beyond forest and non forest fd

Upload: zafeersaqib

Post on 07-Apr-2018

224 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/4/2019 Beyond Forest and Non Forest Fd

    1/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    1

    Beyond forest and non-forest: MODIS & AVHRR in mapping& quantifying the evolution of forest to cropland types in

    the seasonal tropics of Bolivia

    Danny Redo1

    and Andrew C. Millington2

    The main objective of this study was to utilize MODIS and AVHRR NDVI, supplemented with Landsat

    TM and CBERS-2 imagery and biophysical and phenological data, in order to map/quantify the

    conversion of forest to cropland in southeastern Bolivia from 1986 to 2007. Analyses focus specifically

    on the Tierras Bajas region, one of the most important agricultural region-deforestation hotspots in

    South America and located in some of the most sensitive and poorly understood ecosystem ecosystems in

    the world. Here, we seek to answer the following questions: what types of forest are being converted to a

    particular cropping regime? What are the potential consequences of change for the regions ecology?

    Phenology curves from NDVI and a crop calendar were used as inputs in a decision tree for classifyingcropland in 2007 according to a particular growing season. We classify forests in 1986 according to

    several biophysical characteristics such as seasonal variations in precipitation, soil types, and elevation

    data in order to overcome problems associated with previous approaches.

    During the 21 year study period, cropland cultivated in summer and winter as well as summer only

    caused the loss of nearly 8,500 km2 of forest in an area of only 21,800 km2. 52.8% (4,442 km2) were

    classified as seasonal while another 25.5% (2,143 km2) were classified as dry/seasonal. Proportionally,

    68% (1,760 km2) of all wet/seasonal forest present in 1986 was converted to cropland by 2007 compared

    to only 48% seasonal (4,442 km2), 38% dry (66 km2) and 37% dry/seasonal forest (2,143 km2). The

    significance of these findings is that farmers appear to be seeking out lands which receive the most

    rainfall the fragile transition zone between the Chaco and Chiquitano ecosystems. By overlaying a

    DEM, soils map, and mean annual precipitation data, it also evident that these parcels also contain thebest soils and the most level terrain in addition to the highest rainfall totals. Farmers are consciously

    aware of the most suitable lands for agriculture and attempting to maximize production by producing two

    crops per year soybeans in the summer and sunflower or sorghum in the winter. By assessing between

    changes in various types of forest and crop classes, we provide planners and conservationists with more

    than simply quality, accurate forest cover and change maps. This methodology provides decision-makers

    with more detailed insight as to the proximate causes or driving forces of change.

    Keywords: remote sensing, MODIS, AVHRR, land-use and land-cover change, Bolivia

    Introduction

    A primary goal of remote sensing and land change science is to map land-use and land-cover (LULC) anddetermine how much of it is in a state of expansion, decline, or resistant to change (Hansen et al. 2008a). To

    achieve this objective, scientists have turned to Landsat MSS, TM, and ETM+, owing to their near-global

    and longer temporal coverage of 30+ years. For this reason, they are the preferred choice the workhorses

    for use in mapping the conversion of forest to agriculture. A major drawback, however, is that Landsat is

    unable to capture seasonal variations due to infrequent temporal coverage and cloud contamination, two

    issues common in the tropics (DeFries and Belward 2000). It is therefore ineffective for use in mapping

    types of forest or agriculture classes in many tropical areas. The usual remedy is AVHRR or MODIS with

    1Corresponding author: Ph.D. Candidate and Graduate LecturerMailing Address: Department of Geography, Texas A&M University, 810 O&M Building, College Station,TX 77843 USA

    2 Corresponding author:Advisor, Director of Environmental Programs in Geosciences, and Professor of GeographyMailing Address: Department of Geography, Texas A&M University, 810 O&M Building,College Station, TX 77843 USA Texas A&M University, USA

  • 8/4/2019 Beyond Forest and Non Forest Fd

    2/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    2

    their greater image acquisition frequency and areal coverage, but the comparatively coarser spatial

    resolution is also problematic as most change occurs at scales below 0.25 to 1-km resolution (Hansen et al.

    2008b). In either case, the end result is that one or more types of LULC classes are often collapsed into a

    generic forest and non-forest classification scheme giving rise to pluralistic interpretations and anecdotal

    evidence about the changes that are occurring or possibly, concealing them altogether (Robbins 2001).

