Effectiveness of Generative AI for Post-Earthquake Damage Assessment

After an earthquake, assessing the seismic vulnerability of buildings is essential for prioritizing emergency response, guiding reconstruction, and ensuring public safety. Accurate and timely evaluation of structural damage plays a vital role in mitigating further risks and facilitating in-formed decision-making during disaster recovery. This study investigates the performance of various Generative Artificial Intelligence (GAI) models, developed by different companies with diverse model sizes and context windows, in analysing post-earthquake images. The primary objective was to evaluate the models' effectiveness in classifying structural damage according to the EMS-98 scale (with 5 levels of damage), which ranges from minor damage to total destruction. For masonry buildings, the correct classification rates varied widely across models, from 28.6% to 64.3%, with mean damage grade errors ranging from 0.50 to 0.79. In the case of reinforced concrete buildings, correct classification rates ranged from 37.5% to 75.0%, with mean damage grade errors between 0.50 and 0.88. The use of fine-tuning could probably improve the results substantially. Improving the accuracy of GAI models could significantly reduce the time and resources needed to assess post-earthquake damage, namely when comparing with more traditional approaches. The results achieved thus far demonstrate the promise of GAI models for rapid, automated, and ac-curate damage evaluation, which is critical for expediting decision-making in disaster response scenarios.