Spatiotemporal modeling of the expansion of mechanized agriculture in the Bolivian lowland forests

详细信息    查看全文

• 作者：Robert M&#xfc ; ller^a ; ^{robemule@yahoo.com} ; Daniel M&#xfc ; ller^b ; ^c ; Florian Schierhorn^c ; Gerhard Gerold^a
• 关键词：Bolivia ; Soy bean ; Deforestation ; Land use change ; Spatial analysis ; Amazon
• 刊名：Applied Geography
• 出版年：2011
• 出版时间：April 2011
• 年：2011
• 卷：31
• 期：2
• 页码：631-640
• 全文大小：1253 K

文摘

Since the 1980s, mechanized soybean production in Bolivia has caused extensive deforestation in the northeast of Santa Cruz de la Sierra in the eastern Bolivian lowlands. We analyze the spatial and temporal dynamics of deforestation due to mechanized agriculture with spatially explicit logistic regression models. Deforestation patterns are derived from the classification of Landsat imagery by Killeen et al. (2007) and include five time steps (1976, 1986, 1992, 2001 and 2005). We associate deforestation with geophysical and socioeconomic determinants. Our model controls for spatial autocorrelation and temporal dependencies, and we assess the robustness of the results for several model formulations.

The expansion of mechanized agriculture is concentrated in areas with favorable environmental conditions, good market access and close proximity to prior deforestation. While overall dynamics remained relatively stable over time, the expansion of mechanized agriculture between 2001 and 2005 became more tolerant to excessive rainfall and less dependent on fertile soils. This mirrors the increasing penetration of mechanized agriculture into humid and less fertile Amazonian rainforests in the northern portion of the study area. The map of deforestation probability substantiates these patterns and shows the highest propensities for future deforestation in the north. Our study demonstrates the value of spatial regression models to better understand the development of deforestation dynamics over a 30-year time span, and contributes to the formulation of policies that aim to reduce deforestation. Yet the results are sensitive to hidden correlations between independent variables, and we therefore advocate a careful evaluation of regression results for different model formulations.