Ecosystem services, agriculture, and rural poverty in the Eastern Brazilian Amazon: Interrelationships and policy prescriptions

Transcription

1 available at Ecosystem services, agriculture, and rural poverty in the Eastern Brazilian Amazon: Interrelationships and policy prescriptions Jan Börner a, Arisbe Mendoza b, Stephen A. Vosti c, a International Center for Tropical Agriculture (CIAT), Amazon Initiative Consortium, Belem, Brazil b University of Bonn, Center for Development Research (ZEF), Germany c Department of Agricultural and Resource Economics, Center for Natural Resources Policy Analysis, University of California, One Shields Avenue, Davis, CA, , United States ARTICLE INFO Article history: Received 19 May 2006 Received in revised form 25 February 2007 Accepted 1 March 2007 Available online 18 April 2007 Keywords: Biodiversity Agriculture Carbon Brazil Deforestation Payment for ecosystem services JEL classification: Q12;Q18;Q16;Q56 ABSTRACT Policymakers seeking to modify financial incentives to increase the flows of ecosystem services in and around tropical moist forests must consider where to focus their attention and what collection of incentives can effectively achieve policy objectives. In most cases, policymakers focus on extensively forested areas where the flows of ecosystem services between agriculture and the environment is generally characterized by massive flows of carbon and soil nutrients from forests to agriculture. In these forest margin areas the stock of primary forest is eventually exhausted and the cheap ingredients provided by nature to agriculture become increasingly scarce. At this point, policy interest generally wanes, and agriculture and the environment begin slow declines in ecosystem service exchange, often with negative consequences for rural poverty. How does one promote increased flows of ecosystem services from agricultural lands without increasing poverty when forests and soils have been depleted? Can the standard instruments, e.g., payments for ecosystem services, be effective in such situations, and if so, do the costs to society of securing these services increase? Here we focus on the flows of ecosystem services at the end of the cycle of converting primary forest to agriculture. Primary data from the Bragantina area in the southeastern Brazilian Amazon, an area cleared of primary forest decades ago, are used to characterize smallholder production systems, to describe the flows of ecosystem services into and from these systems, and to develop a bioeconomic model of smallholder agriculture capable of predicting the effects of several types of policy action on ecosystem services provided by and to agriculture, and on-farm household incomes and food selfreliance. Of particular interest is the Proambiente Pilot Program in Brazil, which uses smallholder payment schemes to induce farmers to manage land and forest resources in ways that generate more ecosystem services. Baseline results suggest that smallholder agriculture leads to a gradual loss of ecosystem services (mainly above-ground and root carbon) provided by secondary forest fallows, and that reduction in fallow age leads to reductions in plant diversity. Intensifying agricultural activities accelerates this process, but Paper prepared for the Special Issue of Ecological Economics focusing on Ecosystem Services and Agriculture. Corresponding author. Fax: address: vosti@primal.ucdavis.edu (S.A. Vosti) /$ - see front matter 2007 Elsevier B.V. All rights reserved. doi: /j.ecolecon

2 357 considerably increases smallholder incomes. Paying farmers for ecosystem services linked to the retention of secondary forests and the Proambiente program both increase area in forest fallow, but the latter substantially reduces farm income because of input use restrictions. In general, programs aiming to promote the production of ecosystem services should not limit farmers' choices of ways to provide them. Employment and food selfreliance issues associated with policy options for increasing on-farm stocks of carbon and plant biodiversity are also explored Elsevier B.V. All rights reserved. 1. Introduction Tropical moist forests are regarded as important providers of ecosystem services, such as biodiversity, carbon sequestration, and the regulation of water and nutrient cycles. Most of these services contribute to and sustain human life at local, regional and even global scales (Metzger et al., 2006). Even when disturbed, e.g., via the complete removal of forest cover, these ecosystems continue to provide, at least for awhile, services that are critical to agriculture (Hoelscher, 1997; Blanche et al., 2006). For example, the conversion of tropical moist forests to agriculture is generally characterized by massive flows of carbon and soil nutrients from forests to agriculture (Palm et al., 1996; Fujisaka et al., 1996). Depending on product mix and management strategies, some carbon is re-sequestered and some measures of biodiversity can increase, but the reduction in overall ecosystem service provision following conversion of forest to cropland and rangeland is generally large and permanent (Sanchez et al., 2005; Tomich et al., 2005; Gockowski et al., 2001; Vosti et al., 2001b). Whether or not the local population in these areas can escape poverty depends on the absolute amounts and qualities of natural capital (primarily forests, soils, and water) available for transformation into other forms of capital, and the terms of trade available for this transformation, which are given by the net proceeds from agricultural production relative to the values of other forms of desired capital (e.g., housing, health, education) and the returns provided by them. Eventually, the stock of primary forest is exhausted and initially cheap ingredients provided by nature to agriculturebecomescarceresources.atthispointintheforestto-agriculture conversion process, secondary forest fallows managed by smallholders generally become the primary source of natural soil enrichments, and in the absence of purchased inputs to enhance soil fertility, agricultureandtheenvironment begin a slow decline in ecosystem service exchange, often with increases in rural poverty (Wunder, 2001). With primary forest cover exhausted, policymakers charged with managing forest ecosystems move on to areas with more extensive forest