Smallholder policy adoption and land cover change in the southeastern Peruvian Amazon: A twenty-year perspective

Applied Geography 53 (2014) 223e233

Contents lists available at ScienceDirect

Applied Geography journal homepage: www.elsevier.com/locate/apgeog

Smallholder policy adoption and land cover change in the southeastern Peruvian Amazon: A twenty-year perspective
vez a, *, Eben N. Broadbent b, c, d, e, Ange
lica M. Almeyda Zambrano b, c, e Andrea B. Cha
nica de Madre de Dios, UNAMAD, Instituto de Investigacio
n de Recursos Naturales y Medio Ambiente, Universidad Nacional Amazo Ciudad Universitaria, Biblioteca Segundo Piso, Puerto Maldonado, Peru1 b Sustainability Science Program, Harvard Kennedy School of Government, Harvard University, Cambridge, MA 02138, USA c Department of Global Ecology, Carnegie Institution, Stanford University, 260 Panama Street, Stanford, CA 94305, USA d Department of Biology, Stanford University, Gilbert Hall, Stanford, CA 94305, USA e Spatial Ecology and Conservation (SPEC) Lab, Department of Geography, University of Alabama, Tuscaloosa, AL 35487, USA a

a b s t r a c t Keywords: Peru Amazon Land use and land cover Deforestation Policies Change trajectories

The Peruvian Amazon has undergone extensive changes in land-use and land-cover changes in the last decades related to policy implementation at local to national scales. Understanding the complexity of such changes is one of the more important challenges at present and requires research approaches capable of spanning temporal and spatial scales and academic disciplines. Here, we investigate the impacts of agriculture incentives and infrastructure development in the Southeastern Peruvian Amazon using such an approach. We integrate Landsat satellite derived land-cover maps spanning the years 1986 and 2006 to understand the land-use/land-cover changes, including forest, crops and pasture, and secondary vegetation, and their implications stemming from voluntary policy adoption along the ~ apari-Iberia portion of the Inter-Oceanic highway. This road portion is one component of the broader In Initiative for the Integration of Regional Infrastructure expansion, which is resulting in rapid and extensive socio-economic and biophysical changes in the region. Results from this research highlight that changes in land-cover are associated with the farmers' voluntary adoption of agricultural policies, and that policies associated with cattle expansion and credit incentives, among others, have greatly influenced forest conversion. Although land-use/land-cover change causes are manifold and linked to more than policy events, the method used in this study improves the understanding of the effects of complex policy processes in this biodiversity and culturally rich region of the Amazon. © 2014 Elsevier Ltd. All rights reserved.

Introduction Rapid deforestation throughout the tropics has global implications, including for biodiversity conservation, sustainable development, and global climate change (Angelsen & Kaimowitz, 1999; Fearnside, 2000; Geist & Lambin, 2002; Malhi & Grace, 2000; Wood & Porro, 2002). Understanding the complex drivers of land-use and land-cover changes (LULCC) continues to be a major challenge for global change research and the land change science community (Geist & Lambin, 2002; Gutman et al., 2004; Lambin &

* Corresponding author. Universidad Nacional Amazonica de Madre de Dios, CETEGERN e INRENMA, Biblioteca Segundo Piso, Ciudad Universitaria, Puerto Maldonado e Madre de Dios, Peru. Tel.: þ1 352 331 6796.
vez). E-mail address: [email protected] (A.B. Cha 1 Tel.: þ51 989 655 036. http://dx.doi.org/10.1016/j.apgeog.2014.06.017 0143-6228/© 2014 Elsevier Ltd. All rights reserved.

Geist, 2006). Comprehensive discussions of tropical deforestation have pointed to a number of underlying drivers that include state policies as they affect incentives for landholders to clear forests (Lambin & Geist, 2006; Turner, Geoghagen, & Foster, 2004). This is of essential significance for frontier regions, which in Latin America often preserve tropical forest cover while at the same time promote frontier expansion (Schmink & Wood, 1992; Walker et al., 2009). A suite of LULCC studies have contributed to a better understanding of the specific impacts on rates of deforestation based on various socio-economic explanatory factors (Aldrich, Walker, Arima, & Caldas, 2006; Caldas et al., 2007; Fox, Rindfuss, Walsh & Mishra, 2003; Hecht, 2005; Moran & Ostrom, 2005; Turner et al., 2004; Walsh & Crews-Meyer, 2002; Wood & Porro, 2002). However, despite the important contributions by various disciplines, much work remains to be done to comprehend the implication of socio-economic processes and their relationship with deforestation. Further, policy analyses of LULCC are still scarce

224


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

and often limited to what-ifs scenarios (Andersen, Granger, Reis, Weinhold & Wonder, 2002; Fox et al., 2003; Rudel, 2005). A thorough understanding of the specific impacts of diverse policies on rates of deforestation remains limited (Lambin & Geist, 2003; Mather, 2006; Oliveira et al., 2007). Given this gap, there is a pressing need for empirical detailed policy analysis of LULCC outcomes. In this study, we seek to address this gap through a detailed assessment of LULCC following implementation and adoption of agricultural policies at the local and regional scale for a study area located in the southeastern Peruvian Amazon. Specifically, we ask how LULCC has been affected by the adoption of agricultural policies between the years 1986 and 2006, which allows for comparing LULCC across periods with distinctive policy regimes and therefore the identification of direct causal linkages. Given that under even ideal conditions, indirect effects may constrain LULCC dynamics and landholders do not always participate in the studied policies, we compare LULCC dynamics within, and between policy periods and among voluntary adopters and non-adopters. We hypothesize that the main factor driving land conversion from forest to nonforest is the voluntary adoption of agricultural policies such as cattle insemination, agricultural mechanization and credits during
vez, 2013). For example, specific presidential administrations (Cha the voluntary adoption of seed improvement and/or copoasu plantation policies would be linked to land conversion from forest to non-forest. Likewise, the non-adoption of non-forest induced policies would affect the transformation to forest cover areas. While this approach does not control for all possible effects, it offers an advance over pure multi-temporal analysis and the information from this study provides baseline information useful for assessing the direct causal effects of policy incentives at the landholding level. Methodology Study design This study constructs a LULCC history of the Inter-Oceanic~ apari to Iberia in the southwestern Highway corridor from In Peruvian Amazon from the mid-1980s to 2006, focusing on policies related to agricultural expansion in three specific time periods (Fig. 1). Two main comparisons are important in this study: (a) The comparison of LULCC dynamics among policy periods; and (b) the comparison within each period of LULCC dynamics of policy adopters and non-adopters, with voluntary policy adoption taking place during different policy periods. Multi-date Landsat satellite data is used to provide estimates of land-cover change between ~ apari and 1986 and 2006 along the road connecting the towns of In Iberia, Madre de Dios, Peru, where our study households are located. Study area Our study area lies within the region of Madre de Dios and ~ apari and Iberia in the corresponds politically to the districts of In province of Tahuamanu in Southeastern Peru (Fig. 2). The province of Tahuamanu covers an area of 21,197 km2, out of which 2040 km2 are considered in our study area. The focus of this study is the path of the 2006-paved Inter-Oceanic Highway, which links the Brazilian southern state of Acre with Peru and covers a road extension of 1580 km from the Brazilian border to the Peruvian ports in the Pacific Ocean. Specifically, this study is focused on the road con~ apari to Iberia, which has a length of approximately necting In 70 km.

