Abstract This article explores the effect of neighborhood land use on land use change in Mato Grosso, Brazil, one of the world’s most dynamic agricultural frontiers. Using an innovative combination of spatial statistics and remotely sensed data, this research shows that the likelihood of an area being converted to agriculture is closely tied to how land is used in a location’s surroundings. The results also suggest that this spatial spillover effect cannot be tied exclusively to the distribution of natural suitability characteristics. Rather, the likelihood of an area being converted to agriculture appears to reflect the dynamic socio-economic conditions of a location’s surroundings. The findings imply that (i) agricultural agencies or experts seeking to support developing agricultural regions should recognize the importance of returns to scale and local clustering and that (ii) land use modeling can be improved by accounting for suitability and land uses in nearby locations. 1. Introduction There is nowhere in Brazil, or perhaps anywhere in the world, where agriculture has expanded at a pace or breadth more astounding than that of Mato Grosso. Since 1990, the Brazilian State of Mato Grosso has more than quadrupled its area of soybean production (1.5 to 8.6 mha), increased its total soybean harvest by nearly 900% (3 million to 26.5 million tons) and raised its average yield from two to over three tons per hectare (IBGE, 2016). On its own, Mato Grosso now produces 8 to 10% of the world’s total supply of soybeans, or what amounts to one-third of Brazil’s yearly harvest. The impacts of Mato Grosso’s soybean expansion have been dramatic. Soybeans have been associated with not only gains in school quality (VanWey et al., 2013), but poverty reduction (Weinhold et al., 2013) and, since 2001, as much as one-half of the state’s economic growth and new formal sector employment (Richards et al., 2015). The environmental costs have been just as profound. From 2000 to 2005, during the height of the state’s soybean expansion period, more than 50,000 km2 of primary, tropical Amazon1 forest were razed in Mato Grosso (INPE, 2015). Still more were cleared in the Cerrado ecosystem (Morton et al., 2006; Macedo et al., 2012; Garrett and Rausch, 2016). During this period, soybeans were linked to forest loss through both the direct conversion of forests to cropland, and indirect land use change (Barona et al., 2010; Lapola et al., 2010; Arima et al., 2011; Richards et al., 2014). Given the pace of agricultural growth in Mato Grosso, and the impact of this growth on the physical and socio-economic environments in the state, Mato Grosso offers a tremendous laboratory in which to examine the spatial expansion of an emerging agriculture sector. Social scientists and land use modelers have already broadly examined the growth of the state’s soybean sector to better understand agricultural suitability, and to predict the likelihood of future land use change. Most often, this work has focused on variables such as local institutions, edaphic qualities, precipitation, topography and access to transportation (Cohn, Gil, et al., 2016; Morton et al., 2016; Soares-Filho et al., 2006). This article builds on these efforts, but goes further by exploring a factor which, to date, remains little understood in land use research, namely the roles of economies of scale and spatial clustering in shaping agricultural landscapes. Economic clustering refers to similar firms’ tendencies to gravitate towards similar locations, or to concentrate or cluster (Isard, 1956; Krugman, 1991; Henderson et al., 1995; Henderson et al., 2001). By clustering, firms can share access to support services, infrastructure development and specialized institutions. This can lead, in turn, to efficiency gains and higher profit margins (Delgado et al., 2010). Agriculture, however, is rarely associated with economic clustering. Rather, agriculture is largely seen as fixed to land and dependent first and foremost on geographic suitability. When farms do cluster, it is seen as the manifestation of landowners responding to similar land characteristics (e.g. corn farming is clustered in Iowa, due to the state’s ideal conditions for growing corn, just as wine production is concentrated in Napa Valley or Tuscany). This article takes a different view of the spatial processes of agricultural change. Notably, it views farmers and their investment capital as potentially mobile, and able to relocate as opportunities emerge. This article is constructed in eight sections. The first engages with existing work on economies of scale and agriculture. The second section provides background and context for the analysis, and considers the development and distribution of farmland in Mato Grosso, Brazil. The third section develops a theoretical framework for understanding the importance of returns to scale within a rent-based model. This leads to the fourth section, where the key specifications are described and developed. The fifth through eighth sections detail and interpret the estimates obtained in the specifications and describe the results of several robustness tests. Notably, the results suggest that areas surrounded by 40–50% cropland (e.g. land used for soybean, corn or cotton) are nearly 20% more likely to be converted to cropland than areas where agriculture is not already present nearby. Approximately one quarter of this effect may be tied to agglomeration economies. The concluding section then suggests that (i) attaining returns to scale, at a regional level is critical for agricultural development and that (ii) land use modeling is significantly improved by accounting for local economic context. 2. Economies of scale and agricultural clustering Work at the nexus of economics and geography has historically recognized the importance of agglomeration economies as driving factors behind the development and clustering of economic activity (Isard, 1956; Pred, 1973; Krugman, 1991; Fujita et al., 1999; Henderson et al., 2001; Fujita, 2002). Clustering enables firms to more easily benefit from knowledge spillovers (Audretsch and Feldman, 1996; Audretsch, 2004; Delgado et al., 2014), or from a ready and more stable supply of labor (Krugman, 1991). By clustering, firms are also more likely to benefit from advancements in technological production and innovations, and access to and knowledge of suppliers (and the relative qualities of their products). These advantages can decrease risk, increase efficiency and reduce transaction costs (Howells and Bessant, 2012). Clustering also offers benefits in the form of shared access to external upstream and downstream input and purchase markets, and to public infrastructure and institutions (Porter, 2000). Shared infrastructure, which includes not only roadways, but schools, trade unions, technical support or extension, and the public collection of information, offers critical competitive advantages, at a regional scale, for firms competing in a global marketplace (Anselin et al., 1997). These advantages can be significant enough to overcome advantages associated with location (e.g. physical suitability), and lead to the concentration of similar industries or land uses. This, in turn, has led to the advancement of the so-called cluster theory as a development paradigm (Porter, 2000; Malmberg and Maskell, 2002; Martin and Sunley, 2003). Agricultural landscapes, however, have rarely been conceptualized as shaped by advantages associated with economic clustering. Farmers, unlike service providers or manufacturers, are often conceptualized as site-stable, with limited ability to move to take advantage of opportunities in other locations. Per this perspective, the farmer chooses the most profitable crop for his or her land rather than the most profitable land for his or her skill set. This conceptualization is pervasive across historical work on the geography of agriculture and agricultural change (von Thünen and Hall, 1966; Dunn, 1967) and in more recent work in land economics (Plantinga et al., 2002; Goodwin et al., 2003; Lubowski et al., 2006). Recent ethnographic work in Latin America, however, is challenging this conceptualization, especially as it regards soybean farmers. Mato Grosso’s own development into a soybean frontier was powered, in part, by agricultural migrants from southern Brazil, where soybeans were already being grown. More recently, many of these same farmers, or their children, have purchased, cleared and planted new lands in the state (Richards, 2015). Some of these farmers have since moved into the Paraguayan Chaco, the Matopiba region at the eastern edge of the Amazon Basin, and even to Sub-Saharan Africa (Spera et al., 2014; Gasparri et al., 2016; de Waroux et al., 2016; Morton et al., 2016). These farmers are relocating themselves, and their capital, in search of lands suitable for soybean production. As they do so, they take into account, and react to, benefits accrued (or likely to accrue) through economic clustering and agglomeration. In Mato Grosso, agglomeration economies are visible at two scales. First, farmers cluster around key cities to take advantage of access to inputs, more developed downstream markets and amenities. These cities serve as nodes in the supply chain, and facilitate the movement of inputs, support surrounding farms with financial and mechanical services and collect and process the harvest for export (Weinhold et al., 2013; Richards and VanWey, 2015; Richards et al., 2015). Second, farms cluster at a micro level. This tendency is a partial function of spatial correlation in suitability and institutional variables; for example, flat land and title discrepancies are spatially correlated, owing to historical zoning decisions or long-term geological processes. However, micro-level clustering also enables farmers to leverage their collective political and financial influence to ensure the improvement or development of physical infrastructure, share information on how to improve efficiency and input selection and more efficiently adapt to local conditions. Scalar economies in regional or local soybean production are ultimately a function of volume, regardless of the size of the farm(s) on which this volume is produced. If a farm is large enough, it may be capable of meeting scalar thresholds on its own. A single large farm, for example, may have in-house access to specialized technical or financial services. However, absent very large farms, these scalar thresholds can only be met through the concentration of many medium or smaller sized farms. 