Modelagem espaço-temporal do desmatamento na Amazônia brasileira: uma abordagem via modelos de regressão com coeficientes variando no espaço e no tempo

Abstract

Deforestation in the Brazilian Amazon is a critical environmental issue with significant impacts on climate, biodiversity, and the carbon cycle. Modeling the spatiotemporal dynamics of this process is crucial for understanding its patterns and identifying associated factors. Therefore, this study proposes a spatiotemporal modeling approach for the proportion of deforested area in the Amazon biome, exploring different methodological approaches. Initially, the Geographically Weighted Regression (GWR) model is employed to analyze spatial variability, and the Geographically and Temporally Weighted Regression (GTWR) model is used to assess the spatiotemporal variability of the effects of variables associated with deforestation. Based on these analyses and the limitations of GWR and GTWR models, the need for a more robust approach was identified, leading to the adoption of a spatiotemporal regression model with coefficients varying across space and time. Specifically, a normal model is considered for the logit of the proportion of deforested area, where the spatial variability of the coefficients is captured by Conditional Autoregressive (CAR) models and the temporal variability by dynamic linear models. As an alternative to the normal model, a beta regression model is also proposed, which directly models the proportion of deforested area in its original scale. The evaluation of these models was performed using applications on both simulated and real data for municipalities in the state of Pará. The normal model was also applied to real data from municipalities across the entire Amazon biome. The inference followed a Bayesian approach, employing Markov Chain Monte Carlo methods to estimate the parameters. The results provide a deeper understanding of the factors influencing deforestation and present methodological tools that can contribute to environmental studies and the formulation of public policies.