Integrating geospatial techniques and machine learning for assessing soil erosion and associated geomorphic risks

Abstract

Soil erosion is a critical environmental issue that threatens sustainability, agriculture, and infrastructure, necessitating a precise evaluation of geomorphic risks. This study employs an integrated approach combining geospatial techniques and machine learning models to assess soil erosion exposure in the Colima area. A comprehensive dataset of geospatial parameters, including inclination angle, land use/land cover (LULC), topographic wetness index (TWI), geological formation, proximity to roads and waterbodies, terrain elevation, aspect, sunlight exposure (HillShade), soil classification, vegetation index (NDVI), terrain ruggedness index (TRI), and sediment transport index (STI), was utilized to capture the spatial variability influencing erosion processes. Machine learning models—CatBoost, AdaBoost, Convolutional Neural Networks (CNN), and Stacking—were evaluated for their predictive performance using sensitivity, specificity, specificity, F1 score, recall, and precision metrics. Among these models, CNN achieved the highest accuracy, with sensitivity, specificity, F1 score, recall, and precision values of 0.84, 0.92, 0.90, 0.84, and 0.93, respectively. The CNN model's ability to capture complex spatial relationships and patterns in the dataset underscores its suitability for erosion risk assessment. The findings indicate that terrain elevation, LULC, and NDVI are key determinants of soil erosion susceptibility. High-risk areas were associated with steep slopes, sparse vegetation cover, and proximity to waterbodies. Canoas, Camoltan De Miraflores, Chandiablo, and Jalipa have been identified as areas facing significant soil erosion risks. These regions exhibit steep slopes, sparse vegetation cover, and proximity to water bodies, making them highly susceptible to erosion. The study's findings provide valuable insights for implementing targeted erosion control measures and sustainable land management strategies in these vulnerable areas. The study provides an efficient framework for identifying erosion-prone zones and prioritizing mitigation measures, contributing to sustainable land management strategies. By leveraging advanced machine learning techniques and geospatial analysis, this research advances the predictive modelling of geomorphic risks and aids in the development of targeted erosion control interventions.