Improving ShakeMap performance by integrating real with synthetic data: Tests on the 2009 Mw=6.3 L'Aquila earthquake (Italy)

The ShakeMap software automatically generates maps of the peak ground motion parameters (shakemaps) and of instrumental intensity soon after an earthquake. Recorded data are fundamental to obtaining accurate results. In case observations are not available, ShakeMap relies on ground motion predictive equations, but due to unmodelled site conditions or finite fault effects, large uncertainties may appear, mainly in the near-source area where damage is relevant. In this paper, we aim to account for source effects in ShakeMap by computing synthetics to be used for integrating observations and ground motion predictive equations when near-source data are not available. To be effective, the computation of synthetics, as well as of the finite fault, should be done in near real time. Therefore, we computed rapid synthetic seismograms, by a stochastic approach, including the main fault features that were obtained through inversion of regional and teleseismic data. The rapidity of calculation is linked to a number of assumptions, and simplifications that need testing before the procedure can run in automatic mode. To assess the performance of our procedure, we performed a retrospective validation analysis considered as case study of the M w = 6. 3 earthquake, which occurred in central Italy on April 6, 2009. In that case, the first shakemaps, generated a few minutes after the earthquake, suffered large uncertainties on ground motion estimates in an area closer to the epicenter due to the lack of near-field data. To verify our approach, we recomputed shakemaps for the L'Aquila earthquake, integrating data available soon after the earthquake at different elapse times with synthetics, and we compared our shaking map with the final shakemap, obtained when all the data were available. Our analysis evidences that (1) when near-source data are missing, the integration of real data with synthetics reduces discrepancies between computed and actual ground shaking maps, mainly in the near-field zone where the damage is relevant and (2) the approach that we adopted is promising in trying to reduce such discrepancies and could be easily implemented in ShakeMap, but some a priori calibration is necessary before running in an automatic mode