Grosso, we believe this to be an adequate treatment. For     117    ozone, the AOT40 index increases around 2025 and then decreases to just above 2000 levels by 2050 (Figure 3.2.b). The average annual air temperature increases 0.2 degrees every decade (Figure 3.2.c). Regional precipitation shows no strong trend over the study period but high interannual variability (Figure 3.2.d). 3. Results 3.1. Land-cover and land-use change In 2006, one-third of Mato Grosso is already in pasture or cropland (Table 3.1), with croplands accounting for 29% of all agricultural land. Of the remaining area, there is 378,735 km2 of intact forest and 231,487 km2 of cerrado (Table 3.1). The two BAU scenarios project that 49% of Mato Grosso was converted to agriculture by 2050 (Figure 3.3); one-third of the forest area is lost by 2050, and cerrado is reduced by 7 percent. In the BAUCrop scenario, cropland areas accounted for over 50% of all agricultural land, an increase of 162% over the study period. Pastures cover 80% of all agricultural land by 2050 in the BAUPasture scenario, increasing by 140,813 km2 over the study period. For the GOV scenarios, there was a 10% loss of forest areas and a 3% loss from cerrado by 2050. Agriculture covers 37% of the state by 2050 in the GOV scenarios. The increase in cropland areas are more modest in the GOVCrop scenario (46%), compared to the BAUCrop scenario, and account for 37% of all agricultural land. Pasture areas in the GOVPasture scenario increased 18%. 3.2 Greenhouse gas budgets and land-use change     118    3.2.1. Carbon budgets In a BAU scenario, net carbon exchange shows a large carbon loss of 3.0 and 3.1 Pg C for BAUPasture and BAUCrop, respectively. For GOV scenarios, total carbon losses range from 0.6 to 0.7 Pg C over the study period. With no further change in land-use after 2006 (CONST), Mato Grosso would become a small net carbon sink, gaining 0.2 Pg C by 2050. Starting in the year 2014, annual NCE values in the CONST scenario become positive or close to zero but it takes until 2033 for Mato Grosso to switch from a carbon source to a carbon sink. 3.2.2. Carbon lost through land-clearing of natural land covers Of all the sources of carbon loss, land clearing is the largest contributor to the net carbon emissions (Table 3.2). During land clearing, a portion of the carbon lost from the natural ecosystems (“Subtotal” in Table 3.2) is immediately released to the atmosphere and a portion is transferred to product pools (Table 3.2). We find that from 2006 through 2050, the BAUCrop scenario loses the most carbon from land clearing, followed by BAUPasture. The GOVCrop and GOVPasture scenarios lose roughly 75% less carbon compared to their BAU counterparts due to the land area cleared (Table 3.1). By land-use transition type, we find that transitions from forest to crop have the highest carbon losses by land-use trajectory, due to the high biomass of the forest and combustion of all slash when clearing. Cerrado to pasture transitions are a minor source of carbon. 3.3 Greenhouse gas budget of agricultural systems 3.3.1. Carbon dynamics in agricultural systems     119    Within the agricultural systems, carbon storage varies by land use type (Table 3.3). We estimate the largest change in agricultural carbon stocks in the BAUPasture scenario, with all other scenarios losing within 0.05 Pg C of the CONST scenario. Pastures accumulate carbon in vegetation biomass in all scenarios (Table 3.3). Because all crop vegetation biomass is harvested or moved to soil organic matter pools, there is no net increase in vegetation carbon stocks for single or double crops. There is a long-term legacy of prior land-use conversions, as product pools decay over decades and centuries (Table 3.3). For soil carbon, areas of single cropping patterns loose carbon in all scenarios but areas of double cropping patterns show both increases and decreases (Table 3.3). Under double cropping, the greatest increases in soil carbon are in the BAUPasture and GOVPasture scenarios, where most of the cropland areas (99%) are double cropped by the end of the study period. The soil carbon losses appear high for pastures, as the slash from the original land clearing is transferred to the soil organic pool and slowly decays through time. For croplands, this amount of carbon was lost through burning at conversion. 3.3.2 Nitrous oxide emissions from fertilized cropland Annual nitrogen fertilization in TEM ranged from 35.36 to 79.90 kg N ha-1 year-1. For all scenarios, the average annual N fertilization rates increased by 20 kg N ha-1 year-1 between the first decade of the study and the last decade. The total fertilized cropland area in each scenario largely determined the N2O emissions. The BAUCrop scenario had the highest N2O emissions (0.35 Pg CO2-e),     120    twice as high as the lowest scenario (GOVCrop). Emissions of N2O from fertilized croplands are the largest source of anthropogenic N2O (Table 3.4). 3.3.3 Methane from cattle Net methane emissions from cattle with no increases in pasture area (CONST, BAUCrop, GOVCrop) range from 19.23 (low stocking rate) to 60.09 (high stocking rate) Tg C, or an average of 0.44 to 1.37 Tg C year-1. Under the BAUPasture and GOVPasture scenarios, emissions range from 27.28 to 85.26 Tg C and 21.86 to 68.30 Tg C, respectively. 3.4 Greenhouse gas budgets in natural systems Areas that were intact natural vegetation in 2006 (CONST), have the potential to uptake 0.77 Pg C from 2006 to 2050 (Table 3.5). With changes in land cover and land use in the GOV and BAU scenarios, the potential carbon sink is reduced to a range of 0.65- 0.72 Pg C over the study period. In all scenarios, changes in the vegetation carbon stock accounted for roughly 85% of carbon sink (Table 3.5). Intact forest biomes took up the most carbon, largely from tropical deciduous forests. Cerradão took up the most carbon of the cerrado biomes. Nitrous oxide emissions from soils in forested areas decline with decreasing forest area, so it is no surprise that these emissions are lowest in the BAU scenarios. Annual average emissions decline from 0.09 Tg N year-1 to 0.06 to 0.08 Tg N year-1 by 2050. In all scenarios, forest soils are the largest source of N2O emissions.     121    3.5 Greenhouse gas budget for the region Overall, net greenhouse gas emissions are the highest in the BAUPasture scenario, over five-times greater than the emissions under the CONST scenario (Table 3.4). Anthropogenic emissions from land clearing are the largest source of greenhouse gas emissions. The BAUCrop and BAUPasture scenarios similar net emissions, illustrating the dominant role of deforestation in the net greenhouse gas budget. Natural emissions of nitrous oxide from forest soil are slightly higher, in CO2- e, than carbon uptake by intact natural vegetation. After deforestation, the next largest anthropogenic greenhouse gas emissions are CH4 from cattle and N2O from fertilized cropland. Net greenhouse gas emissions under the GOVPasture scenario were the most conservative, but were still almost twice as high as the CONST scenario. 4. Discussion 4.1. Carbon and nitrogen dynamics from land clearing and agricultural management Land-use choices have global-to regional-scale biogeochemical consequences. In the global context, the annual greenhouse gas emission estimated under the scenarios presented here may account for approximately 1% of future global emissions, largely from deforestation [Forester et al., 2007]. This demonstrates the importance of land clearing in the global anthropogenic greenhouse gas budget, although emissions do not end when clearing does. Our work shows a sizable contribution to emissions from post-clearing land management     122    activities. This work underscores the importance in accounting for post-clearing management in both regional and global scale studies. Agricultural use determines post-clearing greenhouse gas emissions and can have a lasting legacy. Through the land-use legacy of agriculture established prior to 2006, Mato Grosso has already committed to a portion of future emissions (2006 to 2050), 4.19 Pg CO2-e. Greenhouse gas emissions related to agricultural management (croplands, pasture, and cattle) are not trivial, as they will account for 18 to 34% of future net emissions from this region. Anthropogenic methane and nitrous oxide alone contribute 9 to 19% of all emissions in the future. Post-clearing emissions could be even higher under agricultural systems that integrate both pasture and croplands in an intra-annual rotation, as is being experimented on small scales in Mato Grosso. On local to regional scales, carbon and nitrogen losses from agricultural systems have consequences for agricultural sustainability. It is in the best interests of agriculture to minimize the potentially large fluxes of C and N to the atmosphere, as it can cause soil degradation. One other study has estimated future soil carbon budgets related to land-use change in the Amazon, examining soil carbon stocks in 2000 and 2030 based on current deforestation rates and FAO predictions of land- cover and land-use change in 2015 and 2030 [Cerri et al., 2007]. Cerri et al. [2007] used the Century model, RothC model and the IPCC method and found a 7% decline in soil carbon overall, with increasing soil carbon in soybean croplands (double-cropped) and well-managed pasture. Our estimates agree for trends in soil carbon stocks for double cropping, but show differing results for pasture. On closer     123    inspection, Cerri et al. [2007] acknowledge that pastures with little management inputs, as we assume for pastures, will lose soil carbon. The impacts of different levels of pasture management could be further studied with TEM. Globally, ruminants are a methane source of 48.7 to 74.9 Tg C year-1 [Wuebbles and Hayhoe, 2002]. This work estimates that Mato Grosso emissions are currently on the order of 1% of global ruminant methane emissions, but will increase up to 2-fold with future land use. Total methane emissions will vary greatly with cattle stocking rates. A study by Lerner and Matthews [1988] used government statistics tracking the number cattle and a methane production rate of 54 kg CH4 year-1 per cattle head and estimated that the center-west region of Brazil (Mato Grosso, Mato Grosso do Sul, Goiás, and the federal capital, Brasilia) emitted 1.6 Tg C year-1 in the mid-1980s. This work suggests that: 1) CH4 emissions from this region have grown in recent decades and 2) development of pastures in Mato Grosso may be a large contribution to regional emissions now and in the future, as well as, to the global methane budget. Nitrous oxide emissions from fertilized croplands (double cropping areas) in Mato Grosso account for 3% of global agricultural emissions of N2O [Denman et al., 2007a]. We estimate that crop emissions will increase 3 to 6 fold by 2050, depending on the amount of area in double-cropping patterns, as fertilization only occurs with the second crop associated with the double-cropping pattern. This result signifies that the region may have a disproportionally high contribution to global anthropogenic N2O emissions by unit area. For N2O emissions from fertilized croplands, further agricultural research on reducing emissions, as well as, improved     124    field data on observed emissions (e.g. Carvalho et al. [2009] and others underway) will improve future regional estimates of N2O. Emissions of N2O from fertilized crops and pasture will increase in the future but are a minor part of the regional greenhouse gas budget. Melillo et al. [2001] elucidated the temporal dynamics of N2O emissions when pastures are created from forests, showing that emissions in the first few years under pasture are quite high. Previous studies in other tropical pastures (e.g., [Keller and Reiners, 1994] also support the methods for pasture N2O estimates presented here. Our work shows that the long-term effect of pasture N2O emissions is a minor contribution to the regional greenhouse gas budget. However, these emissions are not negligible and should not be ignored as the impacts of N2O emissions are great for global climate and ozone damage, and nitrogen losses may have serious consequences for soil nutrient depletion. 4.2. Carbon and nitrogen dynamics in natural systems Tropical forests have long been considered a large carbon sink, particularly the Amazon, but their contemporary and long-term role of the tropics in global carbon cycling are uncertain given the high rates of land clearing [Schimel, 2007]. The question has been raised: will the tropics be a carbon sink or will climate change impacts make them a source [Gullison et al., 2007]? We find that intact natural tropical areas will persist as a carbon sink, perhaps even enhanced by CO2 fertilization, but deforestation negates uptake by natural vegetation in the net C budget. The real question becomes, will the tropics be a carbon sink or will land-use change make them a source? We find a lost carbon sequestration potential of 0.4 to     125    2.9 Pg C due to land-cover and land-use change (2006 to 2050). Land clearing is the largest source of carbon emissions from this region and, even with no change in clearing rates, emissions from deforestation and degradation will increase as CO2 fertilization increases carbon storage in intact natural ecosystems and temperature rises increase CO2 emissions through decomposition. We find that intact forests of Mato Grosso are a sizable contribution to natural global N2O emissions, 1.5% in 2006, while compromising less than 1% of the earth’s land surface [Denman et al., 2007a]. Using an empirical relationship of soil respiration and N2O production [Garcia-Montiel et al., 2004], we find that natural forests account for over 90% of regional N2O emissions in all scenarios. Recent meta-analysis suggests this relationship may be robust across tropical regions [Xu et al., 2008]. In the future, these forest emissions of N2O will decline slightly with forest clearing but at the sacrifice of carbon storage from tropical forests. We emphasize that the reduction of forest N2O emissions does not result in a substantial reduction in the net N2O emissions. Rather, all our scenarios show that increased N2O emissions from fertilized croplands largely compensates for reductions in N2O emissions from declining forest areas. With these methods and the benefit of further field studies, N2O emissions could be extrapolated across the Amazon Basin. 4.3. Net greenhouse gas budget with scenarios and ecosystems modeling Together, scenarios and models help us to quantify the biogeochemical consequences of land-cover and land-use change. Adjusting the historical land-use change estimates presented by Galford et al. [In Review-a] to account for CH4 and     126    N2O emissions as in this paper, we estimate a net historical and future greenhouse gas emissions range (38 to 47 Pg CO2-e) due to differences in deforestation. These finding suggest moratoriums on deforestation for new agricultural cultivation could drastically lower regional carbon emissions. Since 2006, the Brazilian soy industry has voluntarily agreed to a moratorium on new Amazon deforestation for croplands that has been renewed each year and some cattle buyers are now imposing no new deforestation for pastures as well. More work should be done to understand potential unintended consequences of indirect land use impacts, such as a shift to increased land clearing in other regions, like the cerrado. We emphasize the importance of including carbon and nitrogen dynamics associated with the function of natural intact vegetation, the clearing of vegetation, and the post-clearing land use. These factors have typically been addressed separately by previous modeling exercises [DeFries et al., 2008; Fearnside et al., 2009; Potter et al., 2009] but this work shows the importance of an integrated assessment. Field-based biogeochemical studies of land covers and land uses laid important foundations in understanding the primary sources of greenhouse gas emissions (e.g., Steudler et al. [1996] highlight the relative importance of different sources of methane). Further exercises in modeling could include other scenarios, such as a range of climate scenarios instead of the “best case” type scenario presented here. Different deforestation and land-use scenarios could be included as patterns and rates of land-cover and land-use change evolve or laws and economic incentives change.     