Pseudo-prospective earthquake forecasting experiment in Italy based on temporal variation of the b-value of the Gutenberg–Richter law

We present the BVAL method, designed to forecast potentially damaging earthquakes (Mw ≥ 5.0) in Italy based on temporal variations of the b-value of the Gutenberg–Richter frequency–magnitude distribution. The b-value is used as an indicator of stress within the Earth's crust, with lower b-values associated with higher stress levels and an increased likelihood of significant seismic events. This method issues alarms when the b-value falls below a critical threshold. It is optimized using the HOmogenized instRUmental Seismic catalogue data from 1990 to 2004 and validated pseudo-prospectively using data from 2005 to 2022. Our analysis uses the recently developed b-positive (b+) method to compute the b-value from magnitude differences, providing resilience against data incompleteness. We compare the performance of the BVAL method with two established models: the Epidemic Type Aftershock Sequence (ETAS) model, which forecasts earthquake rates based on the epidemic principle that each shock triggers subsequent shocks, and the FORE model, which relies on the occurrence of strong foreshocks. Additionally, we evaluate two ensemble models that combine BVAL and FORE through additive (EADD) and multiplicative (EMUL) strategies to balance false alarms and missed events. The EADD model declares an alarm when either BVAL or FORE signals it, while the EMUL model triggers alarms only when both methods agree. We assess the predictive efficiency of these models using the area skill score, derived from Molchan diagrams, which plot the miss rate against the fraction of space-time occupied by alarms. Our results demonstrate that BVAL is less effective than FORE and ETAS at high space-time fractions, but it outperforms ETAS at low fractions (τ < 2–4 per cent), indicating its potential utility in scenarios where minimizing false alarms is critical. This comprehensive comparison highlights the strengths and limitations of each method, suggesting that integrating multiple forecasting strategies can enhance the reliability of earthquake preparedness and response efforts in Italy.