    The degree to which Landsat, the newly launched CBERS-2, AVHRR and MODIS imagery cancomplement one another in overcoming these issues is underexplored. Our aim is to utilize imagery from

    these four sensors as well as biophysical and phenological data (soil types, rainfall, elevation, and crop

    calendars) in order to map/quantify the types of forest to cropland conversion that have occurred in lowland

    Bolivia between 1986 and 2000. Analyses focus specifically on the Tierras Bajas region, one of the most

    important agricultural region-deforestation hotspots in South America and located in some of the most

    sensitive and poorly understood ecosystem ecosystems in the world (Lepers et al. 2005). Here, we seek to

    answer the following questions: what types of forest are being converted to a particular cropping regime?

    What are the potential consequences of change for the regions ecology?

    Bolivias Tierras Bajas

    The Corredor Biocenico is a continental transportation artery used to move agricultural products betweenChile and Brazil. In southeastern Bolivia, the corridors main highway and railroad bisect a 21,800 km2 area

    of level terrain overlaid with rich alluvial soils known as the Tierras Bajas (Fig. 1). Mechanized,

    commercial production is concentrated in the Tierras Bajas owing to the regions superior soils, higher

    rainfall, level terrain, proximity to the city of Santa Cruz, as well as a highway and railroad to transport

    produce to the Hidrova Paran-Paraguay and beyond. To meet demand, generally two crops are sown per

    year soy, wheat, or maize in the rainy, summer months and soy, sunflower, rice, sugar cane or sorghum in

    the drier, winter months (ANAPO 2007).

    Most natural vegetation in the region has been converted to cropland, but along the southern and eastern

    periphery of the region are the once widespread remnants of Chaco forest. Based on precipitation, soils, and

    topography, vegetation is dry/seasonal, receiving less moisture and/or situated on poorer, sandier soils

    compared to seasonal or transitional types, which are wet for approximately half the year and dry the other.

    Towards the north-central part of the study area is the Chiquitano ecosystem, forming a transition zonebetween the humid forests of the Amazon Basin and the drier Chaco ecosystem (Kennard 2002).

    Characterized by its deciduousness, for more than half the year Chiquitano forest is tropical wet (Pennington

    et al. 2000). These forests are wet/seasonal or seasonal as they receive more moisture and/or also have

    better soils and poorly-drained topography compared.

    Previous approaches

    Since the mid-1980s, AVHRR has been successfully used to map land cover at the regional and global

    scales (e.g. Tucker et al. 1985; Millington and Townshend 1988; DeFries et al. 1998; Hansen et al. 2000;

    Loveland et al. 2000). The overarching aim was to assess forest resources for use in climate change studies

    or energy balance models resulting in land cover being categorized according to ecosystems (e.g. evergreen

    forest) biomass, or simply aggregated into a forest class. Consequently, land use was ignored and

    categorized as some variation of agriculture. Another factor was spatial as land use normally occurs below

    AVHRRs finest resolution of 1km2. Since 2000, the classification of land-use types has improved with the

    use of 250m2 and 500m2 MODIS data, but once again, spatial limitations have restricted use to regions of

    large-scale commercial agriculture such as western Brazil (e.g. Wessels et al. 2004; Brown et al. 2007) and

    the US Great Plains (e.g. Wardlow 2005, 2006, 2007). Using phenology curves derived from NDVI, these

    studies illustrate the efficacy of MODIS for use in classifying detailed types of land-use.

    LULC data sets available for South America and the Bolivian lowlands (Table 1) lack the level of detail

    found in the U.S. and Brazil, due largely to the exclusive use of Landsat. Vegetation is classified as forest

    and in the case of Davies (1993) according to forest succession (e.g. primary forest). The continental

    assessments (e.g. Eva et al. 2004; Dinerstein et al. 1995) use an ecosystem scheme to classify natural

    vegetation into either Chaco or Chiquitano forest. The problem with this approach is separability of classes

    owing to seasonal and annual variations in precipitation, which together create shifting mosaics. It is in thistransition zone that the Tierras Bajas is situated. Land use, on the other hand, was given little attention and

  • 8/4/2019 Beyond Forest and Non Forest Fd

    3/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    3

    labeled as either non-forest or agriculture. The final factor to consider in these studies is modernity; most, if

    not all, are dated. The two which focus exclusively on the Tierras Bajas (Davies 1993; Steininger et al.

    2001a,b) proceed only to the year 1991 and 1998, respectively, at a time when the greatest rates of forest

    loss in this region had yet to be attained (Redo et al. 2009). This paper addresses these knowledge gaps by

    building and expanding on previous LULC mapping carried out in the U.S., Brazil, and Bolivia through a

    protocol outlined below.

    Data description and mapping protocol

    Mapping Crop Types from 250-m MODIS NDVIMOD13Q1 NDVI (16-day L3 Global 250-m) coverage for two tiles (h11v10 and h12v10) for the entire

    calendar year of 2007 (23 scenes per tile) was acquired from NASAs Land Processes Distributed Active

    Archive Center (LP DAAC). Characteristics of MODIS and NDVI are well known (e.g. see Huete et al.