127    Both carbon storage and future carbon uptake should be regarded as conservation incentives for protecting intact forest ecosystems, in addition to other ecosystems services such as water cycling and climate regulation. Biodiversity and ecosystems services of the cerrado that cannot be accounted for through biogeochemical modeling should be regarded as incentives for conservation. Current deforestation scenarios focus on Amazon deforestation but previous work shows that cerrado regions in Mato Grosso host more cropland than forest regions [Galford et al., In Review-b]. Losses to biodiversity and endemic species of in the cerrado are of high concern, considering this region is a biodiversity hotpot with only 2% of all cerrado lands in conservation [Klink and Machado, 2005]. Some of the other metrics of ecosystem function and health, beyond the biogeochemical work presented here, have already been evaluated for future deforestation scenarios but do not account for agricultural uses after clearing. Such work on the impacts of deforestation include understanding the influence of future deforestation on hydrologic cycles [Coe et al., 2009] and other regional ecosystems consequences. Such studies may place a heavier significance on landscape characteristics that are important environmental impacts of land-cover and land-use change that are not part of the terrestrial biogeochemical story, such as land cover heterogeneity, proximity of clearings to waterways, and post-clearing albedo. Together with the work presented here, such studies present the complete picture for the consequences of future deforestation and agricultural land-uses. 5. Conclusions     128    Land-use change will continue in this region and the ultimate fate of natural vegetation and future agricultural management will largely determine greenhouse gas emissions. Scenarios reduce the uncertainty of future emission sources while elucidating the range of outcomes. They present the trade-offs, in greenhouse gas emission terms, of different levels of deforestation control and management decisions, such as land use type. Mato Grosso and the Amazon are at a crossroads for development, where now is the time to weigh agricultural development goals and environmental sustainability. Both deforestation and post- clearing agricultural land-use must be considered so that future agricultural development can minimize unintended negative consequences and maximize long- term agricultural sustainability. This type of bottom-up approach reduces uncertainty in tropical sources of greenhouse gases and could be applied to wider regions (e.g., the Amazon) or other hotspots of agricultural change. Acknowledgements This work was supported by NASA’s Earth and Space Science Fellowship (G.L. Galford) and NASA’s Large-Scale Biosphere Atmosphere Experiment in Amazonia (Grant no. NNG06GE20A). We thank the Amazon Scenarios Project for open access to the scenarios used here.     129    References Adams, J. B., D. E. Sabol, V. Kapos, R. A. Filho, D. Roberts, M. O. Smith, and A. R. Gillespie. 1995. Classification of multispectral images based on fractions of endmembers: Application to land-cover change in the Brazilian Amazon. Remote Sensing of Environment 52:137-154. Alves, D. S. 2002. Space-time dynamics of deforestation in Brazilian Amazonia. International Journal of Remote Sensing 23:2903-2908. Anderson, L. O., Y. E. Shimabukuro, R. DeFries, and D. Morton. 2005. Assessment of land cover and land use changes in the Brazilian Amazon using multitemporal fraction images derived from Terra MODIS: Examples from the state of Mato Grosso. IEEE Transactions on Geosciences and Remote Sensing 2:315-318. Avissar, R., and D. Werth. 2005. Global hydroclimatological teleconnections resulting from tropical deforestation. Journal of Hydrometeorology 6:134-145. Barona, E. 2008. Indentifying the role of crop production in land cover change in Brazil, 1990-2006. McGill University, Montreal. Boko, M., I. Niang, A. Nyong, C. Vogel, A. Githeko, M. Medany, B. Osman-Elasha, R. Tabo, and P. Yanda. 2007. Africa. Pages 433-467 in M. L. Parry, O. F. Canziani, J. P. Palutikof, P. J. van der Linden, and C. E. Hanson, editors. Climate Change 2007: Impacts, adaptation and vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, UK.     130    Boserup, E. 1965. The conditions of agricultural growth: The economics of agrarian change under population pressure. George Allen and Unwin Limited, London, United Kingdom. Bradley, B. A., R. Jacob, J. Hermance, and J. F. Mustard. 2007. A curve-fitting technique to derive inter-annual phenologies from time series of noisy satellite data. Remote Sensing of Environment 106:137-145. Bradley, B. A., and J. F. Mustard. 2007. Comparison of phenology trends by land cover class: a case study in the Great Basin, USA. Global Change Biology 14:334-346. Broch, D. L., and R. S. Pedroso. 2008. Custo de Produção da Cultura do Milho. Page 1 in F. MS, editor. Fundação MS, Maracaju, MS. Browder, J. O., M. A. Pedlowski, and P. M. Summers. 2004. Land use patterns in the Brazilian Amazon: Comparative farm-level evidence from Rondônia. Human Ecology 32:197-224. Brown, C. J., M. Koeppe, B. Coles, and K. P. Price. 2005. Soybean production and conversion of tropical forest in the Brazilian Amazon: The case of Vilhena, Rondônia. Ambio 34:462-469. Brown, J. C., W. Jepson, J. H. Kastens, B. D. Wardlow, J. M. Lomas, and K. P. Price. 2007. Multitemporal, moderate-spatial-resolution remote sensing of modern agricultural production and land modification in the Brazilian Amazon. GIScience & Remote Sensing 44:117-148.     131    Buschbacher, R., C. Uhl, and A. S. Serrao. 1988. Abandoned pastures in eastern Amazonia II: Nutrient stocks in the soil and vegetation. Journal of Ecology 76:682-699. Bustamante, M., E. Medina, G. P. Asner, G. B. Nardoto, and D. C. Garcia-Montiel. 2006. Nitrogen cycling in tropical and temperate savannas. Biogeochemistry 79:209-237. Carvalho, J. L. N., C. E. P. Cerri, B. J. Feigel, M. Piccolo, V. P. Godinho, and C. C. Cerri. 2009. Carbon sequestration in agricultural soils in the Cerrado region of the Brazilian Amazon. Soil and Tillage Reserach 103:342-349. Castro, E. A., and J. B. Kauffman. 1998. Ecosystem structure in the Brazilian Cerrado: A vegetation gradient of aboveground biomass, root mass and consumption by fire. Journal of Tropical Ecology 14:263-283. Cerri, C. C., M. Bernoux, C. E. P. Cerri, and C. Feller. 2004. Carbon cycling and sequestration opportunities in South America: the case of Brazil. Soil Use and Management 20:248-254. Cerri, C. E. P., M. Easter, K. Paustian, K. Killian, K. Coleman, M. Bernoux, P. Falloon, D. Powlson, N. Batjes, E. Milne, and C. C. Cerri. 2007. Simulating SOC changes in 11 land use change chronosequences from the Brazilian Amazon with the RothC and Century models. Agriculture, Ecosystems & Environment 122:46-57. Coe, M. T., M. H. Costa, and B. S. Soares-Filho. 2009. The influence of historical and potential future deforestation on the stream flow of the Amazon River--     132    Land surface processes and atmospheric feedbacks. Journal of Hydrology 369:165-174. Conservation International. 2008. Cerrado. Biodiversity hotspots. www.biodiversityhotspots.org 20 January 2008. Costa, M. H., S. N. M. Yanagi, P. J. O. P. Souza, A. Ribeiro, and E. J. P. Rocha. 2007. Climate change in Amazonia caused by soybean cropland expansion, as compared to caused by pastureland expansion. Geophysical Research Letters 34. Crutzen, P. J., A. R. Mosier, K. A. Smith, and W. Winiwarter. 2008. N2O release from agro-biofuels production negates global warming reduction by replacing fossil fuels. Atmospheric Chermistry and Physics 8:389-395. Cruz, J. C., F. T. F. Pereira, I. A. P. Filho, and A. M. Coelho. 2005. Resposta de Cultivares de Milho à Adubação Nitrogenada em Cobertura in P. e. A. Ministério de Agricultura, editor. Embrapa Milho e Sorgo, MG, Sete Lagoas, MG, Brazil. Dale, V. H., S. M. Pearson, H. Offerman, and R. V. O'Neill. 1994. Relating patterns of land-use change to faunal biodiversity in the Central Amazon. Conservation Biology 8:1027-1036. DeFries, R., D. Morton, G. van der werf, L. Giglio, G. J. Collatz, J. Randerson, R. A. Houghton, P. Kasibhatla, and Y. E. Shimabukuro. 2008. Fire-related carbon emissions from land-use transitions in southern Amazonia. Geophysical Research Letters 35.     133    DeFries, R. S., M. Hansen, J. R. G. Townshend, and R. Sohlberg. 1998. Global land cover classifications at 8 km spatial resolution: The use of training data derived from Landsat imagery in decision tree classifiers. International Journal of Remote Sensing 19:3141-3168. Denman, K. L., G. Brasseur, A. Chidthaisong, P. Ciais, P. M. Cox, R. E. Dickinson, D. Hauglustaine, C. Heinze, E. Holland, D. Jacob, U. Lohmann, S. Ramachandran, P. L. da Silva Dias, S. C. Wofsy, and X. Zhang. 2007a. Couplings between changes in the climate system and biogeochemistry. in S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H. L. Miller, editors. Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Edgar, J. 2007. IFB CropWatchers: FarmWeek 11/30/2007. Embrapa. 2008. Cultivares de milho, feijão, girassol, e mandioca para o Agrest Sergipano com foco na Agricultura Familiar e no Agronegócio. Page 31 in H. W. L. Carvalho, I. R. Olivera, E. R. C. Donald, S. R. R. Ramos, C. A. P. Pacheco, W. M. G. Fukuda, M. J. Peloso, C. G. P. Carvalho, K. E. O. Melo, L. F. Feitosa, and A. F. Menezes, editors. Documentos / Embrapa Tabulerios Costeiros. Embrapa Tabuleiros Costeiros, Aracaju, SE. Ewers, R. M., J. P. W. Scharlemann, A. Balmford, and R. E. Green. 2009. Do increases in agricultural yield spare land for nature? Global Change Biology 15:1716-1726.     134    Fearnside, P. M. 2001. Soybean cultivation as a threat to the environment in Brazil. Environmental Conservation 28:23-28. Fearnside, P. M., C. A. Righi, P. M. L. A. Graça, E. W. H. Keizer, C. C. Cerri, E. M. Nogueira, and R.I.Barbosa. 2009. Biomass and greenhouse-gas emissions form land-use change in Brazil’s Amazonian “arc of deforestation”: The states of Mato Grosso and Rondônia Forest Ecology and Management In Press. Fisher, J. I., and J. F. Mustard. 2007. Cross-scalar satellite phenology from ground, Landsat, and MODIS data. Remote Sensing of Environment 109:261-273. Fisher, J. I., J. F. Mustard, and M. A. Vadeboncoeur. 2006. Green leaf phenology at Landsat resolution: Scaling from the field to the satellite. Remote Sensing of Environment 100:265-279. Foley, J. A., R. DeFries, G. P. Asner, C. Barford, G. Bonan, S. R. Carpenter, F. S. Chapin, M. T. Coe, G. C. Daily, H. K. Gibbs, J. H. Helkowski, T. Holloway, E. A. Howard, C. J. Kucharik, C. Monfreda, J. A. Patz, I. C. Prentice, N. Ramankutty, and P. K. Snyder. 2005. Global consequences of land use. Science 309:570-574. Forester, P., V. Ramaswamy, P. Artaxo, T. Berntsen, R. Betts, D. W. Fahey, J. Haywood, J. Lean, D. C. Lowe, G. Myhre, J. Nganga, R. Prinn, G. Raga, M. Schulz, and R. Van Dorland. 2007. Changes in Atmospheric Constituents and in Radiative Forcing. In S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H. L. Miller, editors. Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate     135    Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Fundação Agrisus. 2006. Rally da Safra 2006: Situação do plantio direto e da integração lavoura-pecuária no Brasil. Florianópolis. Galford, G. L. unpublished. Field observations, GIS maps and farm records. Collected 2005, 2007. Galford, G. L., J. M. Melillo, D. W. Kicklighter, J. F. Mustard, T. W. Cronin, C. E. P. Cerri, and C. C. Cerri. In Review-a. Carbon emissions and uptake from 105 years of land-cover and land-use change at the agricultural frontier of the Brazilian Amazon. Journal of Geophysical Research. Galford, G. L., J. M. Melillo, J. F. Mustard, C. E. P. Cerri, L. VanWey, and C. C. Cerri. In Review-b. Global frontier of land-use change: The extensification and intensification of croplands in the Amazon and consequences for greenhouse gas emissions. Galford, G. L., J. F. Mustard, J. Melillo, A. Gendrin, C. C. Cerri, and C. E. P. Cerri. 2008. Wavelet analysis of MODIS time series to detect expansion and intensification of row-crop agriculture in Brazil. Remote Sensing of Environment 112:576-587. Garcia-Montiel, D. C., J. M. Melillo, P. Steudler, H. Tian, C. Neill, D. W. Kicklighter, B. Feigl, M. Piccolo, and C. C. Cerri. 2004. Emissions of N2O and CO2 from terra firme forests in Rondonia, Brazil. Ecological Applications 14:S214-S220. Green, R. E., S. J. Cornell, J. P. W. Scharlemann, and A. Balmford. 2005. Farming and the fate of wild nature. Science 307:550-555.     136    Gregory, P. J., and J. S. I. Ingram. 2000. Global change and food and forest production: future scientific challenges. Agriculture, Ecosystems & Environment 82:3-14. Gregory, P. J., J. S. I. Ingram, R. Andersson, R. A. Betts, V. Brovkin, T. N. Chase, P. R. Grace, A. J. Gray, N. Hamilton, T. B. Hardy, S. M. Howden, A. Jenkins, M. Meybeck, M. Olsson, I. Ortiz-Monasterio, C. A. Palm, T. W. Payn, M. Rummukainen, R. E. Schulze, M. Theirn, C. Valentin, and M. J. Wilkinson. 2002. Environmental consequences of alternative practices for intensifying crop production. Agriculture, Ecosystems & Environment 88:279-290. Gullison, R. E., P. C. Frumhoff, J. G. Canadell, C. B. Field, D. C. Nepstad, K. Hayhoe, R. Avissar, L. M. Curran, P. Friedlingstein, C. D. Jones, and C. Nobres. 2007. Tropical Forests and Climate Policy. Science 316:985-986. Hansen, M. C., S. V. Stehman, P. V. Potapov, T. R. Loveland, J. R. G. Townshend, R. DeFries, K. W. Pittman, B. Arunarwati, F. Stolle, M. K. Steininger, M. Carroll, and C. Dimiceli. 2008. Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data. PNAS 105:9439-9444. Houghton, R. A., K. T. Lawrence, J. L. Hackler, and S. Brown. 2001. The spatial distribution of forest biomass in the Brazilian Amazon: a comparison of estimates. Global Change Biology 7:731-746. Houghton, R. A., D. Skole, C. A. Nobre, J. L. Hackler, K. T. Lawrence, and W. H. Chomentowski. 2000. Annual fluxes of carbon from deforestation and regrowth in the Brazilian Amazon. Nature 403:301-304.     137    Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83:195-213. IGBE. 2002. Pesquisas Agropecuárias.in M. d. Planejamento, editor. Ministério do Planejamento, Orçamento e Gestão, Rio de Janeiro. IGBE. 2009. Banco de Dados Agregados: Sistem IBGE de Recuperação Automática (SIDRA). INPE. 2008a. Sistema DETER. Instituto Nacional de Pesquisas Espaciais. INPE. 2008b. Taxa de desmatamento annual: Estimativas Anuais desde 1988 até 2008 Projecto PRODES: Monitoramento da floresta Amazônica Brasilera por satélite. Instituto Nacional de Pesquisas Espaciais. ITT. 2008. ENVI Reference Guide, version 4.6. ITT Visual Information Solutions. Jantalia, C. P., D. V. S. Resck, B. J. R. Alves, L. Zotarelli, S. Urquiaga, and R. M. Boddey. 2007. Tillage effect on C stocks of a clayey Oxisol under a soybean- based crop rotation in the Brazilian Cerrado region. Soil and Tillage Research 95:97-109. Jasinski, E., D. Morton, and R. DeFries. 2005. Physical landscape correlates of the expansion of mechanized agriculture in Mato Grosso, Brazil. Earth Interactions 9:1-18. Jenson, J. R. 2005. Thematic map accuracy assessment. Pages 495-513 Introductory Digital Image Processing: A Remote Sensing Perspective. Pearson Pretice Hall, Upper Saddle River.     138    Jepson, W., C. J. Brown, and M. Koeppe. 2008. Agricultural intensification on Brazil's Amazonian soybean frontier. Pages 73-92 in A. Millington and W. Jepson, editors. Land Change Science in the Tropics. Springer Publications, Boston. Justice, C. O. 1998. The moderate resolution imaging spectroradiometer (MODIS): Land remote sensing for global change research. IEEE Transactions on Geosciences and Remtoe Sensing 36:1228-1249. Keller, M., and W. A. Reiners. 1994. Soil-atmosphere exchange of nitrous oxide, nitric oxide, and methane under secondary succession of pasture to forest in the Atlantic lowlands of Costa Rica. Global Biogeochemical Cycles 8:399- 409. Keys, E., and W. H. McConnell. 2005. Global change and the intensification of agriculture in the tropics. Global Environmental Change 15:320-337. Klink, C. A., and R. B. Machado. 2005. Conservation of the Brazilian Cerrado. Conservation Biology 19:707-713. Laurance, S. G. W., W. F. Laurance, H. E. M. Nasciemento, A. Andrade, P. M. Fearnside, E. R. G. Rebello, and R. Condit. 2009. Long-term variation in Amazon forest dynamics. Journal of Vegetation Science 20:323-333. Lerner, J., and E. Matthews. 1988. Methane emission from animals: a global high resolution data base. Global Biogeochemical Cycles 2:139-156. Mar, G. D., M. E. Marchetti, L. C. F. Souza, M. C. Goncalves, and J. O. Novelino. 2003. Fertilidade do solo e nutricao de plantas: Producao do milho safrinha     139    em funcao de doses e epocas de aplicacao de nitrogenio. Bragantia 62:267- 274. Marland, G., T. A. Boden, and R. J. Andres. 2008. Global, regional, and national CO2 emissions. Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy, Oak Ridge, Tenn., USA. Melillo, J. M., J. Reilly, D. W. Kicklighter, A. C. Gurgel, T. W. Cronin, S. Paltsev, B. S. Felzer, A. P. Sokolov, and C. A. Schlosser. 2009. Indirect emissions from biofuels: how important? Science Express. Melillo, J. M., P. Steudler, B. J. Feigl, C. Neill, D. Garcia, M. C. Piccolo, C. C. Cerri, and H. Tian. 2001. Nitrous oxide emissions from forests and pastures of various ages in the Brazilian Amazon. Journal of Geophysical Research 106:34,179-134,188. Mello, F. 2007. Estimativas dos estoques de carbono dos solos nos Estados de Rondônia e Mato Grosso anteriores à intervenção antrópica. Universidade de São Paulo, Piracicaba, Brazil. Ministry of Science and Technology. 2004. Carbon Dioxide Emissions and Removals from Forest Conversion and Abandonment of Managed Lands. Brazilian Ministry of Science and Technology, Brasilia. Ministry of Science and Technology. 2006. Carbon dioxide emissions and removals from forest conversion and abandonment of managed lands First Brazilian Inventory of Anthropogenic Greenhouse Gas Emissions. Brazilian Ministry of Science and Technology, Brasilia.     140    Miranda, A., H. S. Miranda, J. Lloyd, J. Grace, R. J. Francey, J. A. McIntyre, P. Meir, P. Riggan, R. Lockwood, and J. Brass. 2008. Fluxes of carbon, water and energy over Brazilian cerrado: an analysis using eddy covariance and stable isotopes. Plant, Cell & Environment 20:315-328. Morton, D., R. DeFries, and Y. E. Shimabukuro. 2009. Cropland expansion in cerrado and transition forest ecosystems: Quantifying habitat loss from satellite-based vegetation phenology. In C. A. Klink, editor. Cerrado Land Use and Conservation, Advanced Applied Biodiversity Science. Conservation International, Washington, D.C. Morton, D. C., R. DeFries, J. Randerson, L. Giglio, W. Schroeder, and G. van der werf. 2008. Agricultural intensification increases deforestation fire activity in Amazonia. Global Change Biology 14:2262-2275. Morton, D. C., R. S. DeFries, Y. E. Shimabukuro, L. O. Anderson, F. D. B. Espirito- Santo, M. Hansen, and M. Carroll. 