Spatial Support in Landslide Hazard Predictions Based on Map Overlays

This contribution considers the predictions of mass movements that have yet to take place and the support provided by spatial databases for the prediction. Conventional hazard maps tend to compile the neighbourhoods and dynamic types of the known past landslides, assuming that they satisfactorily represent the locations of future landslides. The latter, however, may also occur elsewhere. Spatial prediction models, instead, isolate the spatial relationships of all distinct mass movements and the presence of elementary or basic map units that accompany those movements. An operational strategy is presented that has the power to assess the degree of support provided by the combination of individual map units in the spatial databases used to compute the predictions. In particular, using validation techniques on the predictions, three aspects of spatial support assessment are considered: (1) How to interpret a prediction in terms of future landslides; (2) How much scarp area should be used to generate a prediction, or similarly, how the scarp area definition affects the analysis; and (3) How to extend a prediction to a neighbouring area. The strategy is based on the spatial validation of the prediction maps using a variety of temporal or spatial subsets of the distribution of the past landslide trigger areas. Two different application examples on the same spatial database are used to document a strategy for temporal and spatial predictions that can be adopted also in situations or models of more general nature than landslide hazard mapping