Multiple pathways of commodity crop expansion in tropical forest landscapes

We synthesized the few recently published case studies of rapid commodity cropland expansion in tropical forest regions that present spatially-explicit land use/cover trajectory matrices. The small number of cases existing prevented a formal meta-analysis; rather, we compiled quantitative land use/cover trajectory variables, and assessed the influencing factors quantitatively and qualitatively. Analyses from the original papers were combined with expert knowledge from case study authors to explain recent dynamics of cropland expansion. Cases differ in geographic extent, boundary definitions, and methods. As is customary in meta-analyses in land change science, we relied on informed decisions by the original authors to define study areas characterized by homogeneous land change processes or areas which can be considered as a single land use system. We did not compare each factor individually across cases, but considered the whole configuration and interactions of variables in each land use system.

For each case, based on data from the original studies and additional sources, we first calculated indicators of land use/cover changes: (i) gross deforestation rate (figure 2); (ii) post-deforestation land use; (iii) gross and net area changes for the target commodity crop and other agricultural land uses; and (iv) land sources for commodity cropland (figure 3). Land use displacement implies causal links between commodity crop expansion in one place and land use change elsewhere. For each case, we thus discuss the underlying land change processes to assess whether land use displacement could have occurred over the study period. Then, based on the case studies, we identified the main factors that affected local crop expansion pathways (table 1), and measured them for each case using various sources and expert knowledge: (i) availability or scarcity of forests versus previously cleared land, measured as the proportion of different land covers at the start of each period, and rural population density and rate of change; (ii) biophysical, accessibility and technical constraints on expansion, including specific crop requirements; (iii) land use zoning, measured as the percentage of forested land covered by a zoning scheme strictly or partly restricting agricultural expansion (i.e., protected areas, indigenous lands, logging concessions, forestry lands zoned for various purposes), and a qualitative ranking of the enforcement of land use policies within these zones; (iv) land tenure and its security, and land markets; (v) types of agents—i.e. smallholders or large-scale actors—active in the various agricultural land uses; and (vi) agricultural intensification, measured as change in average yields of the target crop (see definitions and details in the SI text).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To evaluate the selection biases of the set of cases, we performed a representativeness analysis using the Global Collaborative Engine or GLOBE system, an online collaborative land change database ([28, 29]; http://globe.umbc.edu/). We compared the frequency distribution of global gridded variables relevant for our study in our set of cases, compared with all tropical lands. This comparison shows whether the set of case studies can be considered as resulting from a random sampling of locations within tropical lands, and which ranges of values or categories of the global variables are under- or overrepresented in the sample. Then, we performed the same analysis for the set of deforestation case studies present in GLOBE, using these same variables, compared with all tropical lands (details and results in the SI).

From 2000−05, large-scale, intensive soy agriculture expanded rapidly in this forest frontier (500 915 km²), mainly replacing low-productivity pastures, but also forests [30]. Pastures expanded into forests. From 2006–09, high deforestation rates decreased and gross agricultural expansion declined, with soy expanding almost exclusively into previously cleared lands [22]. During this period, agricultural markets conditions changed, and six implemented measures possibly influenced deforestation rates: expansion of the protected areas network; stronger enforcement of the Brazilian Forest Code which limits deforestation on private properties; creation of a land registry; restrictions on credit for illegal deforesters; satellite-based monitoring of deforestation; and two voluntary moratoria discouraging the sale of cattle and soybeans produced in newly-deforested areas [22, 31–35].

Since the mid-1990s, oil palm plantations have expanded in this landscape (2134 km²) dominated by pastures and swidden cultivation [23]. Private companies developed large-scale plantations mainly on mature forest. Supported by public incentives but facing capital constraints, smallholders established small-scale plantations on diverse land covers, including mature forest, and secondary forest, pastures, and other mixed agriculture (including young fallows). None of the study area is under formal protection [36]. Property rights of smallholder land holdings are generally informal, but can also be registered officially [37]. A dense road network suggests that accessibility is not a strong constraint on expansion [38].