2005. Rapid assessment of annual deforestation in the Brazilian Amazon using MODIS data. Earth Interactions 9:1-22. Morton, D. C., R. S. DeFries, Y. E. Shimbukuro, L. O. Anderson, E. Arai, F. d. B. Espirito-Santo, R. Freitas, and J. Morisette. 2006. Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. PNAS 103. Myers, N., R. A. Mittermeier, C. G. Mittermeier, G. A. B. d. Fonseca, and J. Kent. 2000. Biodiversity hotspots for conservation priorities. Nature 403.     141    Neill, C., J. M. Melillo, P. A. Steudler, C. C. Cerri, J. F. L. d. Moraes, M. C. Piccolo, and M. Brito. 1997. Soil Carbon and Nitrogen Stocks Following Forest Clearing for Pasture in the Southwestern Brazilian Amazon. Ecological Applications 7:1216-1225. Nepstad, D. C., C. M. Stickler, and O. T. Almeida. 2006. Globalization of the Amazon soy and beef industries: Opportunities for conservation. Conservation Biology 20:1595-1603. Nepstad, D. C., C. M. Stickler, B. S. Soares-Filho, and F. Merry. 2008. Interactions among Amazon land use, forests and climate: prospects for a near-term forest tipping point. Philosophical Transactions of the Royal Society 363:1737-1746. Nogueira, E. M., P. M. Fearnside, B. W. Nelson, R. I. Barbosa, and E. W. H. Kiezer. 2008. Estimates of forest biomass in the Brazilian Amazon: New allometric equations and adjustments to biomass from wood-volume inventories. Forest Ecology and Management 256:1853-1867. Potter, C., S. Klooster, A. Huete, V. Genovese, M. Bustamante, L. G. Ferreira, R. Cosme de Oliveira Junior, and R. Zepp. 2009. Terrestrial carbon sinks in the Brazilian Amazon and Cerrado region predicted from MODIS satellite data and ecosystem modeling. Biogeosciences Discussions 6:947-969. Ramos da Silva, R., D. Werth, and R. Avissar. 2008. Regional impacts of future land-cover changes on the Amazon Basin wet-season climate. Journal of Climate 21:1153-1170.     142    Rezende, P. J. L. A. 2007. Cultivares, doses de fertilizantes e densidades de semeadura no cultivo de milho safrinha. UFLA, Lavras. Roy, S. B., and R. Avissar. 2002. Impact of land use/land cover change on regional hydrometeorology in Amazonia. Journal of Geophysical Research 107. Rudel, T. K., O. Coomes, E. Moran, A. Angelsen, F. Achard, E. Lambin, and J. Xu. 2005. Forest transitions: towards an understanding of global land use change. Global Environmental Change 14:23-31. Saatchi, S. S., R. A. Houghton, R. D. Alvala, J. V. Soares, and Y. Yu. 2007. Distribution of aboveground live biomass in the Amazon basin. Global Change Biology 13:816-837. Saatchi, S. S., R. A. Houghton, D. S. Alves, and B. W. Nelson. 2009. LBA-ECO LC- 15 Amazon Basin Aboveground Live Biomass Distribution Map: 1990-2000. Oak Ridge National Laboratory Distributed Active Archive Center. Sanchez, P. A. 2002. Soil fertility and hunger in Africa. Science 295:2019-2020. Scanlon, T. M., and G. Kiely. 2003. Ecosystem-scale measurements of nitrous oxide fluxes for intensely grazed, fertilized grassland. Geophysical Research Letters 30:1852-1856. Schimel, D. 2007. Carbon cycle conundrums. PNAS 104:18353-18354. Shukla, J., C. Nobre, and P. Sellers. 1990. Amazon Deforestation and Climate Change. Science 247:1322-1325. Skole, D., and C. Tucker. 1993. Tropical deforestation and habitat fragmentation in the Amazon: Satellite data from 1978 to 1988. Science 260:1905-1910.     143    Snapp, S. 1998. Soil nutrient status of smallholder farms in Malawi. Communications in Soil Science and Plant Analysis 29:2571-2588. Soares-Filho, B. S., D. C. Nepstad, L. M. Curran, G. C. Cerqueira, R. A. Garcia, C. A. Ramos, E. Voll, A. McDonald, P. Lefebvre, and P. Schlesinger. 2006. Modeling conservation in the Amazon basin. Nature 440:520-523. Souza, E. F. C., and R. P. Sorrato. 2006. Efeito de fonts e doses de nitrogênio em cobertura, no milho safrinha, em plantio direto. Revista Brasileira de Milho e Sorgo 5:395-405. Steudler, P. A., J. M. Melillo, B. J. Feigel, C. Neill, M. C. Piccolo, and C. C. Cerri. 1996. Consequence of forest-to-pasture conversion on CH4 fluxes in the Brazilian Amazon Basin. Journal of Geophysical Research 101:18,547- 518,554. Stickler, C. M. unpublished. Field validation data for the Xingu Basin. Collected 2006. Turner II, B. L., and A. M. A. Ali. 1996. Induced intensification: Agricultural change in Bangladesh with implications for Mathus and Boserup. PNAS 93:14984- 14991. Van der werf, G., J. Randerson, G. J. Collatz, and L. Giglio. 2003. Carbon emissions from fires in tropical and subtropical ecosystems. Global Change Biology 9:547-562. Van der werf, G., J. Randerson, L. Giglio, G. J. Collatz, P. Kasibhatla, and A. F. J. Arellano. 2006. Interannual variability of global biomass burning emissions     144    from 1997 to 2004. Atmospheric Chemistry and Physics Discussions 6:3175- 3226. Wuebbles, D. J., and K. Hayhoe. 2002. Atmospheric methane and global change. Earth-Science Reviews 57:177-210. Xu, X., H. Tian, and D. Hui. 2008. Convergence in the relationship of CO2 and N2O exchanges between soil and atmosphere within terrestrial ecosystems. Global Change Biology 14:1651-1660.     145    Tables Table 3.1 Land covers and land uses by scenario   Potential vegetation types [Mello, 2007], their original extent before land use and extent of natural vegetation and land use by 2050 in each scenario (km2). GOVPasture BAUPasture GOVCrop BAUCrop CONST Natural Natural vegetation extent Tropical 137,670 123,936 65,845 65,845 112,012 112,012 Evergreen Forests Tropical 371,876 229,054 163,417 163,417 206,513 206,513 Deciduous Forests 38,602 25,745 21,175 21,175 24,658 24,658 Riparian Forests Cerrado stricto 264,572 166,854 157,012 157,012 163,266 163,266 senso (savanna) Cerradão 41,191 18,753 16,611 16,611 18,104 18,104 (dense savanna) Campo Limpo 65,738 45,880 41,900 41,900 45,017 45,017 (savanna grassland) - 54,159 47,560 523 50,289 523 Single cropping - 35,374 186,947 89,010 80,091 89,010 Double cropping - 221,092 221,092 366,066 221,092 261,939 Pasture Other 5,576 4,378 3,666 3, 666 4,183 4,183 (water, unclassified)     146    Table 3.2 Carbon losses from land clearing Net carbon changes from land clearing (Pg C) from 2006 to 2050. GOVPasture BAUPasture GOVCrop BAUCrop Forest to crop -3.02 0.00 -0.80 0.00 Forest to pasture 0.00 -2.74 0.00 -0.73 Cerrado to crop -0.10 0.00 -0.03 0.00 Cerrado to pasture 0.00 -0.08 0.00 -0.03 Sub-total -3.12 -2.83 -0.83 -0.75 Product Pools 0.81 0.81 0.22 0.22 Total -2.31 -2.01 -0.61 -0.54     147    Table 3.3 Carbon dynamics in agricultural lands For 2006 to 20505, the changes in carbon pools (Pg C) for land uses (single cropping, double cropping and areas that shifted from single to double cropping; pasture) and from the decay of agricultural (Ag) and woody (10 year and 100 year). The “Net ∆” represents the net losses of carbon from post-clearing land use over the study period. GOVPasture BAUPasture GOVCrop BAUCrop CONST    Single    ∆ Veg C 0.00 0.00 0.00 0.00 0.00 ∆ Soil C -0.04 -0.15 -0.03 -0.07 -0.03 Double ∆ Veg C 0.00 0.00 0.00 0.00 0.00 ∆ Soil C 0.00 -0.04 0.09 0.00 0.09 Single to Double ∆ Veg C 0.00 0.01 0.00 0.00 0.00 ∆ Soil C 0.00 -0.13 -0.01 -0.04 -0.01 Pasture ∆ Veg C 0.02 0.02 0.08 0.02 0.04 ∆ Soil C -0.22 -0.22 -0.80 -0.22 -0.39 Product pools ∆ Ag 0.00 0.02 0.01 0.01 0.01 ∆ 10 year -0.16 -0.76 -0.76 -0.33 -0.33 ∆ 100 year -0.14 -0.21 -0.21 -0.16 -0.16 Net ∆ -0.55 -1.45 -1.64 -0.80 -0.79     148    Table 3.4 Net greenhouse gas budgets Primary sources of greenhouse gas emissions from land cover and land use in Mato Grosso, 2006 to 2050. We present the average value for pasture emissions from cattle and soil, where both a high and low estimate were available. GOVPasture  BAUPasture  GOVCrop  BAUCrop  CONST  Natural land cover             Land clearing of natural land cover (Pg CO2‐e)  ‐0.00  ‐11.44  ‐10.37  ‐3.04  ‐2.76             N2O from forest soils (Pg CO2‐e)  ‐3.95  ‐2.95  ‐2.95  ‐3.64  ‐3.64            Uptake by intact natural vegetation (Pg CO2‐e)  2.82  2.39  2.39  2.64  2.64  Croplands             Carbon dynamics (Pg CO2‐e)  ‐0.14  ‐0.71  0.20  ‐0.26  0.20             N2O emissions from fertilized crops (Pg CO2‐e)  ‐0.07  ‐0.35  ‐0.20  ‐0.18  ‐0.20  Pastures   ‐0.73  ‐0.73  ‐2.66  ‐0.73  ‐1.28            Carbon dynamics (Pg CO2‐e)            N2O from pasture soils (Pg CO2‐e)  0.00  0.00  ‐0.11  0.00  ‐0.03            Methane emissions from cattle (Pg CO2‐e)  ‐0.99  ‐0.99  ‐1.41  ‐0.99  ‐1.13  Product Pools  ‐1.12  ‐0.47  ‐0.53  ‐0.99  ‐1.00  Total (Pg CO2‐e)  ‐4.19  ‐15.25  ‐15.63  ‐7.21  ‐7.20      149    Table 3.5 Carbon in intact natural ecosystems Changes in carbon storage (Pg C) for areas of intact natural vegetation, 2006 to 2050. ∆ Veg C ∆ Soil C Total CONST 0.64 0.13 0.77 BAU 0.55 0.10 0.65 GOV 0.60 0.12 0.72     150    Figures 6. Figure legends Figure 3.1. The state of Mato Grosso with the extent of contemporary land use (2006), BAU and GOV deforestation scenarios, and natural ecosystems. Figure 3.2. Projected changes in environmental factors, as used in TEM, including atmospheric CO2 concentrations (A), mean regional AOT40 ozone index (B), regional annual mean air temperatures (C) and regional annual precipitation (D). Figure 3.3. Change in land use areas by scenario.     151    Figure 3.1 Extent of natural land covers, land use in 2006     152    Figure 3.2 Climate data sets     153    Figure 3.3 Changes in agricultural area by scenario     154    CONCLUSIONS Land cover and land use change Remote-sensing derived, regionally tuned land-cover and land-use change information is essential in understanding rates and trajectories of human-induced environmental changes. This work shows that large-scale agricultural processes, including clearing and subsequent management, can be resolved with moderate resolution remote sensing data. The cropland detection techniques of Chapter 1 and Appendix A can be applied across wide regional scales. These methods are adaptable to changes in planting and harvesting dates that vary with climate gradients and individual farm management. This flexibility in handling the regional heterogeneity of cropland phenologies was not capable with previous techniques in remote sensing [Bradley et al., 2007; Brown et al., 2007; Fisher and Mustard, 2007]. Chapter 2 provides a historical reconstruction of land use that approximates the spatial-temporal dynamics of agricultural development for use in ecosystems modeling (Chapters 2 and 3). Additional efforts could be made to improve the spatial accuracy of such a reconstruction by using coarser resolution historical remote sensing data sets extending further back in time with meso-scale government records instead of state-level. Identifying the location of land-use change is important for understanding the ecological impacts. As shown in Chapters 2 and 3, the clearing of forest ecosystems produces much greater carbon losses than clearing from cerrado. Other potential ecological applications of land-cover and land-use change information (Chapter 1)     155    include: 1) quantifying the extent of natural habitat fragmentation, 2) estimating nutrient run-off into streams and rivers using information on the proximity of land use to water bodies, and 3) modeling regional climate changes from land use. Dynamics & trade-offs of extensification and intensification Agricultural extensification and intensification have large roles in meeting global demands for agricultural production [Gregory and Ingram, 2000], but local environmental impacts and their corresponding contributions to global environmental change need to be understood. Theoretically, agricultural intensification may slow or halt extensification [Boserup, 1965]. Intensification has been cited as one way to reduce carbon emissions from deforestation activities, particularly in the Amazon [DeFries et al., 2008]. A recent global-scale study shows that intensification may not slow extensification, except in certain cases where external product demands are constant or where crops are for subsistence only [Ewers et al., 2009]. The relationship between extensification and intensification needs to be tested empirically in rapidly changing and heavily mechanized regions, such as the agricultural frontier of the Brazilian Amazon. The detailed information presented in Chapter 1 can be applied to analyzing the trade-offs between extensification and intensification in this region. Then, understanding the avoided extensification due to cropland intensification, the avoided emissions from deforestation could be estimated. Advocates of intensification in Mato Grosso and the Amazon suggest that cropland intensification is land-sparing. That is to say, that one can meet the same     156    production targets with a smaller land area if one practices more intensive cropping. This argument implicitly assumes that farmers have a production target and that cost is not an obstacle. By turning these assumptions around, one could assume that farmers will increase production to the extent possible given costs and that the tradeoff for them between intensification and extensification is determined by relative costs. A trade-off ratio could be calculated based on costs and quantity of labor required for clearing, costs of land, machinery used for clearing, planting, harvesting and other associated activities, fuel, and fertilizer. There are further sociological applications of the information on extensification and intensification from Chapter 1. These may include examining the underlying factors that cause intraregional differences in choosing extensification or intensification. One could also estimate the natural land conserved from clearing through investments in cropland intensification (single to double cropping) or intensification from pasture to croplands. The information from Chapter 1 provides important spatial and temporal information for understanding social behavior and decision making. Scenarios There are many incentives for future development of new agricultural lands, as well as for the conservation of intact forests in the Amazon. While agricultural land use is sure to increase, the exact timing, location and extent remains uncertain, and the conservation of remaining ecosystems may depend on government creation and enforcement of environmental regulations, economic incentives for conservation     157    and public acceptance of regulations and incentives. The balance of agricultural development and conservation clearly affects carbon emissions (Chapter 3), which is just one of many ecological factors motivating conservation. The extent of natural areas in the future also influences carbon uptake and long-term carbon storage. We demonstrate that the CO2 fertilization effect observed by Laurance et al. [2009] in Amazon forest stands over the last few decades will persist into the future but that the exact magnitude of this sequestration depends on the extent of remaining forests. The carbon impacts of these scenarios could be further assessed for their economic differences in light of possible carbon credit scenarios for avoided deforestation. Greenhouse gas emissions Carbon emissions from land clearing activities are the largest regional source of greenhouse gases. However, Chapter 3 projects the important role of methane and nitrous oxide from pastures and croplands in the future greenhouse gas budget. Previous field studies show that tropical forest soils emit substantial amounts of nitrous oxide, which we also simulate in our model. While deforestation decreases N2O emissions from forest soils, these reductions are off-set by increases in N2O emissions from croplands. Although large amounts of N2O are released in the first few years when forests are converted to pastures, we find that (1) the initial pulse of N2O is large enough to be included in the regional greenhouse gas budget but that (2) it is not one of the largest sources of emissions. Methane emissions from cattle are large and will depend on the stocking rates of pastures. Well-managed pastures     158    might be cared for with occasional (every several years) application of nitrogen fertilizer, which would increase emissions from pastures. A few sites in Mato Grosso are experimenting with “integration” agriculture that involves single or double cropping during the wet season and growing rich pasture grasses during the dry season with high cattle stocking rates (6-7 AU ha-1). If this system became widespread, emissions would likely increase. Different types of pasture and cropland management could be evaluated through the methods presented here, provided field data exists to sufficiently understand the expected trends under these land managements. Moving beyond the Amazon The work presented in this dissertation is as much a proof of methods and concept as it is a regional case study. It is an integrated approach to assessing changes in agricultural land uses, natural ecosystem dynamics and impacts of changing climate. As land-cover and land-use changes are one of the primary anthropogenic sources of greenhouse gases, there are many potential applications of this approach around the world. This type of spatial-temporal analysis is needed to assess greenhouse gas budgets with regional specificity not provided in global products of land cover or land use. In Chapters 2 and 3, we suggest that other rapidly developing tropical agricultural regions could benefit from this type of integrated land-cover and land-use change analysis, although the exact methods may need adaptation for applications in different systems.     