Supplementary information to “Amazon Conservation Scenarios” Roads to be paved The most important determinant of future patterns of deforestation in the Amazon (defined as the Amazon river watershed, the Legal Amazon in Brazil, and the Guiana region) is the paving and construction of highways. Several paving projects are currently planned by the Brazilian government. A 700-km section of the BR-163 highway is slated for paving from the border of Mato Grosso and Pará states to Itaituba, linking the soy production region of Mato Grosso with the river transport system of the Amazon. Other paving projects planned for the Brazilian Amazon include the BR-230 (Transamazon Highway), BR-319 (ManausPorto Velho Highway), BR-156 from Amapá state to French Guyana, BR-401 from Roraima state to Guiana, as well as many others of secondary importance (Fig. S1). Outside of Brazil, highway paving is planned across the Andes, linking the lowland Amazon with Pacific ports including Callao in Peru and Arica in Chile. One of these highways links Assis Brasil, in Acre, Brazil, to Puerto Maldonado and Cuzco or Puno in Peru; the other would link Cruzeiro do Sul, Acre, via Pucalpa, to Lima. Paving is also planned for the road from Cárceres, Mato Grosso state in Brazil to Santa Cruz city in Bolivia (Fig. S1). Santa Cruz is a burgeoning population center located inside the Amazon Basin with economic importance even greater than the capital La Paz owing to its large natural gas fields. The Cárceres-Santa Cruz corridor would become the shortest route linking the industrial and highly populated southeastern Brazil, through its agro-business central region, to the Pacific Ocean. The impacts of proposed highway paving on land use change, migration patterns, and the human populations that already live along these roads will depend upon the effectiveness of regional planning processes currently underway1. Other infra-structure investments are also contemplated for the Amazon, including river channelization for fluvial transportation, port construction, hydroelectric plants, and gas pipelines2,3,4. Our analyses focus on the effect of road-paving on future trajectories of land-use change, since the effects of other infra-structure investments is highly uncertain. To incorporate the influence of this planned road paving into the simulations, we established a schedule of likely dates at which paving will be completed based upon analyses of government documents and conversations with government officials, setting the years of completion of 1 all roads slated for paving within the next three decades (Table S1). These new all-weather highways will exert an effect on deforestation not only augmenting the regional rates, but also initiating new deforestation frontiers. Table S1 Road paving schedule key code road name 1-1' 2-2' 3-3' 4-4' 5-5' 6-6' 7-7' 8-8' 9-9' 10-10' 11-11' 12-12' 13-13' BR-230 BR-230 BR-230 BR-163 BR-163 Transamazônica Transamazônica Transamazônica Cuiabá-Santarém Cuiabá-Santarém BR-319 BR-319 BR-210 BR-210 BR-401 BR-156 14-14' BR-158 15-15' 16-16' MT-130 17-17' BR-364 18-18' 19-19' 20-20' 21-21' GO-255 22-22' TO-280 23-23' 5N 24-24' 26 B 25-25’ 3S BR-364, 26-26’ 16B, 5N 27-27’ BR-364 tracks to be paved from Araguatins (TO) to Itupiranga (PA) from Itupiranga (PA) to BR-163 from TO-040 to GO-118 and associated tracks in MA and TO from intersection to Colíder (MT) to BR-230 (Transamazônica) from BR-230 (Transamazônica) to Santarém between Alta Floresta (MT) and BR-364, near Ariquemes (RO) Manaus-Porto Velho from 160 km south of BR-174 southwards Manaus-Porto Velho from 195 km south of BR-174 southwards from kilometer 75 to kilometer 175 from kilometer 175 from Bomfim (RR) to Queenstown (Guyana) Cayenne-Oiapoque from 150km north of Santana (AP) to Matoury (French Guiana) east of BR-158 southwards Vila Rica (MT) between 220km north of Nova Xavantina (MT) and 250km south of Redenção (PA) east of BR-158, from 220km north of Nova Xavantina (MT) from 23km north of Primavera do Leste (MT) to the intersection of MT-110 from Chapada dos Guimarães (MT) to BR-174; from 85km east of MT-170 northwards from 90km west of MT-170 northwards Cáceres-Santa Cruz between Montero and San Matias (Bolivia) from Paranã (TO) to TO-280 from GO-255 to TO-040 from 10km west of Vila Rica eastwards Tingo María (Peru) to 16 N intersection from Cuzco to Puerto Maldonado (Peru) and from 45km north Transamericana of Puerto Maldonado to Iberia (Peru) from 100km to 300km westwards Cuzco (Peru) paving completion 2008 2012 2025 2008 2008 2025 2012 2018 2008 2025 2012 2018 2008 2008 2008 2008 2012 2012 2018 2012 2008 2008 2018 2008 from Cruzeiro do Sul (AC) to Mayobamba (Peru) 2008 2018 from 90km north of Senador Guiomard (AC) to Feijó (AC); 2008 Acronyms for Brazilian States: To - Tocantins, PA - Pará, GO - Goiás, MT – Mato Grosso, RO – Rondônia, RR- Rorâmia, AP – Amapá, AC – Acre. Road paving phases comprise: 2001-2008, 2008-2012, 2012-2018, 2018-2025, 2025-2051. Town names are in italic. 2 Fig. S1 – Amazon Basin, its major cities, paved and major dirt roads, and existing and proposed protected areas. Road to be paved are indicated by numbers keyed to the paving schedule of Table S1. 3 General approach The model architecture embodies coupled models developed within two spatial structures: (1) subregions defined from socioeconomic stratification (Fig. S2) and (2) raster cells. Forty-seven subregions were defined using an anthropogenic pressure index, which was developed to measure the potential for deforestation as determined by socioeconomic and demographic growth5 An upper model projects the deforestation rates for the subregions, processing data on deforestation (Table S2), road paving (Table S1), and existing and proposed protected areas (Table S2), and passes them to a spatially explicit simulation model that uses cartographic data for infrastructure (roads, railways, gas pipelines, waterways, and ports), administrative units (state and national boundaries and protected areas), and biophysical features (topography, soil, and vegetation) within a raster grid map of 3144x4238 cells at 1 km2 resolution. Each subregion therefore has a unique spatial model with customized parameters, consisting of 1) an cellular automata type model that simulates the spatial patterns of deforestation, incorporating a probability map depicting the integrated influence of cartographic data on the location of deforestation, and 2) a road constructor model that projects the expansion of secondary road network, and thereby incorporates the effect of road expansion on the evolving spatial patterns of deforestation. We ran the model for eight scenarios encompassing 50 annual time steps starting in 2001. The base-line scenario, referred to as “business as usual” (BAU), considers the deforestation trends across the basin, projecting regional rates by using 2001-2002’s figures and their average yearly derivatives determined from 1997 to 2002 (Table S2), and adding to them the effect of paving a set of major roads. The