Analysis of amplifying effect of antecedent drought on rainfall-induced shallow landslides

Abstract

Rainfall-induced shallow landslides are typically attributed to short-term meteorological triggers such as rainfall intensity and duration. However, the influence of antecedent hydrological conditions, particularly prolonged dry spells, has received limited attention despite their potential to alter soil behavior significantly. This study investigates the role of antecedent dry spells on landslide occurrence in the clay-rich terrains of the eastern Black Sea region of Türkiye. Four intense rainfall events from two study sites were examined. At both sites, the later event followed a prolonged dry period and produced a substantially larger mapped landslide inventory despite lower landslide-day rainfall than the earlier event, which occurred under wetter antecedent conditions. To investigate the mechanisms underlying this behavior and evaluate whether it was evident across the event inventories, we combined mechanistic numerical simulations with a data-driven statistical analysis. Numerical simulations indicate that a configuration representing crack-related hydraulic and mechanical changes enhances infiltration and reduces matric suction, resulting in strength reduction and lower slope stability. Additionally, a Generalized Additive Model incorporating landslide-day precipitation, the 30-day Standardized Precipitation Index (SPI-30), local relief, slope steepness, and aspect reveals that landslides are more likely to occur when intense rainfall coincides with drier antecedent conditions rather than under persistently wet conditions. The model showed consistent discriminatory performance under grouped tenfold cross-validation, with an overall ROC AUC of 0.836 based on combined validation-fold predictions. These findings suggest that in clay-rich environments, conventional rainfall thresholds used in predictive models may underestimate landslide occurrence probability when preceded by dry spells. Incorporating antecedent drought indicators into forecasting frameworks may improve the representation of landslide probability, although validation across additional events and locations is required.