The climate is hot and tropical, seasonally humid, with abundant rains from October to March with a short dry season from June
vez, 2009). The vegetation is composed of lowto September (Cha land rainforests located both along low terraces bordering major rivers on higher grounds than the alluvial terrains and water bodies and distinguished by diverse forest types on solid ground (i.e., Terra firme). Although the tree species diversity of Amazonian moist forests is generally among the highest of any forest type in the world (Gentry, 1988), mono-dominant stands of bamboo (Gadua spp.) or palm (Mauritia flexuosa or Jessenia bataua) develop in some areas (IIAP-CTAR, 2001). The province of Tahuamanu is known for preserving big-leaf mahogany, one of the most critically endangered forests species (ITTO, 2005), and for extensive stands of Brazil Nut trees (Bertholletia excelsa). The Tahuamanu Province had an estimated human population of 10,742 in 2007 (INEI, 2007). Land tenure comprises a mixture of indigenous reserves, small and middle-sized private holdings (mainly agriculture and ranching), state land, timber concessions, and unclaimed land (IIAP-CTAR, 2001). Land-use patterns are associated with major productive activities and consist of slash and burn-based subsistence agriculture (mainly rice, beans, and maize), forest extraction, and to a lesser extent cattle ranching (INADE-OEA, 1998). Although areas with agricultural potential represent 28% of the total area, only 0.6% is devoted to agricultural activities, an indication of the poor returns to farmers from cultivation and the lack of market access (INRENA, 1999; Mora, 1993). Despite the low population density, the region of Madre de Dios has experienced the highest migration and population growth rates in Peru (annual population growth rate of 6%), generally concentrated around transportation routes (Tahuamanu, 2001). Historically, the Madre de Dios Region has experienced LULCC through national policies ~ a”), gold, and timber boom-bust based on rubber, Brazil nut (“castan
vez, 2013; Cha
vez &Perz, 2012). Economic economic cycles (Cha prosperity has been slow, since rural and urban areas have required adequate communication and transportation infrastructures, and most government incentives have lacked the focus on competitive market opportunities, trading mechanisms, and price stability (Ch
avez, 2009). This area is therefore ideal for the analysis of the interaction between policy shifts, policy adopters, and LULCC. Policy periods and participation Figure one illustrates the three specific policy periods (IeIII) corresponding to expected land transformation outcomes as measured from remote sensing. Landholder households may choose to either participate or not in the offered incentive of each studied policy influencing the outcome of LULCC differences (Ch
avez &Perz, 2012). In general, policy adoption by households increases the probability of land conversion (Perz, 2003). For this reason, we compared land conversion rates for between these two groups within each policy period. In order to account for time lags between policy participation and altered LULCC we analyzed forest clearing or reforestation occurring within a time span of approximately five years following initiation of the policy period (Ch
avez, 2013). Land conversion rates are measured using field data and satellite images chosen to correspond to each policy period. Specifically, the policy periods are: (I) Policies that favored the establishment of annual crops and pasture from the year 1985 through 1990. This period is tied to the first presidency of Alan Garcia and represents the time when credit for annual crops and incentives for cattle were available. During this time, the government facilitated expenditure opportunities among farmers through the Agrarian Bank, which provided financial support and guaranteed the purchase of farmer's crops at preset retail values (Escobal D'Angelo, 1992); (II) Policies that favored reforestation from the


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

225

Fig. 1. Relevant policies linked to specific Landsat images and their expected land transformation outcomes.

~ apari - Iberia, Madre de Dios, Peru, showing the two towns, the Inter-Oceanic Highway, the locations of titled farms, landholdings of policy adopters, Fig. 2. Study site road axis: In and training samples used in classification.

year 1990 through 2000. This period includes two presidency terms of Alberto Fujimori and abandonment of agricultural supports alongside the introduction of reforestation and seed improvement policies. During this period, new economic liberalization reforms and new agricultural markets designed to increase crop yields were

established (INRENA-OEA, 1998; MINAG, 2005; Trivelli, Shimzu, & Glave, 2003); and (III) Policies that favored farm diversification from the year 2001 through 2006. This period comprises the presidential office period of Alejandro Toledo and the execution of added decentralized policy initiatives for cattle insemination, fish


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

226

farming, perennial crops and agricultural mechanization (CTAR, 2002; INADE-PEMD, 2004). It is worth noting that these policies are most representative for this study area, other national policies
vez, may have been representative in other parts of the country (Cha 2013). Specifically, we hypothesized that voluntary policy adopters or non-adopters of: (a) agricultural credit incentives, including seed improvement programs [which focused on coffee, cacao, citrus, pepper, maize, and plantains (CTAR, 2002)], would respond by increasing rates, and extent, of forest conversion to agricultural land, and/or through intensification of agricultural activities. Agricultural mechanization policies, however, would result in both forest conversion and abandonment with their relative importance varying with the years since policy adoption; (b) cattle-acquisition credit incentives, such as government-sponsored cattle insemination programs, would respond by increasing rates, and extent, of conversion of both forest and regrowth to pasture areas; that (c) policies assisting the establishment of fish farms and/or copoasu (Theobroma grandiflorum) plantations would result in reduced forest conversion to crop or pasture (Fig. 1); and that (d) other policies focusing on agroforestry systems would increase rate and extent of forest areas. Household surveys Detailed land-use history and policy adoption e or not e was
vez, recorded from 2003 to 2006 via extensive field surveys (Cha
vez &Perz, 2012). Individual farmers and their titled 2009; Cha farmland were first randomly chosen drawing on the 2006 digital agricultural cadastral boundary file provided by the Peruvian land titling agency (PETT) for a total of 124 households. From this selection we subset households meeting the criteria of: (a) experience up to a minimum of two of policy periods (i.e., residing in the region

for no less than 10 years), which permitted evaluation of land use across at least two presidential administrations; and (b) a combination of migration trajectories (e.g., considering both migrants who relocated from other parts of Peru under colonization programs as well as natural residents of the Tahuamanu Province). The total area of studied landholdings was 11,280 ha, which corresponds to nearly 20% of all registered farms in the study area (50,059 ha) and distinguished by policy and non-policy adopter's
vez, 2009). farms (Fig. 3) (Cha Field surveys consisted of structured questionnaires and covered segments related to land-use practices, migration history and landuse records, adoption of agricultural policy incentives, soil fertility,
vez land tenure, credit history, and governmental support (Cha &Perz, 2012 ). See Appendix D for an English translation of the Spanish questionnaire. Information was verified during consecutive field visits in 2005 and 2006. Households in which the head of household was not born in Madre de Dios region were considered ‘immigrant households’ for the purposes of our analyses. In discussing land tenure we principally consider whether households had an official title to their landholding (s). A more detailed description of
vez andPerz (2012) . In household data collection can be found in Cha addition to descriptive statistics of our sample households (see Table 1 for overview), we used a qualitative hierarchical binomial decision tree approach to identify core groups of households having similar policy decisions (Table 2) e henceforth referred to as policy decision groups. We then used ANOVA analyses to identify significant differences in household variables among these groups. Remote sensing Landsat satellite images for the study region were used for the years 1986, 1991, 1996, 2001, and 2006 (WRS Path/Row 003/068) (Ch
avez, 2013). These Landsat TM (1986, 1991, 1996, and 2006) and

Fig. 3. Land cover classification of the study area for year 2006 showing the landholdings owned by households adopting policies within titled farm plots.


