Speaker
Description
This study investigates to what extent the organization of regional seismicity, viewed as a driven and dissipative system, is encoded in the correlations of its activity field, and whether such a description carries predictive value. The analysis is based on the USGS Southern California catalogue (M ≥ 2.5, 1984–2023), discretized into a daily, spatially coarse-grained record of event counts. The diagonal of the resulting spatiotemporal correlation tensor is examined first and displays several near-critical signatures: broad frequency–size and waiting-time distributions, growing fluctuations under temporal coarse-graining, and avalanche-like bursts. These are interpreted as consistency checks rather than as evidence of self-organized criticality. The off-diagonal sector is then used to construct a correlation network, which, when tested against a per-cell time-shuffled null, proves substantially more structured than chance. Its communities, recovered without recourse to fault information, coincide with the Eastern California Shear Zone and other principal active corridors. The same representation further supports short-term forecasting: histogram-based gradient-boosted regression trees, trained under a Poisson likelihood on lagged, rolling, and neighbour-coupling features, predict weekly per-cell rates appreciably above a persistence baseline. Taken together, these results show that the correlations of the activity field encode both tectonic structure and predictive information.