OWHK: Operational volumetric water content forecasting model for shallow rainfall-induced landslides in Hong Kong

Abstract

Rainfall-induced landslides result from complex hydrological and geotechnical interactions, with one of the key challenges being the accurate estimation of infiltrated rainfall. This study introduces a unique operational model for predicting shallow volumetric water content (VWC), a critical parameter for assessing rainwater infiltration. Using Hong Kong as a case study, the proposed approach overcomes spatiotemporal limitations in existing predictive models by accounting for the randomness of rainfall-triggering mechanisms. The methodology integrates a unique data preparation technique, independence-oriented time series windowing, with the predictive power of deep learning (DL), specifically, Long Short-Term Memory (LSTM) networks, and deterministic seepage modeling via GeoStudio SEEP/W. The model is developed using field data from 15 sensors across three sites in Hong Kong (Pa Mei, Tung Chung, and Tsing Shan) at two depths (0.5 m and 1.5 m), complemented by a numerically validated case at Fei Ngo Shan Reservoir. 33 field samples were collected from eleven different locations to validate initial hypotheses. This study is among the initial systematic evaluations comparing the accuracy of operational versus site-specific models. The outcome is OWHK (Operational VWC Forecasting Model for Shallow Layers in Hong Kong), a user-friendly tool demonstrating predictive performance with mean absolute error (MAE) below 0.6 %, coefficient of determination (R²) exceeding 0.92, and a mean Monte Carlo dropout standard deviation (SD) of 0.3 %.