    2002) and will not be repeated here.

    Individual 16-day composites in adjacent tiles were mosaicked to provide coverage of the study area

    (Fig. 2). We performed geometric correction of each mosaic using a 2006/07 CBERS-2 (20-m) classified

    product (99% accuracy Redo et al. 2009) resulting in a range of RMS errors between 0.15 and 0.31 pereach 16-day composite. All analyses were restricted to a 50-km north-south buffer centered on the main

    highway with endpoints in the west at the Andean front range/Rio Pira and the western extent of the

    Brazilian Shield/Quimome River in the east (Fig. 1). Masks were created from the 2006/07 CBERS-2 (20-

    m) imagery to remove areas not cultivated for agriculture (forest, water bodies, natural bare ground, and

    urban regions).

    We classified types of cropland using decision trees. Decision trees have been used to classify and detect

    change in LULC data since the 1990s, and have proved accurate in a number of studies (e.g. Friedl and

    Brodley 1997; Zhan et al. 2002; Wardlow 2006, 2007; Hansen et al. 2008a). For a complete description of

    characteristics, see Hansen et al. (1996) and DeFries et al. (1998). Using this method, composites were first

    classified into two discrete categories: green (cultivated) or fallow (bare) cropland. Decisions were based on

    phenology curves, which were produced through random selection of twenty agricultural fields identified on

    both MODIS and CBERS imagery. One 250-m pixel centrally located within a randomly selected

    agricultural field was chosen as a representative. A single pixel was used instead of a pixel window orsquare (e.g., 2 x 2 pixels) in an attempt to reduce, and in some larger fields, eliminate the inclusion of mixed

    edge pixels (Wardlow 2005, 2006). Analysis of the resulting phenology curves revealed a threshold NDVI

    value of 5,500 (original MOD13Q1 NDVI imagery is scaled from 1 to 10,000), which was used to

    determine field status for a particular composite.

    Further analysis of phenology curves coupled with a local crop calendar also revealed the presence of two

    distinct growing seasons in the Tierras Bajas: summer, wet (December 19-March 5) and winter, dry (May

    25-July 27). Only 16-day composites which fell within the peak growing season were selected for final

    analysis (Fig. 2). For example, the peak winter cropland map included only the following: May 25-June 9;

    June 10-25; June 26-July 11; July 12-27. After selecting composites for growing seasons, a singular

    classified map of each growing season was created using maximum value compositing (MVC). The MVC

    technique selects, pixel-by-pixel, the highest NDVI value for multiple raster images to create a single

    composited image (Holben 1986). A pixel was classified as green or fallow in the new winter and summerrecomposites based on the majority rule in each of the final growing season maps. For example, if a pixel

    was initially classified as fallow from May 25-June 9, but green in the periods June 10-25, June 26-July 11,

    and July 12-27, it was labeled as cultivated during the peak winter growing season. Finally, the winter and

    summer maps were composited and majority rule employed once more to create a singular crop map with

    four cropland classes: summer and winter cultivation, summer only, winter only, and fallow.

    Accuracy assessment was initially scheduled for October, 2008, but the recent outbreaks of violence in

    Pando, and continued protests and threats in Santa Cruz over proposed land expropriation have prevented

    field work. A future visit is tentatively scheduled for May, 2009 and will consist of interviewing 5 types of

    land managers with varying degrees of ownership and modernization. They will be prompted with the maps

    generated in this study and asked to recall aspects of forest clearance and types of crops grown.

  • 8/4/2019 Beyond Forest and Non Forest Fd

    4/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    4

    Mapping Forest Types from 1-km AVHRR NDVIAVH13C1 NDVI (daily, 1-km NDVI) scenes for the entire calendar year of 1986 were acquired from

    NASAs Long-Term Data Record (LTDR). Daily scenes were aggregated into monthly composites using

    the MVC technique (see above section for description of technique) in order to reduce data volume,

    eliminate cloud contamination, and because monthly intervals can still capture seasonal variations in forests

    (Fig. 3). We performed geometric correction of each monthly composite using a 1986/89 Landsat TM (30-m) classified product (Redo et al. 2009) resulting in a range of RMS errors between 0.17 and 0.31 per each

    monthly composite. To compare changes between forest and crop types, the spatial extent of the study area

    was kept consistent (see above section). Masks were created from the 1986/89 Landsat TM (30-m) imagery

    in order to remove non-forest areas (cropland, pasture, bare ground, and water bodies).

