Searcharxiv⌕ Search

arXiv subjects

Ricardo Mansilla Sánchez

Publications and source records attributed to Ricardo Mansilla Sánchez.

1 recordsLinked to original sources

Pooling ETAS triggering parameters across subduction margins: forecast transfer, estimator stability and sequence-limited skill in Mexico

Pooling earthquake-triggering parameters can reduce estimation variability on margins with few recorded sequences, but its forecast value must be distinguished from the stability of simply freezing a fitted law. We fit the ETAS model to the Mexican national catalog above magnitude 5.0, using the Slab2 coverage mask as the domain. On a 4.7-year holdout the model gains 1.45 bits per earthquake over a time-independent spatial Poisson reference (cluster-resampled interval [-0.03, +2.75] bits), with two clusters carrying 90% of the signed gain. Using the Global CMT catalog we fit eight disjoint subduction margins and a joint model on 3,210 earthquakes sharing six triggering parameters with a background rate per margin; BIC favors the joint model by 184 units, though a parametric bootstrap rejects its exact shared specification. Transferred into Mexico with the local background fixed, six of seven foreign triggering sets score within 0.07 nats per event of the home fit. In leave-one-margin-out evaluation, the excluded-pool law's event-weighted paired gain is +0.0334 [+0.0002, +0.0670] nats per event; its lower limit sits barely above zero and crosses zero under alternative cluster-linking rules, so we read it as neither conclusive superiority nor guaranteed noninferiority. Across 28 overlapping annual evaluations, local refitting minus a pooled law frozen before the first origin has mean gain -2.07 nats/year (moving-block interval [-6.92, +1.15]), while local productivity varies by a factor of 5.4. Coverage of these intervals is unmeasured, and a measured analogue fell short of nominal; the paired median contrast's significance depends on a Gutenberg-Richter slope that is conventional rather than measured. These results support pooling as a regularized alternative to local fitting, short of equivalence, a universal law, or an optimal update policy.

physics.geo-ph↗