best-case “governance” scenario also considers the paving of a set of major highways and the current deforestation trends across the basin, but now the rate projection assumes an inverted U-curve to reflect the gradual increase of governance throughout the Amazon1,6. In these scenarios, road paving follows a predefined schedule (Table S1) and its effect on accelerating deforestation is empirically estimated comparing density of deforested land with mean distance from current paved roads within Brazilian municipalities. Within the governance scenario, deforestation cannot surpass 50% of the forest cover outside of protected areas as required by governmental regulations, while in the business-as-usual scenario this limit is set to 85%. The minimum area of forest remnants in the business-as-usual (15%) and governance (50%) scenarios is 4 lower than that currently required by the Brazilian government, but we determined that these minima more realistically bracket the range of forest remnant values that will be attained (Fig. S3). Notice that Brazil is the only country to possess such restriction on deforestation on private land, although Venezuela has a moratorium on deforestation and logging for its Amazonian region – specifically the State of Amazonas7. The governance scenario also assumes that the network of protected areas will be expanded in the Brazilian Amazon as proposed in the “Áreas Protegidas da Amazônia” (Protected Areas of the Amazon) program (ARPA)8. Full protection of conservation areas is guaranteed in the governance scenario, whereas, in the business-as-usual scenario, existing protected areas may lose as much as 40% of their original forest cover due to lax enforcement9. As a result, deforestation declines as the percentage of deforested land within a subregion approaches these preset limits. Six intermediate scenarios were also run by varying the following assumptions for the extreme-case scenarios: 1) governance scenario without further road paving 2) governance scenario without the inclusion of ARPA, 3) BAU with expansion of the protected area network to include ARPA plus strict enforcement to guarantee their integral protection, 4) BAU without ARPA plus strict environmental enforcement within protected areas, and 5) BAU with ARPA in a lax environmental enforcement scenario, in which protected areas may lose as much as 40% of their original forest cover, and 6) historical, which assumes only the deforestation historical trend (Table S3). 5 Table S2 Input data for the subregions country Sub area 1 43,775 2 207,144 3 56,179 4 40,748 5 269,210 6 85,834 7 175,963 8 39,236 9 102,103 10 52,980 11 30,617 12 88,403 13 138,838 14 60,846 15 69,300 16 163,545 17 37,934 18 191,112 19 30,111 20 38,437 21 73,766 22 205,397 23 54,334 24 106,149 25 40,972 26 53,804 27 1,647,690 28 227,186 29 247,891 30 348,886 31 123,495 32 137,145 Brazil’s total 5,189,032 33 147,479 Suriname 34 184,265 Venezuela 35 116,947 Ecuador 36 215,409 Guyana 37 473,714 Peru 38 308,544 39 106,404 40 84,861 Peru’s total 973,523 Bolivia 41 63,756 42 174,898 43 214,509 44 235,287 Bolivia ’s total 688,450 Colombia 45 240,938 46 204,147 Colombia’s total 445,085 47 85,301 F. Guiana Amazon 8,045,491 Brazil forest deforested nonforest 2001 gross 2001 net annual protected pr. forest forest + ARPAS 2001 2001 deforest. deforest. derivative 30,768 10,780 2,227 515 1.67% 5.44% 1,166 22,326 189,637 8,788 8,719 274 0.14% 15.71% 107,922 167,687 2,756 21,117 32,306 383 13.89% 2.21% 304 335 1,158 3,097 36,492 11 0.94% 2.68% 63 63 162,883 56,170 50,157 1558 0.96% 19.43% 61,412 78,548 71,849 8,665 5,320 834 1.16% 8.76% 44,472 62,821 110,351 4,316 61,297 381 0.35% 6.72% 70,765 80,626 3,093 8,240 27,903 63 2.03% 2.11% 63 63 74,971 20,207 6,925 1178 1.57% 3.08% 21,482 54,648 49,804 2,098 1,078 119 0.24% 6.03% 18,631 24,875 19,978 7,512 3,128 443 2.22% 6.63% 6,275 10,365 9,157 3,896 75,350 79 0.87% 1.40% 2,804 2,821 123,394 8,366 7,078 439 0.36% 7.04% 39,049 56,894 30,976 22,462 7,408 1111 3.59% 2.73% 20,471 20,652 30,445 9,251 29,603 421 1.38% 1.03% 15,437 20,948 124,632 7,736 31,176 452 0.36% 12.27% 36,327 79,395 15,422 20,456 2,056 718 4.66% 3.95% 5,059 5,059 22,222 20,250 148,640 531 2.39% 1.94% 4,721 5,020 16,271 13,297 544 759 4.66% 23.88% 890 6,630 20,486 14,095 3,857 462 2.26% 1.86% 17,007 17,007 25,785 32,654 15,327 807 3.13% 3.73% 3,446 7,289 186,245 15,409 3,742 831 0.45% 2.99% 51,972 87,649 598 5,862 47,875 27 4.47% 0.48% 24 24 71,947 16,476 17,726 541 0.75% 8.64% 12,432 30,347 23,784 14,181 3,007 1160 4.88% 11.77% 2,515 3,130 23,459 12,623 17,722 355 1.51% 3.94% 14,859 14,901 1,481,503 27,080 139,107 1373 0.09% 6.84% 552,217 716,897 43,877 122,608 60,702 1513 3.45% 8.98% 19,313 19,313 204,721 24,898 18,272 1475 0.72% 6.54% 71,420 81,264 82,577 53,262 213,047 1598 1.93% 2.48% 19,235 27,709 73,870 34,869 14,756 2137 2.89% 2.37% 9,559 17,347 15,139 37,046 84,961 719 4.75% 1.86% 4,187 4,437 3,343,757 667,766 1,177,508 23266 0.70% 1,235,497 1,727,090 0.27% 133,119 2,086 12,274 242 0.18% 17,265 17,275 2.24% 160,130 12,776 11,359 553 0.35% 57,165 57,165 2.96% 94,745 8,540 13,663 388 0.41% 22,513 22,513 2.24% 182,233 7,390 25,786 210 0.12% 6,035 6,235 405,179 24,825 43,710 510 0.13% 0.40% 72,507 72,507 95,215 37,979 175,349 260 0.27% 2.96% 39,045 39,045 95,789 5,664 4,952 128 0.13% 0.27% 47,424 47,424 80,865 1,246 2,751 331 0.41% 2.96% 40,603 40,603 677,048 69,713 226,762 1,230 0.18% 199,578 199,578 56,278 1,358 6,120 75 0.13% 2.31% 7,739 7,739 80,315 6,475 88,108 665 0.83% 2.31% 29,845 29,845 66,075 1,301 147,133 83 0.13% 2.31% 8,919 8,919 127,955 30,187 77,144 1,077 0.84% 2.31% 14,781 14,781 330,623 39,322 318,505 1,900 0.57% 61,284 61,284 231,848 511 8,579 292 0.13% 0.40% 95,478 95,478 158,658 28,791 16,698 650 0.41% 2.96% 44,205 44,205 390,506 29,302 25,276 942 0.24% 139,683 139,683 78,760 285 6,257 0.27% 143 0.18% 0 0 5,390,921 837,180 1,817,389 28,882 0.54% 1,739,024 2,230,901 areas in km2, annual derivative (∆fdt.) is an average calculated from the difference between 1997-2000, 2000-2001, and 2001-2002 annual deforestation rates. 6 Fig. S2 – Stratification of the Amazon Basin, depicting annual deforestation and forest decline from 2001 to 2050 forecasted for the subregions within the BAU scenario. Numbers are keyed to subregions’ data in Table S2. Table S3 Scenario assumptions assumptions scenarios governance (GOV) governance without further road paving governance without ARPAS BAU with ARPAS, strict enforcement BAU without ARPAS, strict enforcement BAU with ARPAS, lax enforcement historical (no further road paving) business-as-usual (BAU) road paving pressure added to the deforestation trend yes no yes yes yes yes no yes ARPA degree of minimum % rates rates included protection of forest projected asymptotically in for protected reserve on by using projected by protected areas private land yearly using yearly areas derivatives derivatives yes 100% 50% no yes yes 100% 50% no yes no 100% 50% no yes yes 100% 15% yes no no 100% 15% yes no yes 60% 15% yes no no 60% 15% yes no no 60% 15% yes no 7 Fig. S3 – Density of deforested land % (deforested land/(municipality’s area - nonforest)), deforestation density % (deforestation/municipality’s area), and anthropogenic pressure index for the Brazilian Amazon’s municipalities. Land cover change from PRODES10. 