Evaluation of Morphometric Conditioning Factors and Antecedent Rainfall in the Occurrence of Torrential Flows in Colombian Andean Watersheds

Torrential flows, a broad category of rapid hydrogeomorphic processes that in the Colombian Andes includes debris flows, mudflows, and hyperconcentrated flows, pose a major hazard in tropical mountain regions. This study used two complementary binary classification models to examine geomorphometric conditioning and antecedent rainfall triggering of torrential flow occurrence. A 12.5 m ALOS PALSAR DEM and 42 years of daily rainfall data (1981–2023) from IDEAM rain gauges and CHIRPS v2 were analyzed in a GIS-based regional framework. Antecedent rainfall variables were aggregated at watershed scale using zonal statistics. The conditioning dataset comprised 642 watersheds (321 with documented events and 321 controls). Gradient boosting ranked first in the preliminary grouped holdout comparison, whereas the uncalibrated random forest achieved the highest mean score under spatial leave-one-province-out validation and was selected as the final conditioning model (mean ROC-AUC = 0.747 ± 0.052). Basin scale and relief were the leading morphometric associations. In the rainfall trigger model, previous day IDEAM mean rainfall and previous day IDEAM maximum rainfall were the two leading permutation importance predictors, followed by monthly CHIRPS rainfall; the 90-day IDEAM maximum accumulation ranked fourth. This ordering indicates that immediate rainfall dominated the fitted model, while longer antecedent wetness retained a secondary contribution. The results support watershed prioritization and regional hazard assessment; because operational rainfall thresholds were not derived, they should not be treated as a ready-to-use early-warning model.