Expansion of large-scale, export-oriented, intensive crop production, predominantly of bananas and pineapples, began in the 1990s in this landscape (6617 km²) of forest and pasture [24]. From 1986 to 1996, deforestation was common in mature and secondary forests, and only ∼15% of the forests in the study area were officially protected as riparian or protected zones [39]. Since 1996, the Forest Law mandated a nation-wide ban on deforestation and expanded a fund for payments for environmental services, including tree planting and forest protection in specific areas [39]. Consequently, from 1996 to 2011 the loss of mature forests declined sharply, although clearing of unprotected secondary forests accelerated. After 1996, most new cropland was sourced from pastures. Banana expansion, which is constrained by access to roads, and the need for fertile soil and large capital investments, is concentrated in fertile river floodplains. By contrast, pineapple can grow on poor-quality soils and is mainly constrained by road access.

Dak Lak study area (7478 km²) experienced a coffee boom and major deforestation in the 1990s. Deforestation decreased in the early 2000s with the coffee bust, and then increased again from 2005 to 2010 as coffee prices slowly recovered [25]. Shifting cultivation was the main land use after forest clearing as well as the primary land source for coffee expansion [25]. Coffee expansion by well-capitalized smallholder migrants resulted in spatial displacement and marginalization of poor migrants and ethnic minorities, who resorted to shifting cultivation on increasingly marginal forestland. In Dak Nong (6513 km²), rubber expansion by large-scale actors, especially former state forest enterprises, directly encroached into forests [26]. Zoning subdivides land into protection, special-use and production forests. On the latter, local administrations sometimes tolerate subsistence agriculture. Rubber expansion is authorized in 'poor quality' production forests, encouraging a sequence of forest logging, followed by clearing for rubber. Long-term certificates grant agricultural land rights to households, while most forestry lands remain under the control of forest enterprises.

In the Ketapang study region (12038 km²), oil palm expansion began in the early 1990s, when logging concessions were converted to large-scale plantations, with support from state policies [27]. From 1996–2005, moderate oil palm expansion occurred mainly into logged and intact (hereafter referred to as 'secondary' and 'mature') forests. From 2005–08, land sources shifted; oil palm expanded rapidly onto swidden agricultural lands, while only 5.4% of expansion cleared mature forests. Strikingly, >90% of 1989–2008 deforestation resulted from intentional and drought-related fire, especially during the 1997–98 El Niño Southern Oscillation-associated drought. Forests are now concentrated within protected areas and peatlands, while rural communities and their swidden mosaics are concentrated on mineral soils. Until 2002, all lands were controlled by the central State. Today, specific land zones are controlled by district, provincial, and national agencies. Land use plans often bear little relation to field conditions: agrarian communities are frequently enclosed within the forest estate, where legally agriculture is restricted. Because formal land ownership comes with high transaction costs and rarely excludes state and private sector interests, most smallholders forgo land titles. Rural communities lack the capital to invest in the infrastructure required for palm oil processing, and must therefore sell fruit to company mills.

Across cases and periods, 1.7–89.5% of new commodity cropland was sourced from forestlands. The remaining 10.5–98.3% was sourced from existing agricultural or other lands (figure 3). We identified four types of factors that contribute to bias in commodity crop expansion towards forest or existing agriculture (figure 4).

Zoom In Zoom Out Reset image size

First, a scarcity of suitable forestland, determined by a combination of forest area, biophysical or accessibility constraints, and land use policies, was associated with a higher share of conversion of existing agricultural land to commodity cropland. In Ketapang, few mature forests on mineral soils remained outside protected areas after ∼2005 [27]. This influenced the increased proportion of swidden lands sourced for oil palm plantations over this period. In contrast, the greater Kalimantan region harbors extensive mature forests vulnerable to oil palm expansion [40]. In Dak Lak, remaining forests concentrated on steep slopes and high elevation areas lacking water for coffee irrigation. Constraints on expansion change over time. Technological progress and road expansion may increase the pool of land available for commodity agriculture. In the second analytical period in Mato Grosso and Saraquipi-San Carlos, implementation of new land use policies constrained commodity crop expansion into already-cleared lands or secondary regrowth (figure 3) [22, 24]. In Dak Lak, zoning for forest protection restricted coffee expansion [25]. By contrast, in Pucallpa, lack of legal constraints on forest clearing enabled disproportionate expansion of large-scale oil palm into forests. However, land use policies may also induce expansion to concentrate on forestlands: state policies supported expansion of rubber into degraded forests in Dak Nong [26], and of oil palm into logging concessions in West Kalimantan [41].

