Spatially Local Pre-Seismic Magnitude Variance Compression: Evidence from the JMA Unified Catalog and Eight Additional Earthquake Catalogs

Abstract Whether detectable precursory changes in seismicity occur before large earthquakes remains one of the most debated questions in earthquake science. Here I introduce a simple, reproducible indicator: the rolling standard deviation of earthquake magnitudes computed within a spatial radius around each target mainshock. Using the JMA unified hypocenter catalog (2000–2023, M≥3.0, 178,577 events in the Japan region), I identify 288 M≥6.0 mainshocks isolated by ±60 days and 200 km from other M≥6.0 events. Within 100 km of each mainshock, the rolling standard deviation of magnitudes decreases by 7–11% at lags of 21, 14, 7, and 3 days before the event (p < 10⁻⁵, Benjamini–Hochberg FDR-significant at 4 of 5 tested lags). The suppression strengthens as the measurement radius shrinks from 300 km (−5%) to 50 km (−11%), consistent with a near-field preparation process. A quartile-based four-state classification reveals that the deeply compressed state (S4) is occupied 34% of the time before mainshocks versus 25% during background periods, and that once the system enters S4 before a mainshock, it persists there with significantly elevated probability (p < 0.00005). After the mainshock, 84% of post-event windows occupy the relaxed state (S1). The same pipeline applied to eight additional catalogs spanning subduction, strike-slip, continental collision, back-arc extension, oblique subduction, global, and intraplate settings reveals significant suppression in the San Andreas (−10.6%, p = 0.007) and Chile (−5.6%, p = 0.027) catalogs and a confirmed null in the intraplate control. These observations are consistent with progressive deformation localization and stress self-organization, and suggest that magnitude variance may provide an independent, complementary channel for earthquake monitoring.