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.

Urban flooding is a complex natural hazard, driven by the interaction between several parameters related to urban development in a context of climate change, which makes it highly variable in space and time and challenging to predict. In this study, we apply a multivariate analysis method (PCA) and four machine learning algorithms to investigate and map the variability and vulnerability of urban floods in the city of Tangier, northern Morocco. Thirteen parameters that could potentially affect urban flooding were selected and divided into two categories: geo-environmental parameters and socio-economic parameters. PCA processing allowed identifying and classifying six principal components (PCs), totaling 73% of the initial information. The scores of the parameters on the PCs and the spatial distribution of the PCs allow to highlight the interconnection between the topographic properties and urban characteristics (population density and building density) as the main source of variability of flooding, followed by the relationship between the drainage (drainage density and distance to channels) and urban properties. All four machine learning algorithms show excellent performance in predicting urban flood vulnerability (ROC curve > 0.9). The Classifications and Regression Tree and Support Vector Machine models show the best prediction performance (ACC = 91.6%). Urban flood vulnerability maps highlight, on the one hand, low lands with a high drainage density and recent buildings, and on the other, higher, steep-sloping areas with old buildings and a high population density, as areas of high to very-high vulnerability.

The SISE-Eaux database of water intended for human consumption, archived by the French Regional Health Agency (ARS) since 1990, is a rich source of information. However, more or less regular monitoring over almost 30 years and the multiplication of parameters lead to a sparse matrix (observations × parameters) and a large dimension of the hyperspace of data. These characteristics make it difficult to exploit this database for a synthetic mapping of water quality, and to identify of the processes responsible for its diversity in a complex geological context and anthropized environment. A 10-year period (2006–2016) was selected from the Provence-Alpes- Côte d’Azur region database (PACA, southeastern France). We extracted 5,295 water samples, each with 15 parameters. A treatment by principal component analysis (PCA) followed with orthomax rotation allows for identifying and ranking six principal components (PCs) totaling 75% of the initial information. The association of the parameters with the principal components, and the regional distribution of the PCs make it possible to identify water-rock interactions, bacteriological contamination, redox processes and arsenic occurrence as the main sources of variability. However, the results also highlight a decrease of useful information, a constraint linked to the vast size and diversity of the study area. The development of a relevant tool for the protecting and managing of water resources will require identifying of subsets based on functional landscape units or the grouping of groundwater bodies.

The application of GRACE satellite-derived Total Water Storage (TWS) data for local water management is constrained by its coarse spatial resolution (100-300 km). To address this limitation, a Random Forest-based model was employed to downscale GRACE TWS data from 100 km to 1 km resolution over Morocco, a drought-prone region, covering the period from 2002 to 2022. The input datasets included precipitation (GPM, 10 km), NDVI (MODIS, 1 km), land surface temperature (LST, MODIS, 1 km), evapotranspiration (MODIS, 500 m), elevation (SRTM, 30 m), and the Normalised Difference Snow Index (NDSI, MODIS, 500 m). While downscaling improves the spatial resolution of GRACE data, validating these higher-resolution outputs presents challenges. In this study, the downscaled data were validated using three complementary approaches: statistical validation, groundwater level in-situ data validation, and validation against known aquifer dynamics. Statistical validation demonstrated strong model performance, with a Nash-Sutcliffe Efficiency (NSE) of 0.80, a low RMSE of 0.82 cm, and MAE of 0.57 cm, along with an R^² of 0.80 between original and downscaled data. Cross-validation confirmed the model’s consistency, yielding mean, median, and maximum R^² values of 0.56, 0.64, and 0.89 respectively. Error metrics remained consistently low throughout the study period, with MAE values ranging from 0.36 cm to 0.6 cm and RMSE values between 0.5 cm and 0.8 cm. Comparison with in-situ groundwater levels showed significant improvements, with correlation coefficients increasing for 63% of the 139 analysed wells. The 1 km TWS data revealed localised variations and clearer trends across different aquifers, with aquifer systems within the same structural domain exhibiting similar TWS patterns. These findings highlight the potential of the downscaling model to enhance local water management by capturing finer hydrological variations. The proposed approach effectively overcomes GRACE’s spatial resolution limitations, as demonstrated through comprehensive validation. This methodology shows particular promise for water resource monitoring in drought-vulnerable regions such as Morocco, providing decision-makers with higher-resolution data for improved water management strategies.

This study examines urban flood vulnerability in Tetouan city, Northern Morocco, using four machine learning models—Classification and Regression Tree (CART), Support Vector Machine (SVM), Logistic Regression (LR), and Factorial Discriminant Analysis (FDA)—to identify and map flood-prone areas. The primary goal is to enhance flood prevention efforts and minimize losses by determining the most vulnerable zones. The analysis highlights consistent flood risk along the Martil River and eastern plains, areas characterized by low-lying topography, dense drainage, proximity to canals, and recent urban development. Despite some spatial variation among the models, all consistently indicate low and very high vulnerability zones, with FDA identifying the highest proportion of very high risk areas (58%), followed by CART, SVM, and LR (39%, 38%, and 37%, respectively). In terms of model accuracy, SVM and LR outperform others, demonstrating their effectiveness in flood risk delineation. The findings offer valuable insights for urban planners and decision-makers in flood risk management, contributing to more informed resource allocation in Tetouan-Martil and potentially guiding similar strategies in comparable regions globally.