Economic and technical characteristics of agricultural systems also influence land cover sources for commodity crop expansion. In Mato Grosso, production of crops and cattle destined for exports outside the region, and the dynamic of land markets, have caused progression of the agricultural frontier to follow Von Thunen's location rent model at the regional scale, centered on the major export points and corridors [42, 43]. Lower-value cattle ranching occupied remote locations with scarce labor and low land rent, while higher-value soy advanced into high-rent already-cleared land, where supply chain infrastructure permitted agglomeration economies [44]. During the early 2000s, with highly favorable conditions for exports, soy fields 'leapfrogged' pastures at greater rate than before, and expanded directly into forests [30, 45]. Coffee expansion into shifting agriculture in Dak Lak followed the same Thunian pattern. In Saraquipi-San Carlos, post-ban commodity crop expansion was dominated by pineapple, which already expanded mainly into low-fertility land outside mature forests before the deforestation ban. Banana expansion, which occurred mainly into mature forests before the ban, decreased immediately after the ban but then recovered by targeting fertile pastures [24]. Different requirements in soils and market accessibility largely explain these trajectories. Intensification and changes in yields may also affect trajectories of expansion. But understanding the role of yield changes is complicated by the sensitivity of yields to climate fluctuations, and long time lag between crop establishment and first harvest. Rapid expansion of a crop can decrease mean yields temporarily, as in Mato Grosso and Indonesia.

Third, expansion behaviors of small-scale and large-scale actors differ due to different constraints and opportunities associated with farm size. Smallholders tend to use their already-cleared agricultural lands to develop commodity crops. In Pucallpa, smallholders, often planted oil palm in degraded pastures and secondary forests, with government support, thereby increasing the value of these lands. By contrast, large companies preferentially planted oil palm into state-owned forests, likely to minimize transaction costs and social conflicts associated with consolidating a large number of small plots from multiple smallholders [23]. In West Kalimantan, plantation development since 2007 has been skewed toward peatlands, despite higher costs of drainage and land preparation [27, 36]. It has been argued that avoiding disputes with local communities over land tenure rights could be a motivation explaining this trend, but there is not yet any conclusive evidence to support this hypothesis [46]. Large-scale operators can reduce transaction costs of establishing large landholdings by dealing with only one, often public, land provider, especially in regions with loose legal frameworks and limited recognition of customary rights on forestlands [47, 48]. Large companies can finance their own infrastructure (e.g., roads) beyond existing agricultural lands [16]. In Laos, only 20% of the land targeted by large-scale investment deals for agricultural projects was already cultivated [49].

Finally, costs and benefits of forest clearing differ across geographical contexts, further influencing the rent of different land pools. Conversion of already-cleared land generally requires lower capital investments. But higher quality soils under forests and benefits from timber harvest, especially from high-value dipterocarp forests in Asia, can favor forest clearing. In West Kalimantan, companies commonly finance plantations with profits from residual timber [27, 50]. This initial revenue pulse is particularly important for perennial crops, which become productive only after several years. The broader economic context modifies this incentive: in Indonesia, the overcapacity of wood-based industries required large amounts of timber, encouraging forest clearing for oil palm plantations during the first study period [41].

Displacement of existing agriculture replaced by commodity cropland can be local (i.e., within the study area) or distant (i.e., outside the study area through teleconnections).

Locally, within Amazonian Mato Grosso in the early 2000s, some deforestation might be attributed to pastures displaced by expanding soy fields, but this remains unquantified [43, 51]. After 2006, pasture expansion into forests declined nearly seven-fold despite increasing soy area, suggesting that local displacement became unlikely [22]. Land use policies discouraged deforestation, while the strength of the Brazilian Real decreased the cost competiveness of Brazilian soy production relative to US soy, and increases in variable production costs also reduced the profitability of soy [22, 45]. Cattle intensification was promoted by institutional changes and land use policies, and was suggested as a way to reduce displacement associated with conversion of pastures to cropland, but its role remains unclear. Between 1975 and 1996, across Brazilian Amazon municipalities, increased stocking rates were associated with pasture expansion, suggesting that intensification did not always reduce local expansion. Yet, the relation reversed in many states, including Mato Grosso, from 1996–2006 [52]. In Saraquipi-San Carlos, from 1996–2011, pasture area remained relatively constant, despite pastures being replaced by cropland. Assuming complete displacement, as much as 10–50% of deforestation by pasture can be related to cropland expansion in 1986–2005. Local displacement was likely reduced in 2005–10, as cropland expansion into pasture exceeded deforestation for pasture. In Dak Lak, local displacement of shifting cultivation, pushed by expansion of coffee and other market crops, was the main direct cause of deforestation. This displacement was enabled by incomplete enforcement of zoning, with local authorities recognizing the lack of alternatives for marginalized smallholders. In contrast, rubber in Dak Nong and large-scale oil palm in Pucallpa expanded almost exclusively into forests, with no discernable displacement of agriculture. Small-scale oil palm in Pucallpa spread preferentially into degraded pastures and secondary forests. While converting abandoned pastures is unlikely to drive displacement of land use, some cacao or annual croplands converted to oil palm may have been displaced further into forests. In West Kalimantan, there is little evidence of displacement of smallholder agriculture. Outside of protected areas, few forests remained on mineral soils, and biophysical and financial constraints on cultivating peatlands likely prevented displacement of swidden croplands into these lands. Beyond agricultural displacement, commodity crop expansion may lead to other forms of iLUC. For example, in West Kalimantan, although the causes of fires could not always be discerned, oil palm plantations are considered to be major contributors to regional fire prevalence.