159    Brazilian sugar cane Brazil is one of the largest sugar cane producers in the world. Much of this crop goes to supplying domestic alcohol for fuel and accounts for 40% of Brazil’s consumption of domestically-produced combustible fuels. Traditionally, sugar cane harvests have required high inputs of manual labor to burn and cut the cane, posing hazards to human health from smoke inhalation and poor air quality even far from the fields, while contributing large amounts of carbon dioxide to the atmosphere. Recent advances in farm mechanization now make it possible, within certain topographical constraints, to harvest sugar cane without first burning the fields. This shift in crop management has large impacts on the carbon budget. Preliminary field studies are under way by CENA/USP to quantify long-term differences in soil carbon stocks and above-ground biomass at different sites representing traditional burning and new mechanized harvests. Remote sensing methods based on Chapter 1 can be used to detect these differences in management. Combining field data, crop management information from remote sensing and ecosystems modeling, the impacts of these shifts in management could be assessed locally and for larger regions, such as São Paulo state, the largest sugar-producing region in the country. Food scarcity and soil degradation in sub-Saharan Africa Sub-Saharan Africa is under increasing pressure to support a rapidly growing population with food and fuel [Melillo et al., 2009; Sanchez, 2002]. The region currently has a very high population growth rate (2.3%), and the U.N. predicts that the current population will almost double by 2050. Most of sub-Saharan Africa has a     160    net soil nutrient depletion as nutrient inputs of fertilizers, manures, biological nitrogen fixation and atmospheric deposition are outweighed by nutrient export through crop harvest, leaching, erosion and other pathways [Sanchez, 2002]. Sub-Saharan Africa annual depletion rates of nitrogen (N) (estimated to average 22kg N/ha) are comparable to the rates of N fertilizer input for countries that have successfully undergone the Green Revolution, such as in the Brazilian Amazon where maize crops receive 24 kg N/ha in mineral fertilizer. The sub-Saharan nutrient deficit severely limits subsistence cereal crop productivity to 1 ton/ha or less [Sanchez, 2002]. Sub-Saharan Africa is beginning a new Green Revolution that calls for getting nitrogen fertilizer into severely depleted soils and managing it well. The methods presented here can be applied to this region to understand past, current and projected land-cover and land-use changes by combining ongoing research in the region on soils (African Soil Information Service—www.africasoils.net) with new remote sensing and modeling efforts . To date, there are no spatially explicit data on regional variation in crop yield increases or associated changes in nutrient cycling with nitrogen fertilizer additions beyond the plot or village level. Remote sensing and ecosystems modeling are powerful tools in scaling these local measurements to regional estimates. Maps produced from remote sensing of crop yield responses to fertilizer will allow others to address national- to local- scale economic strategies of how to use fertilizer most efficiently by understanding where fertilizer does the most good, how local management practices may improve yields, and how to minimize N losses from the landscape. As in Chapter 3, future scenarios for this region can     161    estimate crop productivity for continued and potential management adaptations to minimize the negative effects of expected increases in regional climate variability [Boko et al., 2007]. The methods from Chapters 1 and 2 can be used to address spatial patterns of historical and contemporary crop production from intermediate (landscape) to large scales (national) by linking remote sensing, field data and modeling with a special emphasis on how these patterns are affected by major drivers – climate variability, soil type, landscape position and management regime (e.g., nitrogen fertilizer application). As I show in Chapter 1 and Appendix A, applications of remote sensing tools and data for crop detection show that crop types and management can be successfully separated even at coarse resolutions. For sub-Saharan Africa, a GIS framework incorporating land-cover and land-use change remote sensing products and land-use change drivers could explain factors underlying the spatial-temporal patterns. While there are field-level and aggregated government records on the effectiveness of nitrogen fertilizer additions in many parts of sub-Saharan Africa, there are no spatially explicit (gridded) records of crop yields. The land-cover and land-use change products can then be used to drive the Terrestrial Ecosystems Model, TEM, to assess if agriculture is being conducted in a way such that N is conserved. This time series approach illustrates the interplay of N additions and weather extremes, such as recent droughts. New information collected from the Africa Soil Information Service describes soil profiles (type, texture, nutrients, etc.), represent a vast improvement over previous data sets for the region and can be directly input     162    to TEM. Understanding the historical dynamics of agricultural systems and their drivers are crucial to projecting future scenarios in this region. The next step would be to project future crop production in response to alternative scenarios of climate change and management regimes using TEM, as presented in Chapter 3. There are several unanswered questions that are key to sustainable development in sub-Saharan Africa: 1. How will crop productivity change under increased climate variability? 2. What is the role of “best management” practices in future adaptation? 3. Can long-term support of fertilizer additions, such as through subsidies, increase soil fertility? To examine question 1, crop productivity can be estimated using the improvements made to TEM under the historical and contemporary analysis using a suite of future climate scenarios that predict increased variability in sub-Saharan Africa [Boko et al., 2007]. Insights on future productivity can help government and aid agencies plan for adaptation. As to question 2, TEM can also be applied to estimate future responses to fertilizer inputs, identifying areas that may continue to be most responsive to fertilizer inputs and areas that might be further marginalized. Agricultural management practice scenarios could examine the effectiveness of agricultural adaptations in the form of alternative crops or biofuels for marginal landscapes, while estimating any additional nutrient requirements. Regarding question 3, it has been suggested that the extreme depletion of C and N in soils may not be remedied through fertilization or enhanced crop biomass, as most biomass is removed and even burned [Snapp, 1998]. Using TEM, estimates of the N pool and     163    crop production after 10 years of fertilization may be relevant to government agencies, as it identifies the potential time frame for rehabilitating soils to a point that fertilization applications could be reduced. Further, C and N stocks can be estimated 100 years into the future under a range of scenarios of crop types, fertilization rates and climate variability. Through a quantitative assessment of TEM runs, the biogeochemical impacts of land-cover and land-use change resulting from fertilizer interventions and future scenarios of land use and climate for sub-Saharan Africa can be assessed. This work could suggest which adaptation strategies are effective at which scales, such as, nationally, where to apply fertilizers and, locally, which practices should be used. This information would be useful for decision-makers seeking the greatest pay-off in the transition from the necessary quick fixes of nitrogen fertilization to long-term sustainability. Final Thoughts New sensors and data sets have demonstrated that the land-cover and land- use change community is ready and adept at developing new techniques to transform remotely-sensed data into useful products to answer the most pressing questions in earth system science. There are limitless applications of existing and future data sets, and it is crucial to pursue new analytical techniques to forward our understanding of the earth system. Increased computing power and open access to many data achieves has made long-term phenological studies more feasible, and the power of phenology in discerning land covers, land uses and changes in these     164    systems should be further exploited. Such analyses are particularly useful in ecosystems modeling, where a continuous time series is much preferred over snap- shots of change provided by analysis of just a few images. New methods and applications may present novel innovations in remote sensing but they should also serve to extend beyond the remote sensing community to address issues of concern to other fields of science and to society. The methods developed in Chapter 1 and Appendix A made this entire project possible, and this work exists both as an important characterization of a global frontier of land-cover and land-use change for agricultural uses and as a data set with applications in ecosystems ecology, biology, hydrology, meteorology and sociology. This extension of remote sensing analyses into other fields, particularly in addressing environmental and agricultural sustainability, is one of the most powerful contributions this science can make to society. The capacity for modeling changes in biogeochemistry is greatly enhanced with detailed regional land-cover and land-use change information. Anthropogenic emissions of greenhouse gases is one of the greatest political, social and environmental crises of our time. Modeling presents a way to understand the processes and outcomes from the human environment. Understanding the historical and contemporary patterns of agricultural land-use change emissions is an essential baseline. Scenarios of future emissions allow us to consider the implications of different types of agricultural development and price we will pay for them in greenhouse gas emissions. They also have the potential to inform positive changes in greenhouse gas regulation and soil fertility, such as carbon uptake by natural     165    ecosystems or restoration projects, or changing management practices to increase crop productivity while minimizing negative impacts. By improving our understanding of current biogeochemical responses to agricultural land-use changes, we can project future trends in agricultural productivity, greenhouse gas emissions and soil fertility while considering a range of climate scenarios. References Adams, J. B., et al. (1995), Classification of multispectral images based on fractions of endmembers:  Application to land‐cover change in the Brazilian Amazon, Remote Sensing of Environment, 52, 137‐ 154.  Alves, D. S. (2002), Space‐time dynamics of deforestation in Brazilian Amazonia, International  Journal of Remote Sensing, 23(14), 2903‐2908.  Anderson, L. O., et al. (2005), Assessment of land cover and land use changes in the Brazilian  Amazon using multitemporal fraction images derived from Terra MODIS: Examples from the state of  Mato Grosso, IEEE Transactions on Geosciences and Remtoe Sensing, 2(3), 315‐318.  Avissar, R., and D. Werth (2005), Global hydroclimatological teleconnections resulting from tropical  deforestation, Journal of Hydrometeorology, 6, 134‐145.  Barona, E. (2008), Indentifying the role of crop production in land cover change in Brazil, 1990‐2006,  McGill University, Montreal.  Boko, M., et al. (2007), Africa, in Climate Change 2007: Impacts, adaptation and vulnerability.  Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on  Climate Change, edited by M. L. Parry, et al., pp. 433‐467, Cambridge University Press, Cambridge,  UK.  Boserup, E. (1965), The conditions of agricultural growth: The economics of agrarian change under  population pressure, George Allen and Unwin Limited, London, United Kingdom.  Bradley, B. A., et al. (2007), A curve‐fitting technique to derive inter‐annual phenologies from time  series of noisy satellite data, Remote Sensing of Environment, 106(2), 137‐145.  Bradley, B. A., and J. F. Mustard (2007), Comparison of phenology trends by land cover class: a case  study in the Great Basin, USA, Global Change Biology, 14(2), 334‐346.  Broch, D. L., and R. S. Pedroso (2008), Custo de Produção da Cultura do Milho, edited by F. MS, p. 1,  Fundação MS, Maracaju, MS.  Broch, D. L., and R. S. Pedroso (2009), Milho Safrinha e Culturas de Inverno 2009, edited by F. MS, p.  1, Fundação MS, Maracaju, MS.  Browder, J. O., et al. (2004), Land use patterns in the Brazilian Amazon: Comparative farm‐level  evidence from Rondônia, Human Ecology, 32(2), 197‐224.  Brown, C. J., et al. (2005), Soybean production and conversion of tropical forest in the Brazilian  Amazon: The case of Vilhena, Rondônia, Ambio, 34(6), 462‐469.  Brown, J. C., et al. (2007), Multitemporal, moderate‐spatial‐resolution remote sensing of modern  agricultural production and land modification in the Brazilian Amazon, GIScience & Remote Sensing,  44(2), 117‐148.      166    Buschbacher, R., et al. (1988), Abandoned pastures in eastern Amazonia II: Nutrient stocks in the soil  and vegetation, Journal of Ecology, 76, 682‐699.  Bustamante, M., et al. (2006), Nitrogen cycling in tropical and temperate savannas, Biogeochemistry,  79, 209‐237.  Carvalho, J. L. N., et al. (2009), Carbon sequestration in agricultural soils in the Cerrado region of the  Brazilian Amazon, Soil and Tillage Reserach, 103(2), 342‐349.  Castro, E. A., and J. B. Kauffman (1998), Ecosystem structure in the Brazilian Cerrado: A vegetation  gradient of aboveground biomass, root mass and consumption by fire, Journal of Tropical Ecology,  14(3), 263‐283.  Cerri, C. C., et al. (2004), Carbon cycling and sequestration opportunities in South America: the case  of Brazil, Soil Use and Management, 20, 248‐254.  Cerri, C. E. P., et al. (2007), Simulating SOC changes in 11 land use change chronosequences from the  Brazilian Amazon with the RothC and Century models., Agriculture, Ecosystems & Environment, 122,  46‐57.  Coe, M. T., et al. (2009), The influence of historical and potential future deforestation on the stream  flow of the Amazon River‐‐ Land surface processes and atmospheric feedbacks, Journal of  Hydrology, 369(1‐2), 165‐174.  Conservation International (2008), Cerrado, in Biodiversity hotspots, edited,  www.biodiversityhotspots.org 20 January 2008.  Costa, M. H., et al. (2007), Climate change in Amazonia caused by soybean cropland expansion, as  compared to caused by pastureland expansion, Geophysical Research Letters, 34(L07706).  Crutzen, P. J., et al. (2008), N2O release from agro‐biofueld production negates global warming  reduction by replacing fossil fuels, Atmospheric Chermistry and Physics, 8, 389‐395.  Cruz, J. C., et al. (2005), Resposta de Cultivares de Milho à Adubação Nitrogenada em Cobertura  edited by P. e. A. Ministério de Agricultura, Embrapa Milho e Sorgo, MG, Sete Lagoas, MG, Brazil.  Dale, V. H., et al. (1994), Relating patterns of land‐use change to faunal biodiversity in the Central  Amazon, Conservation Biology, 8(4), 1027‐1036.  DeFries, R., et al. (2008), Fire‐related carbon emissions from land‐use transitions in southern  Amazonia, Geophysical Research Letters, 35.  DeFries, R. S., et al. (1998), Global land cover classifications at 8 km spatial resolution: The use of  training data derived from Landsat imagery in decision tree classifiers, International Journal of  Remote Sensing, 19, 3141‐3168.  Denman, K. L., et al. (2007a), Couplings between changes in the climate system and  biogeochemistry, in Climate Change 2007: The Physical Science Basis. Contribution of Workding  Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, edited  by S. Solomon, et al., Cambridge University Press, Cambridge, United Kingdom and New York, NY,  USA.  Denman, K. L., et al. (2007b), Couplings between changes in the climate system and  biogeochemistry, in Climate Change 2007: The Physical Science Basis. Contribution of Working Group  I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, edited by S.  Solomon, et al., Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.  Edgar, J. (2007), IFB CropWatchers: FarmWeek 11/30/2007, edited.  Embrapa (2008), Cultivares de milho, feijão, girassol, e mandioca para o Agrest Sergipano com foco  na Agricultura Familiar e no Agronegócio, in Documentos / Embrapa Tabulerios Costeiros, edited by  H. W. L. Carvalho, et al., p. 31, Embrapa Tabuleiros Costeiros, Aracaju, SE.  Ewers, R. M., et al. (2009), Do increases in agricultural yield spare land for nature?, Global Change  Biology, 15, 1716‐1726.      167    Fearnside, P. M. (2001), Soybean cultivation as a threat to the environment in Brazil, Environmental  Conservation, 28(1), 23‐28.  Fearnside, P. M., et al. (2009), Biomass and greenhouse‐gas emissions form land‐use change in  Brazil’s Amazonian “arc of deforestation”: The states of Mato Grosso and Rondônia Forest Ecology  and Management, In Press.  Fisher, J. I., et al. (2006), Green leaf phenology at Landsat resolution: Scaling from the field to the  satellite, Remote Sensing of Environment, 100, 265‐279.  Fisher, J. I., and J. F. Mustard (2007), Cross‐scalar satellite phenology from ground, Landsat, and  MODIS data, Remote Sensing of Environment, 109(3), 261‐273.  Foley, J. A., et al. (2005), Global consequences of land use, Science, 309, 570‐574.  Forester, P., et al. (2007), Changes in Atmospheric Constituents and in Radiative Forcing., in Climate  Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment  Report of the Intergovernmental Panel on Climate Change, edited by S. Solomon, et al., Cambridge  University Press, Cambridge, United Kingdom and New York, NY, USA.  Fundação Agrisus (2006), Rally da Safra 2006: Situação do plantio direto e da integração lavoura‐ pecuária no Brasil, Florianópolis.  