A coherence time-series framework for earthquake-induced building damage detection using SAR

Earthquake-induced building damage detection using interferometric SAR coherence is complicated by multiple decorrelation mechanisms unrelated to structural damage. Most existing approaches rely on coherence changes extracted from a limited number of image pairs, which makes it difficult to isolate structural damage signals from other temporally varying contributions. This study proposes a sequential coherence time-series framework that explicitly models decorrelation components across pre-, co-, and post-seismic phases according to their distinct spatiotemporal characteristics, and derives two complementary damage indicators: the average pre-to-co-seismic coherence ratio (ρ0) and the average pre-to-post-seismic coherence ratio (ρ1). Both indicators combine candidate pixel selection and temporal averaging to suppress damage-irrelevant coherence variability while amplifying step-type transitions associated with building damage. ρ0 supports rapid initial assessment using the first post-event acquisition, whereas ρ1 provides higher specificity to persistent structural damage by excluding the co-seismic pair from the indicator computation. The framework is validated using Sentinel-1 time series over Mashiki Town and Wajima City, affected by the 2016 Kumamoto and 2024 Noto Peninsula earthquakes. Compared with three representative coherence-based damage indicators, the proposed approach consistently reduces false alarms in both urban and non-urban areas. ρ1 demonstrates substantially improved precision–recall performance under class imbalance conditions, with AUPRC values of 0.56 and 0.66 at the two sites, compared to 0.22–0.39 for the reference methods. These results indicate that explicit temporal modeling of coherence decorrelation provides a physically interpretable and operationally practical basis for post-earthquake damage mapping.