Quantitative spatial analysis of deforestation in legal Amazonselected topics

Resumo

The dissertation covers five topics on deforestation in Legal Amazon. The first study investigates spatial heterogeneity of deforestation determinants at municipality level. Spatial differences are assessed by geographically weighted regression. The distances between regression points are measured in travel time. The computation is done programmatically. Different drivers of deforestation emerge in different locations of Legal Amazon. For Pará and its surroundings, cattle market is an especially strong driver of deforestation. Crop cultivation leads to forest clearings only in a relatively small area, located in southeastern Pará and northeastern Mato Grosso. Rural credit constraints are effective in curbing deforestation in Pará. Here less deforestation happens where more forests are legally protected, where precipitation levels are favorable for agriculture and at lower altitudes. U-shaped environmental Kuznets curve is concluded for the entire region. However, significant links are found only in Amazonias, Roraima, Pará and its proximities. Timber value motivates deforestation in most parts of the Amazon biome. Official roads contribute to deforestation in Amazonias, Roraima and their surroundings. Adverse effect of unofficial roads on extant forests is especially evident in northern Rondônia and northeastern Pará. Links between rural population and deforestation are very strong for western parts of Rondônia and Mato Grosso, but are very weak in Pará. The implementation of economic distances relative to Euclidean distances changes the results significantly for some regions. The second article investigates whether sugarcane expansion in southern Brazil exports deforestation into the Amazon. This indirect land use change is captured using spatial Durbin model. The parameters are estimated by fixed-effects regression. The results indicate that sugarcane expansion exported 16.3 thousand km2 (12.2%) of deforestation during period 2002-2012, which is equivalent of 189.4 Mg of carbon emissions. The third study contributes to the polemics of whether rural population is linked with deforestation on forest edges. Empirical strategy is as follows: Pará state is partitioned into 5x5 km grids, only cells that classify as forest frontier are retained, links between deforestation and its covariates (including rural population) are investigated both parametrically (fractional logistic regression) and non-parametrically (regression tree). The results confirm positive link between the size of rural communities and deforestation on forest frontiers. Both methods suggest that deforestation is positively linked with cattle herd size and distance to the most proximate river and negatively linked with forest cover and precipitation. Regression tree also reveals that deforestation within protected areas is substantially lower. The fourth paper quantifies avoided deforestation in Pará’s protected areas, on their edges and in their peripheral areas (buffer zones) by matching. Location characteristics are converted into a single propensity score by the means of logistic regression. Pará avoided ~2900 km2 of deforestation during 2000-2004. Space has huge implications: conservation units in remote regions do not avoid deforestation, whereas protected areas near deforestation hotspots save substantial areas of forests. Avoided deforestation is positive in buffer zones located to the west of highway BR-163 and on the banks of Amazon River, and negative in buffers located in eastern Pará. Boundaries of conservation units are well protected from edge effects. The last study maps deforestation at 5x5 km grids in selected territory in Rondônia from past and time-fixed factors. Eigenvector-based spatial filtering is applied to solve spatial autocorrelation problem and to improve mapping accuracy. Output values of trained artificial neural network satisfactory correlate with actual values (correlation coefficient is 0.79).