ABSTRACT

The subjectivity in the artificial selection of calculation rules for seismicity parameters is an important factor that constrains the scientific results of calculations and their application in other fields. In this article, we apply the idea of a data‐driven approach to seismicity parameter calculation and propose a parameter‐free method to estimate the spatial distribution of parameters for the Ogata–Katsura 1993 (OK1993) model. This model is currently widely used and expressed in continuous function form. To verify the feasibility of this data‐driven approach and to explore the rules of model selection, we generate a synthetic catalog according to three sets of different parameters for the OK1993 model and perform large‐scale calculations and method testing. In the calculation, we use a total of 5,049,000 randomly generated models with various numbers of Voronoi nodes (ranging from 2 to 100) and cell meshes generated from 1000 random throws for each number of nodes. Next, we construct a penalized function to select the optimal model based on the Bayesian information criterion (BIC). The results of the median (Q2) and median absolute deviation for the β, μ, and σ values in the OK1993 model show that the initial input parameters can be well recovered for a synthetic catalog. Additionally, we see that our calculation is optimized with an effective node number Nv=20100 and the best 1000 BIC values. We apply this data‐driven approach and the corresponding model selection rules to an actual earthquake catalog in the Shimian–Mianning–Xichang region in southwest China. The results show that the Gutenberg–Richter magnitude–frequency b‐values, converted from the moment‐frequency β values, are quite different from results obtained by previous studies using traditional methods. The low β values in the northern segment of the Anninghe fault and in the Daliangshan fault may indicate a higher stress accumulation in these regions. This new data‐driven and nonparametric calculation method for the OK1993 model can provide estimations of seismicity parameters that are potentially useful in many fields, including earthquake engineering, seismic hazard assessment, and geodynamics.

You do not currently have access to this article.