3. Mato Grosso's Agricultural Frontier As late as the 1970s the Brazilian government viewed Mato Grosso, with its acidic and highly weathered soils, intense heat, prolonged dry seasons and few roads or navigable waterways, as largely unsuitable for agriculture (SUDAM, 1976). Over the last 40 years, however, farmers have transformed the state into one of the world’s most productive tropical farmlands. This transformation (arguably) began in the early 1970s with a series of tax incentives, road building projects, land redistributions and colonization efforts designed to draw both people and investment capital into the Amazon. By the early 1980s, tens of thousands of colonists, land speculators and informal migrants were arriving to Mato Grosso in search of land, resources and opportunities (Branford and Glock, 1985; De Almeida, 2010). Early on in this process, Mato Grosso’s farmers struggled to adapt crops from southern Brazil, including beans, rice, corn, coffee and soybeans, to the unique growing conditions of the state. However, over the course of the following decades, with assistance from local and national research institutes (see, e.g. the Mato Grosso Foundation), farmers developed new production strategies and technologies. Soybeans were adapted to shorter tropical daylight periods (Warnken, 1999). Acidic soils were corrected with locally mined lime. Chemical fertilizers rich in phosphorous were developed and applied regularly (Riskin et al., 2013; Roy et al., 2016). Conservation agricultural practices, to preserve scarce organic soil matter and limit erosion, were widely adopted (Hecht and Mann, 2008). Technical innovations were complemented by a series of infrastructure investments and policy changes. Improvements in transportation and trade policy helped to develop new trade linkages, particularly with Asia (DeFries et al., 2010; Garrett et al., 2013; Oliveira, 2016). Travel times between Mato Grosso and the Atlantic coast declined as roads were paved and widened, and as maintenance improved (Walker et al., 2009). Road paving off the state’s main highways also reduced farm to highway costs. Policies put into place since the end of Brazil’s military government in 1985 also benefitted soybean farmers. Price supports for domestic staples were dismantled in the late 1980s and 1990s, tilting the balance of incentives toward export crops such as soybeans (Helfand and Rezende, 2004; Goldsmith and Hirsch, 2006). Export crops were also exempted from ICMS taxes (Tax on Circulation of Goods and Services), a widely applied value added tax in Brazil, creating further advantages (dos Santos and Marta, 2014). Policies such as these were particularly important for soybean farmers, given their reliance on international markets. Initially, Brazil’s soybeans headed primarily to Europe; however, today, the majority of the crop is shipped to East Asia (DeFries et al., 2010; Garrett et al., 2013; Oliveira, 2016). The stabilization of the Brazilian real in the late 1990s, and its subsequent devaluation was also highly beneficial to Mato Grosso’s soybean farmers (Richards et al., 2012). In the early 2000s, the devaluation of the Brazilian real raised soybean profits to record levels. Farmers responded by investing profits back into agriculture, spurring a dramatic increase in production areas in the early 2000s. From 2000 to 2005 alone, soybean areas increased from 2.9 to 6.1 mha (Richards et al., 2012; Oliveira, 2016). Today, Mato Grosso’s soybean farmers are clustered around a handful of key cities (see Figure A1 in the Online Appendix available as Supplementary data). These cities provide the critical inputs needed for farming, as well as more developed downstream markets and lifestyle amenities. They serve as nodes in the supply chain, and host the urban-based financers, salesmen, mechanics and transporters who ensure that farmers have the inputs and services they need to function (Weinhold et al., 2013; Richards and VanWey, 2015; Richards et al., 2015). Once the harvest is reaped, these same cities help to process it, and send it forth for consumption. Massive grain silos tower over the peripheries of growing urban areas, or key highway junctures. Outside are parked hundreds of trucks, each awaiting a fresh load of grain. Some of these trucks head 500 km westward to barge loading facilities to the Madeira River; others, 1000 km northward along a precarious roadway to the Cargill and Bunge ports at Miratituba or Santarém; the remainder shuttle their loads nearly 2000 km south to ports at Paranagua or Santos, on the Atlantic coast. Farms are also clustered off major highways. By clustering, farmers are more likely to observe what strategies their neighbors use to plant, and the outcome of their decisions. They may band together to maintain or improve roads. Many farmers expand their production locally, given the likelihood that soil and precipitation conditions on these areas are more likely to be correlated with their own, easing the learning curve of farming new lands. The heady expansion of Mato Grosso’s soybean sector in the early 2000s eventually did slow. Real soybean prices declined in the mid-2000s as the Brazilian real strengthened against the US dollar, which reduced profits (Richards et al., 2012; Oliveira, 2016). Environmental enforcement increased in the late 2000s, as the Brazilian government sought to slow deforestation (Assunção et al., 2012; Arima et al., 2014; Nepstad et al., 2014). A private-led (and as of 2016, publically backed) moratorium on planting soybeans in recently deforested lands in the Amazon was also put into place (Gibbs et al., 2015). Some have also argued that the best land has already been occupied, diminishing incentives for further expansion (Spera et al., 2014; Morton et al., 2016). Meanwhile, new opportunities emerged elsewhere and Mato Grosso’s farmers took note. In the 2000s, Mato Grosso’s farmers began investing in places as varied as the Colombian llanos, the Paraguayan Chaco, northeast Brazil and even Sub-Saharan Africa (de Waroux et al., 2016). Each time Mato Grosso’s farmers moved they took with them their networks, their skills and their business models. 4. Soybean returns as a function of local and neighborhood effects In Mato Grosso, efforts to model land use change have focused on the relationships between accessibility, biophysical suitability and the location of agriculture. This work has emphasized access to infrastructure and changing returns to agriculture (or cattle production) as drivers of land use change (Vera-Diaz et al., 2008; Mann et al., 2010; Cohn et al., 2014). Often it has recognized or controlled for the importance of natural variables such as slope, topography and rainfall (Jasinski et al., 2005; Spera et al., 2014; Cohn, Gil, et al., 2016; Cohn, VanWey, et al., 2016). Other work has focused on the importance of institutional constraints, including Brazil’s forest code, and constraints on new land clearings (Morton et al. 2016). This article focuses on another anthropogenic driver of land use change, namely agglomeration economies. The importance of agglomeration economies (or clustering) to agricultural expansion can be illustrated through the widely used rent model: Ri= piqi- civi  where Ri refers to returns per unit of land in locationi and piqi and civi refer to price (p) times output (q) and costs (c) times inputs (v), respectively. In much of the work based on the rent model, landowners are conceived as land static profit seekers who select crops which offer the highest returns. This article relies on a different conceptualization of farm mobility. Notably, it assumes that farmers are mobile and can expand production by moving to (or into) new regions. Where they expand depends on which area offers the highest potential return to soybean production over an extended time horizon, or how piqi and civi vary over space. The spatial variance in piqi and civi stems from several possible sources (Equations 2–3). First and foremost, it is a function of local suitability conditions, including physical (topography, soils) and institutional factors (property size, rights), which can affect farm profits (we write these as a vector, Xi). Second, it is a function of market prices for output goods (this is written as Pi) and traded inputs ( Wi), which will also vary spatially, presumably as a function of transportation costs (Jasinski et al., 2005; Walker et al., 2009; Lapola et al., 2010; Mann et al., 2010). Third and finally, spatial variance in piqi and civi is a function of economic spillovers tied to agglomeration economies and spatial clustering (the focus of this article). This latter variable is written as Si, which refers to the density of a particular agricultural system in a defined area. Formally, piqi and civi can thus be stated as: piqi= f(Pi,Xi, Si)  civi= f(Wi, Xi, Si)  Given advantages associated with agglomeration economies, agricultural performance and returns are assumed to increase in more densely planted areas, as farmers take advantage of knowledge transfers, access to labor and access to robust and stable markets for inputs and outputs. The effect of Si is illustrated in Figure 1, which shows the positive relationship between Si and piqi and a negative relationship between Si and civi. Figure 1 View largeDownload slide Regional scale of production and returns to agriculture. Notes: Theorized changes in income ( piqi), costs ( civi) and returns ( Ri) with respect