Erosion Susceptibility Mapping and Conservation Planning for Agricultural Watersheds in Anambra State Using GIS and Machine Learning

This study develops an erosion susceptibility map for agricultural watersheds in Anambra State, South East Nigeria, using integrated GIS analysis and machine learning algorithms, and applies the outputs to develop spatially targeted conservation planning recommendations. Anambra State is one of Nigeria's most erosion-prone states, with active gully systems threatening agricultural land and rural infrastructure across the Udi Ridge and lower Niger River zones. Effective conservation planning requires spatially explicit understanding of erosion susceptibility that traditional USLE-based approaches may inadequately capture due to non-linear relationships between erosion factors. This study uses random forest and gradient boosting machine learning classifiers applied to multi-source geospatial data including SRTM DEM, Sentinel-2 satellite imagery, soil survey data, rainfall erosivity data, and land use maps for Anambra State. Training data are generated from 2,500 mapped erosion occurrence locations derived from field surveys and satellite image interpretation. The ensemble model is calibrated using 70 percent of training data and validated on the remaining 30 percent. Conservation planning scenarios are developed for high-susceptibility zones. Findings reveal that the gradient boosting model achieves area under the ROC curve of 0.91, significantly outperforming traditional USLE-based mapping. Slope gradient, proximity to existing gullies, and land use type are the most important susceptibility predictors. High and very high susceptibility zones cover 31 percent of the state's agricultural land. The study recommends prioritised conservation investment in the mapped high-susceptibility zones.

Keywords: erosion susceptibility mapping, GIS, machine learning, Anambra State, gully erosion.