8 Basin stratification Deforestation rates vary greatly across the basin due to regional differences, including soils, climate, socioeconomic organization, government systems, public policies, environmental laws and degree of enforcement, population characteristics and dynamics, as well as types and age of frontiers. Thus, it is unrealistic to employ a model that projects a single deforestation rate for the entire basin. Instead, the Amazonian basin must be stratified into subregions representative of a network of cities and their surrounding zones of influence. In light of Christaller’s central place theory11, we interpret the geographical organization of the Amazon as a hierarchy of regions attracted to central urban markets that possess the greatest supply of services and thus higher economic potential. To address this issue, we developed a method for stratifying the Brazilian Amazon into subregions, which utilizes a synthetic anthropogenic pressure index, tertiary economy level, and regional migratory fluxes5. This stratification is developed by first classifying the municipalities according to their intrinsic anthropogenic pressure, an index we developed to measure the potential for deforestation as determined by socioeconomic and demographic growth12. It is calculated by applying the Grade of Membership (GOM) fuzzy classification method13 to demographic, socioeconomic and agriculture census data, such as population density and growth rate, urbanization level and rate; gross domestic products, municipal income taxes and budget; number and types of agricultural implements; production from animal husbandry, agriculture, and forestry; and; education, habitation and health parameters. These data were stratified into a five-dimensional space, with axes that we named: (1) demographic concentration and dynamics; (2) economic development; (3) agrarian infrastructure; (4) agricultural and timber production; and (5) social development, which were combined to produce the anthropogenic pressure index for each municipality (Fig. S3). A positive effect on the anthropogenic pressure index is ascribed for the first four dimensions, and a negative effect for the fifth. In a second step, regional development centers were identified and ranked with respect to their supply of services14, referred to here as the “tertiary economy”, as follows. 9 TI i = TDPi * (1 − e GDP i ln( 0.05) *TDP i TDP ref ) (1) where TIi is a ratio between the tertiary economy domestic product (TDPi) and the gross domestic product (GDPi) of a municipality i, standardized by a reference tertiary economy domestic product (TDPref), specifically the largest regional TDPi. Once a hierarchy of regional poles is established, which can include a varying number of economic centers depending on a chosen cut-off threshold, the interaction between a center and a municipality is calculated by the following equation: Iv IJ = PI (1 + API I ) * PJ (1 + API J ) d ij ξ (2) where Ivij represents the gravitational interaction between center i and municipality j, given by their populations (Pi and Pj) and anthropogenic pressure indices (APIi and APIj), weighted by the distance between them raised to the power of ξ, an attrition coefficient so that: ⎞ ⎛ ln(0.001) ⎜ *vmt IJ ⎟⎟ ref ⎠ ξ = 1 + e⎜⎝ vmt (3) where vmtij is the overall migratory flux between pole i and municipality j and vmtref is the reference migratory flux, namely the largest intermunicipal migratory flux. Thus Ivij measures the dependence of a municipality upon a regional center defined as the attraction exerted by the center’s population plus its anthropogenic pressure. This dependence is strengthened by two-way migratory fluxes and weakened by the geographical distance. The stratification is achieved by assigning to a particular regional center all municipalities where its respective Ivij is greatest. As shown in Fig. S2, the regionalization map for the Brazilian Amazon is comprised of 32 subregions, to which were added 15 additional subregions, defined for the other countries based only on geographical criteria of contiguity and basin interiority due to paucity of census data. 10 Data for deforestation projections Input data for each subregion consist of deforestation rate and its average annual derivative, as well as areal extent of remaining forest, deforested land and protected areas (Table S2). Land cover map for the entire basin is a composite of 2001’s PRODES10, 2000’s SPOT Vegetation Map of South America15, classified 2001’s MODIS vegetation continuous field16, and Bolivia deforestation maps17. For the Brazilian Amazon, PRODES data from three time-periods (1997-2000, 2000-2001, 2001-2002) were employed to derive the 20012002’s deforestation rates and their average annual derivatives within the 1997-2002 period. For Bolivia’s subregions, 2001-2002’s deforestation rates and their annual derivatives were extrapolated from data compiled18 from two deforestation mapping projects17,19. Because systematic deforestation map series are not available for the remaining subregions that occur in other countries, deforestation rates and their annual variation were assigned by applying figures from subregions of Brazil that were considered similar in frontier type and age (Table S2). The deforestation projection model This upper model was implemented in VENSIM, a system-thinking software20. This model is designed to project deforestation for each subregion, processing data on historical deforestation, road paving, and existing and proposed protected areas (Tables S1 and S2). Therefore, it generates the deforestation scenarios under which the lower spatial simulation model runs. Deforestation, at a time t, for a basin’s subregion is calculated as follows: deforestationt = forest t * fd t (4) where forestt is the remaining forest within each subregion and fdt is the net deforestation rate at a time t such that: fd t = fd t −1 (1 + ∆fd t * (1 + acc _ f * ddratiot ) * sat t (5) Initial net deforestation rate (fd2001/2002) is obtained dividing 2001-2002’s deforestation by 2001’s remaining forest (Table S2). The term ∆fdt represents the average annual derivative of the deforestation rate; acc_f is a constant, between 0 and 1, used to impose a delay in adjusting the deforestation rates in response to the surging pressure coming from road paving, as represented in the equation by ddratiot. Thus, we designed the net deforestation estimate to incorporate a time-lag between the completion of road paving within a 11 subregion and the deforestation that it stimulates. Simulations employ acc_f =4.3, set to make the BAU projection approximate the average forecast deforestation of sensitivity analysis (Fig. S4). In turn, the term satt represents an asymptotic saturation