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

The Normalized Difference Vegetation Index (NDVI) is widely used as an indicator of irrigation activity in arid and semi-arid agricultural regions. This study evaluates how NDVI extraction scale and statistical metric influence groundwater level prediction accuracy in the irrigated Tadla Plain, MoroccoA total of 96 Long Short-Term Memory (LSTM) models were developed by combining six NDVI extraction scales, from the well pixel to 20,000 m buffers, and four statistical metrics: mean, median, maximum, and minimum. Results demonstrate that extraction scale is a critical factor controlling model performance. Across the four monitored wells, larger buffers (≥2500 m) generally outperformed smaller ones (≤1000 m), with the optimal scale occurring at 15,000 m for three wells, while one well achieved its best performance at the 1000 m scale. The best models achieved RMSE values between 0.09 and 0.625 m and R² values ranging from 0.94 to 0.997. Maximum NDVI provided the highest predictive accuracy for three wells, whereas minimum NDVI performed best for one well. Statistical analyses further confirmed that extraction scale generally exerts a stronger influence on prediction performance than the choice of NDVI metric. Spatial validation revealed that irrigated areas more than doubled between 2001 and 2017, and model performance improved as NDVI captured this broader irrigation footprint. These findings suggest that groundwater level variations in the Tadla Plain are more strongly associated with NDVI signals extracted at broader spatial scales than with strictly local vegetation conditions around individual wells, highlighting the importance of optimizing both NDVI extraction scale and metric for groundwater forecasting in irrigated semi-arid regions.