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

2001 version). Using the rectified 2001 image, we registered the remaining 1986, 1991, 1996, and 2006 scenes employing the nearest-neighbor resampling technique and maintaining a RMS
vez, 2013). We performed radiometric caliunder 0.5 pixels (Cha bration as described in the CIPEC calibration method (Green, Schweik, & Randolph, 2005). During 2003e2005, fieldwork was conducted in the districts of ~ apari and Iberia, concentrating along the road axis and secondary In dirt roads, where most of the agricultural land is located. Ground data sites were split into training and validation data sets to establish four land-cover classes: forest, crops and pasture,
vez, 2013). Consecutive field regrowth, and built/non forest (Cha visits in 2006 served to correct misclassifications and identify causes of classification errors. Field reconnaissance included the classification of tree height and the recognition of dominant groundcover species, which facilitated a better understanding of
vez, 2009). Indicator species (e.g. forest regrowth stages (Cha Cecropia spp, large bamboo (Gadua spp.), and palms) helped assign regrowth classes. We recorded 295 training plots pertaining to the four established land-cover employing a Global Positioning System
vez, 2009) (Fig. 2). (GPS) device (Cha To estimate LULC between 1986 and 2006, tasseled-cap (TC)
vez, 2013; indices were computed for multi-date images (Cha Ch
avez Michaelsen et al., 2013). Our TC classification is based on studies reported by Fiorella and Ripple (1993) and by Guild, Cohen, and Kauffman (2004) for tropical forests. Brightness, greenness, and wetness (BWG) indices, a methodology used extensively in landscape classification (Crist & Cicone, 1984; Crist, Lauren, &

Table 1 Descriptive statistics for household and landholding variables in 2005e06 (N ¼ 122 households). Household and landholding variables

Mean*

Category

Variable

Units

Household Household Household Household Household Landholding Landholding Landholding Landholding

Years on landholding Years on landholding with lag Immigrant household Land titled (i.e, tenure status) Household size Distance to market Total landholding size Reported landholding size Soil fertility

Years Years No/Yes*** No/Yes*** # Km Ha Ha Low/High****

14.84 13.91 0.44 0.86 3.92 10.22 95.56 90.81 0.57

227

Std. Dev. 8.31 8.32

1.92 5.60 93.94 97.57

* Proportion Yes; ** 1 ¼ 1985e1990; 2 ¼ 1990e1995; 2.2 ¼ 1995e2000; 3 ¼ 2001e2006; *** No ¼ 0/Yes ¼ 1; **** Low ¼ 0/High ¼ 1.

ETM þ images (2001) were selected for analysis in order to link four land-cover change trajectories (1986e1991; 1991e1996; 1996e2001, and 2001e2006) with the three policy periods (Fig. 1). The time interval of Landsat images chosen here, correspond to the beginning and ending of the five-year interval policy periods represented by presidential administrations. Image preprocessing and subsequent classification analysis utilized ERDAS Imagine 9.3. We first chose the ETMþ 2001 scene as the base image, reprojected the image to the UTM WGS 1984 Zone 19S coordinate system and performed a geometric rectification and registration to a
fico, 1:100,000-scale topographic map (Instituto Nacional Geogra Table 2 Hierarchical decision tree of household policy adoption from 1985 to 2006. Policy adoption decision hierarchy A. Policy decision tree 1985e90 (1)

1990e95 (2)

1995e00 (2.2)

2001e06 (3)

Yes

Yes

Yes

Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No

No No

Yes No

No

Yes

Yes No

No

Yes No

C. Total percentage

B. Count 1985e90 (1)

1990e95 (2)

1995e00 (2.2)

2001e06 (3)

3

62

24

24

18 0 0 0 0 2 13 23 1 4 0 2 0 0 11 44

0 38

1990e95 (2)

1995e00 (2.2)

2001e06 (3)

37.8%

14.6%

14.6%

11.0% 0.0% 0.0% 0.0% 0.0% 1.2% 7.9% 14.0% 0.6% 2.4% 0.0% 1.2% 0.0% 0.0% 6.7% 26.8%

0.0% 23.2%

1.2% 22.0%

4.3%

3.0% 1.2%

33.5%

0.0% 33.5%

2 36

2 62

7

5 2

55

0 55

1 Rank

1985e90 (1)

37.8%

Rank

3

D. Incremental percentage 1985e90 (1)

1990e95 (2)

1995e00 (2.2)

2001e06 (3)

50.0%

38.7%

100.0%

75.0% 25.0% 0.0% 0.0% 0.0% 100.0% 36.1% 63.9% 20.0% 80.0% 0.0% 100.0% 0.0% 0.0% 20.0% 80.0%

0.0% 61.3%

5.3% 94.7%

2 50.0%

11.3%

71.4% 28.6%

88.7%

0.0% 100.0%

1


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

228

Cicone, 1986; Ivits, Lamb, Langar, Hemphill, & Koch, 2008), were first generated for all acquired Landsat images using Landsat 5 TM coefficients for the 1986, 1991, and 1996 images and Landsat 7 at
vez, 2013; Cha
vez satellite reflectance for the 2001 image (Cha Michaelsen et al., 2013; Crist & Cicone, 1984; Huang, Wylie, Homer, Yang, & Zylstra, 2002; Kauth & Thomas, 1976). We generated a final land-cover map employing a maximum likelihood supervised classification technique. Our final land cover classes correspond to: forest (consisting of alluvial, terra firme, and mono-dominant bamboo stands), agriculture (consisting of crops and pasture), regrowth, built/non-forest (consisting of urban areas, roads, and water), and haze/clouds
vez, 2013) (Fig. 3). Initial efforts to classify separately areas of (Cha crops and pasture showed them to be spectrally similar and therefore were merged into the agriculture class. Areas of regrowth were assigned approximate stand ages using the multi-temporal analysis and extensive field surveys of: (a) 1e3 years; (b) 4e7 years; and (c) 8e15 years (Ch
avez, 2013). The accuracy of the classification approach was tested using a stratified random sampling scheme encompassing 30 training
vez samples for each land cover within the 2001 image (Cha Michaelsen et al., 2013). Knowledge of the history of land-use gained during the 2003e2005 field visits, and training samples collected in 2003, allowed for ground-data and satellite data as
vez, 2009). This analysis showed the sociation back to 2001 (Cha 2001 classification to have an overall accuracy of 73.06% with a kappa score of 0.664. Detailed results are provided in Appendix A and Appendix B. Statistical integration Statistical analysis integrated remote sensing and household survey information to model the rate of deforestation and reforestation on each surveyed household's landholdings. We selected input variables for a multi-variate-regression analysis predicting deforestation and reforestation at the household landholding scale. Input variables were transformed to best fit a normal distribution, following which all subsets regression was used to identify

significant combinations of predictor variables for deforestation and reforestation within the four 5-year policy periods to identify models having the best balance between predictor variables and the adjusted R2. We also categorically compared deforestation and reforestation among the core policy decision groups using ANOVA with Tukey HSD post-hoc analyses of deforestation and reforestation variables (ha and % landholding) following log transformation to obtain a normal distribution. Statistical analyses were performed in R (R Core Team, 2013) and JMP (JMP, Version 7. SAS Institute Inc., Cary, NC, 1989e2007). Results Results encompass the integration of household surveys and remote sensing evidenced by statistical data and classification maps. A total of 124 households participated in the study. Of these sufficient data was obtained from 122 households for statistical analysis. Table 1 provides descriptive statistics of these households. On average, households had lived for 15 years on their landholdings totaling 95 ha in area and located 10 km from the nearest market. Households were evenly divided between immigrant and nonimmigrant households, and consisted of 4 members, with 86% having the title to their landholding (s). Fifty-seven % of households considered their landholdings to have high soil fertility (versus low). The number of households adopting 0e4 policies was 44 (35.5%), 36 (29.0%), 19 (15.3%), 7 (5.6%), and 18 (14.5%), respectively. Whereas our sample households were evenly divided between policy adopters and non-adopters in period 1 (1985-90 yrs), analyses of subsequent policy adoption decision making identified three distinct policy decision groups of households (Table 2). The most abundant groups were: (1) those that did not adopt any policies (44), i.e., non-adopters; followed by, (2) those that adopted a policy in period 1 only (23), i.e., initial adopters; and then, (3) those that adopted all policies (18), i.e., adopters. Once a household adopted a policy, the likelihood that it would adopt the subsequent policy increased greatly from 11.3% for initial non-adopters to 38.7% for initial adopters for the second period. The distance to the nearest market had no significant relationship with the number of