factor, as described in equation (10), and the annual derivative of the deforestation rate (∆fdt ) is given as follows: ∆fd t = lg_f t * ran(h _ ∆fd , abs(mean _ ∆fd − h _ ∆fd )) (6) business historical 50% 75% 95% 100% 60,000 45,000 30,000 15,000 0 2001 2013 2026 Time (Year) 2038 2050 Fig. S4 - Deforestation forecast for the Brazilian Amazon. Output from sensitivity analysis varying forest remnant percentage from 0.1 to 0.2, percentage of protected forest core from 0.6 to 0.8, and acc_f from 0 to 1. acc_f = 0.43 was set to approximate the mean forecast deforestation (black line). Due to few time periods available to estimate the deforestation rate derivative, subregion’s values for this variable were approximated to a regional mean (mean_∆fdt), used together with its historical average (h_∆fdt , Table S2) as input parameters - mean and variance - to a random number generating function – ran. For the Brazilian subregions, regional mean annual variations were derived from PRODES deforestation data for the Brazilian states from 1997 to 200210, and 1.6% was used for all the other countries. In the business-as-usual scenario, the logistic factor (lg_ft) is set to 1, leaving ∆fdt constant with only minor random oscillations. For the governance scenario, ∆fdt is projected using lg_ft output by a logistic 12 curve, that varies as a function of time t (equation 7). In this manner, the model considers that all the measures incorporated within this scenario1 gradually reduce the current deforestation trend. lg _ f t = −1 + 0.076 (1 + exp(0.067 * (−2050 + t ) ) 7) The effect of paving a major road through a subregion on its deforestation rate is expressed by the term ddratiot, which is a ratio between an expected density of deforested land owing to the average proximity to paved road and the subregion’s current density of deforested land. dd ratiot = exp_den_def deforested t /(forest t+deforested t ) (8) where deforested and forest represent, respectively, the current areal extents of deforested land and remaining forest for a subregion. The term exp_den_def stands for the density of deforested land expected to occur within a certain subregion, incorporating its mean distance to a paved road (Fig. S5), which is preset according to the model’s sequence of road paving (Table S1). A regression analysis supplies the coefficients for equation (9), in which mean_d2paved_road is the mean distance to paved road in kilometers for a subregion. In this manner, road paving produces an upward effect on deforestation, since mean distance to paved road diminishes over time as new road tracks are paved (Fig. S1). exp _ den _ def = 1 /(0.050873)* mean _ d 2 paved _ road + 1.00762) (9) The asymptotic saturation factor (satt) in equation (5) is calculated as follows: sat t = forest t − min _ forest forest 2001 + min _ forest * forest t + min _ forest forest 2001 − min _ forest (10) where forestt and forest2001 represent, respectively, the extent of remaining forest for time t and 2001 and min_forest is given by: min _ forest = % for _ re * ( forest 2001 − prot _ for ) + % prot _ for _ core * prot _ for (11) The asymptotic saturation factor (satt) is introduced in order to compute the influence of protected areas (prot_for) and the minimum percentage of forest remnants (%for_re), as preset for each scenario, in slowing deforestation. 13 In sum, the governance scenario departs from the business-as-usual scenario through: (1) the minimum percentage of forest remaining outside of protected areas, reflecting a range of both government land use policies and their enforcement; (2) the expansion, or not, of the Brazilian protected area network to include ARPA; (3) the enforcement of protected areas; and (4) the gradual reduction of deforestation rates below historical rates (Table S3, Figs. S6 and S7). Fig. S5 – Percent of deforested land as a function of distance to paved roads, derived for Brazilian Amazon’s municipalities using PRODES 2001 and mean distance to current paved roads. Brazil’s deforestation 40,000 30,000 20,000 10,000 0 2001 2008 2015 2022 2029 2036 2043 2050 Time (Year) GOVERNANCE WITHOUT ROAD PAVING GOVERNANCE WITH ROAD PAVING BUSINESS-AS-USUAL HISTORICAL TREND (WITHOUT ROAD PAVING) Fig. S6. Forecast deforestation for the Brazilian Amazon for various scenarios. 14 total deforestation forecasted by 2050 2,000,000 Brazil Amazon km2 1,500,000 1,000,000 500,000 0 governance governance governance BAU w ith BAU w ithout BAU w ith w ithou w ithout ARPAS, ARPAS, ARPAS, lax further ARPAS strict strict enforcement paving enforcement enforcement historical business-asusual (BAU) scenarios 70% % of deforestation reduced compared with the one of BAU for 2050 60% Brazil Amazon 50% 40% 30% 20% 10% 0% governance governance w ithou further paving governance BAU w ith BAU w ithout BAU w ith w ithout ARPAS, strict ARPAS, strict ARPAS, lax ARPAS enforcement enforcement enforcement historical business-asusual (BAU) scenarios Fig. S7 – Total new deforestation* forecasted by 2050 for 8 scenarios and percent of deforestation reduced for each scenario by 2050 using the BAU scenario as a base line. *Because the resolution, quality and availability of data sets vary greatly across the basin with Brazil having much better data available than the other countries, the model’s results should be viewed as average thresholds that may be reached over the analyzed period rather than as absolute figures. 15 Spatially explicit simulation The spatial model aims to simulate the evolving spatial patterns of deforestation taking into consideration proximate and biophysical variables21. Spatially-explicit simulations of deforestation therefore attempt to quantify and to integrate the influences of variables, representing biophysical, infrastructure, and territorial features (e.g. topography, rivers, vegetation, soils, climate, proximity to roads, towns and markets, and land use zoning), on the spatial prediction of deforestation9. To incorporate these spatial variables into the simulation, we have developed a cartographic database consisting of a land cover map and ancillary cartographic layers structured into one subset of static data layers and a second subset of dynamic data layers (Fig. S8). The land cover map for the entire basin, used as the initial landscape in the simulation, is a composite of 2001’s PRODES10, 2000’s SPOT Vegetation Map of South America15, classified 2001’s MODIS vegetation continuous field16 and 1993’s Bolivia deforestation map17. For Brazil, PRODES 2001 map, at an original resolution of 60 meters, was vectorized and stamped on a 1-km2-resolution raster. This procedure ensured the capture of fine spatial patterns of deforested land with only minor distortion (Fig. S9). The same procedure was applied to Bolivia data and the resulting composite map was either updated or data gaps were filled in with the SPOT Vegetation map and a deforested mask derived from the 2001’s MODIS vegetation continuous field. Finally, a non-forest mask, obtained from vegetation maps22,23, was laid over the land cover map composite. Dynamic data layers include: distance to previously deforested land, distance to non-paved roads, and distance to paved roads. Hence, the cartographic database comprises two layers of roads, one of non-paved and another of paved roads. Due to its semi-dynamic character, the latter variable was represented by five different layers depicting sequential phases of road paving, as defined by the simulation paving schedule (Table S1). Roads compiled from various sources (Table S4) were updated by visual interpretation of ortho-rectified Landsat images made available by Tropical Rain Forest Information Center (TRFIC)24. Using this database, spontaneous roads (also known as endogenous) were extensively mapped for all the Brazilian Amazon and added to the non-paved road layer. 