Seismic risk assessment of critical infrastructure requires a detailed site‐specific investigation of local geological conditions. However, site‐specific investigations at millions of spatially distributed assets, such as the Kanto region in Japan or the European Seismic Risk Model 2020 (ESRM20) are impractical. A more practical approach for such regional‐scale risk assessments has been to use regional site‐response maps. Such maps rely on site‐specific ground‐motion observations at a few well‐characterized sites in the region, whose empirical site‐responses are correlated to widely available “proxy” parameters, e.g., topographic slope and geological era in ESRM20, and then used to predict site‐responses at every location in the region. In some cases, geostatistical models (e.g., Kriging) are also used in combination with proxy parameters. Although such proxy‐based regional site‐response maps are quite practical, there are at least three issues: 1) Inefficient or insufficient proxies lead to incorrect ergodic site‐response predictions, 2) Epistemic and aleatory uncertainties of predicted site‐responses do not reflect the availability and variability of site‐specific observations, and 3) Extrapolated and interpolated predictions are restricted by the imposed stationarity of spatial correlation of observed site‐responses. Proxy‐based maps, with and without Kriging, share the last two issues. All three issues can be addressed using Uncertainty Projected Mapping (UPM). This study demonstrates the application of UPM to the Kanto region in Japan, evaluates its strengths and weaknesses against proxy‐based and Kriging‐based site‐response maps, and discusses the potential of UPM as a tool for regional site‐response mapping.
Chakraborty et al. (Sun,) studied this question.