The paper presents a two-stage model of Amazon deforestation. The first stage, at município level, is a dynamic model of the socio-economic determinants of major classes of land use (pasture, permanent crop, temporary crop, and forest areas). Estimation of bayesian spatial autocorrelation models based upon panel data of Agricultural an Demographic Censi information from 1970 to 1995 complemented by geo-ecological and transport infrastructure information available for the same period.
The second stage is a static cross section model of the geo-ecological (soil, vegetation, precipitation, etc.) and transport infrastructure (distance, cost, etc.) determinants of the distribution of different land uses among census tracts within each município. Estimation of bayesian autocorrelation models are based upon 1995 Agricultural Census data complemented by the road infrastructure and geo-ecological information available at Census tract level.
The coupling of these models allows the simulation of future land uses and their spatial distribution for different scenarios of infrastructure development.