Quantifying a potential bias in probabilistic seismic hazard assessment: Seismotectonic zonation with fractal properties

摘要

The accurate definition and characterization of relevant seismic sources are critical steps in probabilistic seismic hazard assessment (PSHA). This is particularly challenging in low-seismicity regions because observation periods are relatively short, seismicity is often diffuse, and active faults are difficult to identify. In such regions, seismogenic sources are typically represented as areal sources, zones with equal seismic potential. However, observed seismicity is never truly uniformly distributed but clusters at all scales, a reflection of the seismotectonic forces and tectonic structures. In this study, we use a fractal scaling approach to explore a more realistic characterization of the seismicity distribution within each source zone. For a hypothetical square source zone, we compute hazard curves and hazard maps resulting from simulations of uniformly distributed seismicity (fractal dimension D = 2:0), and we compare these with those resulting from simulations of clustered seismicity (1:0 < D < 2:0). We find that the assumption of uniform distribution of events leads systematically to a higher estimate of hazard within the source zone. This overestimation is increasing for lower probability levels. Equally important, the assumed uniform distribution underestimates the uncertainty of the hazard by up to a factor of three. We perform an extensive sensitivity study of PSHA input parameters, illustrating the complexity of the interaction between the specific setup and the resulting impact on the PSHA. We apply the fractal scaling approach to the seismicity of Switzerland and measure the fractal dimension of instrumental seismicity for the past 30 yr (D 1:5); using this value for synthetic catalogs, we build a fractal seismic zonation and hazard model. We find that, in general, the assumption of uniform distribution of events overestimates the mean hazard in Switzerland by 3% to 20%, and the uncertainty distribution estimation is 50% to 100% narrower than in fractal distribution, depending on the location and the probability level of interest.