to changes in regional production levels ( Si). Farm returns rise as the size (or density) of agriculture use in a neighborhood increases. Figure 1 View largeDownload slide Regional scale of production and returns to agriculture. Notes: Theorized changes in income ( piqi), costs ( civi) and returns ( Ri) with respect to changes in regional production levels ( Si). Farm returns rise as the size (or density) of agriculture use in a neighborhood increases. Combined, these relationships carry two implications. First, they imply that returns increase as a function of the density of a particular agriculture use. For example, a lone and isolated farm plot might be unprofitable for farming, even if the land is biophysically and institutionally appropriate. Conversely, a more biophysically marginal farming area in a densely planted farming district might be more likely to be converted to agriculture. Second, they imply that parcels are more likely to be converted to agriculture as the density of a particular agriculture use increases in the surrounding area. In theory, each cropland expansion progressively increases the likelihood of another new expansion nearby. This research seeks to highlight the relationships between local economies of scale and returns to soybean farming, and estimate the impact of these relationships in shaping Mato Grosso’s agricultural landscapes. Estimating this impact, however, is complicated by the difficulty of parsing the economic benefits of agglomeration from the effect associated with the spatial correlation in biophysical and institutional variables. To overcome this obstacle, this article follows work by Robalino and Pfaff (2012), and uses both a two-stage least squares (2SLS) and an ordinary least squares (OLS) approach to estimate the exogenous and endogenous sources of spatial correlation shaping agricultural change. 5. Specifications Following much of the past work on land use change in Mato Grosso, this analysis frames the probability of a location’s conversion to agriculture as a function of returns. Returns, in turn, are framed as a function of location, and the relative variation in locational suitability. The probability of a location which is not in use for agriculture at the beginning of an initial period (t - 1), being used for agriculture at a later period (t), is then written as Ci. Ci is a function of a vector of local conditions, Xi, and neighborhood land use, which is defined as the density of a particular agriculture use in the surrounding areas and written as Si, or as Si = ∑i=1NYn∑i=1NNn. The summation ∑n=1NYn refers to the sum of agricultural area located in location i’s neighborhood and ∑i=1NNn refers to the total population of neighbors of that location. The subscript n indicates that a location is a neighbor of location i and Yn signifies that that neighboring parcel is used for agriculture. The superscript N refers to the neighborhood extent, and is a measure of connectivity (e.g. distance of time), meaning that ∑i=1NYn∑i=1NNn denotes the density of an agriculture use within N distance of location i. In Equation  the relationship between Ci and ∑i=1NYn∑i=1NNn is written as non-linear. εi refers to randomly distributed unobservable factors. Ci=βXi+θ1∑i=1NYn∑i=1NNn + θ2∑i=1NYn∑i=1NNn 2+ɛi  The estimate of θ1will likely be positive for two reasons. First, as argued in the opening sections, farmers will reap economic benefits from clustering. Thus, a parcel of forest or pasture surrounded by more of a given agricultural use is more likely to be converted to that agricultural use. Second, many of the conditions that affect Yn, namely the exogenous suitability conditions, will be spatially correlated (e.g. Irwin and Bockstael, 2001; Moffitt, 2001). Together, these effects comprise what would likely be observed as spatial correlation. Unfortunately, estimating the impact of the economic benefits associated with spatial clustering on the likelihood of agricultural growth is complicated by the difficulty of parsing the economic benefits of agglomeration from the effects attributable to spatial correlation in the biophysical and institutional variables (which also incentive farmland expansion). For example, consider that the probability of each neighbor using their land for agriculture could be defined as: Yn=βXn+θ1∑nn=1NYnn∑nn=1NNnn+ θ2∑nn=1NYnn∑nn=1NNnn2+ɛn  or as both a function of a neighbor’s own local suitability characteristics (e.g. slope, institutions, accessibility, written together as Xn) and of how land is used in areas surrounding that neighbor’s land (the subscript nn refers to the neighbors of location n). Because Xi is also necessarily included in Xnn, and Xn will co-vary with Xi, location i is more likely to be converted to agriculture, simply because the suitability conditions in location n that are favorable for agriculture are also likely to be present in location i. Given Equations  and , any estimate of Ci directly, using Equation , would capture both the unbiased effects associated with physical (or institutional) suitability and the potentially endogenous effects associated with economic clustering (e.g. the shared benefits associated with potential agricultural expansion). This article, following recent work by Robalino and Pfaff (2012), parses these effects using a 2SLS approach. Namely it estimates the exogenous influence of local suitability effects using the suitability conditions present in a neighbor’s location and in each neighbor’s neighborhood as an exogenous predictor of the density of an agriculture. This is shown as Equation : Yn=Π1Xn+Π2∑nn=1NXnn+Π3∑nn=1NXnn2+vn  where vn is a randomly distributed error term, Π1– Π3 are coefficients to be estimated and the summation ∑nn=1NXnn is a vector of exogenous suitability conditions relevant to the neighbor’s neighborhood. The coefficients estimated in Equation  predict the likelihood of each location being used for agriculture. The likelihood of agriculture for each set of neighbors is then used to generate the exogenous instrument, namely the predicted density of agriculture use around each location in 2001, ∑i=1NYn^∑i=1NNn. Whereas ∑i=1NYn∑i=1NNn, the actual neighborhood density of agriculture use captures both the exogenous and endogenous effects associated with economic spillovers, the instrument, ∑i=1NYn^∑i=1NNn, captures only the spillover effects associated with local suitability. The instrument ∑i=1NYn^∑i=1NNn is valid if it can be demonstrated that (i) ∑i=1NYn^∑i=1NNn co-varies with ∑i=1NYn∑i=1NNn, and that (ii) covariance between ∑i=1NYn^∑i=1NNn and εi equals zero (e.g. the exclusion factor). Given that legions of land use change research, including work in Mato Grosso using the same datasets used in this analysis, have suggested that soil and slope alone are highly predictive of the likelihood that a given location would be used for agriculture (Spera et al., 2014; Morton et al., 2016), the exogenous suitability conditions used to predict Y^n should have a high predictive power. For the second condition, the exclusion condition would be violated if the suitability conditions used to predict the probability of a location being used for agriculture in location n were directly affecting conditions in location i. For example, if mountains were tall enough in location n to shade location i, or if topographic features directed excess rainfall from a neighbor n into location i and routinely flooded it, there might be a direct effect associated with a neighbor’s land quality. Given the size of the neighborhoods used (e.g. 3–57 min of travel time), however, it would be difficult for a neighbor’s suitability conditions to directly affect local suitability. 6. Data and definitions For this analysis, a random sample of 200,000 points was drawn from across Mato Grosso2 and associated with a range of suitability indicators. The sample points were then used to extract location-specific information on land use and a range of other physical and institutional attributes. The key dataset, the extent of soybean, corn and cotton production in Mato Grosso in 2001 and 2011, was based on 250 m resolution, MODIS vegetation time series classifications (for a full description of the methods, see Spera et al., 2014). Areas identified as soybean–corn, soybean–cotton (double crop rotation), soybeans (single crop) or cotton were designated as cropland. Any point which was classified as cropland in 2011 but not in 2001 was classified as a cropland expansion. In total, 7200 locations, or 5% of the sample, were used as agriculture in 2001; 5800 additional locations were converted to agriculture during the 2001 to 2011 period. Several controls for physical suitability were used in this analysis. Notably, NASA’s 1 arc second (∼30 m) resolution Shuttle Radar Topography Mission data (Figure A2a in the Online Appendix available as Supplementary data) was used to extract data on elevation and slope (USGS, 2006). A soils map obtained from EMBRAPA (the research arm of Brazil’s Ministry of Agriculture) provided information on soil type and texture (Figure A2b in the Online Appendix available as Supplementary data). The Global Forest Change data product (Hansen et al., 2013) offered an estimate of forest density in 2001 (e.g. percent forest cover, see Figure A2c in the Online Appendix available as Supplementary data). Two measures were also calculated to control for relative accessibility. First, the geometric distance between each sample location and the city of São Paulo, and between each location and the nearest major road, were calculated to control for relative accessibility within Brazil. Second, travel times between each location and the nearest major agricultural city were also estimated.3 Each sample point was also associated with several institutional variables. First, data on protected areas was downloaded from Brazil’s Ministry of the Environment (MMA, 2016). Locations in protected areas were designated accordingly to capture their negative impact on the likelihood of land use