Rainfall Induced Shallow Landslide Temporal Probability Modelling and Early Warning Research in Mountains Areas: A Case Study of Qin-Ba Mountains, Western China

Abstract

The rainfall-induced landslide early warning model (LEWM) is an important means to mitigate property loss and casualties, but the conventional discriminant matrix-based LEWM (DLEWM) leaves room for subjectivity and limits warning accuracy. Additionally, it is important to employ appropriate indicators to evaluate warning model performance. In this study, a new method for calculating the spatiotemporal probability of rainfall-induced landslides based on a Bayesian approach is proposed, and a probabilistic-based LEWM (PLEWM) at the regional scale is developed. The method involves four steps: landslide spatial probability modeling, landslide temporal probability modeling, coupling of spatial and temporal probability models, and the conversion method from the spatiotemporal probability index to warning levels. Each step follows the law of probability and is tested with real data. At the same time, we propose the idea of using economic indicators to evaluate the performance of the multilevel LEWM and reflect its significant and unique aspects. The proposed PLEWM and the conventional DLEWM are used to conduct simulate warnings for the study area day-by-day in the rainy season (July-September) from 2016 to 2020. The results show that the areas of the 2nd-, 3rd-, and 4th-level warning zones issued by the PLEWM account for 60.23%, 45.99%, and 43.98% of those of the DLEWM, respectively. The investment in issuing warning information and the losses caused by landslides account for 54.54% and 59.06% of those of the DLEWM, respectively. Moreover, under extreme rainfall conditions, the correct warning rate of the PLEWM is much higher than that of the DLEWM.

Keywords

• landslides
• empirical rainfall threshold
• early warning
• landslide susceptibility index
• artificial neural network