Estimating the quality of landslide susceptibility models.

We present a landslide susceptibility model for the Collazzone area, central Italy, and we propose a framework for evaluating the model reliability and prediction skill. The landslide susceptibility model was obtained through discriminant analysis of 46 thematic environmental variables and using the presence of shallow landslides obtained from a multi-temporal inventory map as the dependent variable for statistical analysis. By comparing the number of correctly and incorrectly classified mapping units, it is established that the model classifies 77.0% of 894 mapping units correctly. Model fitting performance is investigated by comparing the proportion of the study area in each probability class with the corresponding proportion of landslide area. We then prepare an ensemble of 350 landslide susceptibility models using the same landslide and thematic information but different numbers of mapping units. This ensemble is exploited to investigate the model reliability, including the role of the thematic variables used to construct the model, and the model sensitivity to changes in the input data. By studying the variation of the model's susceptibility estimate, the error associated with the susceptibility assessment for each mapping unit is determined. This result is shown on a map that complements the landslide susceptibility map. Prediction skill of the susceptibility model is then estimated by comparing the forecast with two recent event inventory maps. The susceptibility model is found capable of predicting the newly triggered landslides. A general framework for testing a susceptibility model is proposed, including a scheme for ranking the quality of the susceptibility assessment.