Galford, G. L., et al. (2008), Wavelet analysis of MODIS time series to detect expansion and  intensification of row‐crop agriculture in Brazil, Remote Sensing of Environment, 112, 576‐587.  Galford, G. L., et al. (In Review‐a), Carbon emissions and uptake from 105 years of land‐cover and  land‐use change at the agricultural frontier of the Brazilian Amazon, Journal of Geophysical  Research.  Galford, G. L., et al. (In Review‐b), Global frontier of land‐use change: The extensification and  intensification of croplands in the Amazon and consequences for greenhouse gas emissions.  Galford, G. L. (unpublished), Field observations, GIS maps and farm records. Collected 2005, 2007.  Garcia‐Montiel, D. C., et al. (2004), Emisions of N2O and CO2 from terra firme forests in Rondonia,  Brazil, Ecological Applications, 14(S4), S214‐S220.  Green, R. E., et al. (2005), Farming and the fate of wild nature, Science, 307(5709), 550‐555.  Gregory, P. J., and J. S. I. Ingram (2000), Global change and food and forest production: future  scientific challenges, Agriculture, Ecosystems & Environment, 82, 3‐14.  Gregory, P. J., et al. (2002), Environmental consequences of alternative practices for intensifying  crop production, Agriculture, Ecosystems & Environment, 88(3), 279‐290.  Gullison, R. E., et al. (2007), Tropical Forests and Climate Policy, Science, 316(5827), 985‐986.  Hansen, M. C., et al. (2008), Humid tropical forest clearing from 2000 to 2005 quantified by using  multitemporal and multiresolution remotely sensed data, PNAS, 105(27), 9439‐9444.  Houghton, R. A., et al. (2000), Annual fluxes of carbon from deforestation and regrowth in the  Brazilian Amazon, Nature, 403, 301‐304.  Houghton, R. A., et al. (2001), The spatial distribution of forest biomass in the Brazilian Amazon: a  comparison of estimates, Global Change Biology, 7, 731‐746.  Huete, A., et al. (2002), Overview of the radiometric and biophysical performance of the MODIS  vegetation indices, Remote Sensing of Environment, 83, 195‐213.  IGBE (2002), Pesquisas Agropecuárias, edited by M. d. Planejamento, Ministério do Planejamento,  Orçamento e Gestão, Rio de Janeiro.  IGBE (2009), Banco de Dados Agregados: Sistem IBGE de Recuperação Automática (SIDRA), edited.  INPE (2008a), Taxa de desmatamento annual: Estimativas Anuais desde 1988 até 2008 in Projecto  PRODES: Monitoramento da floresta Amazônica Brasilera por satélite, edited, Instituto Nacional de  Pesquisas Espaciais.  INPE (2008b), Sistema DETER, edited, Instituto Nacional de Pesquisas Espaciais.  ITT (2008), ENVI Reference Guide, version 4.6, edited, ITT Visual Information Solutions.      168    Jantalia, C. P., et al. (2007), Tillage effect on C stocks of a clayey Oxisol under a soybean‐based crop  rotation in the Brazilian Cerrado region, Soil and Tillage Research, 95(1‐2), 97‐109.  Jasinski, E., et al. (2005), Physical landscape correlates of the expansion of mechanized agriculture in  Mato Grosso, Brazil, Earth Interactions, 9(16), 1‐18.  Jenson, J. R. (2005), Thematic map accuracy assessment, in Introductory Digital Image Processing: A  Remote Sensing Perspectice, edited, pp. 495‐513, Pearson Pretice Hall, Upper Saddle River.  Jepson, W., et al. (2008), Agricultural intensification on Brazil's Amazonian soybean frontier, in Land  Change Science in the Tropics, edited by A. Millington and W. Jepson, pp. 73‐92, Springer  Publications, Boston.  Justice, C. O. (1998), The moderate resolution imaging spectroradiometer (MODIS): Land remote  sensing for global change research, IEEE Transactions on Geosciences and Remtoe Sensing, 36(4),  1228‐1249.  Keller, M., and W. A. Reiners (1994), Soil‐atmosphere exchange of nitrous oxide, nitric oxide, and  methane under secondary succession of pasture to forest in the Atlantic lowlands of Costa Rica,  Global Biogeochemical Cycles, 8(4), 399‐409.  Keys, E., and W. H. McConnell (2005), Global change and the intensification of agriculture in the  tropics, Global Environmental Change, 15, 320‐337.  Klink, C. A., and R. B. Machado (2005), Conservation of the Brazilian Cerrado, Conservation Biology,  19, 707‐713.  Laurance, S. G. W., et al. (2009), Long‐term variation in Amazon forest dynamics, Journal of  Vegetation Science, 20, 323‐333.  Lerner, J., and E. Matthews (1988), Methane emission from animals: a global high resolution data  base, Global Biogeochemical Cycles, 2(2), 139‐156.  Mar, G. D., et al. (2003), Fertilidade do solo e nutricao de plantas: Producao do milho safrinha em  funcao de doses e epocas de aplicacao de nitrogenio, Bragantia, 62(2), 267‐274.  Marland, G., et al. (2008), Global, regional, and national CO2 emissions, Carbon Dioxide Information  Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy, Oak Ridge, Tenn., USA.  Melillo, J. M., et al. (2001), Nitrous oxide emissions from forests and pastures of various ages in the  Brazilian Amazon, Journal of Geophysical Research, 106(D24), 34,179‐134,188.  Melillo, J. M., et al. (2009), Indirect emissions from biofuels: how important?, Science Express.  Mello, F. (2007), Estimativas dos estoques de carbono dos solos nos Estados de Rondônia e Mato  Grosso anteriores à intervenção antrópica, Universidade de São Paulo, Piracicaba, Brazil.  Ministry of Science and Technology (2004), Carbon Dioxide Emissions and Removals from Forest  Conversion and Abandonment of Managed Lands, edited, Brazilian Ministry of Science and  Technology, Brasilia.  Ministry of Science and Technology (2006), Carbon dioxide emissions and removals from forest  conversion and abandonment of managed lands in First Brazilian Inventory of Anthropogenic  Greenhouse Gas Emissions, edited, Brazilian Ministry of Science and Technology, Brasilia.  Miranda, A., et al. (2008), Fluxes of carbon, water and energy over Brazilian cerrado: an analysis  using eddy covariance and stable isotopes, Plant, Cell & Environment, 20(3), 315‐328.  Morton, D., et al. (2009), Cropland expansion in cerrado and transition forest ecosystems:  Quantifying habitat loss from satellite‐based vegetation phenology, in Cerrado Land Use and  Conservation, Advanced Applied Biodiversity Science, edited by C. A. Klink, Conservation  International, Washington, D.C.  Morton, D. C., et al. (2005), Rapid assessment of annual deforestation in the Brazilian Amazon using  MODIS data, Earth Interactions, 9(8), 1‐22.  Morton, D. C., et al. (2006), Cropland expansion changes deforestation dynamics in the southern  Brazilian Amazon, PNAS, 103(39).      169    Morton, D. C., et al. (2008), Agricultural intensification increases deforestation fire activity in  Amazonia, Global Change Biology, 14(10), 2262‐2275.  Myers, N., et al. (2000), Biodiversity hotspots for conservation priorities, Nature, 403.  Neill, C., et al. (1997), Soil Carbon and Nitrogen Stocks Following Forest Clearing for Pasture in the  Southwestern Brazilian Amazon, Ecological Applications, 7(4), 1216‐1225.  Nepstad, D. C., et al. (2006), Globalization of the Amazon soy and beef industries: Opportunities for  conservation, Conservation Biology, 20(6), 1595‐1603.  Nepstad, D. C., et al. (2008), Interactions among Amazon land use, forests and climate: prospects for  a near‐term forest tipping point, Philosopical Transactions of the Royal Society, 363, 1737‐1746.  Nogueira, E. M., et al. (2008), Estimates of forest biomass in the Brazilian Amazon: New allowmetric  equations and adjustments to biomass from wood‐volume inventories, Forest Ecology and  Management, 256, 1853‐1867.  Potter, C., et al. (2009), Terrestrial carbon sinks in the Brazilian Amazon and Cerrado region  predicted from MODIS satellite data and ecosystem modeling, Biogeosciences Discussions, 6, 947‐ 969.  Ramos da Silva, R., et al. (2008), Regional impacts of future land‐cover changes on the Amazon Basin  wet‐season climate, Journal of Climate, 21(1), 1153‐1170.  Rezende, P. J. L. A. (2007), Cultivares, doses de fertilizantes e densidades de semeadura no cultivo  de milho safrinha, UFLA, Lavras.  Roy, S. B., and R. Avissar (2002), Impact of land use/land cover change on regional  hydrometeorology in Amazonia, Journal of Geophysical Research, 107(D20).  Rudel, T. K., et al. (2005), Forest transitions: towards an understanding of global land use change,  Global Environmental Change, 14(4), 23‐31.  Saatchi, S. S., et al. (2007), Distribution of aboveground live biomass in the Amazon basin, Global  Change Biology, 13, 816‐837.  Saatchi, S. S., et al. (2009), LBA‐ECO LC‐15 Amazon Basin Aboveground Live Biomass Distribution  Map: 1990‐2000, edited, Oak Ridge National Laboratory Distributed Active Archive Center.  Sanchez, P. A., et al. (1982), Amazon Basin Soils: Management for Continuous Crop Production,  Science, 216, 821‐827.  Sanchez, P. A. (2002), Soil fertility and hunger in Africa, Science, 295(5562), 2019‐2020.  Scanlon, T. M., and G. Kiely (2003), Ecosystem‐scale measurements of nitrous oxide fluxes for  intensely grazed, fertilized grassland, Geophysical Research Letters, 30(16), 1852‐1856.  Schimel, D. (2007), Carbon cycle conundrums, PNAS, 104(47), 18353‐18354.  Shukla, J., et al. (1990), Amazon Deforestation and Climate Change, Science, 247(4948), 1322‐1325.  Skole, D., and C. Tucker (1993), Tropical deforestation and habitat fragmentation in the Amazon:  Satellite data from 1978 to 1988, Science, 260(5116), 1905‐1910.  Snapp, S. (1998), Soil nutrient status of smallholder farms in Malawi, Communications in Soil Science  and Plant Analysis, 29(17&18), 2571‐2588.  Soares‐Filho, B. S., et al. (2006), Modeling conservation in the Amazon basin, Nature, 440, 520‐523.  Souza, E. F. C., and R. P. Sorrato (2006), Efeito de fonts e doses de nitrogênio em cobertura, no milho  safrinha, em plantio direto, Revista Brasileira de Milho e Sorgo, 5(3), 395‐405.  Steudler, P. A., et al. (1996), Consequence of forest‐to‐pasture conversion on CH4 fluxes in the  Brazilian Amazon Basin, Journal of Geophysical Research, 101(D13), 18,547‐518,554.  Stickler, C. M. (unpublished), Field validation data for the Xingu Basin. Collected 2006.  Turner II, B. L., and A. M. A. Ali (1996), Induced intensification: Agricultural change in Bangladesh  with implications for Mathus and Boserup, PNAS, 93, 14984‐14991.  Van der werf, G., et al. (2003), Carbon emissions from fires in tropical and subtropical ecosystems,  Global Change Biology, 9(4), 547‐562.      170    Van der werf, G., et al. (2006), Interannual variability of global biomass burning emissions from 1997  to 2004, Atmospheric Chemistry and Physics Discussions, 6, 3175‐3226.  Wuebbles, D. J., and K. Hayhoe (2002), Atmospheric methane and global change, Earth‐Science  Reviews, 57, 177‐210.  Xu, X., et al. (2008), Convergence in the relationship of CO2 and N2O exchanges between soil and  atmospher within terrestrial ecosystems, Global Change Biology, 14, 1651‐1660.            APPENDIX A Wavelet analysis of MODIS time series to detect expansion and intensification of row-crop agriculture in Brazil Gillian L. Galford a,b, John F. Mustard a, Jerry Melillo b, Aline Gendrin a,c, Carlos C. Cerri d, Carlos E.P. Cerri e a) Geological Sciences, Brown University, United States b) The Ecosystems Center, MBL, United States c) Institut d'Astrophysique Spatiale, Orsay, France d) Centro de Energia Nuclear na Agricultura, Universidade de São Paulo, Brazil e) Escola Superior de Agricultura Luiz de Queiroz, Universidade de São Paulo, Brazil Remote Sensing of Environment 112 (2008) 576-587 171  172    0. Abstract Since 2000, the southwestern Brazilian Amazon has undergone a rapid transformation from natural vegetation and pasture to row-crop agriculture with the potential to affect regional biogeochemistry. The goals of this research are to assess wavelet algorithms applied to MODIS time series to determine expansion of row-crops and intensification of the number of crops grown. MODIS provides data from February 2000 to present, a period of agricultural extensification and intensification in the southwestern Brazilian Amazon. We have selected a study area near Comodor, Mato Grosso because of the rapid growth of row-crop agriculture and availability of ground truth data of agricultural land-use history. We used a 90% power wavelet transform to create a wavelet-smoothed time series for five years of MODIS EVI data. From this wavelet-smoothed time series we determine characteristic phenology of single and double crops. We estimate that over 3,200 km2 were converted from native vegetation and pasture to row- crop agriculture from 2000 to 2005 in our study area encompassing 40,000 km2. We observe an increase of 2,000 km2 of agricultural intensification, where areas of single crops were converted to double crops during the period. Keywords: MODIS; EVI; Time-series analysis; Wavelet analysis; Land use and land cover change; croplands and crop detection; row-crop agriculture; Land use in the Brazilian Amazon     173    1. Introduction The southwestern Brazilian Amazon is one of the world’s fastest growing agricultural frontiers. Historically, the clearing of forest and savanna ecosystems to create cattle pastures has been the primary land transformation (Skole and Tucker 1993). This land-use pattern has recently changed. Today, pastures and areas of natural vegetation are being converted to large-scale croplands to grow cash crops, row crops including soybean, maize, and dry-land rice (Instituto Brasileiro de Geografia e Estatistica 2006; Morton et al. 2006). From 1990 to 2000, soybean cover in the southwestern Brazilian Amazon doubled while production has nearly quadrupled due to farm mechanization (CONAB 2004). A major frontier of row crops is in the state of Mato Grosso, home of some of the largest contiguous row-crop plantations in the world. Here, the area planted in soybean has increased on average 19.4% annually since 1999. By 2004 over 5 million hectares, or about 6% of Mato Grosso, was soybean plantations (CONAB 2004). Regional shifts in land-cover and land-use have numerous consequences relevant to both environment and agriculture, including changes in carbon and nitrogen storage, trace gas emissions, quality of surface water and biodiversity (Melillo et al. 1996, 2001; Steudler et al. 1996; Luizão et al. 1989; Neill et al. 1997, 2001; Myers et al. 2000). Determining the physical and temporal patterns of agricultural extensitificaiton, or expansion, and intensification is the first step in     174    understanding their implications, for example, long-term crop production, and environmental, agricultural and economic sustainability.   Since the early 1970s, remote sensing studies have tracked the land- cover and land-use changes in the Brazilian Amazon, initially using Landsat data to identify areas of deforestation (Skole and Tucker 1993). The iconic images of development in the state of Rondonia, off of highway BR364, showed the dramatic impact of deforestation. Recently, the Brazilian Instituto Nacional de Pesquisas Espaciais (INPE; National Institute for Space Research) has used Landsat sensors for monitoring deforestation and detecting fires for the purpose of enforcing environmental regulations (Instituto Nacional de Pesquisas Espaciais 2006). As conversion from natural vegetation and pasture to row crops has become increasingly widespread, the focus has shifted to documentation of the land cover changes in type localities. For example, Brown et al. (2005) illustrated the large-scale development of soybean agriculture with temporal “snap-shots” of Vilhena, Rondônia with Landsat data in 1996 and 2001. Morton et al. (2006) document wide-spread, regional changes in land cover. This new focus on cropland detection is particularly important due to the large spatial scale of individual farms (i.e. a single farm of row-crop agriculture typically occupies more than 2,000 hectares) and regional agricultural intensification (Mueller 2003). Remotely-sensed green leaf phenology is one metric for distinguishing the type of land- cover and land- use change and is suitable for agricultural     175    applications. Croplands present a more complex phenology than natural land cover due to their many peaks resulting from multiple crops planted sequentially within a growing season. Additionally, the uniform cover of green leaves in an agricultural field creates very high observed greenness, especially as compared to the bare soils left after harvest. This large dynamic range of cropland green vegetation through time depends highly on natural factors (e.g. magnitude and temporal variability of precipitation in the region), as well as, management decisions (e.g. time of planting, crop variety). Phenology studies often utilize a curve-fitting algorithm for the observed data sets. A curve-fit simplifies parameterization necessary for identification of metrics such as start of season. Previous studies (e.g. Bradley et al. 2007; Zhang et al. 2003) have identified land cover based on specific properties of the observed green leaf phenology, such as start of season, dry season minimums, and amplitude of maximums. The simplest method for creating a smoothed time series is to use a multi-point smoothing function which may not remove high frequency noise. Other curve fitting methods, (e.g. Bradley et al. 2007) rely on a harmonic curve fit to the annual average phenology in order to characterize inter- annual variability of a time series. A sigmod curve fitting algorithm can be applied to a time series of a single year (Fisher et al. 2006; Zhang et al. 2003) but utilizes a priori knowledge of the system’s seasonality in order to detect the phenological peaks (i.e. the algorithm must be informed with expectation for when to find the phenological peak) (Fisher et al. 2006; Jönsson and Eklundh 2004; Zhang et al. 2003).     