A predictive model of societal landslide risk in Italy

We propose a novel approach to evaluate the spatial and the temporal distribution of societal landslide risk from historical, sparse, point information on fatal landslides and their direct human consequences. We test the approach using a record of 5571 fatalities caused by 1017 landslides at 958 sites across Italy, in the 155-year period 1861-2015. Adopting a Zipf distribution, we model societal landslide risk for the whole of Italy, and for seven physiographic and 20 administrative subdivisions of Italy. Results confirm that the Zipf distribution is adequate to describe the frequency (and the probability) of fatal landslides, and show that societal landslide risk varies in Italy depending on the largest magnitude landslide F, the number of fatal events E, and the scaling exponent of the Zipf distribution s, which controls the relative proportion of low vs. large magnitude landslides. To model societal landslide risk, we then test different grid spacings, g and circular kernel sizes, r finally adopting g = 10 km and r = 55 km. Using such geometrical constraints, we prepare maps of the variables F, E and s, revealing the complexity of landslide risk in Italy, which cannot be described properly with a single metric. For each grid cell, we assign the {F, E, s} variables to the red, green and blue bands of a composite linage to obtain a single view of landslide risk to the population of Italy. Next, we prepare risk scenarios for landslides of increasing magnitudes, which we validate checking the anticipated return period of the fatal events against information on 130 fatal landslides between 1000 and 1860, and eleven fatal landslides between January 2016 and August 2018. Despite incompleteness in the old part of the record for the low magnitude landslides, and the short length and limited number of events in the recent period 2016-2018, the anticipated return periods are in good agreement with the occurrence of fatal landslides in both validation periods. Despite the known difficulty in modelling sparse datasets, the approach provided a coherent and realistic representation of societal landslide risk in Italy. Our results give new insight on the spatial and temporal variations of societal landslide risk in Italy. We expect this to contribute to improve existing zonings of landslide risk in Italy; to foster the efficacy of national and regional landslide early warning systems; and to design and implement better landslide communication, mitigation and adaptation strategies. Our approach is general and not constrained to the information on fatal landslides available for Italy. We therefore expect the approach to be used to model societal landslide risk in other geographical areas for which adequate information is available, and to model the fatal consequences of other hazards.