Polarimetric synthetic aperture radar (PolSAR) data show good performance in near-real-time earthquake/tsunami damage assessment. In this paper, an improved damage level index for earthquake/tsunami damage level mapping of urban areas based on distance metric learning using full PolSAR data has been proposed. First, urban areas are extracted by our proposed classification method, which conjunctively uses multiresolution segmentation and support vector machine. Then, a double-bounce scattering power-based damage level index is adopted for damage level mapping. To solve the problem that the primary result does not match with the truth damage level in heavily damaged areas, distance metric learning method is introduced to calculate the Mahalanobis metric to compose an improved index. Extensive experimental comparison and analysis on L -band ALOS PALSAR data of Tohoku earthquake demonstrate the effectiveness of the proposed index. Both the analysis about damage level assessment map and sample areas’ linear fitting with truth damage level indicate the improved damage level index achieves high accuracy. The same analysis was applied to evaluate the damage induced in Kumamoto earthquake event. The analysis result verified the robustness of the proposed damage level index.