16 Static data layers include soil and vegetation maps, an urban attraction factor, altitude, slope, distance to major rivers, distance to gas lines and railways, and protected areas. Soil, vegetation, gas lines, railways, hydrographic and topographic data come from various sources (Table S4). Soil and vegetation layers are composites of the more detailed available data. Urban attraction factor is meant to represent the influence of urban centers on deforestation and was calculated using a unidirectional gravity-type model, as follows. Uai , j = ∑n Popn d 2 i, j (12) where Uai,j is the urban attraction in a rural cell i,j, exerted by summing the populations from all urban centers (Popn) in the basin and surrounding major South American cities, weighted by their distances (d) to the rural cell i,j. Protected areas include national and state natural reserves, conservation units, parks and indigenous reserves25. For the governance scenario, proposed protected areas by ARPA8 program were added to the exiting network. Fig. S8 - Input, derived, and simulated maps with respect to the spatial model architecture. 17 Table S4 Cartographic data used in the simulation Layer land cover (landscape) subregion boundaries roads major rivers topography towns & population pipelines & railways soil vegetation protected areas Brazil INPE (2004); Eva et al. (2002); UMD/GLCF (2004) IBGE (2000a) GuiaQuatroRodas (2004) ESRI (1992); WHRC/IPAM/CABS (2002) NASA/USGS (2004) IBGE (2000a, b) WHRC/IPAM/CABS (2002) MNE (1973) MNE (1973) MMA (2004); IUCN & UNEP (2003) 60 m other countries Eva et al. (2002); UMD/GLCF (2004); Steininger et al. (2001) ESRI (2002); WHRC/IPAM/CABS (2002) MM (2001); Berndtson & Berndtson (2001). ESRI (1992) NASA/USGS (2004) FRG (2003) WHRC/IPAM/CABS (2002) FAO (1998) WWF (2002) IUCN & UNEP (2003) 1 km2 Fig. S9 –Land cover patterns at 60 meters and 1 km2 resolutions. Simulation platform The spatially explicit simulation runs on DINAMICA software1,26,27. Among other features, DINAMICA incorporates the concept of phase – defined as a set of time steps with its own parameters. Following the highway paving schedule (Table 1), the simulation time span was divided into five phases, uptading for each new phase the “distance to paved road” layer. Because deforestation tends initially to concentrate in corridors along major roads1, the expansion of the paved road network drives deforestation to new radiating axes as a new phase begins. Geographical analyses of deforestation demonstrate that deforestation is both spatially and temporally autocorrelated28,29,30. DINAMICA embodies this feedback effect through the calculation of dynamic variables, i.e. input variables that are updated after each iteration. 18 Three types of dynamic variables are included: frontage distance to land cover class, sojourn time, and distance to roads. The first is used to calculate and update the layer “distance to deforested land”, the second is applied to track time since the cell has changed its state, and the third is output from the road constructor model, a component that drives the expansion of the secondary road network1. As input, the road constructor model employs a layer of all existing roads, a friction map – used to derive an accumulated transport cost surface -, and an attractiveness map that indicates the areas most likely to be reached by new roads (Fig. S8). In this way, the simulation model also incorporates the effect of road expansion on the evolving spatial patterns of deforestation. Road expansion was set to occur in pulses, first by forming lengthy access routes and then a network of side roads, generating as a result the typical fishbone structure commonly found in the Brazilian Amazon. By setting the parameters: 1) average road segment length per step, and 2) number of map quadrants used to plot road terminations, the road constructor model was adjusted to simulate a road network that slightly exceeds the deforestation front. Spatial variables can be used to calculate probability (also referred to as “favorability”) maps of deforestation. Previous analytical modeling of tropical deforestation included mostly methods such as multivariate linear regression31,32, logistic regression28,33,34,35,36 or Weights of Evidence1. We applied the Weights of Evidence method to analyze the effects of spatial variables on the location of deforestation9. This analysis was conducted out in 12 case study areas representative of different types of Amazonian agricultural colonization frontiers, each one comprising a Landsat/TM scene. (Fig. S10). The Weights of Evidence models were assessed comparing simulations of 1977-2000 deforestation, with the favorability maps as input, with PRODES 2000’s map using image similarity tests based on a fuzzy multiple resolution comparison9. Deforestation simulated using Weights of Evidence models achieved spatial agreements up to 83% within a window size of 5x5 cells9, ≈ 1.25 km of resolution. In general, the Weights of Evidence analysis showed that deforestation is attracted by urban centers, avoids both low flooded terrain and elevated and steep slopes; is not influenced by soil quality and vegetation type, and does not necessarily follow the major river network. 