A data-driven method for the estimation of shallow landslide runout

Rainfall-induced shallow landslides cause damages and casualties. Estimating the development of the runout is essential, however, the methods which are traditionally employed to predict the runout distance are either reliable but complex and applicable at a local scale or applicable at a larger scale but highly simplified. The present work aims to develop a method based on data-driven algorithms for predicting the runout of shallow landslides at a large scale, taking into account geological, geomorphological, and land use heterogeneities. The model, which was tested in the Oltrepò Pavese (Italy), a hilly area of over 1000 km2, requires as inputs a set of predictors collected from geological maps, land use maps and freely available satellite images. Different algorithms were tested, identifying the Random Forest algorithm as the best performing, with a Coefficient of Determination of 0.94 and Mean Absolute Error of 4–5.8 m. The size of the source area strongly influences runout estimation, as do land use, lithology, and slope angle. The model provides a probable runout length, which can be estimated, based on past observations in the test area, to propagate along the line of the greatest slope. The main novelties of this work include: a) the development of a methodology to study the previously overlooked runout dynamics, b) the exploitation of remotely-derived, freely available input data, c) the application at a large scale in a heterogenous area, d) the adaptability of the model different study areas and e) the dependency of the model on land use, which allows for land use change scenarios to be made. If coupled with a susceptibility assessment tool to identify where a landslide might develop, it could be used to give a fast yet accurate assessment of the probable runout length and to identify which targets could potentially be affected by the landslide.