Evaluation and modification of geospatial liquefaction models using land damage observational data from the 2010–2011 Canterbury Earthquake Sequence

Abstract

This paper evaluates existing geospatial liquefaction models using the observed land performance across four events of the Canterbury Earthquake Sequence and analyses the influence of region specific input variables on the models' potential to predict liquefaction manifestation. The aggregated performance for each event is assessed using receiver operating characteristic (ROC) analysis and the liquefaction spatial extent (LSE). The assessment results in high ROC performance and shows that the models are able to capture the areas of the most severe liquefaction manifestation observed following each earthquake. Replacing the input variables for distance to closest river, time averaged shear wave velocity over the upper 30 m (Vs 30 ) and water table depth with New Zealand specific datasets does not significantly impact the outcomes of the ROC analysis across the study areas. However, the improved spatial accuracy of the LSE maps based on these specific datasets supports their use, especially for regional or local hazard assessments, and demonstrates the potential for further development of the models. The findings of this study provide an improved understanding of the performance of geospatial liquefaction models and their ability to identify exposed areas for regional and large scale hazard assessments. • Geospatial liquefaction models are tested across four earthquakes in New Zealand. • Some models perform well in aggregate, however spatial accuracy is variable. • Low resolution (global) input data might be unsuitable for regional assessments. • Region specific input data improves spatial accuracy of liquefaction hazard maps.