change. Second, information on public land settlement areas (assentamentos) was downloaded from The National Institute for Colonization and Agrarian Reform (INCRA, 2016). Locations in assentamentos, where property owners were less likely to have a full title or the production size needed to make soybean production competitive, were also designated accordingly (Mier y Terán Giménez Cacho, 2016).4 Third, boundaries for Mato Grosso’s biomes were acquired from Brazil’s Ministry of the Environment (MMA, 2016) and each sample point was annotated with its biome type. In Brazil, legal restrictions on land clearings are particularly stringent in the Amazon, where landowners must maintain up to 80% of their properties in forest. In the Cerrado region, landowners may deforest up to 65% of their land. To estimate the impact of agglomeration economies, each point needed to be tied to a neighborhood of locations. To accomplish this task, this research follows several recent studies and identified neighbors as a function of travel costs (Lesage and Polasek, 2008; Luo and Qi, 2009; Delamater et al., 2012). Travel times were then estimated from on detailed road data for the Brazilian Amazon acquired from the Brazilian Institute of the Environment and Renewable Natural Resources (IBAMA). Travel times for each road segment were based on segment length and quality.5 Travel times between potential neighbors were then calculated as the smallest sum of segment travel times between each set of points. A similar step was taken to estimate travel times between each location and the nearest major agricultural city.6 For the results presented in the next section, the distance threshold for neighbors was limited to a thirty minute cutoff. However, fully recognizing that there is no clear limit for determining neighborhoods, and that no specific time could clearly be argued to be the most appropriate, a range of potential neighborhood sizes are tested in the sensitivity analyses. 7. Estimates The principal objective of this research was to estimate the influence of the relative presence of agriculture in nearby regions on the likelihood of an area being converted to agriculture. This was accomplished through several steps. First, a 2SLS model, using the predicted density of agriculture use (as described above) as an exogenous instrument was used to estimate the extent to which spatial clustering is a function of the spatial correlation of exogenous suitability factors. Second, an OLS approach was used to estimate the entire neighborhood effect (including both spatial correlations in physical suitability and economic incentives from clustering). The second set of models captures the endogenous influence of agricultural clustering, or the extent to which the likelihood of agricultural change increases as a function of nearby agriculture (i.e. not just the presence of nearby lands which might be suitable for agriculture). The difference between the influence of the exogenous instrument (2SLS) and the endogenous (OLS) indicator of neighborhood agriculture amounts to an estimate of the clustering effect. The estimates are presented in three parts. First, the predicted likelihood of each sample point being used for agriculture, using only the suite of suitability variables and a logit model, is presented. Second, the neighborhood effect is estimated using both the endogenous, actual density of agriculture use in a neighborhood (the OLS models) and the exogenous, predicted density (the 2SLS model). Third and finally, a series of robustness checks, including two probit models and estimates based on a range of neighborhood sizes, were tested. 7.1. Predicting agricultural land use To predict the likelihood of each sample point being used for agriculture in time t, the set of local Xi and neighboring Xn physical, institutional and location variables were regressed on Yi, where Yi, is a binary variable, with one signifying cropland. The results suggested that higher elevations were positively associated with an agricultural use. Steeper slopes and density of forest (e.g. portion of neighborhood with forest cover in 2001), in contrast, were negatively associated with agriculture. The three distance variables (to main roads, São Paulo and the nearest major agricultural city) were all negatively associated with the likelihood of a sample point being used for agriculture. For example, the farther a point was away from a key agriculture city, a major road or São Paulo, the less likely it was that that location was used for agriculture. Of the three location variables, distance to a major road had the largest influence on land use change. The set of coefficients used in these models (shown in full in Table 1) were used to predict the likelihood of each specific location being used for agriculture. The sum of the predicted probabilities of each point within a given location’s neighborhood (initially set at thirty minutes of driving time) was divided by the total number of neighbors (for each point) to generate an exogenous instrument for predicting the density of agriculture in a neighborhood. Table 1 Predicting agricultural areas (1) Dependent variable: Land in cropland, 2001 Odds ratios7 Elevation (m) 1.016*** Slope (%) 0.809*** Forest cover (%) 0.948*** Distance to road (km) 0.997*** Distance to SP (km) 1.001*** City dist (min) 0.990*** Soil type Yes Soil texture Yes Biome Yes Assentamento 1.028 Protected 0.0771*** Neighborhood elevation (mean) 0.989*** Neighborhood slope (mean) 0.431*** Neigh Forest cover (mean) 1.001*** Neighborhood protected (dens) 0.0442*** Neighborhood assentamento (dens) 31.64 Constant 1.67e-05*** Neighborhood zize8 30 min Observations 141,586 (1) Dependent variable: Land in cropland, 2001 Odds ratios7 Elevation (m) 1.016*** Slope (%) 0.809*** Forest cover (%) 0.948*** Distance to road (km) 0.997*** Distance to SP (km) 1.001*** City dist (min) 0.990*** Soil type Yes Soil texture Yes Biome Yes Assentamento 1.028 Protected 0.0771*** Neighborhood elevation (mean) 0.989*** Neighborhood slope (mean) 0.431*** Neigh Forest cover (mean) 1.001*** Neighborhood protected (dens) 0.0442*** Neighborhood assentamento (dens) 31.64 Constant 1.67e-05*** Neighborhood zize8 30 min Observations 141,586 ***p < 0.01. 7.2. Predicting agricultural expansion The clustering effect on agricultural expansion was measured through two models. First, a 2SLS model with the exogenous measure estimated from the coefficients in Table 1, was used to estimate of the influence of spatially correlated physical variables on agricultural change. Second, an OLS model, using the actual density of agriculture nearby, was used to estimate agricultural clustering as a function of both exogenous, spatially correlated suitability conditions and peer effects associated with agglomeration economies. The results are presented in Table 2 and as maps in Figure 2. Results suggested that many of the local suitability indicators were congruent with the results from the logistic regression presented in Table 1 (which estimated the likelihood that a location was already in cropland), and with other recent work on agricultural change. Slope, forest cover and locational variables such as distance to São Paulo are all negatively correlated with the likelihood that a location will be converted to agriculture. Table 2 Likelihood of agricultural expansion Dependent variable: agricultural expansion (2001–2011) Mato Grosso Cerrado Amazon Specification (1) (2) (3) (4) (5) (6) 2SLS OLS 2SLS OLS 2SLS OLS Neigh. agriculture (dens) 0.640*** 0.937*** 0.777*** 0.812*** 0.972*** 1.762*** Neigh. agriculture2 (dens) −0.723*** −1.040*** −0.985*** −0.814*** −2.016** −3.049*** Elevation (m) 0.000180*** 0.000138*** 0.000101*** 8.20e-05*** 0.000337*** 0.000302*** Slope (%) −0.00453*** −0.00412*** −0.00355*** −0.00317*** −0.00379*** −0.00355*** Forest cover (%) −0.000400*** −0.000407*** −0.000726*** −0.000729*** −0.000353*** −0.000358*** Distance to road (km) −6.07e-05*** −2.66e-05*** 7.93e-05*** 0.000125*** −1.62e-05 2.74e-05*** Distance to SP (km) 1.36e-05*** −6.00e-06* 9.40e-05*** 7.25e-05*** −6.58e-05*** −6.97e-05*** City dist (min) −1.18e-05*** 5.59e-06** −3.11e-05*** −9.45e-06** 1.77e-05*** 3.75e-05*** Soil type Included Included Included Included Included Included Soil texture Included Included Included Included Included Included Biome Included Included Not included Not included Not included Not included Assentamento −0.000536 −0.000754 0.00044 0.00014 −0.000609 −0.000796 Protected −0.0460*** −0.0334*** −0.0479*** −0.405*** −0.142*** −0.130*** Constant −0.0413*** 0.0253 −0.125*** −0.0656 0.0289* 0.0595*** R2 0.093 0.098 0.123 0.127 0.074 0.084 Neighborhood size (min) 30 30 30 30 30 30 Observations 134,477 134,477 58,971 58,971 66,752 66,752 Dependent variable: agricultural expansion (2001–2011) Mato Grosso Cerrado Amazon Specification (1) (2) (3) (4) (5) (6) 2SLS OLS 2SLS OLS 2SLS OLS Neigh. agriculture (dens) 0.640*** 0.937*** 0.777*** 0.812*** 0.972*** 1.762*** Neigh. agriculture2 (dens) −0.723*** −1.040*** −0.985*** −0.814*** −2.016** −3.049*** Elevation (m) 0.000180*** 0.000138*** 0.000101*** 8.20e-05*** 0.000337*** 0.000302*** Slope (%) −0.00453*** −0.00412*** −0.00355*** −0.00317*** −0.00379*** −0.00355*** Forest cover (%) −0.000400*** −0.000407*** −0.000726*** −0.000729*** −0.000353*** −0.000358*** Distance to road (km) −6.07e-05*** −2.66e-05*** 7.93e-05*** 0.000125*** −1.62e-05 2.74e-05*** Distance to SP (km) 1.36e-05*** −6.00e-06* 9.40e-05*** 7.25e-05*** −6.58e-05*** −6.97e-05*** City dist (min) −1.18e-05*** 5.59e-06** −3.11e-05*** −9.45e-06** 1.77e-05*** 3.75e-05*** Soil type Included Included Included Included Included Included Soil texture Included Included Included Included Included Included Biome Included Included Not included Not included Not included Not included Assentamento −0.000536 −0.000754 0.00044 0.00014 −0.000609 −0.000796 Protected −0.0460*** −0.0334*** −0.0479*** −0.405*** −0.142*** −0.130*** Constant −0.0413*** 0.0253 −0.125*** −0.0656 0.0289* 0.0595*** R2 