Integrating SAR and Optical Imagery Analysis for Liquefaction Phenomenon Identification of Post-Pohang Earthquake 2017, South Korea, Utilizing a Hybrid Deep-Learning Approach

Abstract

An interesting liquefaction event happened following the 5.6-Mw Pohang earthquake on November 15, 2017. Liquefaction affects soil density as a result of earthquake vibrations, causing water to ascend and combine with solid soil. In general, this effect results from increasing water pressure in the buried lower layer. Remote sensing data, particularly those obtained with differential interferometry SAR (DInSAR), can be utilized to assess surface changes and soil moisture levels. This analysis utilizes Sentinel-1 C-band data from 2017 to 2020. Furthermore, the biased time-series identification presented to analyze the anomaly may be present in this study. In addition, optical satellite data were used to estimate changes in water content and soil moisture following Pohang earthquake. A combination of spectral bands sensitive to changes in water content was used to identify anomalies following an earthquake based on Sentinel-2 and Landsat-8 data. Based on the spatial analysis, optical images may detect changes in water content with a spatial accuracy of 60%–80% when compared to field data. Moreover, the liquefaction susceptibility map has been generated using a hybrid dense convolutional neural network (DCNN) architecture and swarm-based optimization algorithm. As a result, the susceptibility model performance was conducted using k-fold cross correlation with an area under the curve (AUC) value of 0.75–0.86. However, this research was the initial effort to determine the potential of liquefaction in the future based on the 2017 Pohang earthquake, and the results can improve our understanding of this phenomenon compression.