Ranking GMPEs using macroseismic intensity data of 20th century mining-triggered earthquakes.

Neefs, Ben [UCL] Vanneste, Kris [Royal Observatory of Belgium] Camelbeeck, Thierry [Royal Observatory of Belgium] Van Noten, Koen [Royal Observatory of Belgium]

From the late 19th century until the end of the 20th century, the Hainaut region in Belgium has been subjected to increased seismic activity, suggested to be related to the local coal mining activities taking place over the same time span. Characterized by low to moderate magnitudes (MW 2.8-4.1), shallow focal depths (< 6 km) and a maximum macroseismic intensity of VII (EMS-98), the seismic hazard in the region is often projected as the second highest in north-western Europe. While lacking a dense instrumental network at the time, fortunately, a rich macroseismic dataset is available for the region. A recent macroseismic compilation (Camelbeeck et al., 2022) shows that intensities attenuate much faster with increasing distances in the Hainaut region than elsewhere in Belgium. The current selection of ground motion prediction equations (GMPEs) for seismic hazard assessment (SHA) is not adjusted properly and overestimates the impact of these events. To resolve this discrepancy between SHA and macroseismic intensity maps, we test the applicability of several GMPEs to the macroseismic intensity data of the Hainaut region. Converting the results of GMPEs to intensities using ground motion to intensity conversion equations (GMICE) allows us to evaluate the goodness of fit between GMPEs and the macroseismic intensity data. We selected the GMICE relation of Atkinson & Kaka (2007) for this study as it allows for magnitude and distance correction terms which leads to the most consistent results between PGA and PGV. By applying several metrics proposed to evaluate the goodness of fit between GMPE and observed ground motion like the LH (likelihood) and LHH (log-likelihood) measures (Scherbaum et al., 2004; 2009) and Euclidean-based Distance Ranking (Kale & Akker, 2013), we ranked a selection of 20 recent GMPEs. This resulted in the best overall fit to our data for two GMPEs for induced seismicity (Atkinson, 2015; Douglas et al., 2013), as well as the NGA-East GMPE from Atkinson & Boore (2006). Even though the fit is not yet perfect between these GMPEs and the macroseismic intensity data, it already delivers a 20-30 % reduction in hazard levels compared to the selection of GMPEs used elsewhere in Belgium. To improve seismic hazard computations, we plan to apply the same methodology to the country as a whole.