176    All of these methods have proven powerful in the systems they have been tested in but will fail in the case of row-crop agriculture in Mato Grosso for three reasons: 1) they fail to remove high frequency noise caused by the long rainy season. A multi-point smoothing function maintains sensitivity to high frequency noise observed during the rainy season which poses problems when trying to detect the maximums defining the cropping system. Using such a smoothing procedure would require a finely-tuned crop detection algorithm that would have to be adjusted for the strong precipitation gradients in the region. 2) They cannot capture the inherent variability in the system. Using an average annual phenology to identify land cover (Bradley et al. 2007) does not work because it does not examine each year separately, a problem since observed annual phenology is not a function of the previous year. In this human environment, the change in phenology from year to year (e.g. when converting from natural vegetation to cropland or intensifying from single crops to double crops) can be tremendous, rendering the average annual phenology of the time series meaningless. The traditional Fourier transform expects periodicity whereas the change in crop behavior from single to double cropping systems in addition to management coupled with climate makes the time series signal non-stationary, which is better handled by the wavelet transform (Sakamoto et al. 2005). 3) The stochastic nature of rainfall and the influence of human management affect the timing and spatial patterns of phenology peaks that make it difficult to precisely predict the timing. This system fails the criteria of a priori knowledge regarding     177    the timing of phenology peaks necessary for some curve-fitting algorithms, such as the sigmod (Fisher et al. 2006, Zhang et al. 2003). We look to the wavelet-based curve-fitting methodology (Wavelet based Filter for determining Crop Phenology, WFCP) presented by Sakamoto et al. (2005) in order to remove high frequency noise while remaining sensitive to annual changes in phenology. This is a necessary step in accepting WFCP as a generalized methodology. The true utility of such a method comes in being able to apply it to various study areas without sacrificing performance or requiring many changes. Our case study presents a robust test for the wavelet methodology-- an area with high variability in phenology patterns with additional noise cause by the tropical rainy season. We are interested in the application of the WFCP in the southwestern Brazilian Amazon because of its curve-fitting capabilities for cropland phenology, as well as, the potential for it to be rapid and highly automated. We implement a wavelet transform for time-series analysis to study these highly dynamic systems. Wavelet analysis provides an efficient method for extracting relevant information from large data sets such as hyperspectral image cubes, sea surface temperature, vegetated land-cover and seismological signals (e.g., Gendrin et al. 2006; Torrence and Compo 1998; Percival et al. 2004; Mallat 1998; Li & Kafatos 2000; Sakamoto et al. 2005, 2006). In agricultural applications, a wavelet-smoothed time series can be used to identify the start of growing season and the time of harvest with low error (11     178    to 14 days, respectively; Sakamoto et al. 2005). Wavelet analysis is capable of handling the range of agricultural patterns that occur through time as well as the spatial heterogeneity of fields that result from precipitation and management decisions because the transform is localized in time and frequency. Using a wavelet analysis for a study area in the Amazon is highly desirable because it removes the high frequency noise caused by the frequent cloud-cover in a highly automated way. MODIS data products offer a great opportunity for phenology-based land- cover and land-use change studies by combining characteristics of AVHRR and Landsat, including: moderate spatial resolution, frequent observations, enhanced spectral resolution and improved atmospheric calibration (Justice et al. 1998; Zhang et al. 2003). MODIS data products have provided global land-cover mapping annually to document land-cover change over time (Hansen et al. 2002; Friedl et al. 2002). These data sets are informative at the global level, but lack relevant regional-scale details about land-cover and land-use classes and change. Recent regional-scale applications of MODIS data to cropland land-use include spectral unmixing to time series to detect subpixel land-cover in croplands (Lobell and Asner 2004). Wardlow et al. (In Press) demonstrate that MODIS vegetation indices in time series are statistically sufficient for distinguishing crop types across a broad region, such as the state of Kansas. In the Amazon, Anderson et al. (2005) have utilized MODIS data sets to document broad changes in land cover and land use.     179    Understanding the degree of extensification and intensification in croplands from remote sensing provides insight into the direction and magnitude of impacts on natural and agricultural environments. In the industrial-scale croplands that are beginning to dominate portions of the Amazon Basin, patterns of cropland extensification and intensification have biogeochemical consequences that affect the natural and cropland sustainability, including soil fertility for decades to come. Our objective is two-fold: to understand 1) the massive transition from natural vegetation and pasture to large-scale row crops and 2) the intensification of cropping systems within existing croplands using MODIS data sets. The purpose of this study is to detect cropping patterns for the cerrado region using a wavelet-smoothed time series. This study evaluates wavelet tools, as presented by Sakamoto et al. (2005), in a new environment and tests the limits of the wavelet model while modeling crop phenologies from MODIS time series data during the 2001-2006 growing seasons. We provide analysis and discussion of the effectiveness of wavelet analysis for the detection of single and double cropping systems and hereby demonstrate the utility of wavelet analysis on time series data with application to a land-use and land- cover change case study in the southwestern Brazilian Amazon. 2. Methods and Approach An overview of the methodology used here is presented in Figure A.1. The general methodology is based on the “Wavelet based Filter for Crop Phenology”     180    (WFCP, Sakamoto et al. 2005). There are four main steps: 1) Data processing, 2) Identification of land use, 3) Field verification and 4) Error Analysis. 2.1. Study Area We selected a region of rapid change in croplands in the state of Mato Grosso for this study (Upper left corner: 12° 15’ 23.61’’ S, 59° 45’ 18.23’’ W; Lower right corner: 13° 59’ 27.86’’ S, 57°57’ 30.8’’ W; Figures A.2, A.3;). The area is 40,100 square kilometers, has annual rainfall from 1,800 to 2,200 millimeters, and a dry season from July-September and rainy (growing) season from November to April. The soils are entisols with 15-25% clay. Dominant native vegetation types range from cerradão (woody savanna) and cerrado (open savanna), referred to from here forward as “cerrado”. In this region, land-use transitions have two major pathways to row crops from cerrado: natural vegetation to pasture to row crops; and natural vegetation directly to row crops. Row crops are subject to a variety of management regimes - types and sequences of crops; types, timing and amounts of fertilizer and other chemicals; and tillage versus no-tillage. 2.2. Field Data Field work conducted in July 2005 provided ground-control points on a fazenda of 41 square kilometers for which a detailed agricultural history has been kept through this study period. Using a hand-held GPS we mapped three management units with different agricultural histories. The histories include     181    information on the type of land cover or land use before row-crop agriculture, the timing of conversion to row-crop agriculture and the cropping patterns used for each year land use was row-crops. These histories were provided to us from records kept by the farm manager. Most of the native vegetation was converted in 2002 or 2003; one unit was previously pasture until conversion in 2002. The crops are generally single crops for the first two to three growing years and then change to double crops. 2.3. Creating a time series Remotely sensed data create a detailed classification of croplands by detecting important characteristics (parameters) of land-cover and land-use change from a smoothed Vegetation Index (VI) time series. For this work, we used MOD09 (V004) 8-day, 500 meter surface reflectance composites data (Figure A.3). The study site was subset from the larger MODIS scene (h12v10). We derived Enhanced Vegetation Index (EVI) products using the standard formulation (Huete et al. 2002): EVI was chosen because it has a greater dynamic range than the more commonly used NDVI and thus is better suited to capture the dynamic crop phenology in this region without reaching saturation (Huete et al. 2002). Combining the EVI images gives us an EVI time series for each pixel, although the time steps are not equally spaced. We used the date of observation flag included in the data product as the day of the year for each observation to create an unevenly-spaced time series for a given pixel. Using the date of observation flags more accurately defines the timing and magnitude of     182    green peaks. The 8-day product without the date of observation flags assumes evenly spaced observations when, in fact, observations can be up to 16 days apart or as few as 2 days apart. The accuracy of the shape of the input data affects the detection of cropping systems. Assuming evenly spaced data from with the original aggregated MODIS data (as with the 8-day product without the observation flags) misrepresents the data. 2.4. Data Processing Data processing for noisy and contaminated pixels consisted of 2 steps: 1) detecting of cloud-contaminated and extremely noisy pixels and 2) replacing bad data points through linear interpolation. Data processing treated each pixel as a one-dimension time series. For each time step, a point in the time series was identified as cloud-contaminated when band 3 (459 – 479 nanometers) reflectance values exceeded 10% (Sakamoto et al. 2006) and were subsequently removed. Extremely noisy data, generally caused by minor cloud contamination, were identified if they exceeded a 0.15 change threshold in EVI from the value at the previous time-step and were also removed. We replaced the missing values through linear interpolation from observed data points. Since the observation dates varied by pixel and were unevenly spaced, we produced a daily time-step EVI time series to avoid an aliasing effect when creating a wavelet-smoothed time series. Then we resampled the daily interpolated data set to 7 day intervals to reduce the size of the data set and the processing time during further analysis.     183    We first separate croplands from other land covers. Lands managed in croplands display a distinctly higher annual standard deviation compared to natural vegetation due to high vegetation density during the growing season and extremely low vegetation density following harvest. Woody cerrado land cover has very little phenological variation and maintains a mean value of 0.6 EVI, rarely exceeding 0.8 EVI (Figure A.4). The open cerrado phenology shows some seasonal variation as it decreases in greenness through the dry season and increases during the rainy season, with an annual mean of 0.25 EVI and a maximum around 0.6 EVI (Figure A.4). Both woody and open cerrado remain above 0.2 EVI through the year. Pasture phenology is similar to the cerrado with a slightly larger dynamic range and a mean of 0.5 EVI (Figure A.4). Croplands have maximum EVI values exceeding 0.8 and EVI minimums often reaching 0.1 or lower (Figure A.4). Because these differences in amplitude of seasonal phenology in EVI we can use the standard deviation of an annual EVI time series to distinguish lands in croplands from native vegetation (Figure A.5). There is a bimodal distribution in the histogram of standard deviations for all pixels in the study area, separating croplands on the right tail (Figure A.5). There is a biomodal distribution of the standard deviation for each pixel. For each year, we used the standard deviation value that separated the two modes (the histogram minimum) as the detection point for croplands—all pixels with a standard deviation higher than this point were classes as croplands. The value of the detection point, or histogram     184    minimum between modes, is included in Figure A.5 for each year of analysis. The mean detection point for all years was at the standard deviation of 0.149. 2.5. Creating a wavelet-smoothed EVI time series Here we use a discrete wavelet transform. A wavelet function φ (t ) is an oscillating function with a finite energy and null mean: +∞ Equation 1 ∫ −∞ φ (t )dt = 0 The wavelet transform W(a,b) is defined by Equation 2: 1 ⎛t −b⎞ Equation 2 W ( a, b)i = a ∫ φ *⎜ ⎝ a ⎠ ⎟s(t )dt where s(t) is the analyzed input signal and φ * is a mother wavelet, or a wavelet basis function. A number of different mother wavelets exist, including Daubechies, Derivative of a Gaussian (DOG) and Coiflet (Torrence and Compo 1998). In this equation, the wavelet width is determined by the scaling parameter a while ifts center is determined by the parameter b. The variable t represents the time-step in the one dimensional time series over which the integration is performed. The wavelet transform has the advantage of retaining information related to the width (scale) and the location (time) of the features present in s(t). This formulation (Equation 3) can be used to reconstruct a signal (Gendrin et al. 2006)     185    The wavelet-created time series W is a summation of wavelets over a number of different widths, x Equation 3 W = ∑W ( a, b)i i =1 where W (a, b)i is the wavelet transform created in Equation 1. Wavelet transforms of decreasing width are summed from i to x, where x is the number wavelet transforms necessary to achieve the user defined number of coefficients retained from the input data. The width of a wavelet transform has half the width of the previous wavelet. It is the sum of the wavelets (W in Equation 3) that is referred from here on as the wavelet-smoothed time series. The wavelet filtering begins by applying a smoothing function on the one- dimensional time series that is evenly-spaced, to avoid aliasing effects, after cloud-removal and interpolation of missing values. First, a discrete wavelet transform removes the residual high frequency noise. The smoothed EVI time series is then reconstructed with an inverse discrete wavelet transform. Applying the wavelet to an EVI time series requires selecting parameters of mother wavelet, order, and power that define the wavelet behavior. We used the Coiflet mother wavelet with order 4 because the wavelet shape is as similar as possible to the peaks in agricultural phenology we are detecting. (See Sakamoto et al. 2005 for performance comparison of mother wavelets). Order is a measure of the wavelet’s smoothness, where a higher order produces a smoother wavelet (Burke et al. 1994).     186    The wavelet requires a power threshold that corresponds to the number of coefficients determining how much of the input EVI time series is retained during the wavelet transform. A higher power or a greater number of coefficients retains more of the original data by forming a narrower wavelet that includes more fine- scale features but may also retain more noise. A lower power, or fewer coefficients, retains less high frequency data by applying a wider wavelet. A low power wavelet may capture trends through the entire time series but may lose phenological detail during a single year. We conducted error analysis cropping patterns detected with the 70%, 80%, 85%, 90% (both 0.3 and 0.4 EVI detection thresholds; see Section 2.6 for further discussion of this threshold) and 95% (0.4 threshold) power wavelets. A random point generator selected 122 verification points within the row-crop agricultural zone. Reference data on cropping patterns for a given year in a given pixel were generated from the input EVI time series with bad pixels removed. These reference data are used to verify the cropping patterns detected from each wavelet-smoothed time series and the results tabulated to calculate overall accuracy and Khat. Each point has five years worth of data and each year was treated individually, essentially multiplying the number of verification points by the number of years, giving a total of 610 verification points. This error analysis shows high overall accuracies and Khat values for all wavelet powers except the 95% power wavelet (Table A.1). From these results we conclude the 90% power wavelet-smoothed time series best captures     187    cropping patterns and will be the focus of our further analysis. We employed the 90% power wavelet (27 coefficients) which minimized RMS error and had the lowest omission and commission errors in detecting cropping patterns. Qualitatively, we observed thousands of pixels where the strong overall performance of the 90% wavelet was apparent when compared to the other wavelet powers. Figure A.6 provides one such example. In areas of single crops, the 90% power wavelet captures the overall data trend well, getting closest to the high observed EVI values and the low values while reducing false detections of peaks. The agricultural system gets more complex where there are double crops. Resonance in the wavelet may create false peaks but they do not go above our threshold of 0.4 EVI. Throughout the time series, RMS error is low, on the order of magnitude of 0.1 for the 90% power wavelet-smoothed time series. The power threshold is a variant on the method of Sakamoto et al. (2005) where multiple frequency thresholds defined the length of plausible growing seasons to remove noise with the wavelet. We chose to use the power variable to remove high frequency noise as it required no assumption of length of growing season. This gives more flexibility in fitting both very narrow peaks, as is often the case in both wide peaks found in single crops and very narrow peaks found in double cropping systems. Application of the wavelet filter can create distortion around the edges of the EVI time series (Sakamoto et al. 2005). To avoid this problem, we augmented the input EVI time series to allow for spin-up, or conditioning of the     188    wavelet. Conditioning the wavelet is a necessary step shown by Sakamoto et al. (2005). For our application, we augmented the one-dimensional EVI time series by replicating the first and last year worth of data ten times at the beginning and end, respectively, of the time series. After applying the wavelet transform, the extra years of data were removed from the wavelet-smoothed time series. 