19 Of special interest, this analysis identified the variables “distance to previously deforested land” and “distance to roads” (including paved and non-paved) to be the strongest predictors of deforestation and demonstrated the importance of indigenous reserves on deterring deforestation along the active frontier. The supplied weights of evidence coefficients (Fig. S10) were employed to calculate a favorability map of deforestation input for the simulation at the basin level -, as follows: e P(i ⇒ j ( x, y ) V ) = ∑ W kni ⇒ j (V ) xy k 1+ ∑e (13) ∑ W kni ⇒ j (V ) xy k ij where V is a vector of k spatial variables, measured at location x,y and represented by its weights W+ 1xy, W+ 2xy, ..., W+ nxy, being n the number of categories of each variable k. In this way, weights of evidence are assigned for categories of each variable represented by its cartographic layer. Spatially-explicit simulations were performed for a subset of scenarios processed by the upper projection model, using a land cover map composed of 3144x4238 cells of 1 km2 resolution, and the model was set to run for a time span of 50 annual time steps starting at 2001. This scenarios subset included the two-extreme case scenarios: 1) governance and 2) business-as-usual (BAU), plus 3) governance without further road paving and 4) historical scenarios. For the BAU scenario, we ran three additional spatial simulations, each isolating the specific effect of paving the following highways: 2) Br-163, 3) Transamericana, and 4) Manaus-Porto Velho, the two first highways with paving completion by 2008 and the latter by 2010. See detailed results of these runs on www.csr.ufmg.br/simamazonia. To avoid the time-consuming calibration process - since simulation duration increases exponentially as a function of the cell resolution (e.g., 2 hours and 42 minutes for 2 km2 and 9 hours and 55 minutes for 1-2 km2 mixed resolutions on a 3.0 gigahertz PC processor) -, DINAMICA can employ multiple resolution data in a single simulation run. In this way, the input data were divided into two groups according to their resolutions: landscape, sojourn time and road layers at 1 km2 and static, friction, attractiveness layers at 2 km2. Besides increasing the performance of some procedures such as the calculation of distance 20 to roads, this feature also reduces the amount of memory required to load the data set. Because DINAMICA’s vicinity-based transition functions are set in hectares, the simulation could be fine-tuned using a multi-scale approach, first at 2-km cell and thereafter applying the final simulation script on the 1 km2 data. As described above, the simulation is structured in two spatial levels: raster and subregions. Using the concept of subregions, DINAMICA retrieves from a single database a subset of data to perform the subregion’s simulation. Each subregion has a unique spatial model with customized parameters, including the coefficients for the Weights of Evidence, transition rates, density and length of road segments per step, and average size and shape of the deforestation patches to be formed by the cellular automata transition functions. This allows a very flexible way to set and conduct simultaneous simulations across the various regions of the Amazon basin. Spatial integration between subregions are attained by computing chorographic variables, such as distances to roads and to previously deforested land, continuously over the entire landscape raster map. In sum, the spatially explicit simulation model features multi-scale vicinity-based transition functions, the concept of phases and subregions, the use of data at various resolutions, feedback through the calculation of dynamic spatial variables, linkage between cellular automata and system thinking software, computation of spatial transition probabilities using Weights of Evidence method, and a component that drives the expansion of the road network. Additional information on the model and its results are available at: www.csr.ufmg.br/simamazonia. 21 weights of evidence 2 0 -2 -4 -6 -8 nonprotected State conservation unit Federal conservation indigenous reserve unit protected areas weights of evidence 2 0 -2 -4 -6 -8 0:0 . 2 0.2 : 3 2.5 2: 1. 0. 0. 0. :3 : 4 0.5 5:0.7 7:0.9 9:1.5 5:2 2.5 4:5 5:6 6:8 8:1 0 distance to roads (km) > 1 10 :12 2 :15 15 weights of evidence 2 0 -2 -4 -6 -8 0:0 . 5 0.5 : 1: 0. 0.7 75:1 1.5 5 1.5 : 2 2:2 . 5 2.5 : 3 3:4 4:5 5:7 7:1 0 > 2 1 10 :15 5 :20 0 :30 30 distance to deforested (km) 22263 22467 22565 22763 22765 22861 22967 23068 23167 23267 22363 average 22863 Fig. S10 – Weights of evidence derived for the case study areas. Numbers refer to orbitpoints of Landsat scenes. 22 Disaggregated effects of highway paving We disaggregated the effects of highway paving by running the model for three additional scenarios, which incorporate all the business-as-usual assumptions, but now have only a single highway paved: 1) the BR-163 (Cuiabá-Santarém) and 2) the Transamericana (Assis Brasil-Cuzco) paved in the year 2008, and 3) the Manaus-Porto Velho highway paved in the year 2010. The influence of paving each highway alone is analyzed, comparing deforestation forecast within their areas of influence with the results of the historical (current trend but no further road paving), governance without further road paving, and governance scenarios (Fig. S11-S13 and tables S5-S7). 2030 Governance no further paving Governance with paving Historical Business as usual Historical Business as usual 2050 Governance no further paving Governance with paving Deforested by 2003 Simulated deforestation Roads by 2003 Simulated roads Forest Protected forest Non Forest Water Fig. S11- Results for BR-163 area of influence. 23 2030 Governance no further paving Governance with paving Historical Business as usual Historical Business as usual 2050 Governance no further paving Governance with paving Deforested by 2003 Simulated deforestation Roads by 2003 Simulated roads Forest Protected forest Non Forest Water Fig. S12 - Results for InterAmericana highway (Assis Brasil- Cuzco) area of influence. 24 2030 Governance no further paving Governance with paving Historical Business as usual Historical Business as usual 2050 Governance no further paving Governance with paving Deforested by 2003 Simulated deforestation Roads by 2003 Simulated roads Forest Protected forest Non Forest Water Fig. S13 - Results for Manaus - Porto Velho highway area of influence. 25 Table S5 Summary statistics for BR - 163 area of influence Deforested Forest Non forest Deforested Forest Non forest Deforested Forest Non forest Deforested Forest Non forest land cover (km2) net change % reduction by GO* 2003 2030 2050 2030 2050 2030 2050 Business as usual, Br-163 Highway paved in 2008 74,747 247,182 399,992 231% 435% -65% -73% 500,668 335,079 182,270 -33% -64% -67% -75% 44,531 37,685 37,685 Historical trend, no further road paving 74,747 217,415 372,580 191% 398% -57% -71% 500,668 364,846 209,681 -27% -58% -60% -73% 44,531 37,685 37,685 Governance plus road paving 74,747 135,618 161,069 81% 115% 0% 0% 500,668 446,643 421,192 -11% -16% 0% 0% 44,531 37,685 37,685 Governance no further road paving 74,747 128,759 153,203 72% 105% 13% 10% 500,668 453,502 429,058 -9% -14% 15% 11% 44,531 37,685 37,685 * Reduction in Governance scenario of total deforestation and forest decline. Table S6 Summary statistics for InterAmericana highway (Assis Brasil- Cuzco) area of influence Deforested Forest Non forest Deforested Forest Non forest Deforested Forest Non forest Deforested Forest Non forest land cover (km2) net change % reduction by GO* 2003 2030 2050 2030 2050 2030 2050 Business as usual, Interamericana paved in 2008 33,044 112,369 212,400 240% 543% -30% -53% 817,299 738,263 638,232 -10% -22% -30% -53% 59,886 59,597 59,597 Historical trend, no further road paving 33,044 94,773 179,856 187% 444% -10% -42% 817,299 755,859 670,776 -8% -18% -10% -42% 59,886 59,597 59,597 Governance plus road paving 33,044 88,535 117,673 168% 256% 0% 0% 817,299 762,097 732,959 -7% -10% 0% 0% 59,886 59,597 59,597 Governance no further road paving 33,044 817,299 59,886 78,706 771,926 59,597 106,587 744,045 59,597 138% -6% 223% -9% 22% 22% 15% 15% * Reduction in Governance scenario of total deforestation and forest decline. 