Retrospective short-term forecasting experiment in Italy based on the occurrence of strong (fore) shocks

In a recent work, we computed the relative frequencies with which strong shocks (4.0 ≤ Mw < 5.0), widely felt by the population were followed in the same area by po- tentially destructive main shocks (Mw ≥ 5.0) in Italy. Assuming the stationarity of the seismic release properties, such frequencies can be tentatively used to estimate the probabilities of potentially destructive shocks after the occurrence of future strong shocks. This allows us to set up an alarm-based forecasting hypothesis related to strong foreshocks occurrence. Such hypothesis is tested retrospectively on the data of a homogenized seismic catalogue of the Italian area against a purely random hypothesis that simply forecasts the target main shocks proportionally to the space–time fraction occupied by the alarms. We compute the latter frac- tion in two ways (i) as the ratio between the average time covered by the alarms in each area and the total duration of the forecasting experiment (60 yr) and (ii) as the same ratio but weighted by the past frequency of occurrence of earthquakes in each area. In both cases the overall retrospective performance of our forecasting algorithm is definitely better than the random case. Considering an alarm duration of three months, the algorithm retrospectively forecasts more than 70 per cent of all shocks with Mw ≥ 5.5 occurred in Italy from 1960 to 2019 with a total space–time fraction covered by the alarms of the order of 2 per cent. Considering the same space–time coverage, the algorithm is also able to retrospectively forecasts more than 40 per cent of the first main shocks with Mw ≥ 5.5 of the seismic sequences occurred in the same time interval. Given the good reliability of our results, the forecasting algorithm is set and ready to be tested also prospectively, in parallel to other ongoing procedures operating on the Italian territory.