cover, turning over agriculture, forest remnants and secondary forest fallows that occupy previously forested areas to those who must now attempt to manage agriculture, often practiced by cashstrapped farmers, on depleted soils. In such settings, ecosystem services are usually forgotten. But that should not and need not always be the case. It may be possible in areas virtually denuded of primary forest to begin to rebuild forest-based ecosystem services and agriculture, simultaneously, and with positive consequences for poverty. Moreover, the set of policies aiming to increase the flow of ecosystem services from agriculture may have symbiotic effects on agricultural productivity, with some of the benefits associated with enhanced ecosystem services accruing to local farmers; if this is the case, farmers may be willing and able to pay for some of the costs of such policies. Before taking action to induce farmers in such situations to provide these ecosystem services, policymakers need to know two things. First, what measurable changes in ecosystem services will likely emerge from alternative land use and land management strategies? Second, since policymakers usually pay for such services indirectly by providing incentives for changes in land use patterns, what will be the expected effects on farmers' land use decisions of an array of regulatory and other available policy options, and with what consequences for poverty (Vosti et al., 2005)? Here we use primary data collected from small-scale agriculturalists in the eastern Brazilian Amazon to examine these issues. The Brazilian Amazon region hosts the world's largest tropical moist forest. To date, roughly 16% of the forest area has been cleared from primary forest and deforestation rates remain high (Fearnside, 2005). In view of the growing demand for crop and livestock products worldwide (Rosegrant et al., 2001), it can be expected that the pressure on natural habitats that provide important ecosystem services will increase and be especially intense in areas amenable to rainfed agriculture. Yet, even if deforestation could be halted, large areas around tropical moist forests will remain covered by agricultural crops, pastures, or secondary forests. Albeit at lower levels, these areas continue to provide global and local ecosystem services. In the context of many tropical and other countries, maintaining ecosystem service provision will therefore require developing ecoagriculture strategies that enhance the contribution of farming systems to ecosystem services (McNeely and Scherr, 2003; Scherr and McNeely, 2003; Swinton et al., 2006). In the agricultural landscapes of the Amazon, secondary forests are the main contributors to carbon sequestration and biodiversity (Kammesheidt, 2002; Palm et al., 2004). They also serve as fallows and provide important local, primarily agronomic, ecosystem services that contribute to agricultural productivity such as nutrient and water recycling from deep soil layers or ph control and fertilization in slash-and-burn systems (Nepstad et al., 1994; Hoelscher, 1997; Paparcikova et al., 1999; Sommer et al., 2000). Yet, the amount and quality of both global and local ecosystem services from these areas depend critically on the land uses that prevail and on the types of technologies used in agriculture. Today, the approximately 0.5 million small-scale farmers in the Brazilian Amazon region can choose from a set of available cropping patterns and production technology options ranging from traditional fallow-based slash-and-burn (S&B), to agroforestry systems, to intensive cropping using modern agricultural

3 358 ECOLOGICAL ECONOMICS 64 (2007) inputs and mechanical land preparation (Vosti et al., 2003; Valentim and Vosti, 2005). Each of these cropping pattern/ production technology options has different streams of costs and benefits, and delivers different amounts and types of ecosystem services. Moreover, many of these options require significant amounts of cash and labor to establish and manage, which can affect adoption by labor- and/or cash-constrained farmers (Reardon and Vosti, 1997). In the Brazilian Amazon region, policymakers are aware of some of the constraints farmers face with regard to natural resource conservation and meeting livelihood security objectives. Several established and innovative policy measures are in place or are being tested to address rural poverty and the environmental consequences of agriculture. Prominent examples are: 1. The Reserva Legal, a (poorly enforced) federal policy that requires farmers to retain a certain amount of farm area under forest cover. Current federal decrees relevant to most Amazonian states mandate that 50 80% of farm area be retained in forest. 2. Proambiente, a federal program that provides farmers with subsidized credit for production practices that are assumed to enhance ecosystem service flows from agriculture. Proambiente is currently being tested across the Brazilian Amazon. 3. Patrulhas mecanizadas, a program that provides agricultural machinery (tractors, plows, and harrows) to farmers at subsidized costs; such machinery is meant to provide a fire-free substitute for traditional S&B methods of clearing secondary forest fallow. With the exception of the Reserva Legal (e.g., Vosti, 2002), little is known about the expected effects of any of these policy options on farm-level land use decisions, and hence on the flows of ecosystem services, on-farm income, or on other factors of importance to smallholders. In this study, we address the following research questions: 1. What measurable ecosystem services are provided by specific smallholder land use practices in the study area? 2. What do these same practices contribute to farm income and to food self-reliance? 3. What will be the effects on smallholder land use decisions (and hence on ecosystem services and indicators of farm household welfare) of the selected of policy instruments, among them the policies noted above? To address these research questions we first provide background information on the study site (next section). We then apply land use system (LUS) analysis to agricultural practices Fig. 1 The study sites in the eastern Brazilian Amazon. The dark grey highlighted municipalities were selected as representative of the light grey region that comprises the Zona Bragantina.