Less evidence exists regarding land use displacement to or from distant places. In 2003–08, soy expansion in Mato Grosso influenced deforestation for cattle in the Amazonian frontier, providing evidence for distant displacement, but the marginal effect remains unquantified [43, 53]. Accelerating rates of Cerrado clearing after 2010, including in areas remote from Mato Grosso and the Amazon (on average ∼7500 km² per year over 2010–12, versus ∼3900 km² per year over 2004–10) ([54], www.lapig.iesa.ufg.br), could also be related to distant land use displacement, but no study has established a causal link. Further, soy and cattle expansion in the Amazon and Cerrado may partly result from their displacement by sugarcane expansion in southeast Brazil [55], and cattle intensification in the Center-West region may have partly compensated for pasture contraction in Mato Grosso [56]. A partial equilibrium model experiment suggests that policies supporting further cattle intensification and taxing extensive cattle ranching in Brazil could spare pasture land, concentrate cropland expansion on pastures, and reduce displacement of pastures into the Amazon [57]. In Pucallpa, local landlords sometimes consolidated oil palm plantations, resulting in previous landowners migrating, mostly to nearby cities like Pucallpa, creating a demand for agricultural products from local and distant sources. Loss of swidden land, as in West Kalimantan, may also be compensated by intensification, off-site seasonal employment, remittances, income generated from land sales to and employment by large-scale companies, and permanent migration, with various effects on land use displacement. In the long run, technological innovations allow for distant geographic displacement of crop booms, as for South American rubber in mainland Southeast Asia [15, 17], and Asian soy in the Cerrado and parts of the Amazon. Macroeconomic factors, including trade policies and currency exchange rates, are also important factors affecting the regional distribution of crop production.

Quantifying displacement is challenging. First, the absence of local displacement does not preclude the possibility of long distance displacement or iLUC, so that the area over which an analysis is conducted can determine whether displacement is detected or not. Fully measuring land use displacement would require accounting for land pools and transitions at a global scale. Beyond humid tropical forests, savanna and dry forests ecosystems like the Brazilian Cerrado may be disproportionately affected by iLUC due to weaker land use policies. Monitoring land use changes in these biomes may require different remote sensing approaches, and they are often ignored because of their comparatively lower carbon stocks. Displacement may not only involve shifting products to a different location, but also product substitutions in the supply chain. Further indirect effects could also be due to changes in consumption of displaced landowners [58]. Second, establishing firm causal links between substitution in one place and expansion in another place requires developing a plausible counterfactual for the state of land use absent either agricultural expansion or the diversion of agricultural output to other uses [59]. The same holds true for assessing land sparing, for which a counterfactual, absent agricultural intensification, is required. Simulation models greatly contribute to this goal, but empirical approaches are also needed to improve the design, calibration, validation and interpretation of simulations. Statistical approaches to build such counterfactuals, including statistical matching and synthetic control methods, are increasingly used in land systems science [60–62], and could be used for assessing land sparing or displacement. Statistical inferences about displacement and iLUC can also be made based on spatial regressions [53] and spatial analyses of land use change patterns [25]. Various analytical approaches, including fixed effects panel analyses or natural experiments, control for unobserved characteristics that may influence outcomes [59]. Empirical studies are also crucially needed to investigate the motivations and decision-making strategies of land users (e.g., households, large farms, and corporations), reconstruct the means by which regional demand for agricultural products is met, and track actors across space via land registries [8].