    Field research (ground and overhead) coupled with biophysical data (a DEM, soils map, mean annual

    precipitation data, drainage network coverage, and Landsat TM imagery were used to identify an NDVI

    threshold separating dry or moist forest. After several iterations, a decision tree was used to classify all

    monthly composites according to this scheme. These were then recomposited into a single yearly

    classification using the majority rule and a second decision tree classifier resulting in five forest classes: dry

    forest, dry/seasonal forest, seasonal forest, wet/seasonal forest, and inundated forest. Accuracy of classes

    was assessed using a variety of techniques. Field visits during the dry seasons of 2006, 2007, and 2008

    helped to identify dry, seasonal, and inundated forest classes on-site. We supplemented field notes with

    stream network, elevation, and soil data too indentify areas on-screen. Areas of interest which could be

    positively identified were denoted on the forest final map and an accuracy assessment was conducted. High

    overall accuracy was achieved at 96.43%.

    Results

    In 2007, 67.5% (7,693 km2) of all cropland in the Tierras Bajas was in both summer and winter production

    (Table 2). Distribution was ubiquitous, lacking any discernable pattern as summer and winter cropland

    appeared in the older, wetter areas of production north of Santa Cruz as well as in newer, more marginal

    agricultural zones such as those south of El Tinto (Fig. 4). 28.5% (3,248 km2) of all cropland was cultivated

    only during the summer growing season and associated with the Mennonite communities of Basilio, Tres

    Cruces, and Pozo del Tigre. Both areas receive relatively less precipitation compared to the northwest andare located on soils less suitable for year-round cultivation (i.e. vertisols, dark clayey soils that sink during

    the dry season). Another possibility is the practice of land lying fallow during the dry season or every four

    years in marginal areas per the advice of Mennonite extension agencies based in Santa Cruz. This may also

    explain the 4% (455 km2) of cropland which remained under fallow throughout both growing seasons

    between Tres Cruces and Pozo del Tigre.

    In 1986, the Tierras Bajas was a largely forested landscape with nearly 18,000 km2 remaining east of the

    Rio Grande. 52.3% (9,317 km2) was seasonal in nature and well-distributed throughout the central and

    southern portions of the study region (Table 2; Fig. 5). Homogenous areas were located in association with

    the large, poorly-drained floodplains of the Rio Quimome and Rio Parapeti floodplains as well as near

    smaller drainage systems northeast of Pozo del Tigre. Nearly one-third (5,279 km2) was dry/seasonal forest

    with much of it confined to one large patch centered on El Tinto. This pattern is the result of comparatively

    less precipitation and the presence of well-drained, rocky soils and outcrops of the western-most extent ofthe Brazilian Shield. 14.6% (2,597 km2) was wet/seasonal forest and concentrated in a single large patch in

    the north where, in contrast to the extreme east, annual rainfall is hundreds of millimeters higher than areas

    further south or east. Finally, 1% (174 km2) of all forest types was classified as dry. Counter intuitively, it

    was found in the usually permanent wetlands northeast of Lago Concepcon and likely should have been

    classified as inundated forest. One possible explanation for this classification error is that more water was

    present than vegetation in 1986 and the absorption of near-infrared energy resulted in consistently low

    monthly NDVI values.

    Crop and forest classes were intersected to map and quantify the types of forest to cropland conversion

    that occurred from 1986 to 2007 (Table 3; Fig. 6). Results show that cropland which was cultivated in both

    summer and winter was responsible for 69% (5,820 km2) of all deforestation that took place during the study

    period. 35.5% (2,987 km2) was classified as seasonal forest, 18.8% (1,583 km2) dry/seasonal and 14.3%

    (1,198 km2) wet/seasonal. The loss of seasonal forest was well distributed throughout the study region, but

    dry/seasonal deforestation mainly occurred in close proximity to the highway and marginal areas such as El

    Tinto and Tres Cruces. The loss of wet/seasonal forest was confined to its two main areas of previous

  • 8/4/2019 Beyond Forest and Non Forest Fd

    5/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    5

    existence: around the north-central and southwest portions of the study region. Cropland cultivated only

    during the summer was responsible for the loss of 15% (1,253 km2) seasonal forest as well as 5.7% (476

    km2) of forest which was classified as wet/seasonal. Concentrated losses of seasonal forest occurred mainly

    around Tres Cruces while loss of wet/seasonal forest to summer cropland was in the same areas as that

    which was lost to cropland cultivated in both summer and winter. Less than 1% (59 km2) of dry/seasonal

    was converted to this cropping regime and was well-distributed. Fallow cropland was responsible for only4% of all forest loss during the study period, but over approximately two-thirds was forest classified as

    seasonal and confined mainly to area north of the highway between Pozo del Tigre and Tres Cruces.

