The SLIP model: A major step towards the application in real time civil protection integrated platforms for landslide prevention.

Abstract

This work has focused on validating, on a large scale, the physically based model, named SLIP (Shallow Landslide Instability Prediction) for its future application in real time civil protection integrated platforms, after almost 15 years from its first formulation. Many works have been carried out by our research group using this model to back analyze occurred events and with the correct calibration of its input data the model always gave good results. In this thesis a further validation on various aspects of the model have been carried out from the prediction of instability of simulated landslides in laboratory flume tests to a real scale analysis. Particularly SLIP has been used to model two case studies, namely the landslide event that hit the Parma Apennines in April 2013 and the event of Giampilieri (ME) occurred the 1st October of 2009. In the first study case a large gathering of information was carried out from both in situ measurements and laboratory tests on the landslides and its soil. Thanks to an already known background for this type of soil, studied in previous works of our research group, the modeling gave good predictive capability. A new technique that extracted spatial land cover classes from pre-event flight images was consequently used. There was a clear improvement in the overall accuracy of the model between the cases in which this differentiation was used and not showing how a better spatial variation of parameters can improve the model predictive capacity. From a temporal stand point both the parameter sets give excellent results remarking the instability pattern that witnesses and local news provided. The results highlight a good prediction although there is a high over prediction ratio in the first set, and some false alerts in the second set. These problems can be related to an incomplete landslide database and spatial errors due to the absence of post event images. The calibration of the input parameters of the Giampilieri event was made by laboratory geotechnical characterization, numerical models for hydraulic parameters and by simulating in a small scale flume the triggering of landslides. Flume tests can be used for multiple purposes, such as to evaluate in back-analysis the initial soil conditions of a reference landslide event, but also to define several input parameters of SLIP model, as well as to analyze in detail the triggering mechanisms of the material potentially susceptible to shallow landslides. Furthermore, the apparatus used in this study is not complex or expensive. With the correct expedients, such as insertion of macro channels, the flume can be used to simulate the real case and to model hypothetical scenarios before they occur in real slopes. The outcome of flume tests underlines the influence of initial soil conditions on times and modalities of slope failure, as well as indicating how a variable rainfall input produces an increase of water infiltration compared to a constant one of same cumulative depth. The output of two models, SLIP and TRIGRS, a well established model, are shown. The results indicate that the models reconstruct quite well the event, both in terms of temporal evolution and spatial distribution of slope instability, and identify substantially the same areas mostly affected by shallow landslides. The comparison confirms the good predictive capability of the SLIP physically-based model, considering that the two maps converge to the same solution in large part of the study area, although SLIP overestimates spatially the instability while TRIGRS underestimates the landslide occurrence and slightly overestimates temporally the duration of the landslide event. SLIP is a more simple model than TRIGRS requiring less input parameters. The results of a two-year daily analysis are shown only using the SLIP model because a yearly analysis with TRIGRS would require elevated computational time. The results show haw the model well predicts instability capturing both the reported events of this time span and producing only one false alert. SLIP model returns the results in a few minutes for large areas. This means that updated triggering scenario maps can be obtained substantially in real-time. This feature is obviously essential considering a possible integration of the approach with an early warning system. Furthermore, if SLIP model was used in this way, it would operate with rainfall inputs forecasted for the next hours, then much more reliable than those estimated with a statistical analysis of historical rainfall data which, furthermore, are not always available for a specific area. Overall, if coupled with forecasted rainfall maps the model could be used as a preliminary early warning system for landslides and could be used to simulate landslide susceptibility over large areas with different rainfall scenarios.