Table 3 Land-cover transition (LCT) analysis from 1986e1991, 1991e1996, 1996e2001, and 2001e2006 for the entire study region (495840 ha) and for our surveyed household landholdings (8269 ha) as calculated from Landsat TM satellite imagery. Transitions are in ha with % change in parentheses. F ¼ forest, B ¼ built/non-forest, C ¼ crops/pasture, R ¼ regrowth, and H ¼ haze. LCT

FeF FeB FeC F-R FeH BeF BeB BeC B-R BeH CeF CeB CeC C-R CeH ReF ReB ReC ReR ReH HeF HeB HeC

1986e1991

1991e1996

1996e2001

2001e2006

Study region

Landholdings

Study region

Landholdings

Study region

Landholdings

Study region

Landholdings

459633.1 (92.7%) 2231.5 (0.45%) 7812.5 (1.58%) 3490.7 (0.7%) 13074.8 (2.64%) 2253.8 (0.45%) 1870.5 (0.38%) 799.9 (0.16%) 134.7 (0.03%) 438.6 (0.09%) 350.1 (0.07%) 530.2 (0.11%) 544.9 (0.11%) 96.7 (0.02%) 253.6 (0.05%) 1787.2 (0.36%) 77.9 (0.02%) 226.3 (0.05%) 75.2 (0.02%) 94.8 (0.02%) 52.2 (0.01%) 7.7 (0%) 3.2 (0%)

6430.8 (77.77%) 170.8 (2.07%) 755.1 (9.13%) 123.2 (1.49%) 471.4 (5.7%) 43.8 (0.53%) 21.8 (0.26%) 26 (0.31%) 4.9 (0.06%) 17 (0.21%) 32 (0.39%) 29 (0.35%) 36.5 (0.44%) 15.4 (0.19%) 42.5 (0.51%) 24.1 (0.29%) 5 (0.06%) 15.4 (0.19%) 3.3 (0.04%) 0 (0%) 1.2 (0.01%) 0 (0%) 0.1 (0%)

447680.2 (90.29%) 1261.1 (0.25%) 3983.3 (0.8%) 8338 (1.68%) 2811 (0.57%) 1534.6 (0.31%) 2281.1 (0.46%) 677.3 (0.14%) 204.1 (13.59%) 13.6 (0%) 538.3 (0.92%) 795.2 (0.36%) 959.6 (0.4%) 73.8 (0.22%) 8.2 (0%) 1.2 (0.61%) 166.7 (0.03%) 89 (0.06%) 333 (0.07%) 7.4 (0%) 2060.8 (2.43%) 3.9 (0.14%) 79.7 (0.12%)

5875.3 (71.23%) 66.7 (0.81%) 291.2 (3.53%) 262.7 (3.18%) 36 (0.44%) 56.3 (0.68%) 100.2 (1.21%) 52.4 (0.63%) 17.5 (0.21%) 0.3 (0%) 328.6 (3.98%) 172.2 (2.09%) 228.2 (2.77%) 102.6 (1.24%) 1.7 (0.02%) 76.4 (0.93%) 16.8 (0.2%) 32 (0.39%) 21.2 (0.26%) 0.4 (0%) 292.4 (3.54%) 129.2 (1.57%) 88.7 (1.08%)

443339.7 (89.41%) 4503.1 (0.91%) 11668.4 (2.35%) 9218.3 (1.86%) 96.2 (0.02%) 1318.5 (0.27%) 3166.3 (0.64%) 1471.3 (0.3%) 211.9 (0.04%) 45.8 (0.01%) 575.3 (0.72%) 295.3 (0.26%) 992.8 (0.4%) 587 (0.12%) 39.5 (0.01%) 295.3 (1.47%) 528.5 (0.11%) 649.6 (0.33%) 899.3 (0.18%) 19.4 (0%) 823.8 (0.57%) 30.3 (0.01%) 48.2 (0.01%)

5440.9 (65.73%) 239.5 (2.89%) 651.9 (7.87%) 285.2 (3.45%) 12.4 (0.15%) 65.9 (0.8%) 273.2 (3.3%) 133.5 (1.61%) 7.8 (0.09%) 4.8 (0.06%) 228.8 (2.76%) 160.6 (1.94%) 246.6 (2.98%) 50.4 (0.61%) 6.1 (0.07%) 201.7 (2.44%) 61.6 (0.74%) 122.1 (1.48%) 42.1 (0.51%) 3 (0.04%) 33.4 (0.4%) 2.9 (0.03%) 3.6 (0.04%)

443328.2 (89.41%) 2116.9 (0.43%) 5560 (1.12%) 7309 (1.47%) 40.2 (0.01%) 3893.8 (0.79%) 3469.8 (0.7%) 1633.4 (0.33%) 517.4 (0.1%) 12.4 (0%) 657.4 (2.15%) 956 (0.39%) 654.2 (0.54%) 557.5 (0.31%) 6.2 (0%) 100.9 (1.84%) 262.8 (0.05%) 91.3 (0.16%) 784 (0.16%) 2.3 (0%) 4.6 (0.01%) 61.5 (0.01%) 54.1 (0.01%)

4762.3 (57.55%) 201.7 (2.44%) 554 (6.69%) 451.9 (5.49%) 0.8 (0.01%) 147.3 (1.78%) 267.2 (3.23%) 255 (3.08%) 68 (0.82%) 0.1 (0%) 426.2 (5.15%) 209.7 (2.53%) 355.4 (4.3%) 166.3 (2.01%) 0.1 (0%) 209.3 (2.53%) 25 (0.3%) 86.3 (1.04%) 65.3 (0.79%) 0.3 (0%) 5.9 (0.07%) 7.4 (0.09%) 9.5 (0.12%)


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

policies the household had participated in. As expected, households with longer residence times on their landholdings had participated in significantly more policies (Policies adopted ¼ 0.7331 þ 0.0412*Years on landholding, Adj-R2 ¼ 0.0535, P ¼ 0.0056, N ¼ 124). Table 3 provides detailed spatial information on land cover transitions across the entire study region and in comparison to the landholdings. Twenty-year land-cover “from-to” change detection analysis was generated producing 25 possible change trajectories and comparing the amount of land under each class in the 1986, 1991, 1996, 2001, and 2006 year land-use and land-cover classifications (Table 3). Areas of haze in any image year were excluded from all image years to enable accurate temporal multi-temporal analyses, resulting in a reduction in the spatial extent of the study area by 388.9 ha, from 11665.2 to 11276.3 ha (3.3%). The majority of the area remained forested throughout the study period. The dominant changes in land cover from 1986 to 2006 were deforestation for agriculture and reforestation of abandoned agriculture and pasture areas (Figs. 4 and 5). The dominant stable land cover class was forest; both for the study region as a whole and within individual properties. The percent of the study area in forest decreased from 1986 to 2006 from 92.7% to 89.41%, respectively. The dominant land cover changes across the entire study area for the policy periods 1e4 was regrowth to forest (0.36%). During the 5 year period from 1991 to 1996, approximately 5% of existing agricultural was abandoned and began forest