26 Table S7 Summary statistics for Manaus - Porto Velho highway area of influence Deforested Forest Non forest Deforested Forest Non forest Deforested Forest Non forest Deforested Forest Non forest land cover (km2) net change % reduction by GO* 2003 2030 2050 2030 2050 2030 2050 Business as usual, Manaus Porto Velho Highway paved in 2010 42,806 216,586 443,278 406% 936% -39% -56% 668,485 502,992 276,300 -25% -59% -41% -57% 49,782 41,496 41,496 Historical trend, no further road paving 42,806 176,743 354,315 313% 728% -21% -43% 668,485 542,835 365,263 -19% -45% -22% -44% 49,782 41,496 41,496 Governance plus road paving (see road schedule) 42,806 148,730 220,194 247% 414% 0% 0% 668,485 570,847 499,384 -15% -25% 0% 0% 49,782 41,496 41,496 Governance no further road paving 42,806 122,601 184,027 186% 330% 33% 26% 668,485 596,977 535,551 -11% -20% 37% 27% 49,782 41,496 41,496 * Reduction in Governance scenario of total deforestation and forest decline. Table S8 Forest loss per watershed losses by 2050 for simulation runs Area (km2) Low Amazon Coastal Amapa Coastal Guiana Coastal Marajo Coastal Maranhao Coastal Para Madeira Negro Orinoco Solimoes Tapajos Xingu 362,230 96,583 439,533 220,872 287,036 original current BAU histori Gover GO no forest all cal nance further (km2) roads (GO) paving 315,431 4% 52% 34% 25% 19% 82,702 6% 63% 55% 22% 22% 402,468 2% 16% 12% 13% 11% 177,609 9% 79% 81% 27% 27% 161,166 78% 92% 92% 82% 82% 137,992 130,391 1,362,132 971,950 769,046 638,116 234,222 214,869 2,156,425 1,887,060 476,754 362,669 614,573 523,302 65% 15% 3% 8% 6% 22% 15% 93% 58% 29% 28% 25% 79% 64% 94% 51% 24% 27% 23% 73% 62% 81% 36% 13% 23% 16% 44% 24% 81% 33% 11% 21% 15% 42% 24% 27 Table S9 Forest losses per ecoregion ECOREGION Apure/Villavicencio dry forests Bolivian Yungas Caqueta moist forests Chiquitania dry forests Cordillera Oriental montane forests Eastern Cordillera Real montane forests Guayanan Highlands moist forests Guianan moist forests Japura/Solimoes-Negro moist forests Jurua/Purus moist forests Madeira/Tapajos moist forests Magdalena Valley montane forests Marajo Varzea forests Maranhao Babacu forests Maranon dry forests Mato Grosso tropical dry forests Napo moist forests Negro/Branco moist forests Orinoco Delta swamp forests Paramaribo swamp forests Peruvian Yungas Purus/Madeira moist forests Rio Negro campinarana Solimoes/Japura moist forests Southwestern Amazonian moist forests Tapajos/Xingu moist forests Tepuis Tocantins-Araguaia/Maranhao moist forests Tumbes/Piura dry forests Uatuma-Trombetas moist forests Ucayali moist forests Xingu/Tocantins-Araguaia moist forests 2 Area original current forest forest 9,744 7,249 3,569 90,544 74,563 70,410 187,852 179,175 170,696 185,542 147,741 95,353 current losses 51% 6% 5% 35% Losses by 2050 for simulation runs Transa Manau BAU Gover GO no histori merica Br163 s Porto all nance further cal na Velho roads (GO) paving 90% 85% 90% 82% 88% 80% 78% 41% 41% 41% 42% 39% 26% 26% 14% 16% 14% 16% 14% 11% 11% 70% 71% 71% 73% 71% 59% 59% 11,902 10,630 8,964 16% 81% 71% 83% 64% 75% 69% 66% 61,229 53,738 44,321 18% 55% 56% 58% 55% 52% 48% 43% 202,349 193,435 188,496 479,064 453,322 445,087 3% 2% 23% 19% 24% 19% 24% 19% 24% 21% 20% 17% 11% 12% 9% 10% 269,167 249,012 246,750 242,958 237,529 236,274 661,984 599,873 486,095 1% 1% 19% 16% 14% 69% 16% 14% 69% 15% 15% 67% 14% 14% 71% 11% 10% 61% 6% 4% 37% 5% 2% 34% 626 82,252 105,771 14,463 414,006 248,389 169,326 3,893 7,724 182,147 174,016 80,863 168,227 209 44,119 18,800 3,148 205,525 216,522 148,828 3,549 5,252 68,329 156,840 51,539 164,039 43% 15% 80% 61% 36% 7% 5% 0% 23% 31% 4% 1% 0% 83% 87% 97% 96% 75% 21% 29% 0% 87% 46% 71% 14% 4% 83% 83% 97% 94% 76% 20% 29% 0% 88% 46% 73% 17% 4% 83% 83% 97% 94% 75% 21% 28% 0% 88% 46% 77% 17% 5% 80% 80% 97% 94% 75% 20% 29% 0% 79% 44% 74% 13% 4% 80% 83% 97% 98% 74% 21% 26% 0% 84% 45% 62% 11% 3% 83% 26% 89% 90% 50% 16% 14% 0% 83% 42% 40% 4% 5% 80% 25% 89% 89% 50% 17% 13% 0% 82% 42% 34% 0% 3% 808,919 737,147 709,470 336,575 323,669 297,501 29,845 28,367 26,742 4% 8% 6% 27% 68% 33% 27% 71% 27% 27% 68% 31% 27% 71% 28% 23% 66% 23% 16% 23% 21% 14% 22% 20% 193,642 180,198 57,367 1,474 577 111 472,996 430,227 410,647 114,654 108,183 93,721 68% 81% 5% 13% 92% 91% 57% 35% 92% 91% 58% 35% 92% 91% 56% 36% 92% 89% 60% 37% 92% 94% 48% 35% 81% 89% 27% 34% 81% 87% 23% 34% 269,485 249,119 172,868 31% 76% 76% 77% 75% 76% 43% 43% 368 52,094 95,699 8,142 322,516 232,419 157,057 3,561 6,802 98,424 163,796 51,910 164,508 area figures in km 28 Mammals Analysis – Brief Methodology A total of 442 non-flying terrestrial mammal species had available Western Hemisphere spatially-explicit geographic range data from NatureServe with ≥1% of their range within the panAmazonian region37. We selected 382 non-flying terrestrial mammal species (comprising 28 families; 116 genera) with ≥ 20% of their ranges within the Amazon (median 95%: Range 100%-20%) (Fig. S14 and Table S10). For each species, the scenarios model results generated for BAU 2050 and GOV 2050 were applied to determine change in forest cover within each species-specific range. Areas of water were excluded from these baseline calculations. All calculations were performed using IDRISI Kilimanjaro software38. Fig. S14 - Overlay of all mammals with >= 20% of range within the Amazon (n=382). 29 Table S10 Number of imperiled Amazon mammals under business-as-usual (BAU) and Governance (Gov) scenarios of future land use. Three hundred eighty-two species were examined, each with at least 20% of their range in the Amazon region. ‘Imperiled’ mammal species (losing ≥40% of forests within their Amazon range) are summarized for each family and scenario. Under the BAU 2050 scenario, 105 species will become imperiled, vs. 41 under the Governance scenario. Family Aotidae Atelidae Bradypodidae Callitrichidae Caluromyidae Canidae Cebidae Cervidae Cuniculidae Dasypodidae Dasyproctidae Didelphidae Dinomyidae Echimyidae Erethizontidae Felidae Hydrochaeridae Leporidae Marmosidae Megalonychidae Muridae Mustelidae Myrmecophagidae Pitheciidae Procyonidae Sciuridae Tapiridae Tayassuidae Total total spp. 7 15 2 28 3 5 10 6 2 11 9 5 1 38 8 9 1 1 26 2 129 5 3 34 6 12 2 2 382 % of all mammals 1.8 3.9 0.5 7.3 0.8 1.3 2.6 1.6 0.5 2.9 2.4 1.3 0.3 9.9 2.1 2.4 0.3 0.3 6.8 0.5 33.8 1.3 0.8 8.9 1.6 3.1 0.5 0.5 100 number of imperiled species imperiled species % of total number of imperiled species imperiled species % of total BAU 2050 2 6 1 16 1 0 3 0 0 2 3 1 0 14 4 1 0 0 9 0 21 1 1 14 0 4 1 0 105 BAU 2050 1.9 5.7 0.9 15.2 1.0 0.0 2.8 0.0 0.0 1.9 2.8 1.0 0.0 13.3 3.8 1.0 0.0 0.0 8.6 0.0 20.0 1.0 1.0 13.3 0.0 3.8 1.0 0.0 100 Gov 2050 0 3 0 8 0 0 1 0 0 0 1 0 0 5 1 0 0 0 2 0 11 0 0 7 0 1 1 0 41 Gov 2050 0.0 7.3 0.0 19.5 0.0 0.0 2.4 0.0 0.0 0.0 2.4 0.0 0.0 12.2 2.4 0.0 0.0 0.0 4.9 0.0 26.8 0.0 0.0 17.1 0.0 2.4 2.4 0.0 99.8 30 References 1. Soares-Filho, B.S. et al. Simulating the response of land-cover changes to road Paving and governance along a major Amazon highway: The Santarém-Cuiabá Corridor. Global Change Biology. 