0.093 0.098 0.123 0.127 0.074 0.084 Neighborhood size (min) 30 30 30 30 30 30 Observations 134,477 134,477 58,971 58,971 66,752 66,752 Note: Results for 2SLS and OLS models, applied across Mato Grosso, and within the Amazon and Cerrado Biomes. ***p < 0.01, **p < 0.05, *p < 0.1. Figure 2 View largeDownload slide Actual (left) and predicted likelihood change in cropland using the 2SLS (center) and OLS (right) models. Figure 2 View largeDownload slide Actual (left) and predicted likelihood change in cropland using the 2SLS (center) and OLS (right) models. The results also show a non-linear relationship between the density of agriculture use in a neighborhood and the likelihood that a given sample point would be converted to cropland. The likelihood of a location being converted to agriculture increases by ∼15% when 40–50% of a neighboring area was used for cropland. These results can be compared with the results produced in the OLS model, where the likelihood of a location being converted to agriculture was estimated as a function of the (actual) density of agriculture use in a location’s neighborhood (presented as Model 2 in Table 2). In the OLS models, when 40–50% of a neighboring area was used as agriculture, that location was about 20% more likely to have been converted to cropland (see Figure 3). This suggests that approximately one-quarter of the observed clustering tendency could be tied to agglomeration economies. Figure 3 View largeDownload slide Relationship between the density of agriculture use in neighborhood and likelihood of agricultural expansion for Mato Grosso, and by Amazon and Cerrado Biomes. Notes: Curves marked 2SLS indicate the influence associated with exogenously determined suitability factors; curves marked OLS estimate the combined influence of both exogenous suitability factors of nearby locations (e.g. more suitable land nearby) and the actual use of the land. The difference between the influence associated with the two curves is an indication of the spatial spillover effect associated with agglomeration economies. The economic benefits of clustering are particularly strong in the Amazon region of Mato Grosso. Figure 3 View largeDownload slide Relationship between the density of agriculture use in neighborhood and likelihood of agricultural expansion for Mato Grosso, and by Amazon and Cerrado Biomes. Notes: Curves marked 2SLS indicate the influence associated with exogenously determined suitability factors; curves marked OLS estimate the combined influence of both exogenous suitability factors of nearby locations (e.g. more suitable land nearby) and the actual use of the land. The difference between the influence associated with the two curves is an indication of the spatial spillover effect associated with agglomeration economies. The economic benefits of clustering are particularly strong in the Amazon region of Mato Grosso. The model results vary significantly by biome. Notably, the agglomeration effect (e.g. the difference between the influence of the endogenous (OLS) and exogenous regressor from the 2SLS model) is significantly larger in the Amazon Biome. In the Amazon, a location surrounded by ∼20–30% cropland is ∼25% more likely to be converted to agriculture; only ∼10% of this likelihood is attributable to spatial correlation in suitability factors (e.g. as estimated in the OLS model). This suggests that clustering may be especially important in the Amazon. In contrast, clustering economies appear to have a lesser influence on land use change in the Cerrado. In the Cerrado, locations surrounded by 40–60% cropland were the most likely to be converted to agriculture. This percentage declined when an area was surrounded by more densely planted farmland; however, not at such a rapid rate as in the Amazon. The difference in agglomeration effects between the Cerrado and the Amazon is likely a function of both remoteness and legal limitations on forest clearing. Notably, the models showed that the likelihood of an area being converted to agriculture in the Amazon declined if the area was already surrounded by more than 30% cropland. In the Cerrado, this likelihood declined only for locations surrounded by more than 60% cropland. These declines roughly correspond to property limitations on land clearing in the two biomes (20% in the Amazon and 65% in the Cerrado). This suggests that the declining marginal impact of neighborhood cropland on agricultural expansion may be closely tied local restrictions on land clearing. While the models are highly suggestive of a significant and positive causal neighborhood effect on the likelihood of a location being converted to agriculture, the explanatory power is relatively weak (R2 for the 2SLS model was only 0.093). To some extent, this is likely attributable to the complex array of factors that explain land use change and the omission of location-specific information on management and ownership characteristics from the analysis. Nevertheless, neighborhood land use decisions are clearly critical in driving this change. As is shown in the following section, accounting for neighborhood effects also greatly improves land use modeling. 7.3. Robustness Tests: reduced form, non-linear estimates and surrounding areas A set of five robustness checks were tested in support of the earlier estimates. This included (i) a reduced form estimate; (ii) an estimate based only on local variables (any neighborhood influence is omitted); (iii) a two-stage probit least squares model (2SPLS); (iv) a probit model; and (v) a series of tests across a range of neighborhood size thresholds. The first robustness check tested a reduced form specification, where the two-stage (2SLS) regression is reduced to a single estimate. This involved regressing the sum of the predicted density of agriculture use directly on the likelihood of a sample point being converted to agriculture. The results, similar to the results from the OLS and 2SLS models from Table 2, suggested that the predicted density of agriculture use in neighborhood (based only on exogenous characteristics) is highly correlated with the likelihood of agricultural change. These estimates broadly supported our findings from the 2SLS models. In the second robustness test, an OLS model was estimated without any controls for neighborhood land uses. When the neighborhood effects were removed, the explanatory power of the model fell significantly. The R2 fell by almost 50% from 0.098 (see Model 2 in Table 2) to 0.052 (Table A1 in the Online Appendix available as Supplementary data, Model 2). This implied that including neighborhood context significantly improved the model. The third and fourth robustness tests included a 2SPLS model and a standard probit model. The 2SPLS model estimates the influence of an endogenous explanatory variable on a binary dependent variable (Lee et al., 1980; Pindyck and Rubinfeld, 1998). For the 2SPLS test, the sum of predicted density of agriculture use and local suitability conditions were regressed on the likelihood of cropland expansion in a linear model, similar to the first-stage specification in Equation . The estimated coefficients were then used to predict the density of neighborhood agriculture. Finally, the predicted densities were used in the standard probit and 2SPLS specifications (shown in Table 3 as specifications 3–4). The results are presented in marginal effects, or the change in the probability of agricultural conversion associated with a one unit change in the density of nearby cropland (while holding all other variables at their means) on the probability of agricultural conversion. Again the estimated effects were positive and significant, and largest when a location was surrounded by 40–50% cropland. Table 3 Estimated change in likelihood of a location being converted to agriculture when 20% of a surrounding region is already in agriculture across a range of neighborhood sizes Neighborhood definition (min) 2SLS Model (%) OLS Model (%) 3 13 14 6 11 15 9 11 15 12 11 15 15 12 15 18 12 15 21 13 15 24 12 15 27 12 15 30 12 15 33 11 14 36 11 14 39 10 14 42 10 14 45 10 14 48 10 14 51 10 14 54 10 14 57 10 13 Neighborhood definition (min) 2SLS Model (%) OLS Model (%) 3 13 14 6 11 15 9 11 15 12 11 15 15 12 15 18 12 15 21 13 15 24 12 15 27 12 15 30 12 15 33 11 14 36 11 14 39 10 14 42 10 14 45 10 14 48 10 14 51 10 14 54 10 14 57 10 13 Note: Neighborhood sizes range from 3 to 57 min of travel. The estimates presented in the earlier models were based on neighborhoods defined by a 30-min cutoff threshold. As a final robustness test, the 2SLS and OLS models (Models 1 and 2 from Table 2) were repeated across a range of neighborhoods definitions. The estimated effects were graphed in Figure 4 and shown in full in Table 3. Figure 4 View largeDownload slide Relationship between the density of agriculture use in neighborhood and likelihood of agricultural expansion, by neighborhood size. Notes: Estimated impacts vary according to neighborhood size. The higher density of agriculture use in a more immediate neighborhood is associated with a greater probability of cropland conversion than a similar density of agriculture use in the neighborhood in a larger surrounding area. Neighborhood sizes are show in minutes and labeled for individual curves. Results correspond with OLS model results from Table 3. Figure 4 View largeDownload slide Relationship between the density of agriculture use in neighborhood and likelihood of agricultural expansion, by neighborhood size. Notes: Estimated impacts vary according to neighborhood size. The higher density of agriculture use in a more immediate neighborhood is associated with a greater probability of cropland conversion than a similar density of agriculture use in the neighborhood in a larger surrounding area. Neighborhood sizes are show in minutes and labeled for individual curves. Results correspond with OLS model results from Table 3. These results consistently suggested that the density of agriculture in a location’s neighborhood was a significant and important predictor of the likelihood of agricultural change, regardless of the neighborhood scale tested. 