2.6. Phenology and Land Cover/Land Use The cropping patterns, or the numbers of crops grown each year, were detected from the wavelet-smoothed EVI time series. Each crop is characterized by one maximum in the wavelet-smoothed EVI time series. To determine if a cropland pixel had a single or double cropping pattern, we detected the number of local maximums in one growing year of the wavelet-smoothed EVI time series. We defined a local maximum as having a higher EVI than the two points before and two points after that point. The wavelet-smoothed EVI time series divided into five growing years from August through July for 2000-2005 and are identified by their harvest year: 2001, 2002, 2003, 2004, and 2005. Wavelets are very sensitive to small maximums in portions of the time series where EVI range is low. The wavelet response to these small local maximums slightly amplifies small real peaks, thereby creating false peaks in phenology. From farm histories, we know that the EVI for crops generally exceeds 0.4. We removed false detections by using a threshold of 0.4 EVI for the 90% power wavelet to minimize false detections to remove these minor false peaks from being detected as phenological peaks of cropland. In some areas we     189    detect two or three real phenological maximums in the EVI time series. In such cases, the first maximum is minor and is likely caused by the early green-up of volunteer crops, weeds or other green cover at the beginning of the rainy season before crops are planted. The second and third maximums, where present, correspond to single and double crops. The early weedy growth or volunteer crops may be distinguished from crops as they do not exceed the 0.4 EVI threshold that prevents us from detecting early green-ups as crops. Utilizing this threshold works with the assumption that every area of row crops has a strong crop phenology (i.e. peaks above 0.4 EVI) but it may exclude very small, real, phenological maximums such as the case of a failing crop that did not exceed the threshold. With these detection criteria, we can identify the cropping patterns and change in cropping patterns that characterize the intensification and extensification of cropland, as first shown by Sakamoto et al. (2006). For verification of the cropping patterns, we used the observed EVI time series with bad pixels removed. A random point generator selected 122 verification points within the cropland areas. For each point, there are five growing seasons (August-July) in the time series. Each growing season was treated individually, essentially multiplying the number of verification points to give a total of 610 verification points. We compared cropping pattern detected from the 90% wavelet-smoothed time series to the original data and, for each point, identified and tabulated misclassifications.     190    Further, comparison of our wavelet-smoothed time series to agricultural history for Fazenda St. Lordes allows us to assess the detection of cropping patterns. There are multiple different land-use histories corresponding to the management units within the fazenda. The agricultural history collected from farm records during a field visit in July 2005 includes the time of conversion as well as sequence and cropping patterns in subsequent years for each management units. We examine how well the process of cleaning the EVI time series and performing the wavelet-transform retains the character of the processes occurring on the ground by comparing the detected cropping patterns to the farm records. 2.7. Error Analysis We performed statistical analysis of the goodness of the curve fitting through residuals and Root Mean Square (RMS) error. We calculated the RMS error of the residual (difference between the raw EVI times series and the wavelet-smoothed time series). The RMS error gives a sense of the magnitude of error with the curve fit. We analyzed error in the land-cover and land-use classes by comparing cropping patterns detected from the wavelet-smoothed time series to observed cropping patterns in the input data. By compiling an error matrix for the classes we could calculate overall accuracy, producer’s accuracy and user’s accuracy as well as a Khat value from KAPPA analysis (see Jensen 1996).     191    To perform the statistical error assessments, we divided our data into four classes. The classes, based on data values, are: not cropland, single cropping system (one maximum), double cropping system (two maximums) and unclassified (more than two maximums detected). As a given pixel may change classes from one year to the next, we considered a pixel’s class for one growing year a test point. We calculated omission and commission errors using two different sets of reference data, the raw EVI time series and spatially and temporally explicit farm history data. Overall accuracy is the total number of test points correctly classed by the total number of test pixels used. Omission error or producer’s accuracy is the total number of correct pixels in a remotely-sensed class divided by the total number of pixels in that class from the reference data. Commission error, or user’s accuracy, is the total number of correct pixels in a remotely-sensed class divided by the total number of pixels in that remotely- sensed class. The Khat statistic comes from KAPPA analysis for discrete multivariate accuracy assessment. Khat incorporates information from the misclassifications recorded in the error matrix and gives a slightly different accuracy assessment than overall accuracy does. Khat would equal zero if the classifications results were completely random. (Jensen 1996) For the raw EVI reference test case, we calculated overall accuracy by comparing the user-detected reference classes to the automated detection of classes from the wavelet-smoothed time series. We used one hundred test points per class. The test points were evenly distributed by year (20 test pixels per year per class) for the not cropland and unclassified classes since there was     192    negligible change in the size of these classes over time. We weighted test points for single and double crop classes by the relative abundance in that year. We randomly located the test points for each year and class using a random sample. We then compared the wavelet-smoothed time series crop detection results to the raw EVI input data (reference test information) that was subjected to the same criteria to create four classes. For the random test points across the entire scene, we use omission and commission errors to understand our accuracy within the classes as well as overall accuracy and the Khat value. We also used reference data (known cropping patterns) for Fazenda Santa Lordes to calculate omission and commission errors over the farm. We verify our results to an independent data source. One limitation to this method is the size of the fazenda. While this is a large fazenda, occupying 25 km2, we are limited to a relatively small sample size (100 pixels) for statistical analysis. Extensification and intensification are two measures of the extent of row crop agriculture. Extensification, or the increase in total row-crop agricultural area, is measured as the annual increase from one growing season to the next. Each year the area of extensification is calculated as the areas detected as row crops that were not previously detected. Intensification of row crops describes the change from a single to double cropping pattern from one year to the next. The concepts and metrics of extensification and intensification allow us to explain the patterns of agricultural development. 3. Results     193    3.1. Agricultural extensification Cropland in the study area increased from 6,255 square kilometers in the 2001 growing season to 9,535 square kilometers in the 2005 growing season, as calculated from cropland detection based on annual standard deviation (Table A.2). This represents a 34% increase in row-crop agriculture to cover a total area of 24% of the study area. Increases in land cover of row crops were largely at the edges of existing croplands (Figure A.8). 3.2. Detection of cropping patterns Statistical analysis of detection errors shows overall good results from the wavelet-smoothed time series. The error matrix shows that majority of pixels considered unclassified (more than two crops detected) were actually false- detections of double crops (Table A.3). For the purpose of tabulating single and double cropping patterns, we considered double crops to be all pixels with two or more maximums, although this may introduce more error. From the error matrix results we can asses our accuracy (Table A.4.a). The omission and commission errors (Table A.4.b) are derived from the error matrix for each classification (Not row crops, ‘Not RC’, includes native vegetation and pastures; Single cropping patterns; and Double cropping patterns). All omission errors are less than 10%, except for Single crops, which have a producer’s accuracy of 77.0%. User’s accuracy is above 84.0% for all classes analyzed (Table A.4.a). The wavelet- smoothed time series gives us an overall accuracy of 88.5% for the entire study region and the corresponding Khat value is 92.1% (Table A.5).     194    The wavelet transform captures land cover and land use at a site on Fazenda Santa Lordes where cerrado was converted to row crops in 2003, with the first crop grown in the 2003-2004 season (Figure A.7). From farm history, we know the first year of cropping was a single soybean crop in 2003-2004 and a double crop in 2004-2005. Statistical error analysis for wavelet-derived classes compared with land-use records for Fazenda Santa Lordes shows high accuracy (Table A.6). Producer’s accuracy is high (low omission errors) the classifications of not row crops (100.0 %) and single crops (81.8%) but low (higher omission errors) for double crops (71.4%). User’s accuracy is high (low commission error) for all classes: not row crops is 94.7%; single crop accuracy is 90.0% and double crops accuracy is 100.0% (greater than 99.99%). The commission and omission errors may not be representative of the entire scene as there were only 5 pixels in this category. The Khat value is 85.7% and the overall accuracy is 94% for land-use classifications on the fazenda. Single and double cropping patterns show a dynamic relationship (Figure A.9). Both cropping patterns have a net increase (Table A.7). Single crops increase from 3,124 to 3,643 km2 (a 14% increase) and double crops increase from 2,283 to 4,443 km2 (a 49% increase) during the study period (Figure A.9). We see a decrease in single crops between the 2003 and 2004 growing years while double crops continue to increase. The increase in single crops from 2002 to 2003 may represent the extensification of croplands into areas that were previously native vegetation. After the first growing year with a single crop many of these areas may have been converted to a double crop, creating a dynamic     195    relationship between the cropping patterns. Agricultural intensification, or increase in the number of crops grown per area, follows a pattern similar to agricultural extensification, radiating outwards from the older row-crop agricultural areas to the periphery with time (Figure A.8). 4. Discussion A major frontier croplands is found in the state of Mato Grosso, Brazil where natural ecosystems of Amazon rainforest and cerrado (savanna) are giving way to some of the largest contiguous row-crop plantations in the world. Agricultural census provides detailed information on cultivated area and crop yields, but is not spatially explicit beyond the level of municipio nor does it tell us the sequence and duration of crops. Census data is compiled annually, but does not elucidate the number of crops grown per parcel, or other land-use practices that impact local biogeochemistry. Remote sensing tools allow us to detect agricultural phenology and derive parameters from which we can construct timeline of land-use and land-management on a per pixel basis. These results have many possible future applications such as for understanding changes in carbon and nitrogen cycling, important from both agricultural and environmental sustainability perspectives, or for estimating crop production rates for economic analysis. Our application of the wavelet transform to an EVI time series (WFCP) captures crop phenological behavior with low error. This was a test of the WFCP methodology under new environmental conditions, namely climate which creates     196    high frequency noise from clouds, and different agricultural phenology than the areas the model was created and validated (Sakamoto et al. 2005, 2006). Occasionally, the wavelet transform exaggerates small phenology peaks from weedy growth or creates a false peak due to wavelet resonance. Small peaks such as these should be classified with double crops. Although the classification of false maximums or weedy growth as double crops is troublesome from a curve-fitting perspective, it is acceptable for the purposes of detecting cropping patterns. Using the 90% power wavelet, many of these false maximums actually fall below the detection threshold for maximums of 0.4 EVI and are not counted, contributing to our high overall accuracy. We have observed distinct spatial patterns in the wavelet-smoothed time series results. New areas of croplands are nucleated around existing areas of croplands (Figure A.8). Small areas of cerrado between large areas of croplands are often filled in with croplands. We observe areas of single crops becoming areas of double crops over time, as seen in Figure A.8. This intensification appears spatially constrained to the center of the zones of agriculture and likely reflects the evolution of farming practices with cropland age. Our knowledge of agricultural practices supports this -- often, in the first years of cultivation, a single soybean crop is grown. After two to three years, double cropping practices emerge, where there may be two soybean crops grown, soybean and secondary cash crop such as corn, or soybean and a soil-conditioning crop such as millet. Increases in crops in the 2002-2004 time period may be related to economic factors, such as the high global market price (Morton et al. 2006). These results     197    show an increase of crops grown per area across the entire study area, either as the development of new croplands or as the intensification of existing lands from a single crop system to a double crop system. 