An hourly shallow landslide warning model developed by combining automatic landslide spatial susceptibility and temporal rainfall threshold predictions

Landslide warning models are important for mitigating landslide risks. The rainfall threshold model is the most widely used early warning model for predicting rainfall-triggered landslides. Recently, the rainfall threshold model has been coupled with the landslide susceptibility (LS) model to improve the accuracy of early warnings in the spatial domain. Existing coupled models, designed based on a matrix including predefined rainfall thresholds and susceptibility levels, have been used to determine the warning level. These predefined classifications inevitably have subjective rainfall thresholds and susceptibility levels, thus affecting the probability distribution information and eventually influencing the reliability of the produced early warning. In this paper, we propose a novel landslide warning model in which the temporal and spatial probabilities of landslides are coupled without predefining the classified levels. The temporal probability of landslides is obtained from the probability distribution of rainfall intensities that triggered historical landslides. The spatial probability of landslides is then obtained from the susceptibility probability distribution. A case study shows that the proposed probability-coupled model can successfully provide hourly warning results before the occurrence of a landslide. Although all three models successfully predicted the landslide, the probability-coupled model produced a warning zone comprising the fewest grid cells. Quantitatively, the probability-coupled model produced only 39 grid cells in the warning zone, while the rainfall threshold model and the matrix-coupled model produced warning zones including 81 and 49 grid cells, respectively. The proposed model is also applicable to other regions affected by rainfall-induced landslides and is thus expected to be useful for practical landslide risk management.