Accounting for satellite rainfall uncertainty in rainfall-triggered landslide forecasting

With an increasing spatial/temporal resolution and accuracy of satellite rainfall products, their potential in landslide forecasting has gradually been recognized. Despite this, few studies have evaluated the effect of satellite rainfall uncertainty on landslide forecasting. In this study, a specific procedure is developed to account for the satellite rainfall uncertainty in rainfall-triggered landslide forecasting. We select the Emilia-Romagna region in northern Italy as the study area, and the NASA GPM-based IMERG Early Run product as the representative of satellite rainfall estimates. An error model for satellite rainfall is first constructed based on the distribution of GR|SR (gauge rainfall | satellite rainfall), which consists of a deterministic distortion component and a stochastic component. With the error model, the satellite rainfall uncertainty is characterized by generating the ensemble of probable “true rainfall” for each rainfall condition responsible for landslides. Then the generated rainfall ensembles are applied to the definition of rainfall thresholds using the bootstrap technique. Finally, the landslide forecasting performance of rainfall thresholds is evaluated with the criteria of hit rate, false alarm rate and the Euclidean distance to the perfect point. Results show that the simulated uncertainty band of the IMERG product encompasses most of the rain gauge measurements. The IMERG uncertainty also propagates to the rainfall thresholds, especially when rainfall events have long durations, the cumulated event rainfall that is likely to trigger landslides has a larger variation range. Besides, the landslide forecasting performance for thresholds that account for IMERG uncertainty varies a lot in terms of false alarm rate from 0.50 to 0.75, demonstrating that it is necessary to take into account the uncertainty associated with the IMERG Early Run product when using it to construct rainfall thresholds for landslide forecasting.