Rainfall-Induced Landslide Prediction Models, Part I: Empirical–Statistical and Physically Based Causative Thresholds

Introduction and Problem Statement: Landslides represent a significant geological hazard worldwide. One of the primary triggers for these landslides is rainfall, which is becoming more intense as a result of climate change. The available literature has produced extensive research. However, this largely overlooks the use of mixed methodologies. Furthermore, a comprehensive review combining empirical, physically based, deterministic, and phenomenological models is still rare. Objective and Method: This study (Part I of a two-part review) addresses this gap by employing a mixed review that integrates quantitative scientometric analysis with a qualitative systematic review. The primary objective of Part I is to deliver a critical assessment, focusing on empirical and physically based causative threshold models. Main Results and Validation: Macroscopically, our analysis reveals that antecedent rainfall is a more robust indicator than classical intensity–duration (I-D) thresholds, though the latter remains widely used due to its simplicity. Physically based models provide a critical bridge when geotechnical data is scarce, correlating rainfall with internal slope responses like displacement. At a microscopic level, hybrid artificial intelligence (AI) models consistently demonstrate superior predictive accuracy by capturing complex, nonlinear relationships missed by simpler models. These findings are validated through a systematic evaluation of performance metrics across the reviewed literature. Main Conclusions and Significance: We conclude that while empirical thresholds offer operational simplicity, the future of accurate prediction lies in sophisticated hybrid AI models trained on extensive monitoring data. This review synthesizes fragmented knowledge into a unified framework, providing a clear roadmap for model selection.