Towards landslide space-time forecasting through machine learning: the influence of rainfall parameters and model setting

Abstract

Landslide susceptibility assessment using machine learning models is a popular and consolidated approach worldwide. The main constraint of susceptibility maps is that they are not adequate for temporal assessments: they are generated from static predisposing factors, allowing only a spatial prediction of landslides. Recently, some methodologies have been proposed to provide spatiotemporal landslides prediction starting from machine learning algorithms (e.g., combining susceptibility maps with rainfall thresholds), but the attempt to obtain a dynamic landslide probability map directly by applying machine learning models is still in the preliminary phase. This work provides a contribution to fix this gap, combining in a Random Forest (RF) algorithm a static indicator of the spatial probability of landslide occurrence (i.e., a classical susceptibility index) and a number of dynamic variables (i.e., seasonality and the rainfall amount cumulated over different reference periods). The RF implementation used in this work allows the calculation of the Out-of-Bag Error and depicts Partial Dependence Plots, two indices that were used to quantify the variables’ importance and to comprehend if the model outcomes are consistent with the triggering mechanism observed in the case of study (Metropolitan City of Florence, Italy). The goal of this research is not to set up a landslide probability map, but to 1) understand how to populate training and test datasets with observations sampled over space and time, 2) assess which rainfall variables are statistically more relevant for the identification of the time and location of landslides, and 3) test the dynamic application of RF in a forecasting model for the spatiotemporal prediction of landslides. The proposed dynamic methodology shows encouraging results, consistent with the actual knowledge of the physical mechanism of the triggering of shallow landslides (mainly influenced by short and intense rainfalls) and identifies some benchmark configurations that represents a promising starting point for future regional-scale applications of machine learning models to dynamic landslide probability assessment and early warning.

Optimization of rainfall thresholds for landslide early warning through false alarm reduction and a multi-source validation

“…Although the major strength of rainfall thresholds is their simplicity, recently, many authors are attempting to include other parameters to better reproduce cause and effect relationships between rainfall and landslides, and among the most investigated parameters, they are using soil moisture or other proxies to indirectly take it into account (Chen et al 2017;Segoni et al 2018b;Zhao et al 2019;Rosi et al 2021;Kim et al 2021). In fact, it is widely accepted in traditional literature that even short-term, low-intensity rainfall can trigger landslides if it occurs during a wet season, therefore, in an already partially saturated soil (Nocentini et al 2023).…”

Space–time landslide hazard modeling via Ensemble Neural Networks

“…This notion is commonly referred to as landslide susceptibility Titti et al, 2021). As for the low number of publications focused on estimating when or how frequently landslides may occur at a given location, the community has produced a number of near-real-time predictive landslide models for rainfall (Intrieri et al, 2012;Kirschbaum and Stanley, 2018;Ju et al, 2020) and seismic Nowicki Jessee et al, 2018) triggers. With regard to characteristics such as velocity, kinetic energy and runout, albeit fundamental to describing a potential landslide threat (Fell et al, 2008;Corominas et al, 2014), these cannot currently be used for data-driven modeling because no observed dataset of landslide dynamics exists to support the modeling and prediction paradigm based on Artificial Intelligence (AI) or statistical approaches.…”

“…The present work expands on the data-driven literature summarized above by proposing a space-time deep-learning model based on an Ensemble Neural Network (ENN) architecture. Neural Networks (NN) are not new to the landslide literature, although they have found the spotlight so far mostly for automated landslide detection (Catani, 2021;Meena et al, 2022), monitoring (Neaupane and Achet, 2004;Wang et al, 2005), and for landslide susceptibility assessment (Lee et al, 2004;Catani et al, 2005;Gomez and Kavzoglu, 2005;Grelle et al, 2014;Montrasio et al, 2014;Catani et al, 2016;Nocentini et al, 2023). Here, the main difference is that our ENN is built as an ensemble made of two elements, i.e., a landslide susceptibility classifier and a landslide density area regression model, both simultaneously defined over the same spatio-temporal domain.…”

Sentinel-1 SAR-based Globally Distributed Landslide Detection by Deep Neural Networks

“…Both Williams et al (2018) and Amatya et al (2023) highlighted the importance of rapid mapping for the 2015 Gorkha and 2021 Haiti MLE coseismic landslides, emphasizing the use of Synthetic Aperture Radar (SAR) data, preferably in an automated pipeline for emergency evacuations. Moreover, a comprehensive understanding of these slope instability processes begins with a spatial assessment of the slope failures for both rainfall (Nocentini et al, 2023;Segoni et al, 2014) and earthquake-induced landslides (Meena and Tavakkoli Piralilou, 2019). Data concerning the location and timing of failed slopes is usually recorded in products also known as landslide inventories (Van Den Eeckhaut et al, 2013).…”