Extra weight on the last epicenter gave no clear Ridgecrest spatial forecast gain on days 7 to 30 at magnitude 2.5 or more

We tested whether extra weight on the most recent epicenter improves the predicted location of the next selected Ridgecrest event. A fixed USGS snapshot has 931 events with preferred magnitude at least 2.5 on days 1 to 30 after the 2019 M7.1 earthquake. The baseline uses Gaussian kernels on all earlier selected locations with a one-day decay scale. Both models predict probabilities for fixed 5 km cells. We chose the blend weight on 373 targets on days 2 to 7, then fixed it at 0.10. On 387 targets on days 7 to 30, mean log score gain was 0.00262 nats per event. Its conditional 95% three-day block interval was -0.02920 to +0.02137. All planned model checks also had intervals that included zero. The result leaves added value uncertain under this simple model. It does not show that locations are independent or that spatial forecasts lack value. One selected sequence, revised data, model form, grid choice, and 23 test days limit inference. We do not forecast event time, magnitude, or hazard. The package gives the source snapshot, hashes, event forecasts, all block draws, code, a figure, an independent audit, and repeat instructions.

Ridgecrest · spatial forecast · kernel smoothing · chronological validation · aftershocks · null result · block bootstrap