5. Conclusions The goal of this study was to apply time series analysis to detect rapid changes in land-cover and land-use choices (single and double cropping patterns). The challenge of this study was to detect crop patterns within croplands. We have tested the wavelet transform to filter noisy EVI time series data. First, we detect the areas of row crops by applying an annual standard deviation threshold to discriminate row crops from other land-cover types. This threshold was selected annually from the local minimum in a bimodal histogram of standard deviation. Identifying areas of row crops on a year-to-year basis allows us to analyze cropping patterns for a pixel only after it has been converted to row crops, thereby reducing processing time. After selecting areas in row crops, we then created wavelet-smoothed time series with the 90% power wavelet. Local maximums, or phenological peaks, were counted as a crop if the time series exceeded 0.4 EVI; this threshold was a means of removing false peaks sometimes created in the wavelet-smoothed time series. We selected this study area to test this model in an area of rapid development of row crops. During the five-year study period, we found an increase in croplands of 3,281 square kilometers, an area larger than the state of Rhode Island. Intensification of row crops is also evident, with increases in row     198    crops coming first in single crops and, subsequently, in double crops, such as in 2003 and 2004. We expected this pattern where there is an extensification in row crops that would typically be grown in a single crop pattern for the first growing season. In subsequent years agricultural intensification, or a shift from growing one crop to two crops, could explain the increase in double crops. Spatially, the extensification of row crops is on the edges of existing areas of agriculture and intensification occurs within the existing areas of croplands. These results show that there is a large increase in cropland in our study area and it is important to understand how this drastic change in land cover will (and does) impact carbon and nitrogen cycling. Distinguishing crop types, such as soybean and corn, is important as different crops have different implications for carbon and nitrogen cycling. Soybean plants fix nitrogen, but most of the fixed nitrogen leaves the system at harvest. Without proper management, over time, the loss of carbon and nitrogen decreases the soil fertility and may have other implications for land-use sustainability and management. Secondary crops, such as corn, may require large inputs of nitrogen fertilizers that increase nitrous oxide emissions. Addition of nitrogen fertilizers also impact local water quality. Knowing the number and type of crops being used allows us to proceed with spatially explicit models of biogeochemical changes associated with this agricultural development and intensification. In this study we demonstrate the stability of the wavelet approach over many years of an EVI time series. Thereby we can apply the WFCP methodology     199    to a wider study region to study regional development patterns of row crops. One future study includes parameterizing the wavelet-smoothed time series to identify crop types. We foresee the ability to use other types of time series in this wavelet method of analysis (WFCP), such as fractions of green vegetation from Spectral Mixture Analysis. Acknowledgements This work was supported by NASA Headquarters under the Earth System Science Fellowship Grant (06-ESSF06-40). Funding for field work in July 2005 was supported by a travel grant from Osvaldo Sala and the Environmental Change Initiative at Brown University. Field work would not have been possible without the help of Vicente Godinho of EMBRAPA (Brazil). Much thanks to Bethany Bradley and Jeremy Fisher for discussions of the project and Shannon Pelkey for help and inspiration. This work was improved by our reviewers. References Anderson, L.O., Y.E Shimabukuro, R.S. Defries, D. Morton (2005). Assessment of land cover and land use changes in the Brazilian Amazon using multi- temporal fraction images derived from Terra MODIS: Examples from the state of Matos Grosso. IEEE Trans on Geoscience and Remote Sensing, 2(3), 315-318.     200    Bradley, B., R. Jacob, J. Hermance, J.F. Mustard (2007). A curve-fitting technique to derive inter-annual phonologies from time series of noisy NDVI satellite data. Remote Sensing of Environment, 106, 137-145. Brown, C.J., M. Koeppe, B. Coles, K.P. Price (2005). Soybean production and conversion of tropical forest in the Brazilian Amazon: The case of Vilhena, Rondonia. Ambio, 34(6), 462-469. Burke, B. (1994). The mathematical microscope: Waves, wavelets, and beyond. In Bartusiak, M., B.Burke, A.Chaikin, A.Greenwood, T.A. Heppenheimer, M.Hoffman, D.Holzman, E.J. Maggio, A.S. Moffat (eds). A positron name Priscilla: Scientific Discovery at the Fronteir. National Academy Press, Washington DC. pp. 196-235. CONAB (Companhia Nacional de Abastecimento) (2004). Homepage: www.conab.gov.br. Fisher, J.I. J.F. Mustard, M.A. Vadeboncoeur (2006). Green leaf phenology at Landsat resolution: Scaling from the field to the satellite. Remote Sensing of Environment, 100, 265-279. Friedl, A.F., D.K. McIver, J.C.F. Hodges, X.Y. Zhang, D. Muchoney, A.H. Strahler, C.E. Woodcock, S. Gopal, A. Schneider, A. Cooper, A. Baccini, F. Gao, C. Schaal (2002). Global land cover mapping from MODIS: algorithms and early results. Remote Sensing of the Environment, 83, 287-302. Gendrin, A. Y. Langevin, J.P. Bibring, O. Forni. (2006). A new method to investigate hyperspectral image cubes: an application of the wavelet     201    transform. Journal of Geophysical Research, 111(E10004), doi:10.1029/2004JE002339. Hansen, M.C. and R.S. DeFries (2004). Detecting long-term global forest chance using continuous fields of tree-cover maps from 8-km Advanced Very High Resolution Radiometer (AVHRR) data for the years 1982-1999. Ecosystems, 7, 695-716. Huete, A., K. Didan, T. Miura, E.P. Rodriguez, X. Gao, L.G. Ferreira (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83, 195-213. Instituto Brasileiro de Geografia e Estatistica (IGBE) (2006). Systematic survey of agricultural production. IBGE website (accessed October 23, 2006). Instituto Nacional de Pesquisas Espaciais (INPE) (2006). Homepage: http://www.inpe.br/. Jensen, J.R (1996). Introductory digital image processing: A remote sensing perspective. 2nd Ed. Upper Saddle River: Prentice Hall, pp. 197-253. Jönsson, P. and L. Eklundh (2004). TIMESAT—a program for analyzing time- series of satellite sensor data. Computers & Geosciences, 30, 833-845. Justice, C.O. (1998). The moderate resolution imaging spectroradiometer (MODIS): Land remote sensing for global change research. IEEE Transactions on Geosciences and Remote Sensing, 36(4), 1228-1249. Lobell, D.B., and G.P. Asner (2004). Cropland distributions from temporal unmixing of MODIS data. Remote Sensing of Environment, 93, 412-422.     202    Li, Z. and M. Kafatos (2000). Interannual Variability of Vegetation in the United States and Its Relation to El Niño/Southern Oscillation. Remote Sensing of Environment, 71, 239-247. Luizão, F., P. Matson, G. Livingston, R. Luizão, P. Vitousek (1989). Nitrous oxide fluxes following tropical land clearing. Global Biogeochemical Cycles, 3(3), 281-285. Mallat, S. (1998). A wavelet tour of signal processing. Academic Press: San Diego, 637p. Melillo, J.M, R.A. Houghton, D.W. Kicklighter, A.D. McGuire (1996). Tropical deforestation and the global carbon budget. Annual Review of Energy and the Environment, 21, 293-310. Melillo, J.M., P.A. Steudler, B.J. Fiefel, C. Neill, D. Garcia, M.C. Piccolo, C.C. Cerri, H. Tian (2001). Nitrous oxide emissions from forests and pastures of various ages in the Brazilian Amazon. Journal of Geophysical Research, 106(D24), 34,179-34,188. Morton, D. C., R.S. DeFries, Y.E. Shimbukuro, L.O. Anderson, E. Arai, R. de Bon Espirito-Santo, R. Freitas, J. Morisette (2006). Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. PNAS, 103(39), 14637-14641. Mueller, C. (2003). Expansion and modernization of agriculture in the Cerrado— the case of soybeans in Brazil’s Center-West. Department of Economics Working Paper 306. University of Brasilia, Brasilia.     203    Myers, N., R.A. Mittermeier, C.G. Mittermeier, G.A.B de Fonseca, J. Kent (2000). Biodiversity hotspots for conservation priorities. Nature, 403, 853-858. Neill, C., J.M. Melillo, P.A. Steudler, C.C. Cerri, J.F.L. Moraes, M.C. Piccolo, M. Brito (1997). Soil organic matter stocks following forest clearing for pasture in the southwestern Brazilian Amazon. Ecological Applications, 7(4), 1216-1225. Neill, C., L. Deegan, S.M. Thomas, C.C. Cerri (2001). Deforestation for pasture alters nitrogen and phosphorus in small Amazonian streams. Ecological Applications, 11(6), 1817-1828. Percival, D.B., M. Wang, J.E. Overland (2004). An introduction to wavelet analysis with applications to vegetation time series. Community Ecology, 5(1), 19-30. Sakamoto, T., M. Yokozawa, H. Toritani, M. Shibayama, N. Ishitsuka, H. Ohno (2005). A crop phenology detection method using time-series MODIS data. Remote Sensing of Environment, 96, 366-374. Sakamoto, T., N.V. Nguyen, H. Ohno, N. Ishitsuka, M. Yokozawa (2006). Spatio- temporal distribution of rice phenology and cropping systems in the Mekong Delta with special reference to the seasonal water flow of the Mekong and Bassac rivers. Remote Sensing of Environment, (100), 1-16. Skole, D. and C. Tucker (1993). Tropical Deforestation and Habitat Fragmentation in the Amazon: Satellite Data from 1978 to 1988. Science, 260(5116), 1905-1910.     204    Steudler, P.A., J.M. Melillo, B.J. Feigel, C. Neill, M.C. Piccolo, C.C. Cerri (1996). Consequence of forest-to-pasture conversion on CH4 fluxes in the Brazilian Amazon Basin. Journal of Geophysical Research, 101(D13), 18,547-18,554. Torrence, C. and G.P. Compo (1998). A practical guide to wavelet analysis. Bulletin of the American Meteorological Society, 79(1), 61-78. Wardlow, B.D., S.L. Egbert, J.H. Kastens (In Press). Analysis of time-series MODIS 250m vegetation index data for crop classification in the U.S. Central Great Plains. Remote Sensing of Environment, XX(XX), XXXX- XXXX. Zhang, X., M.A. Friedl, C.B. Schaaf, A.H. Strahler, J.C.F.Hodges, F. Gao, B.C. Reed, A. Huete (2003). Monitoring vegetation phenology using MODIS. Remote Sensing of Environment, 84, 471-475.     205    Tables Table A.1 Wavelet powers and accuracy The effects of wavelet power on accuracy are reported using overall accuracy and Khat by comparing wavelet-detected cropping patterns to reference patterns generated by the user from the input EVI time series. Overall accuracy is the percent of points accurately identified in a class out of the total number of points sampled. Khat values incorporate misclassifications while assessing classification accuracy.   Wavelet power Overall Accuracy Khat 70% 81.0% 70.3% 80% 83.4% 74.2% 85% 90.0% 84.1% 90% 87.2% 80.4% 90% (0.4 thres.) 88.5% 92.1% 95% (0.4 thres.) 44.8% 25.6%     206    Table A.2 Area in croplands by year The amount of area in row-crop agriculture is reported here in square kilometers by harvest year. The area of row-crop agriculture was measured by the number or pixels having one or more crops in a year of the 90% wavelet-smoothed time series. Year  Area (km2)  2001  6,255  2002  6,799  2003  7,543  2004  8,532  2005  9,535      207    Table A.3 Analysis of unclassified pixels This table presents randomly selected unclassified pixels tabulated by their reference categories. Except where noted, the wavelet-smoothed time series were analyzed with a 0.3 EVI detection threshold. Classification Pixel count Not RC Single Double Unclassified Total Wavelet 70% 0 10 80 9 99 power 80% 0 12 71 17 100 85% 1 10 89 0 100 90% 0 19 73 8 100 90% (0.4 thres.) 2 16 78 4 100 Total 3 67 391 38     208    Table A.4 Accuracy assessments Randomly generated points throughout the study area were used for error analysis of the land use classifications: not row crops (Not RC), single cropping patterns (Single) and double cropping patterns (Double). The wavelet-detected results are compared to the reference data, in this case crop patterns detected by the user from the non-smoothed MODIS time series. An error matrix (A) shows the number of points correctly classified as well as the distribution of misclassified points. The producer’s accuracy and user’s accuracy (B) is low for Not RC Agriculture and Double crops but is rather high for Single crops. A Reference (MODIS) Pixel Counts Not RC Single Double total Wavelet- Not RC 100 0 0 100 detected Single 1 87 12 100 Double 2 26 168 196 total 103 113 180 396 B Producer's Accuracy User's Accuracy Wavelet- Not RC 97.1 % 100.0 % detected Single 77.0 % 87.0 % Double 93.3 % 84.0 %     209    Table A.5 Overall accuracy The overall accuracy of our algorithm was assessed by comparing our automated crop detection techniques preformed on the wavelet-smoothed EVI time series to the input EVI time series data. Overall accuracy is the percent of points accurately identified in a class out of the total number of points sampled. Khat values are also high for all wavelet powers. Reference data Overall Accuracy Khat MODIS 88.5% 92.1% Fazenda 94.0% 85.7%     210    Table A.6 Error matrix for classifications at Fazenda Santa Lordes An error matrix shows the agreement and disagreement between reference data (Fazenda Santa Lordes farm history) and the 90% wavelet-detected cropping patterns (A). The producer’s accuracy and user’s accuracy (B) are shown for each land use classification. A Reference (farm) Pixel Counts Not RC Single Double total Wavelet- Not RC 71 4 0 75 detected Single 0 18 2 20 Double 0 0 5 5 total 71 22 7 100 B Producer's Accuracy User's Accuracy Wavelet- Not RC 100.0 % 94.7 % detected Single 81.8 % 90.0 % Double 71.4 % 100.0%     211    Table A.7 Cropping patterns by year The cropping patterns for each year are presented here. These are the results from the 90% power wavelet-smoothed time series using a 0.4 EVI crop detection threshold, as it performs with the lowest misclassifications (Table 5). The intensification of double crops after 2003 is particularly notable. Area (km2) Cropping Pattern Single Double Year 2001 3,124 3,131 2002 3,251 3,548 2003 4,063 3,480 2004 2,790 5,742 2005 3,643 5,892       212    Figures   Figure A.1 Overview of methodology Overview of methodology, divided into four parts: data processing, crop detection, field verification data and error analysis.       213    Figure A.2 Location map Location map. The study area, shown here in the gray box, is in the southwestern Brazilian Amazon in the state of Mato Grosso.       214    Figure A. 3 False-color infrared MODIS image False-color infrared MODIS image (Red= 859 nanometers, Green= 645 nanometers, Blue= 555 nanometers) for the study area on 28 July 2005 (Upper left corner: 12° 15’ 23.61’’ S, 59° 45’ 18.23’’ W; Lower right corner: 13° 59’ 27.86’’ S, 57°57’ 30.8’’ W). Bright red areas represent dense cerradão native vegetation. Lighter reds to dark greens show the extent of cerrado native vegetation. Bright turquoise blues show mechanized agriculture, and very bright white areas are (bare) agricultural fields. Fazenda Santa Lordes is highlighted with a yellow polygon. N 200 km     215    Figure A.4 EVI time series EVI Time series for 2000-2006 with cloud-contaminated points removed and filled by linear interpolation. Each time series is offset by 1.0 EVI for clarity. The bottom line (A) shows cerrado phenology, which has a low annual mean EVI and exhibits only minor seasonal fluxuations. Forest (B) phenology has only slight changes in greenness between the wet and dry seasons. A pasture (C) has higher annual mean and variance than cerrado, but a lower mean and higher variance than forest. Time series (D) shows an area in cerrado from 2000 to mid 2002. There is a conversion prior to 2003 and a change to agricultural (crop) phenology can been seen from 2003 to the end of the time series. Time series (E) shows an area that exhibits single crop phenology in 2001, 2002, and 2003 and double crop patterns in 2004 and 2005.         216    Figure A.5 Histogram of standard deviations This histogram show how the high standard deviation separates areas of mechanized agriculture from other land cover classes. Mechanized agriculture was identified as areas that have an annual standard deviation great than the local minimum about 0.15 identified for that year. 25000 2001 2002 2003 20000 2004 2005 Number of Pixels 15000 10000 5000 Mechanized Agriculture 0 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Annual Standard Deviation               217    Figure A.6 Input data and residuals The input time series data for an area of single cropping is shown on top (EVI daily time series after processing for clouds and extremely noisy pixels). The residuals, or the difference between the input data and the wavelet-smoothed time series, are shown for each wavelet power with their associated RMS error.         218    Figure A.7 Wavelets Shown here are the 70% power (9 coefficients), 80% power (15 coefficients), 85% power (19 coefficients) and 90% power (27 coefficients) wavelet-smoothed time series where the power used in the wavelet transform determines the amount of detail retained. Circles represent the EVI time series with cloud- contaminated values removed. Using a higher power may fit maxima in the data better but can also create false peaks, indicated here with an arrow. 1 Observed EVI 70% 80% 85% 90% EVI 0 2001.75 2002.25 2002.75 Year             219    Figure A.8 Examples of wavelet-smoothed time series EVI daily time series, after processing for clouds and extremely noisy pixels, is shown in the top plot for an area of double crops (A.8.a and A.8.b). Figure A.8.a shows the wavelet-smoothed time series for each wavelet power tested. These time series are offset by 1 for clarity. In Figure A.8.b, the processed input EVI time series is plotted with the calculated residuals for each wavelet power, offset with clarity, with their associated RMS error.         220    Figure A.9 Example wavelet-smoothed time series A field site at Fazenda Sta. Lordes is represented in this time series. EVI daily time series (2000-2006) after processing for clouds and extremely noisy pixels is plotted with small circles. The solid line shows the 90% power wavelet-smoothed time series. Land cover is converted from cerrado to mechanized agriculture in the middle of 2002. In 2004, a single rice crop was grown. In 2005, two soybean crops were grown.         221    Figure A.10 Extent of croplands by year The 90% power wavelet-smoothed time series results for detecting maximums representative of single and double crops are presented here. Cropping patterns in the study region show an increasing area is cultivated in double crops (black) instead of single crops (gray). The increase in double cropping (black) is notably centered in existing agricultural zones. Extent of mechanized agriculture over time is a result of the standard deviation threshold described in Figure A.3. The extent of mechanized agriculture (shown in black and gray) is observed to be spreading. Change is particularly notable on the edge of the cropland region, such as the areas circled in dotted lines.       222    Figure A.11 Area detected by cropping pattern The 90% power wavelet with a 0.4 EVI threshold for detecting maxima preformed with this lowest misclassifications (Table A.2). These results were used to track the area in single and double cropping patterns by year. The area in croplands increases through the time series. The intensification of double crops after 2003 is particularly notable.  10000 9000 8000 7000 Square kilometer 6000 Double 5000 4000 3000 2000 Single 1000 0 2001 2002 2003 2004 2005 Year