Eroding Confidence in Gully Location Risk Prediction in Mine Pit Rehabilitation

Understanding how rehabilitation measures, such as landform shaping, and surface finishes impact gully erosion risk could assist in designing less erosive rehabilitation. In this study we use high resolution digital elevation models of rehabilitated mine pits to explore risk factors for gully formation. We determine features that pre‐dispose rehabilitated mining pits to gully formation, how topographic features vary across these areas, and generate a binomial regression and a random forest model relating surface features to erosion risk. These models are benchmarked against the ‘topographic threshold’. We found topographic features, like hillslope gradient and curvature, which concentrate and accelerate runoff were overrepresented at gully heads, while at the mine pit level, erosion risk can be predicted by surface area and average pit hillslope gradient, providing a high‐level risk screening tool. The topographic threshold, applied with high resolution accumulation data, outperformed the random forest and logistic regression models in predicting risk from topographic features alone. Despite being based on post‐erosion data, the topographic threshold performs well to predict erosion risk across rehabilitated mining areas. The method balances data requirements with prediction accuracy and is a conservative approach to highlighting risk areas within rehabilitated pits. The random forest and regression approaches underpredicted gully locations, in contrast with the topographic thresholds over prediction, which limits their utility without additional inputs. Operators have opportunities to collect additional data during rehabilitation that could build better predictive models. Informed rehabilitation design offers the opportunity to avoid, where possible, features that occur more commonly in conjunction with erosion.