Space–Time Landslide Susceptibility Modeling Based on Data-Driven Methods

Delineating spatiotemporal variations in landslide susceptibility patterns is crucial for landslide prevention and management. In this study, we present a space–time modeling approach to predict the annual landslide susceptibility of the main island of Taiwan from 2004 to 2018. Specifically, we use a Bayesian version of the binomial generalized additive model, assuming that landslide occurrence follows a Bernoulli distribution. We generate 46,074 slope units to partition the island of Taiwan and divide the time domain into 14 annual units. The binary states of landslide presence and absence are classified by a set of static and dynamic covariates. Our modeling strategy features an initial explanatory model to test for goodness of fit and to interpret the effects of covariates. Then, five cross-validation schemes are tested to provide the full range of the predictive capacity of our model. We summarize the performance of each test through receiver operating characteristic curves and their numerical variation over space and time. Overall, our space–time model achieves satisfactory results, with the mean AUC above 0.8. We believe this type of dynamic prediction is a new direction that eventually moves away from the static view provided by traditional susceptibility models. Meanwhile, we believe that such analyses are only stepping stones for further improvements, the most natural of which are statistical simulations of future scenarios.