Advancing Post-Earthquake Damage Detection: Ground Validation and Deep Learning Approach from the 2023 Mw6.8 Morocco Earthquake

Rapid and automated detection of earthquake-damaged buildings using post-event satellite imagery only can play a fundamental role in effective disaster response and reconnaissance missions. Traditional damage detection methods, which rely on both pre- and post-event imagery, often face challenges due to data scarcity, limitations of single-modality remote sensing, and inability to translate to practical measures on the ground. Here we propose an integrated approach that combines multi-modal remote sensing data with a deep learning framework and ground validation. Our work builds on the Earthquake Engineering Field Investigation Team’s (EEFIT) reconnaissance mission, carried out 8 months after the 2023 Mw6.8 Morocco earthquake. The remote sensing team actively supported itinerary planning and data collection, optimizing the selection of survey areas to capture different damage classes. The mission facilitated the collection of essential ground truth data over 445 structures, covering diverse damage states, necessary to build our neural network. Our deep learning approach leverages RemoteDamageDL, a modular framework designed for earthquake building damage. The framework implements and compares different backbone architectures in early, mid and late fusion strategies. To build the datasets, we utilized multi-modal imagery: SAR data from Umbra Space and Capella Space, and optical imagery from Maxar and Landsat, along with a binary footprint mask for each assessed building. In addition, due to coregistration errors, we performed the alignment of SAR with optical images to ensure spatial matching. The resulting dataset, therefore, contains post-event image patches and labels for destroyed and intact buildings. To evaluate scalability, we implemented cross-training and testing across separate event datasets, such as the 2023 Mw7.8 Turkey-Syria earthquake, in addition to the Morocco event. This study underscores the value of combined field and remote sensing approaches for improving disaster response and the rapid and automated detection of earthquake-damaged buildings and contributes to developing reproducible workflows for global post-earthquake assessment efforts.