    Discussion

    During the 21 year study period, establishment of cropland (cultivated in both growing seasons and summer

    only) east of the Rio Grande caused the loss of nearly 8,500 km2 of forest in the Tierras Bajas an area of

    only 21,800 km2. A reasonable means of contextualization is to compare loss to the size of US states. In

    this case, two Rhode Islands worth of forest was lost in the short time of two decades in an area as small as

    New Jersey. 52.8% (4,442 km2) were classified as seasonal while another 25.5% (2,143 km2) were

    classified as dry/seasonal, but this was mainly due to the fact that these two types covered a combined84.5% of the Tierras Bajas in 1986. In short, absolute values of deforestation can obscure other, more

    meaningful indicators of the underlying drivers of change.

    One such indicator is the proportion of each forest type cleared. 68% (1,760 km2) of all wet/seasonal

    forest present in 1986 was converted to cropland by 2007 compared to only 48% seasonal (4,442 km 2), 38%

    dry (66 km2) and 37% dry/seasonal forest (2,143 km2). The significance of these findings is that farmers

    appear to be seeking out lands which receive the most rainfall the fragile transition zone between the

    Chaco and Chiquitano ecosystems. By overlaying a DEM, soils map, and mean annual precipitation data, it

    also evident that these parcels also contain the best soils and the most level terrain in addition to the highest

    rainfall totals. Clearly, farmers are consciously aware of the most suitable lands for agriculture and

    attempting to maximize production by producing two crops per year soybeans in the summer and

    sunflower or sorghum in the winter.

    Development, however, has come at high cost. The Chaco and Chiquitania ecosystems, both listed among

    the worlds 200 most sensitive (Wassenaar et al. 2007), are currently separated by widening gaps of forestclearance for agriculture. This transition zone, once composed of wet/seasonal forests, was intact as of 1986

    but had shrunk to half its size in 2007 as it was the type most preferred by farmers. Species movement and

    interaction between ecoregions is greatly hampered causing its functioning as an ecosystem to cease at the

    local level.

    To the south, an equally alarming trend is occurring. Results from this study also show that in 2007, the

    frontier of cultivation in the eastern Tierras Bajas was threatening the northern boundary of South

    Americas largest protected area (3.44 million ha), Kaa-Iya National Park and Integrated Management Areas

    (IMA), which together cosset the largest area of tropical dry forest under full-protected area status anywhere

    in the world (Winer 2003). The park is also a model that puts community-based conservation into practice

    through being co-managed by the Bolivian government, the indigenous Izoceo-Guarani and conservation

    organizations. As pressure to continue producing lucrative soybeans and sunflower increases, those

    displaced from development, might well push settlement into the thinly settled northern IMA or even theparks core.

    The current status of southeastern Bolivias remaining forests is also bleak. The Bolivian government is

    nearly finished upgrading the highway through paving and the construction of bridge and drainage

    infrastructure. When complete in 2010, the new highway will be the only paved, all-weather thoroughfare

    in central South America. It will connect the continents agricultural heartlands of Santa Cruz and

    neighboring Mato Grosso do Sul, Brazil more directly to international markets through Brazilian and

    Chilean ports. Imagery analysis (2005-2008) for emerging agricultural colonies to the east shows that

    deforestation rates are accelerating (Redo et al. 2009). If deforestation reaches scales seen in the Tierras

    Bajas, it could be ecologically devastating for the regions indigenous inhabitants and the forests they

    depend on, as well as several noteworthy national parks protecting sensitive ecosystems such as the Pantanal

    wetlands. Analysis of change between 1987 and 2007 for the central and eastern portions of the Corridor

    Biocenico are currently underway. However, more time points need to be incorporated in order to further

    support this evidence. Future work is aimed at acquired AVHRR NDVI for 1993 and MODIS NDVI for

    2000 and 2008 so that trajectories of change can be produced.

  • 8/4/2019 Beyond Forest and Non Forest Fd

    6/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    6

    Conclusion

    This study shows that it is possible to incorporate several data sources and supplement one sensors

    weakness with anothers strength for use in mapping and quantifying changes in detailed land-use and land-cover types. Results of this nature provide better input to spatial models which can tackle issues such as

    trajectory, consequences and future of LULC change. By going beyond classic classification schemes such

    as forest vs. non-forest or ecosystem approaches and assessing between changes in various types of forest

    and crop classes, we provide planners and conservationists with more than simply quality, accurate forest

    cover and change maps. A hybrid methodology provides decision-makers with more detailed insight as to

    the proximate causes or driving forces of change. This information is imperative for raising both

    government and public awareness so that more informed policy responses can be made about landscape

    management and conservation (e.g. planning of future protected areas or effectiveness of existing units). In

    addition, scientists studying human-environment relationships can better understand the dynamic impact

    humans have on the environment. Finally, by focusing on one of the most dynamic regions in the

    Neotropics, we have advanced the basic scientific knowledge of LULC change in southern hemisphere

    semi-arid wooded ecosystems and provided a better understanding of the nature of human-environmentrelationships in one of the most dynamic, contemporary frontier regions in South America.