229

succession. In 2001, forest was lost by almost 3% to built/non-forest, by almost 8% to crops and pasture, and by 3% to regrowth. By 2006, 2.5% of regrowth reverted to forest while a surprising 5.5% changed from crops and pasture to forest. On the other hand, 6.7% of forest changed to crops and pasture and 5.5% changed to regrowth areas. Table 4 provides an overview of statistical analysis integrating remote sensing and household survey information to model the rate of deforestation and reforestation on each surveyed household's landholdings (Table 4). Table 4 complements data presented in Table 3 by presenting deforestation and reforestation results for each policy period for the household landholdings overall, and separately for those households adopting or not adopting each policy. Deforestation was highest from 1995 to 2006 and reforestation was highest from 2001 to 06. Although inadequate sample sizes of policy periods were present, no trends in rates of deforestation correlating with rates of reforestation were found. Forest area decreased more (19.9%) for properties that had adopted agricultural policies (2001e06) as compared to non-adopters (13.6%) and to the study region in general (~4%). In both cases, forest was converted mostly for agricultural expansion. Pronounced differences among adopters and non-adopters were not seen for reforestation during our study period. Table 5 provides Pearson correlations among household socioeconomic and landholding deforestation and reforestation variables. The best household predictor of other household variables

~ apari. F ¼ forest, B ¼ built/non-forest, C ¼ crops/pasture, Fig. 4. Change detection maps of policy adopters for 1986e1991, 1991e1996, 1996e2001 and 2001e2006 for the area of In R ¼ regrowth, H ¼ haze, and ND ¼ No data.


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

230

Fig. 5. Change detection maps of policy adopters for 1986e1991, 1991e1996, 1996e2001 and 2001e2006 for the area of Iberia. F ¼ forest, B ¼ built/non-forest, C ¼ crops/pasture, R ¼ regrowth, H ¼ haze, and ND ¼ No data.

was the years of residence on the landholding, which had significant positive correlations with the total landholding area and likelihood of having their landholding titled. The best household predictor of landholding variables was the distance to market, which had significant negative correlations with all deforestation and reforestation variables. Landholding variables were all significantly and positively correlated with all other landholding variables. The most significant best subsets model of deforestation and reforestation within household landholdings is provided in Table 6 (and see Appendix C for a complete selection encompassing from 1 to 11 predictor variables). The best predictor variable for deforestation and reforestation extent were differences among the policy periods, followed by landholding area, and then distance to market. The best predictor variable for deforestation and reforestation

percentage were policy period differences, followed by distance to market, with a weaker effect of landholding area. Significant differences in deforestation and reforestation were found among the core policy decision groups (Table 7). Non-policy adopting households had the lowest rates of deforestation and reforestation, followed by initial adopters and then adopters. Significant differences among these groups also existed for household variables, with non-adopters having the least years on their current landholdings, having a higher probability of being an immigrant household, and a lower probability of having title to their landholdings. We found no significant effect of distance to market on the likelihood of a household belonging to any policy decision group.

Table 4 Mean ± standard deviation deforestation and reforestation within individual household landholdings for each policy period. Statistics are provided for: (A) all households, and households either (B) adopting, or (C) not adopting, each policy.

The results show a clear connection between LULCC dynamics, policy periods and policy adopters and non-adopters. Furthermore, results confirm distinct household policy trajectory among policy periods and among policy adopters and non-adopters and land change. Policy adopting households had the highest rates of deforestation compared to non-adopters as confirmed by the deforestation and reforestation/extent and percentage best predictor ‘policy period’ and ‘policy period differences’ values. Smallholders adopting agricultural policies, namely credit for crops and cattle expansion policies transformed land uses during the policy period I (1985e1990). For policy adopters' forest changed to agriculture by 9% between 1986 and 1991, one of the highest change rates throughout the study period. Increased agricultural credit and support for cattle acquisition fostered expanded forest clearing, especially for cattle pasture during this time period (Alvarez &
vez &Perz, 2012; Naughton-Treves, Naughton-Treves, 2003; Cha 2004). A change in policy period represented a change in LULCC. The impact of policy adoption or non-adoption is strongly influenced by houshold and landholding characteristics. In general,

Land cover change

Unit Policy period

A. All households Deforestation ha Deforestation % Reforestation ha Reforestation % B. Policy adopters Deforestation ha Deforestation % Reforestation ha Reforestation % C. Non-adopters Deforestation ha Deforestation % Reforestation ha Reforestation %

1985e90 (1)

1990e95 (2) 1995e00 (2.2) 2001e06 (3)

11.02 12.48 1.02 1.21

± ± ± ±

12.54 7.25 ± 7.92 12.45 7.91 ± 7.32 2.19 5.38 ± 5.64 2.53 6.5 ± 6.7

14.27 16.42 3.6 4.07

± ± ± ±

12.31 13.66 ± 11.43 10.48 15.82 ± 9.69 4.17 8.6 ± 7.4 4.33 10.2 ± 7.32

13.73 15.27 1.07 1.34

± ± ± ±

14.01 13.14 1.77 2.42

8.99 8.73 4.96 6.14

± ± ± ±

8.71 7.71 3.99 6.58

15.63 16.80 3.99 3.97

± ± ± ±

12.88 10.90 3.99 3.61

8.19 9.61 1.05 1.12

± ± ± ±

10.24 11.05 2.42 2.57

6.63 7.57 5.52 6.64

± ± ± ±

7.44 7.14 6.11 6.77

13.81 16.27 3.47 4.09

± ± ± ±

12.1 11.39 ± 9.00 10.41 13.56 ± 8.30 4.15 7.69 ± 6.46 4.50 9.67 ± 7.30

17.86 19.94 10.27 11.13

± ± ± ±

14.09 10.71 8.68 7.37

Discussion and conclusion

Table 5 Pearson correlations among household and landholding deforestation and reforestation variables. Variables

PA

Policies Adopted (PA) Distance to Market (DtM) Years on Landholding (YoL) Household Size (HS) Total Landholding Area (TLA) Immigrant Household (IH) Soil Fertility (SF) Tenure Status (TS) Deforestation Ha (Dha) Reforestation Ha (Rha) Deforestation Pct (DPct) Reforestation Pct (RPct) Sig. household corrs. (#) Sig. landholding corrs. (#)

DtM 0.008

0.008 0.022 0.008 0.015 0.025 0.015 0.006 0.022 0.013 0.018 0.01

0.139 0.051 0.035 0.053 0.041 0.036 0.187** 0.277*** 0.17* 0.229** 4

YoL

HS

TLA

IH

SF

TS

Dha

Rha

DPct

RPct

0.022 0.139

0.008 0.051 0.05

0.015 0.035 0.199** 0.082

0.025 0.053 0.034 0.035 0.025

0.015 0.041 0.057 0.08 0.151* 0.064

0.006 0.036 0.189** 0.081 0.121 0.019 0.163*

0.022 0.187** 0.278*** 0.019 0.6**** 0.084 0.103 0.112

0.013 0.277*** 0.28*** 0.061 0.428**** 0.134 0.055 0.022 0.877****

0.018 0.17* 0.143 0.037 0.142 0.068 0.011 0.05 0.543**** 0.557****

0.01 0.229** 0.132 0.043 0.15* 0.118 0.017 0.13 0.381**** 0.622**** 0.823****

0.05 0.199** 0.034 0.057 0.189** 0.278*** 0.28*** 0.143 0.132 2 2

0.082 0.035 0.08 0.081 0.019 0.061 0.037 0.043

0.025 0.151* 0.121 0.6**** 0.428**** 0.142 0.15* 2 3

0.064 0.019 0.084 0.134 0.068 0.118

0.163* 0.103 0.055 0.011 0.017 2

0.112 0.022 0.05 0.13 2

0.877**** 0.543**** 0.381**** 3 3

0.557**** 0.622**** 3 3

0.823**** 1 3

2 3

Significance of each correlation is provided as *,**,*** and **** representing P-values less than 0.1, 0.05, 0.01, and 0.001, respectively. Table 6 Multi-variate regression models predicting total deforestation and reforestation within a policy period [t-ratio (F-ratio) P-value]. Explanatory variables

