Modelagem da distribuição espacial do café em Minas Gerais usando a técnica MaxEnt

Abstract

Coffee is one of the main tropical agricultural commodities, with high economic and social relevance, with Minas Gerais standing out as an important production hub. However, the scarcity of detailed georeferenced records of cultivated areas hinders agricultural planning, production monitoring, and the development of strategies to adapt to climate change. In this context, remote sensing combined with species distribution models represents a promising approach to accurately map and monitor crops. This study evaluated the performance of the MaxEnt model in the spatial modeling of coffee cultivation in Minas Gerais, using satellite imagery from Sentinel-2, Landsat-8 and MODIS, integrated with spectral, topographic, and bioclimatic predictors. Different combinations of variables and training-to-validation ratios were tested, and model performance was assessed using metrics such as AUC, Kappa, accuracy, sensitivity, specificity, and omission rate. The results indicated that sensors with higher spatial and spectral resolution, such as Sentinel-2 and Landsat, provided greater detail in heterogeneous ecosystems, and the use of different predictive variables helps to improve the model and to reduce the omission rate. MaxEnt performance ranged from good to excellent (AUC ranging from 0.782 to 0.936), with low omission rates, demonstrating robustness and strong predictive capability.