An Updated Geospatial Liquefaction Model for Global Application

We present an updated geospatial approach to estimation of earthquake‐induced liquefaction from globally available geospatial proxies. Our previous iteration of the geospatial liquefaction model was based on mapped liquefaction surface effects from four earthquakes in Christchurch, New Zealand, and Kobe, Japan, paired with geospatial explanatory variables including slope‐derived V S 30 , compound topographic index, and magnitude‐adjusted peak ground acceleration (PGA) from ShakeMap. The updated geospatial liquefaction model presented herein improves the performance and the generality of the model. The updates include (1) expanding the liquefaction database to 27 earthquake events across six countries, (2) addressing the sampling of nonliquefaction for incomplete liquefaction inventories, (3) testing interaction effects between explanatory variables, and (4) overall improving the model’s performance. We inspected 14 geospatial proxies for soil density and soil saturation; the most promising of these are slope‐derived V S 30 , modeled water table depth, distance to coast, distance to river, distance to the closest water body, and precipitation. We found that peak ground velocity (PGV) performs better than PGA as the shaking intensity parameter. We present two models which offer improved performance over prior models. We evaluate model performance using the area under the receiver operating characteristic curve, and the Brier score. The best‐performing model in a coastal setting uses distance to the coast but is problematic for regions away from the coast. The second best model, using PGV, V S 30 , water table depth, distance to the closest water body, and precipitation, performs better in noncoastal regions and thus is the model we recommend for global implementation.