An Exploratory Spatial Data Analysis for Deforestation in Brazilian Amazon

Historically, development in Brazilian Amazon was promoted by permits of deforestation and little territory control or supervision. However, due to the importance of this biome for biodiversity and ecosystem balance in a global perspective, Brazilian's government has tighten deforestation control. Since 2004, political, institutional and market mechanisms have been implemented and annuals rates reduced from near 20,000 km² to 6,000 km² in 10 years. Despite several empirical agents behavior models in literature indicate for a spatial process of deforestation, all aspects of the spatial dynamic of deforestation has not been explored yet. In this context, this work studies the spatial pattern of deforestation in a cross-section and time perspective using exploratory spatial data analysis techniques. For these purposes, we used global and local analysis estimating Moran's I statistic and Local Indicator of Spatial Association clusters maps. Data consisted in annual deforestation by municipality for Brazilian's Legal Amazon, composed by 686 municipalities, within a 12 year period: from 2002 to 2013. Statistics were estimated for three variables: absolute deforestation and two indexes calculated annually by municipality. The first one represents the fraction of the total area deforested in each year and the second one is the cumulative sum of the first index. Considering the absolute deforestation analysis we clarify locations that most contribute for deforestation in a macro regional perspective. The purpose of the first index is to identify municipalities with most environmental degradation and the second index tries to capture a maximization behavior of land use agents. Global univariate results indicate the existence of high spatial correlation dependency. Moreover, despite the efficacy of the policy mechanisms adopted for controlling and reducing the level of deforestation in Legal Amazon, Moran's I positive values show persistent spatial concentration of deforestation. This suggests that policy control mechanisms have achieved relative success to reduce the average deforestation but have not influenced its spatial dispersion. As expected, local analysis show clusters in Arch of Deforestation region, known for being the key region of deforestation. Furthermore, the low frequency of divergent neighbors, i. e., high-low and low-high clusters, supports the hypothesis of high spillover effects, as pointed out by high Moran's I values. Comparing annual results show that municipalities with historically high relative deforestation rate presented reduction in forest area conversion, as their detachment in clusters ceased, although we cannot judge if by market, legal or political aspects. Also, we highlight the different implication of cross sectional and temporal analyses for policy selection. Municipalities that detaches in cross section should suffer control and legal action and detachment in temporal analysis indicates necessity of environmental recover and conservation actions.