Search

A common and largely unresolved problem of national-scale landslide early warning systems is their independent evaluation. In a recent paper, Mondini et al. (Nat Commun 14:2466, 2023) proposed a deep-learning system for short-term forecasting of rain-induced shallow landslides in Italy. Here, we independently evaluate the performance of this national-scale system by demonstrating its application between 1 January and 31 May 2021. For the purpose, we use hourly rainfall measurements from the same rain gauge network and different and independent information on the timing and location of 163 rain-induced landslides obtained from the FraneItalia catalogue that occurred in Italy in a period non considered in the construction of the system (https://zenodo.org/records/7923683). Independent demonstration confirmed the good predictive performance of the forecasting system and revealed no geographical or temporal bias in the forecasts. The analysis also showed that the system was more effective at predicting multiple landslides in the same general area than single landslides. This was a good result as multiple landslides are inherently more dangerous than single failures. Analysis of the few misclassified landslides showed that approximately one-third of the landslides were rockfalls, and for approximately another third there was uncertainty about when or where the landslides occurred. We conclude that, despite the inevitable misclassifications inherent in any probabilistically based national-scale landslide forecasting system, the deep-learning system analysed is well suited for short-term operational forecasting of rain-induced shallow landslides in Italy. [ABSTRACT FROM AUTHOR]

Based on a minimum amount of rainfall that when reached or exceeded can trigger landslides, rainfall thresholds are used to predict potential landslide occurrence and are essential parts of many landslide early warning systems. Despite the extensive literature on the definition and use of rainfall thresholds, little attention has been given to examining and comparing the mathematical methods that can be used to define thresholds as lower bounds of clouds of empirical rainfall conditions known to have triggered landslides. When multiple thresholds are available, it is unclear how to combine them. Here, we address both issues. We test and compare four mathematical methods to define event cumulated rainfall—rainfall duration, ED thresholds using 2259 measurements of rainfall duration (D, in hours) and cumulated rainfall (E, in mm) that resulted in mostly shallow landslides in Italy between January 2002 and December 2012. The methods cover a broad spectrum of data driven approaches, including a frequentist least square method, a frequentist quantile regression method, a Bayesian quantile regression method, and a machine-learning symbolic regression method. We apply and compare the methods for three non-exceedance probability levels, p = 0.01, 0.05, 0.10, and we propose a voting strategy to combine the predictions into a single, dichotomous—i.e. ‘sharp’—non-probabilistic landslide prediction that we apply to the available dataset of rainfall measurements. [ABSTRACT FROM AUTHOR]