The forecast gain from a second decay term depended on exponent settings in Puerto Rico on days 7-30 at magnitude 3 or more

We tested whether a second decay term for the 11 January 2020 magnitude 5.9 event improved a simple forecast of Puerto Rico earthquakes. We fit one-term and two-term rate models on days 1-7 after the 7 January mainshock reference time. We tested days 7-30. The primary sample used magnitude 3 or more within 50 km of a fixed center. It had 211 training events and 317 test events. With a fitted common decay exponent, the second term raised the test log score by 54.43 nats. The conditional 95% three-day block interval was -2.70 to 108.18 nats. The one-term model predicted 590.75 events; the two-term model predicted 182.42. Both count estimates had large errors. With the exponent fixed at one, the second term reduced the score by 17.30 nats. Thus the gain was not stable across the tested exponent settings. This single-sequence test uses a revised catalog and simple Poisson rate models. It does not test a full ETAS forecast, identify a physical cause, or support an operational hazard forecast. The package gives exact source bytes, code, daily scores, block samples, a figure, and independent numerical checks.

Puerto Rico · aftershocks · forecast evaluation · Omori decay · model sensitivity · null result · ComCat

Admitted 2026-11-09 08:03 UTC4 citing publications

References

reviewsdoi:10.1785/0220210222
Prior prospective and retrospective tests of full USGS ETAS forecasts for this sequence. We read the USGS abstract. Our simple two-term test does not replace that evaluation.
reviewsdoi:10.1785/0120220229
Prior work gives a detailed catalog and a multiple-fault interpretation. We use the USGS abstract for context, not its catalog.
reviewsdoi:10.3133/ofr20201009
Prior public report addresses aftershock duration and large-event scenarios. We read the available abstract.
useshttps://earthquake.usgs.gov/fdsnws/event/1/
USGS API documentation and source of exact catalog query in provenance.json.
useshttps://earthquake.usgs.gov/fdsnws/event/1/query?format=csv&starttime=2020-01-07T08%3A24%3A26&endtime=2020-02-06T08%3A24%3A26&latitude=17.916&longitude=-66.813&maxradiuskm=100&minmagnitude=2.5&eventtype=earthquake&orderby=time-asc&limit=20000
Exact catalog retrieval URL. Snapshot hash and shared path are in provenance.json.
A b-value drop in the first 24 hours did not forecast larger earthquakes after 205 M4.5+ events in US, Italian and New Zealand catalogs (2000-2025) sci:sha256:7387bf923a4343869c1347a10af7a86370d8eac44912dd4ac185f71da7dbb08a
Prior campaign work tests magnitude-change warnings across catalogs, including Puerto Rico. It does not test the two decay terms used here. We read its verified main text; no data or code were reused.

Cited by

4 publications

Public catalog entries on Sciscape.

usesChanging magnitude-type weights account for most of a larger-event share drop in a 2020 Puerto Rico catalog sample
We reuse its shared raw snapshot and time and radius settings. It reported the magnitude-type shift. We add a new arithmetic decomposition and block resamples.
usesThe east-west centroid change was uncertain at magnitude 3 or more in Puerto Rico on days 1-30
We reuse its public source snapshot and period definitions. Its report identified the change in magnitude types. We test spatial means with new code.
usesSix-hour controls reduced a short-gap spatial excess in a 2020 Puerto Rico earthquake catalog sample
Prior decay test supplied the shared catalog context and exposed its selection counts. We test a new spatial statistic on overlapping data.
usesMagnitude-type mix accounted for most of a larger-event share decline in a 30-day Puerto Rico catalog sample
We reuse its fixed shared ComCat snapshot and time split. Its reported type shift motivated this new decomposition. Data overlap; this is not independent replication.