Avaliação da suscetibilidade à erosão linear e seus fatores condicionantes no cerrado goiano

Abstract

Gully erosion compromises ecosystems, reduces soil fertility, and affects agri-cultural productivity, in disagreement with the United Nations (UN) Sustainable Development Goals (SDGs). Gully erosion susceptibility mapping is essential for soil conservation strategies and territorial planning. This study investigates gully erosion susceptibility in the state of Goiás, employing machine learning techniques. Twenty-three variables related to erosion conditioning factors were considered, inclu-ding topographic attributes, land use and land cover, anthropogenic, geological, and hydrological factors, 11 of which were derived from the FathomDEM Digital Terrain Model (DTM). The dataset comprised 67,078 erosion samples, including gullies and ravines, which were partitioned for model training and testing. The prediction was conducted using a supervised classification approach based on the Random Forest Ranger algorithm (RF Ranger), calibrated through k-fold cross-validation (k = 10) and evaluated using the Areas Under the ROC Curve (AUC). The statistical asses-sment of the RF Ranger model indicated strong predictive performance, yielding an AUC of 0.8657. The variable importance ranking identified the Normalized Diffe-rence Vegetation Index (NDVI) as the most influential factor, followed by pasture and pasture age, reinforcing the role of vegetation alteration in erosion processes. The rainfall erosivity factor (R factor) and topographic variables related to surface roughness and terrain morphology were also relevant. The predictive map shows that areas with highest susceptibility are concentrated in drainage headwaters characterized by steep slopes and more erodible soils, whereas the lower susceptibility are associated with flat terrain and preserved vegetation.