Mapping Gully Erosion Variability and Susceptibility Using Remote Sensing, Multivariate Statistical Analysis, and Machine Learning in South Mato Grosso, Brazil

In Brazil, the development of gullies constitutes widespread land degradation, especially in state South Mato Grosso, where fighting against this degradation has become a priority for policy makers. However, environmental and anthropogenic factors that promote gully are multiple, interact, present complexity can vary by locality, making their prediction difficult. framework, database was constructed Rio Ivinhema basin southern part state, including 400 georeferenced 13 geo-environmental descriptors. Multivariate statistical analysis performed using principal component (PCA) to identify processes controlling variability development. Susceptibility maps were created through four machine learning models: multivariate discriminant (MDA), logistic regression (LR), classification tree (CART), random forest (RF). The predictive performance models analyzed five evaluation indices: accuracy (ACC), sensitivity (SST), specificity (SPF), precision (PRC), Receiver Operating Characteristic curve (ROC curve). results show existence two major erosion. first is surface runoff process, which related conditions slightly higher relief rainfall. second also reflects high conditions, but rather drainage density downslope, close river network. Human activity represented peri-urban areas, construction small earthen dams, extensive rotational farming contribute significantly formation. yielded fairly similar validated susceptibility > 0.8). we noted better (RF) model (86% 89.8% training test, respectively, with an ROC value 0.931). contribution parameters shows erosion not governed primarily single factor, interconnection between different factors, mainly elevation, geology, precipitation, use.