Data-driven spatialbvalue estimation with applications to California seismicity: Tobor not tob

摘要: In this paper we present a penalized likelihood-based method for spatial estimation of Gutenberg-Richter's b value. Our incorporates nonarbitrary partitioning scheme based on Voronoi tessellation, which allows the optimal space using minimum number free parameters. By random placement an increasing nodes, are able to explore whole solution in terms model complexity. We obtain overall likelihood each by estimating values all regions and calculating its joint Aki's formula. Accounting parameters, then calculate Bayesian Information Criterion realizations. investigate ensemble best performing models demonstrate robustness validity our through extensive synthetic tests. apply seismicity California two different time spans Advanced National Seismic System catalog (1984–2014 2004–2014). The results show that last decade, value variation well-instrumented parts mainland is limited range (0.94 ± 0.04–1.15 ± 0.06). Apart from Geysers region, observed can be explained network-related discrepancies magnitude estimations. suggest previously reported variations obtained classical fixed radius or nearest neighbor methods likely have been overestimated, mainly due subjective parameter choices. envision selection criteria used study useful tool generating improved earthquake forecasting models.