Fundamental causes and spatial heterogeneity of deforestation in Legal Amazon

- Cattle ranching causes most deforestation in Pará.
- Generally, crop cultivation is weakly associated with deforestation.
- Rural credit scheme inhibits deforestation in Pará.
- Official and unofficial roads lead to most deforestation in remote areas.
- Travel time as compared to Euclidean distances changes results for some areas.

This study explores the main direct and underlying causes of deforestation in Brazil's Legal Amazon region by considering spatial differences. The computation of localized parameters is based on geographically weighted regression (GWR). The novelty of this paper lies in its incorporation of economic, rather than Euclidean, distances into the GWR. Economic distances are measured by travel time, sourced from Google Inc. A global approach revealed several important factors that affect deforestation, including: rural population, GDP (suggesting a U-shaped environmental Kuznets curve), forest stock, cattle ranching, timber value, and road networks (both official and unofficial). Local analysis uncovered patterns not seen under global models, especially in the state of Pará. Most notably, crop cultivation was found to accelerate deforestation in southeastern Pará and northeastern Mato Grosso, while in some regions (especially in the northeastern corner of Pará), the area covered by crop plantations was negatively associated with deforestation. For Pará, rural credit constraints, larger territories designated as sustainable use areas and indigenous lands, and higher levels of precipitation inhibit deforestation. Further, rural population has a very heterogeneous impact on deforestation across Legal Amazon: it is not a significant factor of deforestation in northern Pará and Amapá, but it has a relatively strong effect in the western parts of Mato Grosso and Rondônia. Also, official and illegal roads create significantly more pressure on forests in remote regions compared to developed areas. Finally, the use of economic distances, as opposed to Euclidean distances, leads to notably different GWR results.