    AcknowledgmentsThe research described in this paper was funded by the National Science Foundation, SBE DDRI Geography

    and Regional Science/Office of International Science and Engineering (Grant: BCS-0802672) and Texas

    A&M University. Special thanks to our driver and friend, Don Lucho Ramirez, who expertly navigated

    long hours on unpaved, dusty, and pot-hole ridden roads, as well as provided unwavering companionship

    during our visits over the last three years.

    References

    ANAPO (Asociacin De Productores De Oleaginosas Y Trigo). 2007. Anuario Estadstico 2007.

    Unidad de Servicios y Communicaciones UsyC, Santa Cruz, Bolivia.

    Brown, J.C., Jepson, W.E., Kastens, J.H., Wardlow, B.D., Lomas, J.M., & Price, K.P. 2007.

    Multitemporal, moderate-spatial-resolution remote sensing of modern agricultural production and land

    modification in the Brazilian Amazon. GIScience and Remote Sensing, 44 (2): 117-148.

    Davies, D. 1993. Estimation of deforestation east of the Rio Grande, Bolivia, using Landsat satellite

    imagery. Unpublished Masters Thesis, Cranfield Institute of Technology, Silsoe College, Silsoe, UK.

    DeFries, R., Hansen, M., Townshend, J.R.G. & Sohlberg, R. 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 (16): 3141-3168.

    DeFries, R. & Belward, A.S. 2000. Global and regional land cover characterization from satellite data:

    an introduction to the special issue. International Journal of Remote Sensing, 21 (6-7): 1083-1092.Dinerstein, E., Olson, D.M., Graham, D.J., Webster, A.L., Primm, S.A., Bookbinder, M.P. & Ledec, G.

    1995. A Conservation Assessment of the Terrestrial Ecosystems of Latin America and the Caribbean.

    Washington, D.C.: The World Wildlife Fund and World Bank.

    Eva, H.D., Belward, A.S., De Miranda, E.E., Di Bella, C.M., Gond, V., Huber, O., Jones, S., Sgrenzaroli,

    M. & Fritz, S. 2004. A land cover map of South America. Global Change Biology, 10: 732745.

    Friedl, M.A. & Brodley, C.E. 1997. Decision tree classification of land cover from remotely sensed data.

    Remote Sensing of Environment, 61: 399-409.

    Hansen, M.C., Dubayah, R. & DeFries, R. 1996. Classification trees: an alternative to traditional land

    cover classifiers. International Journal of Remote Sensing, 17 (5): 1075-1081.

    Hansen, M.C., DeFries, R., Townshend, J.R.G. & Sohlberg, R. 2000. Global land cover classification at

    1-km spatial resolution using a classification tree approach. International Journal of Remote Sensing,

    21 (6&7): 1331-1364.

    Hansen, M.C., Stehman, S.V., Potapov, P.V., Loveland, T.R., Townshend, J.R.G., DeFries, R.S., Pittman,

    K.W., Arunarwati, B., Stolle, F., Steininger, M.K., Carroll, M., & DiMiceli, C. 2008a. Humid tropical

  • 8/4/2019 Beyond Forest and Non Forest Fd

    7/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    7

    forest clearing from 2000 to 2005 quantified by using multi-temporal and multi-resolution remotely

    sensed data. Proceedings from the National Academy of Sciences, 105 (27): 9439-9444.

    Hansen, M.C., Shimabukuro, Y.E., Potapov, P., & Pittman, K. 2008b. Comparing annual MODIS and

    PRODES forest cover change for advancing monitoring of Brazilian forest cover. Remote Sensing of

    Environment, 112: 3784-3793.

    Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X. & Ferreira, L.G. 2002. Overview of theradiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of

    Environment, 83: 195-213.

    Holben, B.N. 1986. Characteristics of maximum-value composite images from temporal AVHRR data.

    International Journal of Remote Sensing, 12: 1147-1163.

    Killeen, T.J., Calderon, V., Soria, L., Quezada, B., Steininger, M.K., Harper, G., Solrzano, L.A. &

    Tucker, C.J. 2008. Thirty years of land-cover change in Bolivia. Ambio, 36 (7): 600-606.

    Kennard, D.K. 2002. Secondary forest succession in a tropical dry forest: patterns of development across

    a 50-year chronosequence in lowland Bolivia. Journal of Tropical Ecology, 18: 53-66.

    Lepers, E., Lambin, E.F., Janetos, A.C., DeFries, R.S., Achard, F., Ramankutty, N. & Scholes, R.J. 2005. A

    synthesis of information on rapid land-cover change for the period 1981-2000. BioScience, 55 (2): 115-124.

    Loveland, T., Reed, B.C, Brown, J.F., Ohlen, D.O., Zhu, Z., Yang, L., & Merchant, J.W. 2000.

    Development of a global land cover characteristics database and IGBP DISCover from 1 km AVHRR

    data. International Journal of Remote Sensing, 21 (6&7): 1303-1330.

    Mertens, B., Kaimowitz, D., Puntodewo, A., Vanclay, J., & Mendez, P. 2004. Modeling deforestation at

    distinct geographic scales and time periods in Santa Cruz, Bolivia. International Regional Science

    Review, 27 (3): 271-296.

    Millington, A. & Townshend, J.R.G. 1988. Biomass Assessment: Woody Biomass in the SADCC

    Region. London, UK: Earthscan Publications.

    Pennington, R.T., Prado, D.E. & Pendry, C.A. 2000. Neotropical seasonally dry forests and Quaternary

    vegetation changes. Journal of Biogeography, 27: 261-273.

    Redo, D., Millington, A.C. & Hindery, D. 2009. Neoliberalism to socialism: Agrarian reform, land-use

    change and deforestation dynamics under Evo Morales. Latin American Perspectives, under review

    January 28, 2009.

    Robbins, P. 2001. Fixed categories in a portable landscape: the causes and consequences of land-cover

    categorization. Environment and Planning A, 33: 161-179.Steininger, M.K., Tucker, C.J., Townshend, J.R.G., Killeen, T.J., Desch, A., Bell, V., & Ersts, P.

    2001a. Tropical deforestation in the Bolivian Amazon. Environmental Conservation, 28 (2): 127-134.

    Steininger, M.K., Tucker, C.J., Ersts, P., Killeen, T.J., Villegas, Z., & Hecht, S.B. 2001b. Clearance and

    fragmentation of tropical deciduous forest in the Tierras Bajas, Santa Cruz, Bolivia. Conservation

    Biology, 15 (4): 856-866.

    Tucker, C.J., Townshend, J.R.G. & Goff, T.E. 1985. African land-cover classification using satellite

    data. Science, 227 (4685): 369-375.

    Tucker, C.J. & Townshend, J.R.G. 2000. Strategies for monitoring tropical deforestation using satellite

    data. International Journal of Remote Sensing, 21 (6&7): 1461-1471.

    Wardlow, B.D. 2005. An evaluation of time-series MODIS 250-meter vegetation index data for crop mapping

    in the U.S. Central Great Plains. Unpublished Ph.D. thesis, University of Kansas, Lawrence, KS.

    Wardlow, B.D. 2006. Using USDA crop progress data for the greenup onset date calculated fromMODIS 250-meter data. Photogrammetric Engineering and Remote Sensing, 72 (11): 1225-1234.

    Wardlow, B.D. 2007. Analyzing of time-series MODIS 250 m vegetation index data for crop

    classification in the U.S. Central Great Plains. Remote Sensing of Environment, 108: 290-310.

    Wassenaar, T., Gerber, P., Verburg, P.H., Rosales, M., Ibrahim, M. & Steinfeld, H. 2007. Projecting

    land use changes in the Neotropics: the geography of pasture expansion into forest. Global

    Environmental Change, 17: 86-104.

    Wessels, K.J., DeFries, R.S., Dempewolf, J., Anderson, L.O., Hansen, A.J., Powell, S.L., & Moran, E.F. 2004.

    Mapping regional land cover with MODIS data for biological conservation: examples from the Greater

    Yellowstone Ecosystem, USA and Par State, Brazil. Remote Sensing of Environment, 92: 67-83.

    Winer, N. 2003. Co-management of protected areas, the oil and gas industry and indigenous

    empowerment the experience of Bolivias Kaa-Iya del Gran Chaco. Policy Matter, 12: 181-191.

    Zhan, X., Solhberg, R.A., Townshend, J.R.G., DiMiceli, C., Carroll, M.L., Eastman, J.C., Hansen, M.C.,

    & DeFries, R.S. 2002. Detection of land cover changes using MODIS 250 m data. Remote Sensing of

    Environment, 83: 336-350.

