Automated b-value traffic lights classified 9 of 65 Japanese M5+ foreshock sequences before the larger event, and over 80% of red alarms were false (JMA, 2000-2024)

Question: Can the b-value foreshock traffic light (FTLS; Gulia and Wiemer 2019) tell in real time that a large earthquake is a foreshock? Approach: We applied the rule automatically and causally to all 1,459 shallow (depth 50 km or less) M5.0+ first events in the JMA unified catalog, 2000-2024 (240 of M6.0+). A first event is the largest event near it in the previous year. The post-event b-value used only events before a decision time of 0.25, 1, 3 or 10 days. The outcome was a larger event within one rupture length in the next 30 days. We varied 72 method settings and used three estimators (FTLS-style Aki-Utsu, common-Mc Aki-Utsu, b-positive). Findings: 65 first events had a larger event within 30 days, and half of these came within 8 hours. Only 8% of first events had enough events for a b-value at 1 day. The FTLS gave a valid class before the larger event in 9 of 65 cases (5 red, 2 orange, 2 green). At the primary setting (1 day), 4 of 35 red alarms (11%) and 3 of 72 non-red classes (4%) were followed by a larger event (Fisher p = 0.16). Red alarms occurred in 31% of sequences with no larger event. Kumamoto 2016 and the 2011 Tohoku foreshock were red, but Noto 2024 M7.6 was also red under b-positive. Over the settings, the signal at 1 day was positive but often not significant, and it was absent at 3 days. A simple magnitude gap (first event minus largest aftershock so far) ranked sequences better (AUC 0.83 versus 0.68). Limitations: few larger events (2-7 per test), a circular source volume instead of a fault volume, early incompleteness, and one catalog.

foreshock · b-value · foreshock traffic light · earthquake forecasting · JMA catalog · Japan · b-positive · Kumamoto · Tohoku · pseudo-prospective test · seismicity