Integrating erosion susceptibility, topographic wetness, and short-duration rainfall intensity thresholds for landslide initiation: A process-oriented framework for the Black Sea Region of Türkiye

Rainfall-triggered landslides result from the interaction between long-term preparatory terrain conditions and short-duration triggering rainfall events. However, most regional susceptibility studies either treat these components separately or represent rainfall using cumulative precipitation metrics that inadequately capture the intensity characteristics of triggering storms. This study proposes a process-oriented monthly framework integrating terrain susceptibility indicators with rainfall erosivity-derived short-duration rainfall intensity metrics to improve representation of landslide initiation processes in the Black Sea Region of Türkiye. Preparatory conditions were characterized using erosion susceptibility indicators derived from the Revised Universal Soil Loss Equation (RUSLE) and the Topographic Wetness Index (TWI), representing terrain instability and moisture accumulation potential. Triggering conditions were quantified using monthly maximum 30-minute rainfall intensity (I30) values extracted from erosive rainfall events identified through the RUSLE rainfall erosivity framework. Because this methodology selectively identifies rainfall events with sufficient intensity and energy to generate significant surface responses, the derived I30 metric provides a physically interpretable representation of rainfall conditions associated with slope destabilization. Seven predictor scenarios combining preparatory and triggering variables were evaluated using four machine-learning algorithms (logistic regression, random forest, XGBoost, and LightGBM). Monthly rainfall-triggering thresholds were independently estimated using probabilistic, ROC-based (Youden index), and percentile-based approaches. Results indicate that scenarios integrating continuous I30 values with erosion susceptibility and wetness indicators consistently achieved the highest predictive performance, with monthly ROC–AUC values frequently exceeding 0.85. Threshold-based representations yielded slightly lower performance but provided physically interpretable estimates of rainfall conditions associated with landslide occurrence. Monthly analyses revealed strong temporal stability in model performance together with pronounced seasonal variability in rainfall-triggering effectiveness. The proposed framework establishes a direct link between rainfall erosivity, terrain susceptibility, and rainfall-triggering thresholds, providing a physically interpretable and operationally relevant approach for regional landslide susceptibility assessment and seasonally adaptive early-warning applications.