Machine Learning and Remote Sensing for Gully Erosion Susceptibility Mapping in Southeast Nigeria: Progress, Challenges, and Research Priorities

Gully erosion remains one of the most destructive forms of land degradation in tropical environments, threatening agricultural productivity, infrastructure, ecosystem services, and human settlements. Southeast Nigeria represents one of the world's most vulnerable gully erosion hotspots owing to its highly erodible geological formations, intense rainfall, rapid land-use change, and increasing anthropogenic disturbances. Recent advances in machine learning (ML), cloud computing, and Earth observation technologies have significantly improved gully erosion susceptibility mapping (GESM); however, the extent to which these innovations have been adopted within Southeast Nigeria remains unclear. This study presents a PRISMA 2020-compliant systematic review of machine learning-based GESM studies published between January 2015 and June 2026. A comprehensive literature search was conducted using structured Boolean search strategies. Random Forest and Extreme Gradient Boosting emerged as the most frequently applied algorithms, and consistently achieving Area Under the Receiver Operating Characteristic Curve (AUC) values ranging from 0.88 to 0.98. Sentinel-2 multispectral imagery and the Shuttle Radar Topography Mission (SRTM) Digital Elevation Model constituted the dominant remote sensing datasets, while slope, vegetation indices, elevation, rainfall, land-use/land-cover, Topographic Wetness Index, and Stream Power Index were the most influential conditioning factors. Despite these global advances, only eighteen studies specifically did research on Southeast Nigeria and three employed ML approach. Furthermore, none integrated Sentinel-1 Synthetic Aperture Radar (SAR) with optical imagery or implemented modern deep learning architectures such as Convolutional Neural Networks or U-Net models. The review reveals major methodological gaps between global developments and current research practices in Southeast Nigeria. Major research gaps include the absence of SAR-optical data fusion, explainable artificial intelligence, deep learning, temporal susceptibility modelling, regional multi-state assessments, and cloud-based Google Earth Engine workflows. A comprehensive methodological framework integrating Sentinel-1 SAR, Sentinel-2 imagery, Google Earth Engine, ensemble machine learning, spatial cross-validation, and SHAP-based model interpretability is proposed to guide future research.