4 359 commonly found at the research site, as well as to some innovative land uses currently being tested, some in the context of the Brazilian Proambiente program. Finally, we develop a whole-farm bioeconomic model to predict the effects of policy changes on land use, ecosystem service flows, and farm household income. We end with a discussion of the policy implications of our LUS and bioeconomic modeling results for the study area and more broadly, with particular focus on the often conflicting objectives of ecosystem service provision and poverty alleviation. 2. Study area and data The Bragantina area, a forest frontier area many decades ago (Penteado, 1967), is located in the northeast of the state Pará in the southeastern Brazilian Amazon, close to the urban center Belém in the eastern part of Legal Amazon (Fig. 1). Field data to support the LUS analysis and the development of the bioeconomic model were collected from 270 randomly selected farm households in three municipalities during the 2002/3 agricultural year. Secondary sources at municipal level were used for site characterization and for price data on agricultural inputs and products (Börner, 2006; Mendoza, 2005). Table 1 reports summary agricultural and regional statistics for three municipalities within the Bragantina area (Igarapé-Açu, Bragança, and Castanhal) and for representative smallholder farms (i.e., b200 ha), and compares these to statelevel statistics. Several key points merit mention. First, the extent of primary forest cover on farms (row one of Table 1) declines markedly from the state to the municipal to the smallholder levels. More than half of the primary forests at state level remain intact, whereas the municipalities in which research was undertaken have long ago been almost denuded of primary forest. Indeed, in the Bragantina area, primary forests cover only about 10 to 30% total farmland and have almost been eliminated on representative smallholder farms (sample and site selection criteria are documented below). Remaining forests are located in protected areas along streams and rivers, or on land unsuitable for agriculture (e.g., waterlogged areas). Hence, the conversion of primary forest for agricultural purposes is no longer an issue; the management of secondary forest fallow, which can generate ecosystem services similar to those provided by primary forests, is the key forest management issue. Second, the extent and use of cleared land (rows two through six of Table 1) in the study sites also differ substantially from state-level patterns. At the state level, cleared land is most commonly dedicated to pasture. At the municipal level within the study area, secondary forest fallow is the predominant land use, followed by pasture, annual crops, and perennial crops. Smallholder land use mimics municipal-level patterns, but pastures are less common among smallholders. The area dedicated to annuals, fallow, perennials, and pasture varies widely within the sample. Although the Bragantina area has supported subsistence and commercial agriculture for over a century, access to markets still varies across municipalities and this varied access is a major factor in determining land use patterns, especially for perennials (mainly black pepper and passion fruit). For example, Bragança is located furthest from the state capital of Belém and has the smallest amount of land dedicated to Table 1 Key characteristics of the study site: State of Pará, municipalities of Castanhal, Igarapé-Açu, and Bragança, and representative smallholdings State Municipality Smallholdings b200 ha Pará Castanhal Igarapé-Açu Bragança Castanhal Igarapé-Açu Bragança Land cover % in primary forest and riparian vegetation % non-primary forest and cultivated area % in annuals (of total) % in fallow/secondary forest (of total) % in perennials (of total) % in pasture (of total) % other (of total) n.a. n.a. n.a. Productivity a Gross value of agricultural output per area (US$/km 2 ) Gross value of agricultural output per inhabitant (total population) (US$/person) Gross value of output per rural inhabitant (US$/person) ,017 25,810 11, Population Population growth (% year, ) % Urban (1996) Population density (inhabitants/km 2, 1996) All values are expressed in 2002/3 US$. Sources: IBGE Censo Agropecuário 1996/97, Censo Demográfico 2002/1996, ZEF/NAEA/Embrapa field data. a State- and municipal-level gross value of total agricultural production during the 1996/97 cropping year, excluding extractive activities.

5 360 ECOLOGICAL ECONOMICS 64 (2007) perennial crops (2.5%) and the largest amount dedicated to annual crops (11.1% of land in the municipality). Smallholder land use patterns differ only for annual crop production, which occupies a larger share of farms in Castanhal than in Bragança. Third, the gross value of agricultural output (the third set of rows in Table 1) also varies significantly across sample municipalities and the groups of selected farm households within them. For example, for the 1996/97 cropping year, the municipality of Igarapé-Açu generated the highest value of agriculture per hectare and per rural inhabitant; the municipality of Bragança was the poorest of the sample municipalities by these measures, while the municipality of Castanhal occupied an intermediate position. Household-level statistics roughly accompany those of their respective municipalities. However, all three municipalities generated value of output per area and value of output per rural inhabitant far above state-level averages. Fourth and finally, state-wide, population density is quite low (4.4 individuals per km 2, fourth set of rows in Table 1), but in the long-settled Bragantina area population density is approximately ten times higher in all municipalities and rises to a high of 114 individuals per km 2 in Castanhal. The very high population density in Castanhal is likely a result of Castanhal's proximity to Belem and a development pattern that has focused on industry and trade. 3. Land use system analysis (LUS) A side-by-side comparison of land uses regarding ecosystem services and the extent to which land uses meet human needs is only possible when based on comparable spatial and temporal scales; land use system (LUS) analysis was developed to meet these needs (Tomich et al., 1998; Vosti et al., 2001b). For the most common land uses in the study area we compiled agronomic, biophysical, and economic data that were collected in the Bragantina region between 1992 and 2003 by the German Brazilian research project Studies on Human Impact on Forest and Floodplains in the Tropics (SHIFT). These data were used to measure the contributions of specific land uses to the provision of two ecosystem services (carbon and plant biodiversity), and to measure the extent to which these land uses fit smallholder circumstances (regarding, e.g., available household labor) and met smallholder income and food security needs. The following commonly encountered and experimental land use systems (each practiced on the same parcel of land, over an 18-year period) were selected for analysis: traditional annual/fallow two-year intercropping of cassava and maize, followed by a fallow period of 6 years with fallow vegetation being cleared using S&B techniques; mechanized annuals continuous cropping of cassava and beans using mechanical plowing and chemical fertilizers, without fallow periods; mulching annual/fallow four-year intercropping of cassava and maize, followed by a four-year fallow that is enriched with fast growing leguminous trees; fallow vegetation is cleared using mechanical mulching machines; secondary forest a 20-year-old stand of forest fallow; black pepper a nine-year cycle of black pepper cultivation followed by 9 years of fallow; passion fruit a two-year cycle of passion fruit production followed by a seven-year fallow period; traditional pasture a continuous, pasture-based, mixed milk beef production system with low stocking rates (approximately 0.5 animal units per hectare) and no purchased inputs. All land uses spanned 18 years (with repeated cycles, in some cases) and include establishment periods and costs; reported results (Table 2) are standardized to a per hectare basis. Measures of the ecosystem services provided by these various land use systems (each identified in column one of Table 2) are reported in columns two through four of Table 2.As expected, above-ground carbon is highest by far in the 20-yearold forest fallow system (60.8 tons/ha) and lowest in the mechanized annual system; all other LUS occupy the low range of approximately 7 to 9 tons per hectare. Plant biodiversity is measured using the Shannon indices of diversity and the evenness (Ludwig and Reynolds, 1988). Both biodiversity indices clearly reflect the positive impact of forest fallow, and specifically fallow length, on plant biodiversity. The final six columns of Table 2 focus on issues related to smallholder welfare and technology adoption. Returns to land and especially returns to farm household labor (columns five and six of Table 2, respectively) are fundamental to meeting household needs and influencing labor allocation decisions. Black pepper and passion fruit dominate other LUS by a wide margin as regards returns to land; traditional cattle/pasture production yields negative returns to land. In calculating the returns to land, market prices are used to value household labor used by each LUS (Vosti et al., 2001b). Returns to labor (wage paid to household labor required to set NPV to zero, Vosti et al., 2001b) are much more uniform across LUS, in part due to the very broad range in the amount of labor required to manage different systems, with passion fruit generating the highest returns. Comparing labor requirements during the operational phase of the LUS (column 7 of Table 2), the pasture system requires the least amount of labor to manage (24 person-days ha 1 year 1 ), followed by the traditional annual/ fallow system, which requires 31 person-days ha 1 year 1.Black pepper and the mechanized annuals system require three and five times (respectively) more labor than the traditional system. For these LUS, hired labor is needed seasonally, and apart from the initial establishment costs, farmers must have fairly advanced crop management skills. Although the Bragantina region has been settled for many decades, food security is still an issue for many smallholders. The third-to-last column of Table 2 identifies the path, or paths, through which a household engaged in a specific LUS must proceed in order to obtain food. At one extreme, some LUS generate food that can be directly consumed (e.g., traditional annual/fallow), while others (e.g., black pepper) rely on product and food markets to secure food. The final two columns of Table 2 provide signals regarding the market and non-market factors that can influence the adoptability of a given LUS by smallholders. Market factors

6 361 Table 2 Ecosystem services and socio-economic indicators for selected land use systems Ecosystem Services Farmer Concerns Land use systems Carbon: aboveground (tons/ha) Plant diversity aboveground species (Shannon indices) Returns to land (US$/ha) Returns to labor (US$/personday of household labor) Labor requirements (persondays/ha/year) Food security entitlement path c Institutional constraints H a EH b Market d Non-market d Traditional $ + consumption r le annual/fallow Mechanized $ + consumption k n annuals Mulching $ + consumption k n annual/fallow Secondary consumption o forest (20-year-old) Black pepper $ k le Passion fruit $ k le Traditional pasture $ + consumption o le All values are expressed in 2002/3 US$. Source: ZEF/NAEA/Embrapa field data and IBGE PAM for additional information on output prices. a Shannon measure of species diversity. b Shannon measure of species evenness. c Path or paths through which households participating in a given LUS secure food: $ = via product and/or food markets, consumption = direct consumption of products produced. d Letters indicate institutional constraints to adoption: inp, input market; o, output market; lb, labor market; k, capital markets; n, information; r, regulatory issues; le, local environmental constraint/problem (e.g., plant diseases). refer to imperfectly functioning output or input markets, especially credit markets. Non-market factors refer to, e.g., the unavailability of information regarding best practices for LUS establishment or management. In the Bragantina region, in general, output and input markets function reasonably well. Yet, lack of information about product and input prices, and lack of bargaining power can put smallholders at a disadvantage in negotiations with traders (hence lowering expected profits). Moreover, more remote areas are less attractive to traders and inaccessibility, particularly during the rainy season, hampers farmers' access to markets. Perennial products are especially affected by large seasonal and inter-annual price variations, so price risk for these products can be large. Chief among non-market constraints is the lack of information and management expertise for the two perennial systems (black pepper and passion fruit) and the two alternatives to traditional slash-and-burn techniques (mechanical mulching and mechanized annual crop production). While the LUS analysis provides important insights regarding the provision of ecosystem services by particular systems, and the extents to which these systems meet household income objectives and food needs and are compatible with labor availability conditions, the analysis stops short of actually identifying which LUS or combination of LUS will be adopted, and how much land will be dedicated to each, and why. Such information is essential to estimating the aggregate environmental and income effects of policy changes. In the next section we develop and use a research tool capable of meeting these challenges. 4. A farm-level bioeconomic model 4.1. A farm household model To determine the effects of alternative policy actions on farmlevel product mix, production technology, the extent and management of secondary forest fallow, and the consequences of all these for ecosystem services, a multi-period bioeconomic optimization model that accounts for fallow-yield relationships and tracks whole-farm root and above-ground carbon and fallow age (an indicator of plant biodiversity) over time was developed (Börner, 2006). We specified, calibrated, and validated the model based on farm household data and technical coefficients collected from farmers and farming experts in 2002/3. The model maximizes the discounted value of nonessential consumption (a food security constraint requires households to meet essential food needs) resulting from the inter-temporal allocation of income from agricultural activities and the sale of household labor, and savings, over a period of 25 years (following Carpentier et al., 2000; Vosti et al., 2001a; Vosti et al., 2002; Carpentier et al., 2005). Aversion to price risk was included in the linear objective function of the model through parameterization of a Mean of Total Absolute

7 362 ECOLOGICAL ECONOMICS 64 (2007) Deviations (MOTAD) risk aversion factor (Hazell and Norton, 1986). The optimization is subject to agronomic, biophysical, and socio-economic constraints. Agronomic constraints involve rotational requirements with respect to possible land use sequences. For example, land has to go into a rehabilitation cycle (forest fallow) after being used for pasture or intensive perennial production. Biophysical constraints require that sufficient nutrients be available during the cropping season to attain maximum potential crop yields. If nutrient deficits occur, crop yields are reduced according to crop-specific nutrient response functions (Witcover et al., 2006). Socio-economic and market constraints limit the amount of hired labor that can be imported onto a farm and the amount of family labor that can be exported off of the farm. The amount of cash that can be borrowed is limited (and all loans must be repaid annually) and the household faces minimum consumption requirements with regard to food and non-food items and leisure, all of which depend on family size and composition and household income. (For details of the model, see the Technical Appendix included at the end of this paper.) For this study, the model was calibrated for one of five types of farm households that were identified using a cluster analysis of characteristics related to land use intensity, household wealth, and market access (Börner, 2006). 51% of the sample households in the Bragantina area belong to this subgroup Baseline simulation A baseline simulation was run for the archetypical farm household in the Bragantina area. This simulation assumes that 10-year ( ) average prices for agricultural outputs and 2002/3 prices for agricultural inputs remain constant over the model's 25-year planning horizon. Land uses that emerge from the baseline simulation correspond to observed land uses in 2002/3 and are hence representative of the types of activities typically found on farms in the Bragantina area. The model predicts that smallholders will produce one or more of the following: cassava, cassava and beans, cassava and maize, beans, passion fruit, black pepper, and low-input pastures in a S&B system. The income and other results that emerge from the baseline scenario are discussed below alongside the results of the policy scenarios Policy scenarios The following scenarios were constructed to measure the response of the archetypical household as regards several key rural development objectives (chief among them household income, food security, and carbon stocks) to a series of policy actions, some of which are already being pilot tested in the Bragantina area. 1. The use of alternative technologies simulation allows the archetypical farmer access to (experimental) mechanized land preparation technologies for all crops and an improved (but not broadly adopted) pasture management system. This scenario includes the provision of a mulching machine service at a subsidized cost. All other technologies are available at market prices. 2. The payment for ecosystem services simulation pays the farmer US$55 annually for each hectare of forest fallow aged 15 years or older that could be converted to agriculture but instead is maintained. Product mix and technology choice options are the same as in the baseline simulation. Table 3 Summary results of baseline and policy simulations Simulation label Farm Food security b Hired labor Value of fallow c Total carbon d Annual income a carbon loss d (average US$/year) (average daily production of calories/household member) (average person-days/ year for system management) (average marginal value of fallow US$/ ha) (average tons/ ha) (average tons/ha/ year) Baseline Use of , alternative technologies Payment for ecosystem services Tax on S&B , Proambiente e Proambiente e modified Reserva Legal f All values are expressed in 2002/3 US$. Source: ZEF/NAEA/Embrapa field data. a Time average of annual non-essential consumption. b Time average of calories from food crops and beef products. c Time average of the marginal value of the constraint on land under fallow. d Time average of root and above-ground carbon. e Higher than baseline due to change in product mix. f Higher than baseline due to very high annual losses in initial years.

8 363 Fig. 2 Land uses and average forest fallow age in the baseline simulation. 3. The tax on slash-and-burn agriculture (S&B) simulation requires the farmer to make a one-time payment of US $104 per hectare of land that is converted from forest fallow to agriculture using S&B technologies. This value was determined through sensitivity analyses undertaken to identify the tradeoffs among tax revenue, ecosystem service provision, and farm income. Product mix and technology choice options are the same as in the baseline simulation. 4. The Proambiente simulation assumes that payments for ecosystem services will be provided in the form of a 40% discount on agricultural credit repayment if farmers use only mechanical mulching for land preparation. Perennial cash crops are less productive, because they require the use of chemical fertilizers, which are not permitted under the current Proambiente program. Under the Proambiente scheme, mechanical mulching services are offered at full market cost. Agricultural credit of up to US$1731 (R$5000, in 2003) is available at a subsidized nominal annual interest rate of 4%; 60% of the loan amount must be paid back within the first 5 years of loan disbursement. 5. The Proambiente modified simulation is similar to the Proambiente simulation, but in this case farmers are allowed to use chemical fertilizers. In addition, mechanical mulching is offered at a subsidized rate of 30% of the estimated service cost. 6. The Reserva Legal simulation requires the farmer to set aside 5 ha of land (approximately 32% of total farm area) for secondary forest fallow beginning in the fifth year of the 25-year planning horizon. 5. Results of the baseline and policy simulations The summary results of the baseline and policy simulations are reported in Table 3. Column headings report, respectively, labels for simulations, farm income (time-averaged net annual household income), food security (the total kilocalories person 1 Fig. 3 Land uses and average forest fallow age in the use of alternative technologies scenario.

9 364 ECOLOGICAL ECONOMICS 64 (2007) Fig. 4 Land uses and average forest fallow age in the payments for ecosystem services scenario. day 1 that is contained in food products produced) IBGE (1999); hired labor (i.e., non-family labor contracted at rural wage rate), value of secondary forest fallow (time-averaged shadow value of secondary forest fallow), total carbon (time-averaged tons of root and above-ground carbon per hectare for the entire farm Sommer et al., 2000; Denich et al., 2000); and annual carbon losses (average annual carbon losses per hectare). Figs. 2 8 that follow Table 3 depict predicted land uses and the average age of secondary forest fallow over the entire 25-year farm household decision time horizon beginning with year one; all scenarios start in year zero with the identical set of farm and farm household initial conditions that were derived from field data The baseline simulation The benchmark for evaluating the results of the policy simulations is the steady-state baseline simulation presented in the second row of Table 3. With an average income of US$ 1407 per year, the average farm household earns slightly more than the Brazilian minimum salary in 2002; this level of income is very stable over the entire 25-year simulation period. The average daily production of calories per person in the baseline exceeds the recommended energy intake of 2500 to 3000 calories per adult (FAO, 1957), indicating that basic nutritional needs can be directly met and that food crops can be produced both for consumption and sale. The farm household relies almost exclusively on family labor, because land, cash, and fallow vegetation are important constraints to production and hence on the demand for hired labor. The time-averaged shadow value of forest fallow is US$ 44 per hectare. Fig. 2 depicts the steady-state nature of the baseline scenario in terms of land uses, even though the stock of root- and aboveground carbon declines by 3.8 tons per year due to the gradual reduction in average fallow age. This reduction has little influence on the overall land use mix because the negative productivity effect of declining fallow age is somewhat Fig. 5 Land uses and average forest fallow age in the tax on slash-and-burn scenario.

10 365 Fig. 6 Land uses and average forest fallow age in the Proambiente scenario. compensated by a shift from cassava production to more fertilizer-intensive bean production and cassava bean intercropping. Highly variable prices for and profitability of perennial crop products such as passion fruit and black pepper prevent risk-averse farmers from dedicating all land to crops that the LUS analysis indicated were the most profitable Use of alternative technologies If the household gains access to mechanized land preparation technologies, the bioeconomic model predicts that these technologies will be adopted and used on part of the farm and that average income will rise by 17% from US$ 1407 to Average daily calorie production would more than triple, in part due to the use of chemical fertilizers. Hired labor for annual crop production increases ten-fold vis-à-vis the baseline. The time-averaged value of forest fallow drops to US$ 3 per hectare because the resulting product mix is dominated by mechanical plowing, a land preparation technique that requires shallow roots and tree stumps be manually removed. This stump and root removal dramatically reduces the re-growth capacity of secondary forest and hence its economic value in terms of nutrient accumulation. Therefore, under this scenario more land is cultivated annually, which reduces secondary fallow cover, increases the area under degraded fallow, and (consequently) annual carbon losses. Fig. 3 helps to further identify the sources of these carbon losses. First, the average age of secondary forest fallows decreases rapidly during the first 6 years of the planning horizon and stabilizes at about 60% of the baseline level. This indicates a reduction in average fallow length, which results in a reduced amount of carbon stored in roots and vegetation, and likely reductions in plant diversity (see Table 2). Moreover, the area under annual crops increases as a result of the adoption of mechanical land preparation at the cost of productive fallows. An additional side-effect of the increased profitability of annual cropping is the elimination of pastures from the land use mix towards the end of the planning horizon. Resources that, in the Fig. 7 Land uses and average forest fallow age in the Proambiente modified scenario.

11 366 ECOLOGICAL ECONOMICS 64 (2007) Fig. 8 Land uses and average forest fallow age in the Reserva Legal scenario. absence of technology change, would have been invested in pasture establishment are now diverted to other land uses Payment for ecosystem services If households were paid US$55 per year for each hectare of forest fallow set aside, and hence maintaining the ecosystem services associated with secondary forests, the bioeconomic model predicts that average annual incomes would increase by US$ 168 per year (over the baseline scenario), but that average annual food crop (kilocalorie) production would fall. Conservation transfers are sometimes criticized for their potentially negative effects on rural labor employment. Our model suggests that the amount of hired labor used on-farm would be unaffected. The reason is a slight increase in labor-intensive perennial cash crop production. Since the payment for ecosystem services represents a risk-free source of income, the additional investment into cash crop production leaves the variability of total farm income almost unaffected if compared to the baseline. Due to the payment for ecosystem services, the timeaveraged shadow value of forest fallow more than triples to US$150 (see Table 3), which leads to an increase in area dedicated to forest fallow, an increase in the average age of secondary fallows (see Fig. 4), and a consequent increase in the amount of on-farm carbon sequestered and in plant diversity. There is little change in the area dedicated to annual crops, if compared with the baseline scenario, but within that category more land is dedicated to fertilizer-intensive bean production and bean cassava intercropping. Moreover, the additional cash made available through the payment for ecosystem services leads to a small increase (vis-à-vis the baseline) in labor-intensive perennial cash crop activities Tax on slash-and-burn If households have access to improved technologies for clearing and preparing land (i.e., mechanical mulching at a subsidized rate and plowing at full service cost), the model predicts that taxing the conversion of fallows using slash-and-burn would eliminate the use of this land clearing practice and slightly reduce income and food production vis-à-vis the use of alternative technology' scenario. Income and food production remain substantially above baseline levels. However, a tax rate of US$ 104 would have only marginal effects on total carbon storage and annual carbon losses when compared to the use of alternative technology' scenario. Fig. 5 depicts the interesting land use patterns that emerge from this simulation. The reduction in fallow age seems to progress less rapidly than in the use of alternative technology scenario; the tax increases the costs of slash-and-burn, which represents an incentive to increase fallow length as a means to raise productivity, as well as to switch to other less-fallow-intensive land uses. In addition, land uses develop into a wave-like pattern that is driven by cyclical investments into black pepper reestablishment, during which less labor and cash are available for annual crop production Proambiente Implementing the Proambiente program (as it is currently written) would cause a precipitous decline in farm income, but food production roughly doubles when compared to the baseline. This outcome is primarily attributable the program's ban on chemical fertilizer use, which causes perennial crop production to become unprofitable (and reduces to zero the land allocated to these crops by about year 8) and causes farm households to expand food cropping. Results of the Proambiente scenario do suggest that the program will increase the amount of carbon retained on the farm and will probably also increase plant diversity (see Table 2), but the cost to farmers in terms of forgone income is very large. The positive environmental effects are captured by Fig. 6 that shows a marked increase in average fallow age and area under secondary forests fallow Proambiente modified If the Proambiente program were modified to allow the use of chemical fertilizers on all crops (especially perennial tree

12 367 crops), farm income would more than double (vis-à-vis the original Proambiente program simulation results). Average daily calorie production would increase to 9649 from 9068 calories per person and annual carbon losses would increase only slightly but still remain far below baseline levels. Fig. 7 depicts the land use changes associated with this modified Proambiente program. Contrary to the original Proambiente program, perennial crops re-emerge as an important land use (and contributor to farm income), while both fallow age and area dedicated to forest fallow decline slightly below original program levels Reserva Legal Lastly, effectively applying the federal forest retention standards to the case of secondary fallows (recall that essentially all primary forests in the study area were removed decades ago) would nearly double the amount of carbon retained on the farm and dramatically increase the age of forest fallow and plant diversity (see Table 2). However, this increase in on-farm ecosystem service provision would come at a cost; household income, calorie production, and use of hired labor would all be reduced compared to the baseline. Fig. 8 suggests that the enforcement of the retention standard would force farmers to scale down annual cropping activities, while holding perennial cash cropping at least constant. This implies an increased vulnerability to inter-annual price fluctuations, which, as mentioned before, are high for perennial cash crop products. The estimated, time-averaged shadow value of forest fallow (US$105) provides a signal of the difficulty policymakers would face in enforcing this land use restriction. 6. Conclusions and policy implications Policymakers in developing countries must wrestle with the dual objectives of reducing poverty and increasing the flows of ecosystem services from rural areas occupied by small-scale agriculturalists. The proper collection of policy actions (e.g., price incentives, land use regulations, new products, new technologies) is not easy to determine, in part because little is known about the effects on ecosystem service flows or on household incomes of changes in land use and land management that alternative policy actions might promote. We address these issues in the context of a long-established forest frontier region in the southeastern Brazilian Amazon, the Zona Bragantina. Our baseline model simulation provides the welcome news that declines in the ability of land uses to generate ecosystem services need not lead to smallholder poverty; farmers can and do adjust product mix and input mix to compensate for less productive (younger) forest fallows, and our results suggest they can be successful in doing so over the medium term, at least. However, smallholder success in such circumstances will be enhanced by policy actions that increase farmer agility in responding to changes in production technology and market opportunities technical assistance with new products and production technologies, agricultural credit, and quick access to information on relative prices all require some form of public policy action in this socioeconomic setting. Many attempts to improve smallholder agricultural production in the Bragantina area have concentrated on laborsaving and cash-intensive land clearing and soil preparation technologies as alternatives to the traditional slash-and-burn technology. Our findings suggest that the relative scarcity of secondary fallow vegetation and labor (during the land preparation season), and the high financial returns to conventional mechanized plowing, will lead farmers to adopt new land preparation technologies, but will also increase the area dedicated to annual food crop production. As a consequence, on-farm carbon stock, which is strongly linked to forest fallow age and the type of land preparation technology used, will decline considerably. Our more static land use system analysis demonstrates that reductions in the age of forest fallow will also lead to reductions in plant diversity. We examine the effects of several types of policy interventions on ecosystem service provision and farm income (among other things); forest set-aside payments, a tax on land converted using slash-and-burn practices, the Brazilian Proambiente program that promotes specific types of land uses, and the effective implementation of the existing federal Reserva Legal program that mandates that farmers retain specific amounts of farmland under forest cover. Forest set-aside payments increased total on-farm carbon stocks, and led to increases in farm income if compared to the baseline. The tax on forest fallow cleared using S&B techniques (while allowing the use of experimental mechanical mulching and plowing) led to a decrease in farm income and a slight reduction in food self-reliance, and did not greatly affect onfarm carbon stocks. The Brazilian Proambiente program as currently written will force unnecessary trade-offs between ecosystem service provision and farm income because chemical fertilizers are not permitted on cultivated lands; relaxing input use restrictions to allow farmers to apply chemical fertilizers will convert trade-offs to synergies as farmers shift out of low-profit annual crops to highprofit perennial crops that require fertilizers. In the absence of evidence on the externality effects of fertilizer use, this change in theproambiente program would seem logical. In general, programs aiming to promote the production of ecosystem services should not limit farmers' choices of ways to provide them. In our simulations, the Reserva Legal program was very successful at increasing the amount of secondary forest cover on farm, but did so at considerable cost in terms of farm household income and food self-reliance. We found no single technology or policy action that would increase carbon stock and plant diversity on farms without bringing about reductions in farm income; single-instrument efforts to deal with environmental problems led to environmentpoverty trade-offs. Some combinations of policies and technologies, on the other hand, were found to be effective at addressing poverty and environmental concerns. For example, model simulations suggest that making mechanical fallow mulching machines available at subsidized rates to farmers who abandon slash-and-burn, while also providing technical assistance and credit (as suggested by the modified Proambiente scenario), would combine to generate long-term increases in income and rural employment. If mechanical mulching can be provided at competitive costs, this policy package could eliminate seasonal smoke and accidental damages from S&B fires.

13 368 ECOLOGICAL ECONOMICS 64 (2007) One important implication of this result is that some very expensive technological innovations (e.g., mulching machines) may not be cost-effective on their own, but may be catalytic in promoting the adoption of a package of forest conversion methods, product mixes, and production technologies that collectively are cost-effective. That said, the taxed S&B scenario demonstrates that local governments need not always offer agricultural machinery services at subsidized rates; farmers may be willing and able to pay market rates for such services in the presence of such a tax. As always, compliance and monitoring issues related to payments for ecosystem services loom large. Policymakers generally pay for changes in land use or land management rather than for ecosystem services directly, and these services can be attributable to different land use characteristics, some of which can be difficult or expensive to measure and monitor. In the Bragantina, forexample,areaundersecondaryfallowisdirectlylinkedto carbon stocks and is easily visible using remote sensing.plant biodiversity, on the other hand, is linked to the age of fallow, which is more difficult to detect from satellite images. Nevertheless, recent evidence suggests that secondary forest age is related to the average height of the highest canopy stratum, which can be estimated using remote sensing techniques (Puig, 2005). Our study area is in many ways unique, even for Brazil. Virtually all primary forest located on arable soils in the Bragantina area was cleared many years ago, and unlike many former frontier areas, smallholders have persisted despite opportunities to migrate to other forest margin areas or to cities, and despite increasingly challenging agricultural circumstances. Nevertheless, the Zona Bragantina may be a window to the future for Brazil and for many other parts of Latin America where primary forest cover is gone, property rights for smallholders are increasingly secure, transportation infrastructure is reducing the distance between farms and markets, smallholder options regarding product mix and production technology are increasing, and especially where ecosystem services are being recognized and valued, and the providers of these services are beginning to be compensated. It may well be that these rapidly expanding former forest frontier areas are the most cost-effective places to begin to rebuild ecosystems and to simultaneously address rural poverty issues. and dedicated much time to explaining it and suggesting ways of adapting and improving it. To these individuals and to the many others who contributed time, effort and expertise, we are indebted. We gratefully acknowledge financial support from the Center for Natural Resources Policy Analysis at the University of California, Davis, the Center for Development Research (ZEF), University of Bonn, and the German Federal Ministry of Education and Research. Field research was supported by the Brazilian Enterprise for Agricultural Research in Belém (Embrapa CPATU), and the Center for Advanced Amazonian Studies (NAEA) of the Federal University of Pará, Brazil. Special thanks go to the rural extension service of the State of Pará (EMATER-PA) and the rural trade unions (STR) of Castanhal, Igarapé-Açu, and Bragança for the time, effort, and expertise they lent to this study. Two anonymous referees and the coeditors of this Special Issue provided very helpful suggestions for improving an earlier draft of this paper. The opinions expressed in this paper are not necessarily those of the supporting agencies. Appendix A. Technical appendix This appendix reports the relevant model equations and constraints. The linear programming model was implemented using the General Algebraic Modeling System (GAMS) and optimized using the Conopt 2 solver package. Fig. FA1 summarizes the basic concept that underlies the model structure. Markets (prices) and climate are given external conditioning factors that influence land use decisions and technology choice at the farm household level depending on the degree of individual risk aversion. As a result of optimizing objective function 13a, the model produces land use and land cover estimates, and internally calculates the related ecosystem services (carbon sequestration) depending on the technologies used for land preparation. Yields of agricultural crops are affected by soil/fallow quality according to Eqs. (5a) (5e) which is a result of farm household decisions on technology choice and product mix. Table TA1 presents the indices used in the model to identify products, inputs used in production processes, and decision time steps. Acknowledgments A large number of individuals contributed to the research project from which this paper is distilled. Tatiana Deane de Abreu Sá provided scientific and logistic support, Socorro and Osvaldo Kato provided key field support and scientific input, Manfred Denich provided important guidance on scientific methods, Francisco de Assis Costa provided input into questionnaire design and testing and guidance on methodological issues, Thomas Hurtienne provided important background information and guidance on methodological issues and research strategy, Ernst Berg and Klaus Frohberg provided thesis supervision and guidance on methodological issues, Bettina Hedden Dunkhorst was project supervisor and in that capacity provided scientific and administrative guidance, and Julie Witcover and Chantal Line Carpentier provided the base linear programming model Fig. FA1 Model structure.