  • 8/4/2019 Beyond Forest and Non Forest Fd

    8/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    8

    Figures

    Figure 1

    Map of the Tierras Bajas and the main elements of the Corredor Biocenico.

    A 50-km buffer north and south of the highway has been used to demarcate the horizontal boundaries of the study area.The western boundary is defined by the Andean foothills and the Pira River to the west while the eastern boundary isdefined by the Quimome River and western ranges of the Brazilian Shield.

  • 8/4/2019 Beyond Forest and Non Forest Fd

    9/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    9

    Figure 2

    Methodology for transforming 2007 MODIS NDVI data into cropland types.

  • 8/4/2019 Beyond Forest and Non Forest Fd

    10/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    10

    Figure 3

    Methodology for transforming 1986 AVHRR NDVI data into forest types.

  • 8/4/2019 Beyond Forest and Non Forest Fd

    11/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    11

    Figure 4

    2007 crop types (classified from MODIS) overlaid onto a 2006-07 CBERS 2 classification

    of forest, infrastructure, water bodies and bare ground for the Tierras Bajas study area.

    Figure 5

    1986 forest classes (classified from AVHRR) overlaid onto a 1986-89 Landsat TM classificationof cultivation, urban, bare ground, and water bodies for the Tierras Bajas study area.

  • 8/4/2019 Beyond Forest and Non Forest Fd

    12/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    12

    Figure 6

    Evolution of forest classes (classified from AVHRR)

    to crop types (classified from MODIS) from 1986 to 2007.

  • 8/4/2019 Beyond Forest and Non Forest Fd

    13/13

    XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

    13

    Tables

    TABLE 1

    Comparison of past land-use and land-cover change studies which have included some portion ofthe Tierras Bajas.Forest Loss

    AuthorTimePeriod

    Total Area(km

    2)

    Sensors Classes AnnualRate

    Total(km

    2)

    Davies 1993 1975-1991

    15,659 Landsat MSS/TM PF, SF, F, A,P

    1.2-18.2%

    2,605

    Tucker & Townshend2000

    1992-1994

    784,759 Landsat TM F, NF NA 28,208

    Steininger et al. 2001a 1975-1998

    700,000 Landsat MSS/TM F, NF ~16.7% 24,703

    Steininger et al. 2001b 1976-1998

    19,533 Landsat MSS/TM F, NF, WA,C

    ~6.0% 9,400

    Mertens et al. 2004 1989-1994

    364,000 Landsat TM F, NF 0.5-4.1% 5,117

    Killeen et al. 2008 1975-

    2004

    720,915 Landsat

    MSS/TM/ETM+

    F, S, G, WE 0.8-6.3% *45,411

    *Including scrub and savanna (9,042 km2), total vegetation loss would increase to 54,453 km2

    Classes:

    PF = Primary Forest NF = Non-Forest S = Scrubland WA = WaterSF = Secondary Forest A = Agriculture G = Grassland C = Cloud

    F = Forest P = Pasture WE = Wetland NA = Not available

    Table 2.2007 cropland types (classified from MODIS) and 1986 forest classes (classified from AVHRR)

    2007 Cropland Types Percentage Area (km2) 1986 ForestClasses Percentage Area (km2)

    Summer & Wintercropland

    67.5 7,693.4 Dry forest1.0 174.7

    Summer only cropland 28.5 3,248.7 Dry/Seasonalforest 32.2 5,729.8

    Winter only cropland 0.0 0.0 Seasonal forest 52.3 9,317.5Fallow 4.0 455.5 Wet/Seasonal 14.6 2,597.0

    -- -- -- Inundated forest 0.0 0.0TOTAL 100.0 11,397.5 TOTAL 100.0 17,818.9

    TABLE 3Evolution of forest classes to crop types (% and km

    2) from 1987 to 2007

    Summer &WinterCropland

    Summer OnlyCropland Winter OnlyCropland FallowCropland TOTAL

    % km2 % km2 % km2 % km2 % km2

    Dry Forest 0.6 50.0 0.2 15.2 0.0 0.0 0.0 0.9 0.8 66.1Dry/SeasonalForest

    18.8 1,583.8 0.7 58.6 0.0 0.0 0.7 58.6 25.5 2,143.2

    Seasonal Forest 35.5 2,987.6 14.9 1,253.6 0.0 0.0 2.4 201.6 52.8 4,442.8Wet/SeasonalForest

    14.3 1,198.8 5.7 475.6 0.0 0.0 1.0 86.2 20.9 1,760.6

    Inundated Forest 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

    TOTAL 69.2 5,820.2 21.5 1,803 0.0 0.0 4.1 347.3 100.0 8,412.7