Análise de potencial de erosão: aplicação da Equação Universal de Perda de Solo Revisada (RUSLE) e de Redes Neurais Convolucionais (CNN) estudo de caso: Distrito de Moatize e Sussundenga

Abstract

This study analyzes the applicability of two models, the Universal Soil Loss Equation (RUSLE) and Convolutional Neural Networks (CNN), for predicting soil erosion susceptibility in two distinct regions of Mozambique: Moatize and Sussundenga. The main objective is to understand the applicability of each approach for mapping soil erosion susceptibility in areas with distinct characteristics, highlighting the potential of integrating empirical methods and artificial intelligence (AI) techniques. The methodology adopted involved processing four factors (R, K, LS, and C) of the RUSLE model, allowing for the estimation of soil loss. At the same time, for the CNN-based model, geoenvironmental variables were carefully selected, ensuring minimization of multicollinearity through the Variance Inflation Factor (VIF), keeping only those with a VIF ≤ 5. Various sets of points were extracted from these images to evaluate the evolution of the CNN model: 870, 4,000, 6,000, and 8,000 points. The data were normalized and used to train the CNN model. The results showed that the RUSLE model offers a direct estimate of soil loss and remains a practical tool for exploratory analyses with low computational cost. In turn, the CNN demonstrated high potential for capturing complex spatial patterns, achieving AUC values ​​of 0.989 in Moatize and 0.984 in Sussundenga, especially when trained with a set of 8,000 points. This approach proves advantageous in contexts that require a higher level of precision and flexibility in results. The conclusions of this study reinforce that both models are applicable for mapping erosion in different contexts. The combined use of these approaches contributes to enriching the methodological basis of erosion studies, providing support and strategies for soil conservation and mitigating environmental impacts.(TRADUÇÃO NOSSA)