A case-based reasoning method of recognizing liquefaction pits induced by 2021 MW 7.3 Madoi earthquake

Abstract

Earthquake-triggered liquefaction deformation could lead to severe infrastructure damage and associated casualties and property damage. At present, there are few studies on the rapid extraction of liquefaction pits based on high-resolution satellite images. Therefore, we provide a framework for extracting liquefaction pits based on a case-based reasoning method. Furthermore, five covariates selection methods were used to filter the 11 covariates that were generated from high-resolution satellite images and digital elevation models (DEM). The proposed method was trained with 450 typical samples which were collected based on visual interpretation, then used the trained case-based reasoning method to identify the liquefaction pits in the whole study area. The performance of the proposed methods was evaluated from three aspects, the prediction accuracies of liquefaction pits based on the validation samples by kappa index, the comparison between the pre- and post-earthquake images, the rationality of spatial distribution of liquefaction pits. The final result shows the importance of covariates ranked by different methods could be different. However, the most important of covariates is consistent. When selecting five most important covariates, the value of kappa index could be about 96%. There also exist clear differences between the pre- and post-earthquake areas that were identified as liquefaction pits. The predicted spatial distribution of liquefaction is also consistent with the formation principle of liquefaction.