Detection of Earthquake-Induced Fractures inon Highways using Multivariate Time-Series Analysis of Optical Satellite Imagery

Reliable and timely remote sensing approaches are increasingly important for the detection and monitoring of road infrastructure damage following major earthquakes. While Sentinel-1 synthetic aperture radar (SAR) has been widely used for surface deformation mapping, it is susceptible to geometric distortions such as layover and foreshortening on steep terrain, which can reduce mapping accuracy in mountainous regions. To support the rapid screening of logistics routes, this study developed a detection framework based on multivariate time-series analysis of Sentinel-2 optical imagery. Fifty-one polygons were delineated over suspected collapse sites, of which 44 sites with sufficient cloud-free observations were retained for analysis, using Sentinel-2 L2A data acquired from 2020 to 2024. Cloud contamination was addressed using the Scene Classification Layer (SCL), and spectral anomalies were computed relative to a pre-earthquake four-year baseline (2020–2023) to reduce the influence of seasonal vegetation variability. Five spectral indices—NDVI, BSI, NDMI, EVI, and MNDWI—were employed to capture post-event surface changes associated with vegetation loss, bare-soil exposure, and variations in surface moisture. At many sites, a concurrent decline in NDVI and an increase in BSI were consistent with vegetation removal and bare-soil exposure, supporting automated preliminary screening. Positive shifts in MNDWI were additionally observed at several sites, potentially reflecting changes in surface moisture or material properties following structural disruption. Cumulative detection performance peaked at F1 = 0.777 (recall = 92.2%) after four post-event observations and reached a maximum recall of 94.1% at 11 observations. Cross-referencing with field-referenced data from 2024 confirmed that the observed anomaly patterns are broadly consistent with early post-earthquake surface change phases. The proposed multi-index anomaly framework demonstrates potential as a complementary tool for the preliminary identification of infrastructure damage in earthquake emergency response contexts.