An automatic b-value foreshock traffic light did not identify shallow Japanese M5+ earthquakes followed by larger events in 2002-2024; most larger events came before a decision

Question: does the Gulia and Wiemer (2019) foreshock traffic light (a 10% drop of the aftershock b-value below the local background b-value means that a larger earthquake is still to come) identify the earthquakes that are followed by an equal-or-larger event? Approach: we applied one fixed, automatic version of the rule, written down before the first run, to all shallow (depth <= 40 km) Mj>=6 (221) and Mj>=5 (910) triggers in the JMA hypocentre catalogue for 2002-2024. The outcome was an equal-or-larger event within 50 km in the 30 days after the decision time. We ran 46 real-time robustness variants and 16 retrospective variants, with Aki-Utsu and b-positive estimators. Results: only 24 M>=6 (11%) and 43 M>=5 (4.7%) triggers had enough data, mainly inland and near the coast. Of all M>=5 triggers, 13% had an equal-or-larger event within 50 km and 30 days. 46% of these came within 0.5 days and 55% within 1.5 days, before the primary decision time. In the primary real-time test, 0 of 23 (M>=6) and 0 of 38 (M>=5) scored sequences were followed, so all 4 and 8 red lights were false alarms. The Gulia-Wiemer style success rate was 0.69 and 0.67, below the 1.00 of a rule that always gives green. No real-time variant had a significant red-light excess (smallest one-sided Fisher p 0.16). In the retrospective test, the b window stopped just before the larger event. With 37-122 scored and 2-12 followed sequences, the AUC of the b-value drop was 0.29-0.73 and every 95% interval included 0.5. Simple features without b-values (magnitude gap, early event count) had AUC point estimates of 0.52-0.74. Kumamoto 2016 was red, orange or too late, depending on the blind time and the window. Limitations: the number of followed testable sequences is small, so a moderate real effect is not excluded. We used circular source volumes and fixed blind times instead of expert-chosen fault boxes. The data are from one catalogue and one region.

foreshock traffic light · b-value · b-positive · foreshocks · earthquake forecasting · JMA catalogue · Japan · aftershocks · Gutenberg-Richter · forecast verification