Key variables for pressure levels in protected areas in the state of Rondônia.
Deforestation in the Amazon, Predictive Modeling, Environmental Governance
Deforestation in the Brazilian Amazon is a complex phenomenon, whose drivers are widely debated in the scientific literature, but frequently analyzed at aggregate scales that obscure local dynamics. This work investigated the drivers of forest loss through a mixed-methods approach, combining a Systematic Literature Review (SLR) and spatially explicit predictive modeling. The SLR, based on 135 articles, revealed a significant methodological bias: 79% of the drivers identified as relevant are concentrated at macro and meso-scales, generating a knowledge gap about the factors operating at the property level and within protected areas. To mitigate this gap, a case study was conducted in the Conservation Units (CUs) of Rondônia, using the Random Forest machine learning algorithm and the SHAP interpretability technique at a microscale (250 m resolution). The model obtained an average AUC of 0.81 and identified "distance to internal deforestation" as the variable with the greatest predictive power, demonstrating that pre-existing environmental liabilities act as catalysts for new conversions through a contagion effect. The results highlighted extreme institutional vulnerability, with the risk concentrated in state-managed and Sustainable Use protected areas, notably the Jaci-Paraná Extractive Reserve, suggesting that internal degradation precedes and justifies political processes of reduction or declassification of these areas (PADDD). It is concluded that the effective protection of protected areas depends not only on legal barriers, but also on the immediate containment of internal logistical expansion and the strengthening of governance at the state level.