Deforestation

Reforestation

Category

Variable name

Units

Ha

%

Ha

%

Time Household Household Household Household Household Household Landholding Landholding Landholding Landholding Model Model Model Model Model Model

Policy perioda Years on landholding Policy adoption Immigrant household Tenure status Years on landholding with lag Household size Distance to market Total landholding area Reported landholding area Soil fertility N Rsquare Rsquare Adj P-value F-Ratio Predictors

Categorical Years No/Yes** No/Yes** No/Yes** Years # Km Ha Ha Low/High***

5.20 (27.05) < 0.0001 3.11 (9.66) 0.0020 3.61 (13.02) 0.0003 1.83 (3.34) 0.0683

4.75 (22.56) < 0.0001 2.17 (4.72) 0.0303 4.10 (16.84) < 0.0001

12.38 (153.34) < 0.0001 2.85 (8.15) 0.0045

12.40 (153.71) < 0.0001 2.41 (5.80) 0.0164

a

3.33 (11.12) 0.0009 1.28 (1.65) 0.1998

2.02 (4.10) 0.0435

5.02 (25.20) < 0.0001 11.86 (140.62) < 0.0001

4.61 (21.29) < 0.0001 2.58 (6.64) 0.0103

5.66 (32.05) < 0.0001 9.27 (85.88) < 0.0001

5.20 (27.04) < 0.0001 3.40 (11.59) 0.0007

488 0.3505 0.3424 <0.0001 43.2577 6

488 0.1289 0.1180 <0.0001 11.8571 6

488 0.3942 0.3879 <0.0001 62.7343 5

488 0.2955 0.2882 <0.0001 40.4353 5

1 ¼ 1985e1990; 2 ¼ 1990e1995; 2.2 ¼ 1995e2000; 3 ¼ 2001e2006;** No ¼ 0/Yes ¼ 1;*** Low ¼ 0/High ¼ 1.

Table 7 Results of ANOVA and Tukey post-hoc analyses comparing deforestation and reforestation variables among core policy decision groups. Significant differences (P-value > 0.05) among groups are provided as letters for each land cover change variable. The binomial household variables of immigrant status, soil fertility and tenure status were tested using Pearson contingency analyses followed by Tukey post-hoc analyses. Land cover change

Unit

Policy groups [mean ± std. dev (median) Tukey] Non-adopters (1)

Initial adopters (2)

Adopters (3)

Others (4)

Deforestation Reforestation Deforestation Reforestation Years on landholding Household size Distance to market Landholding area Immigrants Soil fertility Tenure status

ha ha % % yrs. # km ha No(0)/Yes(1) Low(0)/High(1) No(0)/Yes(1)

8.10 ± 7.87 (6) A 3.36 ± 4.70 (1) A 10.52 ± 9.45 (8) A 4.30 ± 5.93 (2) A 11.41 ± 7.61 (8) C 3.66 ± 1.82 (3.5) 10.66 ± 5.69 (9.3) 85.61 ± 57.46 (73.3) 0.65 A 0.61 AB 0.84

12.89 ± 11.21 (10) B 5.81 ± 6.44 (3) B 14.37 ± 10.88 (12) AB 6.69 ± 6.89 (4) B 20.86 ± 5.61 (23) A 4.10 ± 1.64 (4) 8.13 ± 4.48 (7.5) 94.76 ± 45.40 (95.7) 0.52 AB 0.81 A 0.95

15.49 ± 15.44 (12) B 5.44 ± 6.30 (4) B 14.49 ± 12.24 (12) AB 5.60 ± 6.32 (3) AB 17.72 ± 6.52 (18) AB 4.11 ± 1.78 (4) 11.83 ± 6.11 (10.9) 143.84 ± 206.39 (73.8) 0.33 AB 0.28 B 0.94

12.90 ± 12.09 (9) B 5.12 ± 6.35 (3) B 14.87 ± 10.61 (14) B 6.15 ± 6.67 (4) B 14.05 ± 9.09 (13) BC 4.00 ± 2.29 (4) 9.77 ± 5.36 (9.1) 86.25 ± 53.83 (69.2) 0.21 B 0.5 AB 0.79

Land Cover Change

Deforestation Reforestation Deforestation Reforestation Years on landholding Household size Distance to market Landholding area Immigrants Soil fertility Tenure status

Unit

ha ha % % yrs. # km ha No(0)/Yes(1) Low(0)/High(1) No(0)/Yes(1)

ANOVA results*

Pearson

Adj-R2

F Ratio

P-value

ChiSquare (P)

0.035 0.031 0.028 0.023 0.153

6.8104 6.2468 5.6605 4.8619 8.2045

0.1274 0.0779

6.8381 4.3747

0.0002 0.0004 0.0008 0.0024 <0.0001 0.7546 0.1678 0.1324 0.0003 0.0059 0.2411

18.051 (0.0004) 12.204 (0.0067) 4.245 (0.2362)

Sample sizes for policy groups 1e4 are: 176, 84, 72, and 156, respectively for landholding variables and 44, 21, 18, and 38 for household variables.

232


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

deforestation rates were significantly increased for households having larger landholdings, situated closer to the markets in towns ~ apari and Iberia, and that had adopted more policies. In most of In cases, more established households were in better position to adopt and take advantage of agricultural policies. Households that disclosed more years of residence, increased area of landholdings, and title possession accounted for more deforestation during the first policy period and could most probably have continual advantageous experiences from future policy adoption. As a consequence, timing of policy adoption is crucial and it is here where policy period differences come into play. The presidential administration between 1990 and 2000 abolished the state intervention in the agricultural sector of the previous administration and introduced more free-trade policies. Agricultural policies prioritized improved (hybrid) seed varieties and reforestation for branching out agricultural production via agroforestry. Government programs backed farmers in production, quality improvement, processing and marketing for crops such as coffee, cacao, citrus, pepper, maize, plantain, and agroforestry (CTAR, 2002). The policy reforestation seedling aimed at adding the cultivation of trees to farmer's current land-use practices and included peach palm, teak, rubber, and mahogany timber. During this time period, the total area in agriculture decreased due to low adoption of the new policy. These policies in combination with the state withdrawal of previous cattle expansion and credits for crops can be associated with a decrease in agriculture and built/nonforest in relationship to the previous policy period. Forest continued to be converted to agriculture and regrowth but nowhere as much as during previous policy period thus a clear example of the relationship between land-cover changes and policy adoption changes. Our results show some interesting linkages among land-use decision making after 2000. The policy periods extending from 1995 to 2000 and from 2001 to 2006 show an increase in crops and pasture, built/non-forest, and regrowth over previous periods. A combination of seed improvement and reforestation policies, as well as the new direction of the policies in promoting expanded agricultural production and infrastructure since 2001 may have played a role here (INADE-PEMD, 2004). Further, distance to nearest markets remained a good predictor for forest clearing throughout our policy periods. Policy adopter households located closer to markets strongly impacted land use dynamics either by increasing/reducing deforestation/reforestation, particularly after 2000. Forest area decreased more for properties that had adopted agricultural policies during 2001e6 as compared to non-adopters, a clear indication that road construction planning influenced land change. More accessible lands are likely to change faster, as new road improvements facilitate land change (Perz et al., 2014). Beginning in the 2000s, infrastructure development policies were initiated to improve the region's trade integration (Tahuamanu, 2001) that served to increase migration and new road construction and associate to an increase in built/non-forest after 2001. Agricultural polices implemented since 2001 by a governmental institution working on agriculture extension activities contributed to land change as well. These agricultural policies were part of the Madre de Dios Special Project (Proyecto Especial Madre de Dios, or PEMD) and included: government-sponsored cattle insemination, copoasu plantation, agricultural mechanization, and fish farming. Furthermore, these PEMD indicators related to policies that encouraged agriculture production increase (INADE-PEMD, 2004) may have contributed to forest loss and crops and pasture gain
vez, during 2001e2006 land-cover change trajectory analysis (Cha 2009). Landholding areas that remained cleared from period to period were usually pasture, commensurate with policies promoting cattle

