Sub-basin Prioritisation for Landslide Risk Management via Correlation-Guided Morphometric Feature Selection and Consensus Scoring: Evidence from the Lukaya Watershed, DR Congo

Introduction: Landslide risk management in rapidly urbanising tropical environments often requires decision-ready prioritisation of intervention areas rather than pixel-scale susceptibility maps. We present a reproducible workflow to rank sub-basins for landslide risk management in the Lukaya Watershed (DR Congo) using Digital Elevation Model (DEM)-derived morphometry and an inventory of landslide points for validation.
Methodology: The drainage network was extracted from a hydrologically conditioned DEM using a 3 km² contributing-area initiation threshold. Stream-link sub-basins were delineated and merged downstream to enforce a minimum unit area of 1.5 km², yielding 63 operational sub-basins. Correlation screening reduced candidate predictors to eight parsimonious morphometric variables (area, elongation ratio, compactness coefficient, relief, mean slope, maximum slope, hypsometric integral, and drainage density). Four scoring approaches were implemented: Analytic Hierarchy Process (AHP), equal-weight composite scoring, logistic regression, and random forest. Supervised models were assessed with spatially blocked out-of-fold predictions.
Results: The landslide inventory comprised 75 points across 15 sub-basins. Ranking validation combined discrimination and prioritisation-oriented metrics (ROC–AUC, PR–AUC, Brier score, and success/capture curves). AHP achieved the highest overall discrimination (ROC–AUC 0.786), while random forest captured 50/75 landslide points within the top 10% of sub-basins. A consensus prioritisation (mean rank of AHP, spatially validated logistic regression, and spatially validated random forest) concentrated 46/75 points (61.3%) in the top 10% class and 63/75 points (84.0%) in the top 30% of sub-basins.
Conclusion: Correlation-guided morphometric feature selection combined with multi-method consensus scoring provides a transparent, reproducible, and inventory-validated basis for operational sub-basin prioritisation for landslide risk management.