10, 745-764 (2004). 2. Laurance, W.F. et al. The future of the Brazilian Amazon. Science, 291, 438-439 (2001). 3. Carvalho, G., Barros, A.C., Moutinho, P., Nepstad, D. Sensitive development could protect Amazonia instead of destroying it. Nature. 409, 131 (2001). 4. Nepstad, D. et al. Avança Brasil, the Environmental Costs for Amazônia. (2000). http://www.ipam.org.br/avanca/participen.htm (2004). 5. Garcia, R.A., Soares-Filho, B.S., Sawyer, D.O. Dimensões sócio-econômicas e movimentos populacionais: uma regionalização da Amazônia brasileira. In: XIV Encontro Nacional de Estudos Populacionais (ABEP. Caxambu, Brasil). http://www.abep.org.br/usuario/GerenciaNavegacao.php?caderno_id=030&nivel=1 (2004). 6. Nepstad, D. et al. Frontier governance in Amazonia. Science 295, 629-631 (2002). 7. Bevilacqua, M. et al. The State of Venezuela Forests: A Case Study of the Guayana Region. Fundación Polar; WRI: World Resources Institute; Observatorio Mundial de Bosques. Venezuela (2002). 8. Montiel, F. Programa Áreas Protegidas da Amazônia arpa. Resumo Executivo http://www.mma.gov.br/port/sca/ppg7/doc/arpareex.pdf (2004). 9. Soares-Filho, B.S, Rodrigues, H, Nepstad, D., Voll, E., Alencar, A. Analysis of spatial determinants of deforestation in Amazonia. EarthInteractions (Submitted). 10. INPE - Instituto Nacional de Pesquisas Espaciais. Monitoramento da Floresta Amazônica Brasileira por Satélite - Projeto PRODES. São José dos Campos, BR) http://www.obt.inpe.br/prodes (April 2004). 11. Christaller, W., How I discovered the Theory of Central Places: A Report about the Origin of Central Places. in: English, P.W. Mayfield, R.C., eds., Man Space and Environment. (Oxford Univ. Press, 601-610, 1972). 12. Monteiro, M.P., Sawyer, D.O. Diagnóstico demográfico, socioeconômico e de pressão antrópica na região da Amazônia Legal. In: Capobianco, J.P.R. et al. (Org.). Biodiversidade na Amazônia Brasileira (ISA, São Paulo, 2001). 13. Manton, K., Woodbury, M., Tolley, D. Statistical applications using fuzzy sets. (John Wiley & Sons Inc, New York, USA, 312p, 1994). 14. Lemos, M.B., Diniz, C.C., Guerra, L.P. Pólos Econômicos do Nordeste e suas Áreas de Influência: uma aplicação do modelo gravitacional utilizando Sistema de Informações Geográficas (SIG). Revista Econômica do Nordeste, Fortaleza, Brasil, 30, 568-584, (1999). 15. Eva, H.D. et al. A land cover map of South America. Global Change Biology. 10 (5), 731-744 (2004). 16. Hansen, M. et al. 2002, Towards an operational MODIS continuous field of percent tree cover algorithm: examples using AVHRR and MODIS data. Remote Sensing of the Environment 83, 303-319 (2002). 17. Steininger, M.K. et al. Tropical deforestation in the Bolivian Amazon, 20. Environmental Conservation 28 (2): 127-134 (2001). 31 18. Pacheco, P. Agricultural expansion and deforestation in the Lowlands Bolivia: The Import Substitution versus the Structural Adjustment Model. Land Use Policy. (In press). 19. Rojas, D., Martinez, I., Cordero, W., Contreras, F. 2003. Tasa de Deforestacion de Bolivia. Pages 55. Proyecto BOLFOR, Santa Cruz, Bolivia. 20. Ventana - Vensim software – linking systems thinking to powerful dynamic models. http://www.vensim.com/software.html (November 2004). 21. Geist, H.J., Lambin, E.F. What drives tropical deforestation? A meta-analysis of proximate and underlying causes of deforestation based on subnational case study evidence. (LUCC International Project Office, Belgium, LUCC Report Series, 4 136pp, 2001) http://www.geo.ucl.ac.be/LUCC/lucc.html. (2004) 22. MME - Ministério das Minas e Energia. Projeto Radam. (Levantamento de Recursos Naturais, 1) (Departamento Nacional de Produção Mineral. Rio de Janeiro, Brazil, 1973) 23. WWF (2002) World Wildlife Fund Ecoregions of the Amazon. (The World Wildlife Fund Conservation Science Program 1998 – 1999) In: ESRI Data & Maps. (Environmental Systems Research Institute, Inc. Redlands, California, USA, 2002) (CD 1) 24. TRFIC (2004) The Tropical Rain Forest Information Center - Hosting for LBA-E Program. http://bsrsi.msu.edu/trfic/LBA_E/(April, 2004). 25. World Database on Protected Areas. (IUCN and UNEP - WDPA Consortium. Washington DC, US. 2003) (CD) 26. Soares-Filho, B.S., Pennachin, C., Cerqueira, G.C. DINAMICA – a stochastic cellular automata model designed to simulate the landscape dynamics in an Amazonian colonization frontier. Ecological Modelling, 154, 217-235. (2002). 27. Soares-Filho, B.S., Corradi, L., Cerqueira, G.C., Araújo, W.L. Simulating the spatial patterns of change through the use of the DINAMICA model. In: XI Simpósio Brasileiro de Sensoriamento Remoto. (Instituto Nacional de Pesquisas Espaciais, Belo Horizonte, BR. 721-728). http://iris.sid.inpe.br:1908/col/ltid.inpe.br/sbsr/2002/11.16.12.49/doc/06_257.pdf (2003). 28. Soares-Filho, B.S., Assunção, R.M., Pantuzzo, A. Modeling the spatial transition probabilities of landscape dynamics in an Amazonian colonization frontier. BioScience, 51, 1039-1046. (2001) 29. Alves, D.S. An analysis of geographical patterns of deforestation in Brazilian Amazônia the 1991-1996 period. In: Wood CH, Porro, R. Patterns and Processes of Land use and Forest Change in the Amazon (University of Florida, Gainesville, Florida USA, 95-106. 2002) 30. Alves, D.S. Space-time dynamics of deforestation in Brazilian Amazônia. International Journal of Remote Sensing, 24, 2903-2908 (2002). 31. Reis, E., Margulis, S. "Options for Slowing Amazon Jungle Clearing". In: Dornbusch R., Poterba J. M., eds, Global Warming: Economic Policy Responses. (MIT Press, Cambridge MA, 1991). 32. Pfaff, A.S.P. What drives deforestation in the Brazilian Amazon? Evidence from satellite and socioeconomic Data. Journal of Environmental Economics and Management, 37, 26-43, (1999). 32 33. Ludeke, A., Maggio, R.C., Reid, L.M. An analysis of anthropogenic deforestation using logistic regression and GIS. Journal of Environmental Management, 31, 247-59 (1990). 34. Chomitz, K.M., Gray, D.A. Roads, lands, markets, and deforestation, a spatial model of land use in Belize. World Bank Economic Review, 10, 487-512 (1996). 35. Mertens, B., Lambin, E.F. Spatial modelling of deforestation in Southern Cameroon: spatial disaggregation of diverse deforestation processes. Applied Geography 17, 14368 (1997). 36. Mertens B., Poccard-Chapuis R., Piketty M.G., Lacques A.E., Venturieri A. Crossing spatial analyses and livestock economics to understand deforestation processes in the Brazilian Amazon: the case of São Félix do Xingú in South Pará. Agricultural Economics 27 (2002) 269–294. 37. Patterson, B.D. et al. Digital Distribution Maps of the Mammals of the Western Hemisphere, version 1.0. NatureServe, Arlington, Virginia, USA (2003). 38. Eastman, J.R., Idrisi Kilimanjaro, Clark University, Worcester, MA. USA (2003). 33