In order to estimate the damage distribution immediately after an earthquake, both physical prediction methods and data-driven methods that analyze sensing data obtained from satellites are used. However, the former has the problem of prediction accuracy, while the that analyze sensing data obtained from satellites are used. However, the former has the problem of prediction accuracy, while the that improves the detection accuracy of detailed damage distribution of structures such as total and partial collapse by integrating both methods. As an integration scheme of the two methods, a data assimilation method based on Bayes' theorem is adopted in this study. We proposed a method to update the damage probability of each structure obtained from physical simulations by conditioning it on the observed data obtained from satellite image analysis, and verified its effectiveness.