development in 1986 and 1990. Clearings that underwent forest regrowth indicated a shift in cultivation or abandonment of land when policies no longer encouraged farmers to invest in more crops or when land constraints, such as low profitability forced farmers to relocate. In some cases, adoption of past policy incentives influences future policy trajectories (Chavez & Perz, 2013). For example, if households had a bad experience associated to a first policy adoption, they were less inclined to continue with the land use and would therefore most probably decline future policy opportunities. The dichotomous hierarchical decision tree approach presented in this study sheds light into these assumptions. Eventually, crops and pasture reverted back to forest via regrowth and agricultural activities were basically maintained for subsistence agriculture. Overall, regenerating forest areas were about equally likely to be cleared again or to be continued into forest establishment, depending on farmers' exposure to new policy adoptions and farmers' willingness to risk further transformation of land-uses (Ch
avez, 2009). Furthermore, the effectiveness of policy planning and execution plays a crucial role in future policy adoption (Chavez & Perz, 2013). This study had a number of limitations which we are working to better understand for future projects. The relationship between public policies and land use are complex, in part because there are other causal factors at work or lack thereof behind land-use de
vez, 2009). The inclusion cisions that cause land-cover changes (Cha of a broader suite of context-specific predictor variables, such as those encompassing sociological, ecological, and economic factors, could enable a deeper understanding of the social rationale behind land use decision making. For example, not all farmers will respond to a specific policy at the same time within a policy period, and responses to different policies at different policy periods may overlap. As a consequence, landholdings may have been undergoing important transitions and resulted in reduced policy signal in our models predicting deforestation and reforestation. The understanding of policy response temporality is important for policy
vez, 2013). Furthermore, future reversal processes as well (Cha satellite classification approaches could be improved for identification of critical transition ecosystems, in particular secondary forests. Reforestation and deforestation processes are driven by policy periods. Our findings reveal that government policy incentives, even those from many years before, influence farm-level land use and ultimately deforestation and reforestation dynamics in our study region of Madre de Dios, Peru. Household decision making policies were not universally adopted, and differences in such decision making produced statistically significant differences in landholding forest cover dynamics. Studies of this type may increase the possibility of reducing the impacts of future policies. Acknowledgment This research was financially supported by an international dissertation fellowship from the Compton Foundation, a Tropical and Development Conservation Research Fellowship, and a Tropical and Development Conservation Field Research Grant to Andrea
vez. Eben Broadbent and Ange
lica Almeyda Zambrano were Cha supported through fellowships from the Sustainability Science Program at the Harvard Kennedy School, Harvard University. We are grateful to Michael Binford, Jane Southworth, and Marianne Schmink for commenting on early manuscripts. Great thanks are given to the many individuals, communities, and institutions in Peru and internationally without whose help this research would not have been possible. Two anonymous reviewers greatly improved the manuscript.


vez et al. / Applied Geography 53 (2014) 223e233 A.B. Cha

Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.apgeog.2014.06.017. References Aldrich, S. P., Walker, R. T., Arima, E. Y., & Caldas, M. M. (2006). Land-cover and landuse change in the Brazilian Amazon: smallholders, ranchers, and frontier stratification. Economic Geography, 52, 265e288. Alvarez, N., & Naughton-Treves, L. (2003). Linking national agrarian policy to deforestation in the Peruvian Amazon: a case study of Tambopata, 1986e1997. Ambio, 23, 269e274. Andersen, L. E., Granger, C. W., Reis, E. J., Weinhold, D., & Wonder, S. (2002). The dynamics of deforestation and economic growth in the Brazilian Amazon. New York: Cambridge University Press. Angelsen, A., & Kaimowitz, D. (1999). Rethinking the causes of deforestation: lessons from economic models. The World Bank Research Observer, 14, 73e98. Caldas, M., Walker, R., Arima, E., Perz, S., Aldrich, S., & Simmons, C. (2007). Theorizing land cover and land use change: the peasant economy of Amazonian deforestation. Annals of the Association of American Geographers, 97(1), 86e110.
vez, A. (2009). Public policy and spatial variation in land use and land cover in the Cha Southeastern Peruvian Amazon (Dissertation Manuscript). Gainesville: University of Florida.
vez, A. B. (2013). Landscape dynamics of Amazonian deforestation between Cha 1986 and 2007 in Southeastern Peru: policy drivers and road implications. Journal of Land Use Science. http://dx.doi.org/10.1080/1747423X.2013.807310.
vez, A., & Perz, S. (2012). Adoption of policy incentives and land use: lessons Cha from frontier agriculture in Southeastern Peru. Human Ecology, 40, 525e539. Chavez, A. B., & Perz, S. (2013). Policy adoption, infrastructure upgrades, and land use plans: the case of Southeastern Peru. Geoforum, 50, 138e148. http:// dx.doi.org/10.1016/j.geoforum.2013.09.003.
vez Michaelsen, A. B., Huamani Bricen ~ o, L., Fernandez Menis, R., Bejar Cha Chura, N., Valera Tito, F., Perz, S., et al. (2013). Regional deforestation trends within local realities: land-cover change in Southeastern Peru 1996e2011. Land, 2, 131e157. Crist, E. P., & Cicone, R. C. (1984). A physically-based transformation of thematic mapper data-the TM tasseled cap. IEEE Transactions on Geoscience and Remote Sensing, 22, 256e263. Crist, E. P., Lauren, R., & Cicone, R. C. (1986). Vegetation and soils information contained in transformed thematic mapper data. In Final Proceedings IGARSS’86 Symposium, Zürich, Switzerland, 8e11 September (pp. 1465e1470). Noordwijk, The Netherlands: ESA Publ. Division.
n Regional (CTAR). (2002). Plan estrat
Consejo Transitorio de Administracio egico concertado de desarrollo Madre de Dios 2002-2004. Puerto Maldonado, Perú. Escobal D'Angelo, J. (1992). Impacto de la política de precios y de cr
edito agrícola sobre
n del ingreso en el Perú: 1985-1990. Documento de trabajo 18. Lima, la distribucio Peru: GRADE. Fearnside, P. M. (2000). Global warming and tropical land-use change: greenhouse gas emissions from biomass burning, decomposition and soils in forest conversion, shifting cultivation and secondary vegetation. Climatic Change, 46, 115e158. Fiorella, M., & Ripple, W. J. (1993). Determining successional stage of temperate coniferous forests with Landsat satellite data. Photogrammetric Engineering and Remote Sensing, 59, 239e246. Fox, J., Rindfuss, R., Walsh, S., & Mishra, V. (Eds.). (2003). People and the Environment. Approaches for linking household and community surveys to remote sensing and GIS. Boston, USA: Kluwer Academic Publishers. Geist, H. J., & Lambin, E. F. (2002). Proximate causes and underlying driving forces of tropical deforestation. Bioscience, 52, 143e150. Gentry, A. H. (1988). Tree species richness of upper Amazonian forests. PNAS, 85, 156e159. Green, G. M., Schweik, C. M., & Randolph, J. C. (2005). Retrieving land-cover change information from landsat satellite images by minimizing other sources of reflectance variability. In E. F. Moran, & E. Ostrom (Eds.), Seeing the forest and the trees: Human-Environment interactions in forest ecosystems (pp. 131e160). Cambridge: The MIT Press. Guild, L. S., Cohen, W. B., & Kauffman, J. B. (2004). Detection of deforestation and ^ nia, Brazil using change detection techniques. Interland conversion in Rondo national Journal of Remote Sensing, 25, 731e750. Gutman, G., Janetos, A., Justice, C., Moran, E., Mustard, J., Rindfuss, R., et al. (Eds.). (2004). Land change science: Observing, monitoring, and understanding trajectories of change on the earth's surface. New York, NY, USA: Kluwer. Hecht, S. (2005). Soybeans, development, and conservation on the Amazon Frontier. Development and Change, 36(2), 375e404. Huang, C., Wylie, B., Homer, C., Yang, L., & Zylstra, G. (2002). Derivation of a Tasseled cap transformation based on Landsat 7 at-satellite reflectance. International Journal of Remote Sensing, 23, 1741e1748.
n de la Amazonia Peruana (IAP) e Consejo Transitorio de Instituto de Investigacio
n Regional (CTAR). (2001). Madre de Dios, camino al desarrollo Administracio
n ecolo
gica econo
mica como base para el sostenible: Propuesta de zonificacio ordenamiento territorial. Iquitos, Perú: Impresiones C.E.T.A.

