A susceptibility oriented approach for regional landslide inventory implementation in Sicily

Landslide inventories are mandatory both for picturing the current slope instability status and enabling effective calibration of predictive susceptibility/hazard maps. However, in regions characterized by very high landslide susceptibility conditions, frequently, the inventories are biased as they almost exclusively include landslides which have been noticed and reported for having produced damages, resulting vulnerability/risk oriented. Strategies for landslide hazard management have to be set considering incomplete inventories, so to capturing/representing the huge amounts of phenomena affecting very different geomorphological conditions.. To cope with this limit, a strategy has been adopted in the framework of the SUFRA project, which aims at systematically assessing regional landslide susceptibility conditions in Sicily. In particular, exploiting the currently available incomplete landslide inventory (PAI), susceptibility models are first prepared and susceptibility maps obtained (one for each of recognized typologies). On the basis of these base maps (SUFRA1), for each of the susceptibility classes 30% of area is randomly extracted and submitted to systematic landslide (field/remote) recognition, leading to an increasing of the number of mapped landslides coherent with the spatial distribution of the susceptibility conditions. The Platani river basin extends in central-southern Sicily for approximately 1780 km2, with a geomorphological setting marked by tectonic contacts between brittle (limestones and quartz arenites) and ductile (clays and silty clays) lithologic complexes, in the head sector and smoothed to hummocky long slopes where clays outcropping prevails. Exploiting the PAI inventory, which reported around 600 among earth-flows and earth-slides, two base susceptibility models were prepared for flows and slides. Extracting the 30% of the area of each susceptible classes, systematic landslide recognition allowed to considerably increase the number of mapped landslides (14277 flows and 517 slides) and to prepare two new advanced susceptibility models. To attest the representativeness of the new inventories, base and advanced models were validated with respect to randomly extracted hidden subsets (of landslides. A marked increase both in terms of accuracy, specificity and sensitivity was observed attesting the new inventory together with an obvious magnitude increase allowed to more effectively calibrate the predictive models. Preparing systematic regional landslide inventories configures time costs/money which are not compatible (about twenty-five times the number of mapped events in only the 30% of the test area) with any land planning policy. However, a key point is the suitability of the available inventories for preparing landslide susceptibility models and derived maps. According to the study case, the proposed approach can represent a cost-effective procedure to obtain reliable predictive maps from enriched (still uncomplete) landslide inventories.