However, the threshold of the percentage of pixels with height changes within a building object remains an issue of uncertainty. In the experiment, a building is identified as having collapsed if the percentage of pixels with height changes within the building polygon is greater than the predefined threshold. Otherwise, the building is identified as not having collapsed. Seven threshold-setting scenarios were tested for detecting collapsed buildings using the two difference DEMs. The overall accuracy, average producer accuracy, average user accuracy and kappa coefficient were adopted to assess the accuracies based on the difference DEM as shown in Fig. 11a and b. The result indicated that the assumption of the percentage threshold may have an impact on the detection accuracy. (1) all accuracies and kappa coefficients of detecting buildings that have and have not collapsed using the two difference DEMs reach 100% when the threshold is over 50% and below 60%. This shows that such a range of thresholds is more suitable to the detection of collapsed buildings. (2) When the threshold ranges from 10% to 40%, the accuracies and kappa coefficients for the difference DEM as shown in Fig. 11a are much better than those for the difference DEM as shown in Fig. 11b. The reason for this is that the DEM created from the pre-seismic topographical map is better than that created from the pre-seismic IKONOS stereo pair. This implies that the incorporation of additional accurate GIS data (e.g., ground points with higher accuracy in both planimetry and height in this study) improves the final accuracy of the detection of buildings that have collapsed in an earthquake. However, with an increase in the threshold (e.g., over 50%), the accuracy for both DEMs does not change. (3) In general, the results show the availability of the proposed method to the detection of collapsed buildings using the pre- and post-seismic IKONOS stereo image pairs.