Spatial forecast gains from a preceding-gap rule depend on model settings in two aftershock catalogs

We tested whether the gap between the last two recorded events helps predict the area of the next event. We used fixed USGS Ridgecrest and Monte Cristo catalogs with preferred magnitude at least 2.5 on days 1 to 30. We chose model weights on days 2 to 7 and scored later targets on days 7 to 30. A model with separate last-epicenter weights for preceding gaps of at most 10 minutes and longer gaps was compared with a model with one common weight. The mean log score gains were 0.00569 nats per event for 387 Ridgecrest targets and 0.01071 for 305 Monte Cristo targets. Conditional 95% three-day block intervals were -0.00392 to 0.01577 and 0.00184 to 0.02302. At Monte Cristo the mean gain changed to -0.00254 with a 10 km kernel and -0.00563 at magnitude at least 3.0. The evidence does not give a stable forecast gain across the tested settings. Revised catalogs, mixed magnitude scales, two selected sequences, and conditional block intervals limit inference. The package contains fixed inputs, code, event forecasts, checks, and repeat instructions.

aftershocks · spatial forecast · chronological validation · preceding gap · model sensitivity · null result