8. Discussion and conclusion In recent years researchers have begun to recognize and model the effects of spatial spillovers as influential determinants of when, where and how land use change occurs. Research in Costa Rica, for example, found that protected areas were associated with deforestation in nearby forests (Andam et al., 2008). In the Brazilian Amazon, economists and land change modelers recognized the importance of spatial spillover effects from roads, or from already cleared areas, in driving deforestation (Pfaff et al., 2007; Rosa et al., 2013). Similar research has documented the importance of neighbors’ land use decisions as factors underlying the decision of whether or not to clear new land (Robalino and Pfaff, 2012), or to accept conservation easements (Lawley and Yang, 2015). And in Ghana, recent research suggests that even gold mining, a sector highly tied to the relative concentration of minerals in a given location, is affected by agglomeration economies (Fafchamps et al., 2016). In each of these instances land use change is conceptualized as not only a function of local suitability conditions, but of the economic context in which these local suitability conditions, and the agents responsible for land use change, are located. This article has not only has brought the same concepts to land use change modeling in Mato Grosso, perhaps the most significant agricultural frontier of the present century, but showed robust and significant evidence of effect. It also offered evidence suggesting that the effect cannot be tied to the spatial correlation of suitability variables alone. This work is not the first to recognize that economic clustering is critical to soybean production in Brazil. Notably, Garrett et al. (2013) suggested, as this article does, that soybean farmers in Mato Grosso can overcome relatively remote locations through market efficiencies and agglomeration economies. Other work has also highlighted the capital-intensive production process of soybean farming, the vast opportunities for soybean processing and the utility of soybeans as a base input for luxury food (including meat), fuels and many industrial products, and suggested clear similarities between the soybean production process and manufacturing sectors long associated with agglomeration economies (Oliveira and Hecht, 2016). This article offers two key conclusions. First, it shows that neighborhood effects have an important role in determining where croplands expand and that this effect cannot be explained exclusively by spatial correlation in natural suitability. Second, it shows that predictions of where agricultural expansion is likely to take place can be greatly improved when nearby land uses are accounted for. These results offer several clear implications for policymakers. Notably, they suggest that attaining returns to scale, at a regional level, can be critical for emerging agricultural areas. Returns to scale are also essential for adapting farming systems to specific environments, for ensuring a viable supply chain system and for ensuring competitive markets for farm goods. If a local agricultural sector cannot meet a basic threshold to support needed infrastructure, or to facilitate local crop adaptation, local farmers could well struggle to farm competitively, if at all. This is especially critical today, as Brazilian farmers or other foreign investors are now exploring the viability of developing new areas in Zambia, Sudan, Mozambique and elsewhere, often at the behest of local governments, into centers of soybean production (Stolte 2013; Fulquet and Pelfini, 2015; Gasparri et al., 2016). Incorporating smallholder farmers into future soybean frontiers will be particularly challenging without medium or larger farms available to support the initial supply chain infrastructure. If similar producing farms are clustered in specific regions, they may be able to share the costs of supporting the input and output services and infrastructure needed to farm competitively. Lone farmers, however, if they are not large enough, may not be viable. One potential avenue for soybean development might include a mixture of farm sizes. Large farms, which can internalize many of the support services needed to farm (e.g. input acquisition, financing, technical advice, research and experimentation) could offer initial inroads for soybean production and quickly meet scalar thresholds. Once the support infrastructure is in place, it might then be possible for smallholder farmers to then link into it, and gain access to critical products and markets. The results of this work also offer insights for policymakers concerned with environmental conservation or the displacement of smallholder farmers. Concerns abound, for example, about the extent to which soybean expansion can lead to inclusive economic growth in Mato Grosso. For while evidence does appear to link soybean productions to broad reductions in poverty, it has also contributed to rising inequality (Weinhold et al., 2013). Soybeans have also been central to movements concerned with the environmental impacts associated with chemical and fertilizer inputs (Ezquerro‐Cañete, 2016; Lapegna 2016), or the relationships between new soybean farmers and existing social structures (Adams, 2015, 2015). Soybean agriculture, as discussed at the outset of this article, has also been heavily linked to environmental change. New soybean areas in both the Cerrado and Amazon regions of Brazil have been tied to the devastation of hundreds of thousands of square kilometers of forest since 2000, both through the direct conversion of natural land covers to cropland (Morton et al., 2006; Macedo et al., 2012), and indirectly, through the impact of soybeans on either the location or movement of the cattle sector (Barona et al., 2010; Arima et al., 2011) or on regional land prices (Richards, 2015). Such research has suggested that agricultural expansion can lead to impacts well beyond the immediate loss of forest areas to new cropland (Lapola et al., 2010; Richards et al., 2014). The results in this article suggest that if policymakers are concerned with the potential environmental consequences of agricultural expansion, they should seek to halt new agricultural growth at an early stage. Finally, this article emphasizes that the impact of agglomeration economies on agricultural expansion and landscape formation is closely tied to the premise that farmers themselves are highly mobile. During the recent settlement of Mato Grosso, a process which perhaps evokes the settlement of the American Midwest, farmers were granted access to vast tracts of land. These lands, some of which were already occupied by indigenous or other populations, were distributed or fought over as they transitioned into colonization projects, forest concessions or latifundia style ranch properties. Later, many of these same lands were broken up and sold piecemeal, at low cost, to an emerging class of wealthy farmers. Today, however, this landscape is becoming locked into its current form of protected forests, financially productive croplands and expansive pastures. This tendency may well limit farmer mobility in the future, and limit the potential of new farms to organize spatially in a form conducive to economic clustering. Amidst rising land values, and stronger restrictions on forest clearings in Mato Grosso, and an unprecedented capacity for moving information, material and capital, investors are beginning to export elements of Mato Grosso’s soybean model to new regions. Farmers are already looking abroad for new agricultural opportunities. And while the extent to which agglomeration economies will shape a new wave of agricultural landscapes remains unclear, the enduring demand for standardized, agro–industrial output will continue to privilege economies of scale. Hopefully, future research will continue to explore the extent to which spatial spillovers are shaping agricultural landscapes elsewhere in the world, whether in more developed agricultural regions, such as the American Midwest, or in other emerging tropical agricultural regions. As food becomes more heavily standardized, and as production depends on increasingly complex and valuable inputs (e.g. much of the recent increase in global food production is associated with gains in total factor productivity rather than extensification or intensification, Fuglie, 2010), the importance of agglomeration economies to the next generation of food production will only increase. The success or failure of tomorrow's farmers may well depend on not only their own prowess, but on that of their neighbors. Supplementary material Supplementary data for this paper are available at Journal of Economic Geography online. Acknowledgements This research acknowledges support from National Science Foundation Fellowship in Interdisciplinary Research in Behavioral and Social Award (#1305489), Brazil's New Green Revolution: Capital, Investment, and Agricultural Expansion. The article also benefited from the thoughtful and provocative comments put forth by several anonymous reviewers, and by Andrew Foster, Leah VanWey, and Alex Pfaff. Their comments greatly improved the article. Peter Richards serves as an economist with the Bureau for Food Security at USAID. The views and opinions expressed in this paper are those of the author(s) and not necessarily the views and opinions of the United States Agency for International Development. Footnotes 1 Mato Grosso’s forests pertain to three biomes: the Amazon, the Cerrado and the Pantanal. Tropical forest loss in the Amazon, in this instance, refers to forest loss in the Amazon Biome. Mato Grosso is sometimes referred to as an ‘Amazon State’, owing to the inclusion of the state in the so-called Legal Amazon, a political classification in Brazil. 