Assessing fast and accurate InSAR coherence change detection methods for near real-time earthquake response applications

The U.S. Geological Survey (USGS) provides rapid (within 30 min) estimates of earthquake-induced impacts and ground failure following significant events. These products are based solely on pre-event data and event-specific shaking estimates and do not include direct observations of building damage, casualties, or ground failure following an earthquake. To this end, the USGS is developing an intermediate-timeframe (within days to a week) pipeline for post-earthquake products that combines the current rapid estimation products with post-event observations to identify the most affected areas more accurately. As a vital component of this pipeline, the USGS is developing in-house capabilities to identify post-earthquake building damage and ground failure using Interferometric synthetic aperture radar (InSAR) coherence-based change detection maps (CDMs). We have previously shown that high-quality CDMs—in conjunction with accurate building footprints, prior building damage, and ground failure model estimates—improve upon a priori models of building damage and help differentiate building damage from ground failure effects. However, there is no standardized method for CDM generation, and approaches can vary substantially in computational cost and storage requirements. In this study, we evaluate the trade-offs between different CDM generation methods by assessing: (1) the number of pre-event images and coherence pairs, (2) the specific change detection method, and (3) earthquake-specific factors such as regional climate and timing relative to seasonal cycles. To quantify the accuracy of the different CDM generation methods, we compare our results with direct observations of building damage and ground failure data from three large events: the 2021 Haiti earthquake, the 2023 Morocco earthquake, and the 2023 Türkiye/Syria earthquake sequence. This work is an important step towards incorporating valuable post-event observations into near-real-time USGS earthquake products.