Earth rotation slowing gave no forecast gain for global M≥7 earthquake counts in 2017-2025, and the 1900-2016 link was weak and absent at 10°N-30°S

Question: does a slowing of the Earth's rotation (length-of-day increase) come 5-6 years before more M≥7 earthquakes, as Bilham and Bendick said in 2017? They forecast 25-30 % more M≥7 earthquakes from 2017 for at least five years, with the strongest effect at 10°N-30°S. Approach: we used the IERS EOP C02 length-of-day series (1830-2026) and USGS ComCat M≥7 counts (1900-2025, 1,607 events), with ISC-GEM, M≥7.1 and Gardner-Knopoff declustered counts as checks. We used phase-randomised length-of-day surrogates as the null model, Poisson forecasts fitted on 1900-2016 for 2017-2025, and rolling one-year-ahead forecasts for 1950-2025. Results: 2017-2022 had 12.2 M≥7 earthquakes per year against a 1900-2016 mean of 12.7 (ratio 0.96, 95 % interval 0.70-1.25); 2017-2025 gave 1.03. All 24 models fitted on 1900-2016 forecast 2017-2025 worse than a constant rate. In 1900-2016 the best correlation at a 4-7 year lead was r = 0.22 (surrogate p = 0.07; p = 0.03-0.22 over 48 variants). At 10°N-30°S there was no relation (r ≤ 0.11, p ≥ 0.33). A post hoc relation north of 10°N (r up to 0.64) also failed out of sample. Rolling forecasts gained 0.20 nats over 76 years (surrogate p = 0.14). The fast rotation of 2016-2021 predicted fewer earthquakes in 2021-2025, but the mean was 15.0 per year. Limitations: only about four length-of-day cycles and nine new years, so an effect near 15 % cannot be excluded; we did not repeat the topological renewal-interval analysis.

length of day · Earth rotation · global seismicity · M7 earthquakes · earthquake forecasting · out-of-sample test · surrogate data · ComCat · ISC-GEM · IERS EOP C02 · replication