ABSTRACT
Despite the acknowledged global importance of Brazilian forests, the evidence on the drivers of tree-cover change is mostly based on local analyses. The emphasis on Amazonia also overshadows tree-cover research in the rest of the biomes. This contrasts with widespread human encroachment. This chapter examines the totality of Brazil’s tree-cover change from 1992 to 2001, from the viewpoint of biophysical and socioeconomic impacts. But a large-area study brings forth the issues of spatial variability of the impacts, and local or neighborhood effects. These are dealt with explicitly in our models. The additional problem of sampling, often used to allay the issue of spatial autocorrelation, is examined. The results highlight the strong and signifi cant effect of initial forest cover in predicting subsequent tree loss. In the opposite direction, local effects seem to indicate that forest pixels in forest-only tracts are less likely to suffer tree loss. The historical southeast-northwest direction of deforestation pulses in Brazil is confi rmed, and on average, higher distances to cities tend to protect tree cover. Another strong driver of tree-cover loss seems to be the density of cattle. Considering Amazonia, the Arc of Deforestation (AoD), and Southern Brazil, the signifi cant variables are the same, pointing to similar processes. However, the strength of these variables differs regionally. Temperature and latitude also differ in sign in Amazonia and Southern Brazil. Considering the empirical spatial impacts identifi ed by geographically weighted regression (GWR), steep slopes seem to protect tree cover in northern Amazonia and the Atlantic forest. In Eastern Brazil, higher distances to cities seemed to indicate locations that are detrimental to tree cover. Close spatial matches with the shape of the AoD are given by the density of unpaved roads, whose important role is thus confi rmed, and by human population density. The best match with the AoD is given by the impact of cattle density. The foregoing results support the use of GWR in identifying and measuring empirical patterns of impact, along with spatial autoregressive regressions that measure the strength of local interactions.
