This benchmark approaches earthquake forecasting as a probabilistic and reproducible workflow rather than an exercise in producing single predictions.
Using the JMA catalog, it connects catalog-quality decisions to ETAS modelling, then evaluates forecast behaviour through calibration and reliability-focused validation.
02Method
Reliability starts before the model.
01
Completeness
Assess whether the catalog offers a consistent enough observational basis for modelling.
02
Declustering
Handle event dependence as part of preparing the catalog for analysis.
03
ETAS modelling
Use a framework designed to represent earthquake occurrence and triggering behaviour.
04
Calibration
Check whether predicted probabilities match observed frequencies, and report that honestly.
03Pipeline
Trace every transformation.
Load the JMA catalog
Estimate completeness
Apply declustering
Fit the ETAS model
Validate reliability
Evaluation note: the validation focus is documented, but numerical benchmark results were not published for this write-up, so no model scores are stated.
04Takeaway
Confidence has to be earned.
Reproducibility and calibration matter most exactly where uncertainty is highest — which is the entire point of a probabilistic benchmark.