2 A minimum distance between points was set at 250 m, which coincided with the extent of a MODIS pixel. In total, the sample captured ˜6% of the 3.3 million MODIS pixels in Mato Grosso. 3 The six cities for which the Mato Grosso Institute of Agricultural Economics (IMEA) collected soybean prices in the early 2000s were designated as major soybean cities (the six cities are shown in Figure 1). 4 Protected areas and assentamento boundaries are shown in Figure A2d 5 Travel on dual lane highways was estimated as 150 km/hr, on paved roads as 100 km/h, unpaved roads as 50 km/hr, seasonal unpaved roads as 25 km/hr, bridges as 10 km/hr and ferry crossings as 1 km/hr. 6 A number of the data points were not accessible to a road in the road networks (we generously defined inaccessibility as any random point located more than 5 km for a roadway, given that private access roads can often extend several kilometers off public roadways). In total, 141,725 points in the sample were integrated into the road network. 7 Results are presented as odds ratios. Variables associated with values greater than one are more likely to be used for agriculture. 8 Neighborhood size refers to the size of the area for which each neighborhood variable was calculated, in this case 30 min of driving time. References Adams R. T. 2015 An emerging alliance of ranchers and farmers in the Brazilian Amazon. Tipití: Journal of the Society for the Anthropology of Lowland South America 13: 63– 79. Adams R. T. 2015 Neoliberal environmentality among eElites: becoming “Responsible Producers” in Santarém, Brazil. Culture, Agriculture, Food and Environment 37: 84– 95. Google Scholar CrossRef Search ADS Andam K. S., Ferraro P. J., Pfaff A., Sanchez-Azofeifa G. A., Robalino J. A. 2008 Measuring the effectiveness of protected area networks in reducing deforestation. Proceedings of the National Academy of Sciences 105: 16089– 16094. Google Scholar CrossRef Search ADS Anselin L., Varga A., Acs Z. 1997 Local geographic spillovers between university research and high technology innovations. Journal of Urban Economics 42: 422– 448. Google Scholar CrossRef Search ADS Arima E. Y., Barreto P., Araújo E., Soares-Filho B. 2014 Public policies can reduce tropical deforestation: lessons and challenges from Brazil. Land Use Policy 41: 465– 473. Google Scholar CrossRef Search ADS Arima E. Y., Richards P., Walker R., Caldas M. M. 2011 Statistical confirmation of indirect land use change in the Brazilian Amazon. Environmental Research Letters 6: 024010. Google Scholar CrossRef Search ADS Assunção J., Gandour C. C., Rocha R. 2012 Deforestation Pontífica Universidade Católica (PUC), Rio de Janeiro, RJ, Brazil. slowdown in the Legal Amazon: prices or policies. Climate Policy Initiative Working Paper. Audretsch D. B., Feldman M. P. 1996 R&D spillovers and the geography of innovation and production. The American economic review 86: 630– 640. Audretsch D. B. 2004 Knowledge spillovers and the geography of innovation. Handbook of regional and urban economics 4: 2713– 2739. Google Scholar CrossRef Search ADS Barona E., Ramankutty N., Hyman G., Coomes O. T. 2010 The role of pasture and soybean in deforestation of the Brazilian Amazon. Environmental Research Letters 5: 024002. Google Scholar CrossRef Search ADS Branford S., Glock O. 1985 The last frontier; fighting over land in the Amazon. London: Zed. Cohn A. S., Gil J., Berger T., Pellegrina H., Toledo C. 2016 Patterns and processes of pasture to crop conversion in Brazil: Evidence from Mato Grosso State. Land Use Policy 55: 108– 120. Google Scholar CrossRef Search ADS Cohn A. S., Mosnier A., Havlík P., Valin H., Herrero M., Schmid E., O’Hare M., Obersteiner M. 2014 Cattle ranching intensification in Brazil can reduce global greenhouse gas emissions by sparing land from deforestation. Proceedings of the National Academy of Sciences 111: 7236– 7241. Google Scholar CrossRef Search ADS Cohn A. S., VanWey L. K., Spera S. A., Mustard J. F. 2016 Cropping frequency and area response to climate variability can exceed yield response. Nature Climate Change 6: 601– 604. Google Scholar CrossRef Search ADS De Almeida A. L. O. 2010 The Colonization of the Amazon . University of Texas Press. de Waroux Y. l. P., Garrett R. D., Heilmayr R., Lambin E. F. 2016 Land-use policies and corporate investments in agriculture in the Gran Chaco and Chiquitano. Proceedings of the National Academy of Sciences 113: 4021– 4026. Google Scholar CrossRef Search ADS DeFries R. S., Rudel T., Uriarte M., Hansen M. 2010 Deforestation driven by urban population growth and agricultural trade in the twenty-first century. Nature Geoscience 3: 178– 181. Google Scholar CrossRef Search ADS Delamater P. L., Messina J. P., Shortridge A. M., Grady S. C. 2012 Measuring geographic access to health care: raster and network-based methods. International Journal of Health Geographics 11: 15. Google Scholar CrossRef Search ADS PubMed Delgado M., Porter M. E., Stern S. 2010 Clusters and entrepreneurship. Journal of Economic Geography 10: 495– 518. Google Scholar CrossRef Search ADS Delgado M., Porter M. E., Stern S. 2014 Clusters, convergence, and economic performance. Research Policy 43: 1785– 1799. Google Scholar CrossRef Search ADS dos Santos D. A., Marta J. M. C. 2014 A Lei Kandir e o desenvolvimento de Mato Grosso: análise do período 1990-2009. Revista Brasileira de Gestão e Desenvolvimento Regional 10: 206– 228. Dunn E. S. 1967 The Location of Agricultural Production . University of Florida Press. Ezquerro‐Cañete A. 2016 Poisoned, dispossessed and excluded: a critique of the neoliberal soy regime in Paraguay. Journal of Agrarian Change 16: 702– 710. Google Scholar CrossRef Search ADS Fafchamps M., Koelle M., Shilpi F. 2016 Gold mining and proto-urbanization: recent evidence from Ghana. Journal of Economic Geography 17: 975– 1008. Fuglie K. O. 2010 Total factor productivity in the global agricultural economy: Evidence from FAO data. The Shifting Patterns of Agricultural Production and Productivity Worldwide : 63– 91. Fujita M. 2002 Economics of Agglomeration: Cities, Industrial Location, and Regional Growth . Cambridge: Cambridge University Press. Google Scholar CrossRef Search ADS Fujita M., Krugman P. R., Venables A. J. 1999 The Spatial Economy: Cities, Regions and International Trade . Cambridge, MA: MIT press. Fulquet G., Pelfini A. 2015 Brazil as a new international cooperation actor in sub-Saharan Africa: Biofuels at the crossroads between sustainable development and natural resource exploitation. Energy Research & Social Science 5: 120– 129. Google Scholar CrossRef Search ADS Garrett R., Rueda X., Lambin E. 2013 Globalization's unexpected impact on soybean production in South America: linkages between preferences for non-genetically modified crops, eco-certifications, and land use. Environmental Research Letters 8. Garrett R. D., Rausch L. L. 2016 Green for gold: social and ecological tradeoffs influencing the sustainability of the Brazilian soy industry. The Journal of Peasant Studies 43: 461– 493. Google Scholar CrossRef Search ADS Gasparri N. I., Kuemmerle T., Meyfroidt P., de Waroux Y. l. P., Kreft H. 2016 The emerging soybean production frontier in Southern Africa: conservation challenges and the role of south-south telecouplings. Conservation Letters 9: 21– 31. Google Scholar CrossRef Search ADS Gibbs H., Rausch L., Munger J., Schelly I., Morton D., Noojipady P., Soares-Filho B., Barreto P., Micol L., Walker N. 2015 Brazil’s Soy Moratorium. Science 347: 377– 378. Google Scholar CrossRef Search ADS PubMed Goldsmith P., Hirsch R. 2006 The Brazilian Soybean Complex. Choices 21: 97– 103. Goodwin B. K., Mishra A. K., Ortalo-Magné F. N. 2003 What's wrong with our models of agricultural land values? agricultural land values, government payments, and production (allen featherstone, kansas state university, presiding). American Journal of Agricultural Economics 85: 744– 752. Google Scholar CrossRef Search ADS Hansen M. C., Potapov P. V., Moore R., Hancher M., Turubanova S., Tyukavina A., Thau D., Stehman S., Goetz S., Loveland T. 2013 High-resolution global maps of 21st-century forest cover change. Science 342: 850– 853. Google Scholar CrossRef Search ADS PubMed Hecht S. B., Mann C. 2008 How Brazil outfarmed the American farmer. Fortune 157: 92– 105. Google Scholar PubMed Helfand S., Rezende G. C. d. 2004 The impact of sector-specific and economy-wide policy reforms on the agricultural sector in Brazil: 1980–1998. Contemporart Economic Policy 22: 194– 212. Google Scholar CrossRef Search ADS Henderson J. V., Kuncoro A., Turner M. 1995 Industrial development in cities: National Bureau of Economic Research. Journal of Political Economy 103: 1067– 1090. Henderson J. V., Shalizi Z., Venables A. J. 2001 Geography and development. Journal of Economic Geography 1: 81– 105. Google Scholar CrossRef Search ADS Howells J., Bessant J. 2012 Introduction: innovation and economic geography: a review and analysis. Journal of Economic Geography 12: 929– 942. Google Scholar CrossRef Search ADS IBGE. 2016 Produção Agicola Municipal (Municipal Agricultural Production) 1990-2011 . São Paulo: Instituto Brasileiro de Geografia e Estatística [Brazilian Institute of Geography and Statistics] INCRA. 