233


n de los Estados Americanos Instituto Nacional de Desarrollo (INADE) e Organizacio (OEA). (1998). Plan de desarrollo de las comunidades fronterizas Peruano~ as. In ~ apari-Assis Brasil: CIDI/OEA. brasilen Instituto Nacional de Desarrollo (INADE) e Proyecto Especial de Madre de Dios ~ o 2004. Puerto Maldonado, (PEMD). (2004). Plan operativo institucional para el an Perú: PEMD.
lisis socio
tica (INEI). (2007). Ana Instituto Nacional de Estadísticas e Informa
fico del departamento de Madre de Dios. Lima, Perú: INEI. demogra
n de los Estados Instituto Nacional de Recursos Naturales (INRENA) e Organizacio
n ecolo
gica-econo
mica Yaco-In ~ apari e IbeAmericanos (OEA). (1998). Zonificacio ~ apari. Lima, Perú: Instituto Nacional de Desarrollo-Organizacio
n de ria- In Estados Americanos. Instituto Nacional de Recursos Naturales (INRENA). (1999). Monitoreo de la
n y Zonificacio
n Ecolo
gica Econo
mica en la Amazonia de Perú y Deforestacio Brasil. In Informe Ejecutivo Seminario y Taller Internacional. INRENA, Instituto Nacional de Recursos Naturales, 20e22 Octubre. Lima, Perú: ESERGRAF Impresiones. International Tropical Timber Organization (ITTO). (2005). Evaluation of commercial stocks and strategy for the sustainable management of mahogany (Swietenia macrophylla) in Peru. Project Document. PD 251/03 Rev.3(F). Ivits, E., Lamb, A., Langar, F., Hemphill, S., & Koch, B. (2008). Orthogonal transformation of segmented SPOT5 images: seasonal and geographical dependence of the tasseled cap parameters. Photogrammetric Engineering & Remote Sensing, 74, 1351e1364. Kauth, R. J., & Thomas, G. S. (1976). The tasseled cap-a graphic description of spectral-temporal development of agricultural crops as seen by Landsat. In Final Proceedings: 2nd International Symposium on Machine Processing of Remotely Sensed Data. West Lafayette, IN, USA: Purdue University. Lambin, E. F., & Geist, H. J. (2003). Regional differences in tropical deforestation. Environment, 45, 22e36. Lambin, E. F., & Geist, H. J. (Eds.). (2006). Land-use and land-cover change: Local processes to global impacts. Heidelberg: Springer Verlag. Mather, A. S. (2006). Land use policies. In H. J. Geist (Ed.), Our earth's changing land: An encyclopedia of land-use and land-cover change (Vol. 2). Westport, London, UK.: Greenwood Press. Malhi, Y., & Grace, J. (2000). Tropical forests and atmospheric carbon dioxide. Trends in Ecology and Evolution, 15, 332e337.
n. MinisMinisterio de Agricultura (MINAG). (2005). Plan Nacional de Reforestacio terio de Agricultura (MINAG)/INRENA/PRONAMACHCS/FONDEBOSQUE/BSD/ IIAP.
n de la frontera agrícola en la amazonia peruana. In Mora, C. (1993). Expansio
nchez, & G. Pajares (Eds.), Hacia la recuperacio
n econo
mica y social. Lima, J. Sa Perú: CEDEP, CIED, CIPA, DESCO, CIPCA, Arariwa. Moran, E. F., & Ostrom, E. (Eds.). (2005). Seeing the forest and the trees. Humanenvironment interactions in forest ecosystems. Cambridge, Massachusetts and London, England: The MIT Press. Naughton-Treves, L. (2004). Deforestation and carbon emissions at tropical frontiers: a case study from the Peruvian Amazon. World Development, 32, 173e190. Oliveira, P. J., Asner, G. P., Knapp, D., Almeyda-Zambrano, A. M., GalvanGildemeister, R., Keene, S., et al. (2007). Land-use Allocation Protects the peruvian amazon. Science, 317, 1233e1236. Perz, S. G., Araujo Carvalho, L., Castillo, J., Chavez, A., Cossio, R., Hoelle, J., et al. (2014). Trans-boundary infrastructure, access connectivity, and household land use in a tri-national frontier in the southwestern Amazon. Land Use Science. http:// dx.doi.org/10.1080/1747423X.2014.898104. Perz, S. G., Qiu, Y., Xia, Y., Sun, J., Southworth, J., Marsik, M., et al. (2013). Transboundary infrastructure and land cover change: highway paving and community-level deforestation in a tri-national frontier in the Amazon. Land Use Policy, 34, 27e41. Perz, S. G. (2003). Social determinants and land use correlates of agricultural technology adoption in a forest frontier: a case study in the Brazilian Amazon. Human Ecology, 31, 133e165. R Core Team. (2013). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Rudel, T. (2005). Tropical forests: Regional paths of destruction and regeneration in the late twentieth century. New York: Columbia University Press. Schmink, M., & Wood, C. (1992). Contested frontiers in Amazonia. New York: Columbia University Press. Tahuamanu Municipalidad Provincial. (2001). Plan estrat
egico de desarrollo. Documento de trabajo. Tahuamanu, Madre de Dios, Perú: Municipalidad Provincial Tahuamanu. Trivelli, C., Shimzu, T., & Glave, M. (2003). Economic liberalization and evolution of rural agricultural sector in Perú. L.A.S. Series No. 2. Institute of Developing Economies IDE-JETRO. Turner, B. L., II., Geoghagen, J. G., & Foster, D. R. (2004). Integrated land change sci
n: Final frontiers. Oxford, ence and tropical deforestation in the southern Yucata UK: Clarendon Press/Oxford University Press. Walker, R., Browder, J., Arima, E., Simmons, C., Pereira, R., Caldas, M., et al. (2009). ^ nia in the 21st century. Geoforum, Ranching and the new global range: Amazo 40(5), 732e745. Walsh, S., & Crews-Meyer, K. (Eds.). (2002). Linking people, place and policy: A GIScience approach. London: Kluwer Academic Publishers. Wood, C. H., & Porro, R. (Eds.). (2002). Deforestation and land use in the Amazon. Gainesville, FL, USA: University of Florida Press.