2016 Assentamentos. http://acervofundiario.incra.gov.br/i3geo. INPE. 2015 PRODES - Projeto de Monitoramento do Desmatamento na Amazônia Brasileira por Satélite (Monitoring Deforestation in the Brazilian Amazon by Satelite Project). http://www.dpi.inpe.br/prodesdigital/dadosn/: National Institute of Space Research. Irwin E. G., Bockstael N. E. 2001 The problem of identifying land use spillovers: measuring the effects of open space on residential property values. American Journal of Agricultural Economics 83: 698– 704. Google Scholar CrossRef Search ADS Isard W. 1956 Location and Space-economy: A General Theory Relating to Industrial Location, Market Areas, Land Use, Trade, and Urban Structure . New York: Technology Press of Massachusetts Institute of Technology and Wiley. Jasinski E., Morton D., DeFries R., Shimabukuro Y., Anderson L., Hansen M. 2005 Physical landscape correlates of the expansion of mechanized agriculture in Mato Grosso, Brazil. Earth Interactions 9: 1– 18. Google Scholar CrossRef Search ADS Krugman P. 1991 Increasing returns and economic geography. Journal of Political Economy 99: 483– 499. Lapegna P. 2016 Genetically modified soybeans, agrochemical exposure, and everyday forms of peasant collaboration in Argentina. The Journal of Peasant Studies 43: 517– 536. Google Scholar CrossRef Search ADS Lapola D. M., Schaldach R., Alcamo J., Bondeau A., Koch J., Koelking C., Priess J. A. 2010 Indirect land-use changes can overcome carbon savings from biofuels in Brazil. Proceedings of the National Academy of Sciences 107: 3388– 3393. Google Scholar CrossRef Search ADS Lawley C., Yang W. 2015 Spatial interactions in habitat conservation: evidence from prairie pothole easements. Journal of Environmental Economics and Management 71: 71– 89. Google Scholar CrossRef Search ADS Lee L. -F., Maddala G. S., Trost R. P. 1980 Asymptotic covariance matrices of two-stage probit and two-stage tobit methods for simultaneous equations models with selectivity. Econometrica: Journal of the Econometric Society : 491– 503. Lesage J. P., Polasek W. 2008 Incorporating transportation network structure in spatial econometric models of commodity flows. Spatial Economic Analysis 3: 225– 245. Google Scholar CrossRef Search ADS Lubowski R. N., Plantinga A. J., Stavins R. N. 2006 Land-use change and carbon sinks: econometric estimation of the carbon sequestration supply function. Journal of Environmental Economics and Management 51: 135– 152. Google Scholar CrossRef Search ADS Luo W., Qi Y. 2009 An enhanced two-step floating catchment area (E2SFCA) method for measuring spatial accessibility to primary care physicians. Health & Place 15: 1100– 1107. Google Scholar CrossRef Search ADS PubMed Macedo M., DeFries R. S., Morton D. C., Stickler C. M., Galford G. L., Shimabukuro Y. E. 2012 Decoupling of deforestation and soy production in the southern Amazon during the late 2000s. Proceedings of the National Academy of Sciences 109: 1341– 1346. Google Scholar CrossRef Search ADS Malmberg A., Maskell P. 2002 The elusive concept of localization economies: towards a knowledge-based theory of spatial clustering. Environment and planning A 34: 429– 449. Google Scholar CrossRef Search ADS Mann M. L., Kaufmann R. K., Bauer D., Gopal S., Vera-Diaz M. D. C., Nepstad D., Merry F., Kallay J., Amacher G. S. 2010 The economics of cropland conversion in Amazonia: the importance of agricultural rent. Ecological Economics 69: 1503– 1509. Google Scholar CrossRef Search ADS Martin R., Sunley P. 2003 Deconstructing clusters: chaotic concept or policy panacea? Journal of Economic Geography 3: 5– 35. Google Scholar CrossRef Search ADS Mier y Terán Giménez Cacho M. 2016 Soybean agri-food systems dynamics and the diversity of farming styles on the agricultural frontier in Mato Grosso, Brazil. The Journal of Peasant Studies 43: 419– 441. Google Scholar CrossRef Search ADS MMA. 2016 Dados Georeferenciados. http://www.mma.gov.br/areas-protegidas/cadastro-nacional-de-ucs/dados-georreferenciados: Ministério do Meio Ambiente. Moffitt R. 2001. Policy Interventions, Low-Level Equilibria, and Social Interactions. In Durlauf S., Young P. (eds) Social Dynamics , pp. 45– 82: Cambridge, MA: MIT Press. Morton D. C., DeFries R., Shimabukuro Y. E., Anderson L. O., Aral E., del Bon Espirito-Santo F., Freitas R., Morisette J. 2006 Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. Proceedings of the National Academy of Sciences of the United States of America 103: 14637– 14641. Google Scholar CrossRef Search ADS PubMed Morton D. C., Noojipady P., Macedo M. M., Gibbs H., Victoria D. C., Bolfe E. L. 2016 Reevaluating suitability estimates based on dynamics of cropland expansion in the Brazilian Amazon. Global Environmental Change 37: 92– 101. Google Scholar CrossRef Search ADS Nepstad D., McGrath D., Stickler C., Alencar A., Azevedo A., Swette B., Bezerra T., DiGiano M., Shimada J., Seroa da Motta R., Armijo E., Castello L., Brando P., Hansen M. C., McGrath-Horn M., Carvalho O., Hess L. 2014 Slowing Amazon deforestation through public policy and interventions in beef and soy supply chains. Science 344: 1118– 1123. Google Scholar CrossRef Search ADS PubMed Oliveira G., Hecht S. 2016 Sacred groves, sacrifice zones and soy production: globalization, intensification and neo-nature in South America. The Journal of Peasant Studies 43: 251– 285. Google Scholar CrossRef Search ADS Oliveira G. d. L. 2016 The geopolitics of Brazilian soybeans. The Journal of Peasant Studies 43: 348– 372. Google Scholar CrossRef Search ADS Pfaff A., Robalino J., Walker R., Aldrich S., Caldas M., Reis E., Perz S., Bohrer C., Arima E., Laurance W., Kirby K. 2007 Road investments, spatial spillovers, and deforestation in the Brazilian Amazon. Journal of Regional Science 47: 109– 123. Google Scholar CrossRef Search ADS Pindyck R. S., Rubinfeld D. L. 1998 Econometric Models and Economic Forecasts . Boston, MA: Irwin/McGraw-Hill Boston. Plantinga A. J., Lubowski R. N., Stavins R. N. 2002 The effects of potential land development on agricultural land prices. Journal of urban economics 52: 561– 581. Google Scholar CrossRef Search ADS Porter M. E. 2000 Location, competition, and economic development: local clusters in a global economy. Economic Development Quarterly 14: 15– 34. Google Scholar CrossRef Search ADS Pred A. R. 1973 Urban Growth and the Circulation of Information: The United States System of Cities, 1790-1840 . Cambridge, MA: Harvard University Press. Richards P. 2015 What drives indirect land use change? How Brazil's agriculture sector influences frontier deforestation. Annals of the Association of American Geographers 105: 1026– 1040. Google Scholar CrossRef Search ADS PubMed Richards P., Pellegrina H., VanWey L., Spera S. 2015 Soybean development: the impact of a decade of agricultural change on urban and economic growth in Mato Grosso, Brazil. PloS ONE 10: e0122510. Google Scholar CrossRef Search ADS PubMed Richards P., VanWey L. 2015 Where deforestation leads to urbanization: How resource extraction is leading to urban growth in the Brazilian Amazon. Annals of the Association of American Geographers 105: 806– 823. Google Scholar CrossRef Search ADS PubMed Richards P. D., Myers R. J., Swinton S. M., Walker R. T. 2012 Exchange rates, soybean supply response, and deforestation in South America. Global Environmental Change 22: 454– 462. Google Scholar CrossRef Search ADS Richards P. D., Walker R. T., Arima E. Y. 2014 Spatially complex land change: the Indirect effect of Brazil's agricultural sector on land use in Amazonia. Global Environmental Change 29: 1– 9. Google Scholar CrossRef Search ADS PubMed Riskin S. H., Porder S., Neill C., Figueira A. M. e. S., Tubbesing C., Mahowald N. 2013 The Fate of Phosphorus Fertilizer in Amazon Soya Bean Fields. Philosophical Transactions of the Royal Society of London B: Biological Sciences 368: 20120154. Robalino J. A., Pfaff A. 2012 Contagious development: neighbor interactions in deforestation. Journal of Development Economics 97: 427– 436. Google Scholar CrossRef Search ADS Rosa I. M., Purves D., Souza C.Jr, Ewers R. M. 2013 Predictive modelling of contagious deforestation in the Brazilian Amazon. PLoS ONE 8: e77231. Google Scholar CrossRef Search ADS PubMed Roy E. D., Richards P. D., Martinelli L. A., Della Coletta L., Lins S. R. M., Vazquez F. F., Willig E., Spera S. A., VanWey L. K., Porder S. 2016 The phosphorus cost of agricultural intensification in the tropics. Nature Plants 2: 16043. Google Scholar CrossRef Search ADS PubMed Soares-Filho B. S., Nepstad D. C., Cerqueira G. C., Garcia R. A., Ramos C. A., Voll E., McDonald A., Lefebvre P., Schlesinger P. 2006 Modelling conservation in the Amazon basin. Nature 440: 520– 523. Google Scholar CrossRef Search ADS PubMed Spera S. A., Cohn A. S., VanWey L. K., Mustard J. F., Rudorff B. F., Risso J., Adami M. 2014 Recent cropping frequency, expansion, and abandonment in Mato Grosso, Brazil had selective land characteristics. Environmental Research Letters 9: 064010. Google Scholar CrossRef Search ADS Stolte C. 2013 Brazil in Africa: seeking international status, not resources. Harvard International Review 34: 63. SUDAM. 1976 II Plano Nacional De Desenvolvimento: Programa de Ação do Governo para a Amazônia [II National Development Plan: Government Action Plan for the Amazon] . Belém: Minestry of the Interior, Superintendent of Amazon Development. USGS. 2006 U.o.M. Global Land Cover Facility (ed) Shuttle Radar Topography Mission (SRTM). Maryland: College Park. VanWey L., Spera S., Sa R. d., Mahr D., Mustard J. 2013 Socioeconomic development and agricultural intensification in Mato Grosso. Philosophical Transactions of the Royal Society of London. Series B 368: 20120168. Google Scholar CrossRef Search ADS Vera-Diaz M. d. C., Kaufmann R. K., Nepstad D. C., Schlesinger P. 2008 An interdisciplinary model of soybean yield in the Amazon Basin: The climatic, edaphic, and economic determinants. Ecological Economics 65: 420– 435. Google Scholar CrossRef Search ADS von Thünen J. H., Hall P. G. 1966 Isolated state: an English edition of Der isolierte Staat . Oxford, NY: Pergamon. Walker R., Browder J., Arima E., Simmons C., Pereira R., Caldas M., Shirota R., de Zen S. 2009 Ranching and the new global range: Amazônia in the 21st century. Geoforum 40: 732– 745. Google Scholar CrossRef Search ADS Warnken P. F. 1999 Development and Growth of the Soybean Industry in Brazil . Ames, Iowa: Iowa State University Press. Weinhold D., Killick E., Reis E. J. 2013. Soybeans, poverty and inequality in the Brazilian Amazon. World Development 52: 132– 143. Google Scholar CrossRef Search ADS © The Author (2017). Published by Oxford University Press. All rights reserved. For permissions, please email: firstname.lastname@example.org
Journal of Economic Geography – Oxford University Press
Published: Jan 1, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera