Rapid Damage Mapping after an Earthquake using Sentinel-2 Images (Case Study: Sarpol-e Zahab)

Rapid damage mapping after an earthquake in order to produce damage map is important for relief and rescue operations. Recently, the use of remote sensing images for producing damage maps is considered due to their synoptic view and low cost. In this research, a rapid damage mapping approach according to change detection is proposed, which is applied to the 2018 Sarepol-e Zahab earthquake. In order to assess results, outcomes of the change detection were evaluated using ground truth, which show high accuracy in detecting change areas. On the other hand, our damage map was evaluated using damage map produced by the European Space Agency (ESA), which outcomes depict our proposed method can detect damage areas by an overall accuracy of 84 %. Using the proposed method, damage map of the Sarepol-e Zahab was generated less than 30 minutes. 
Introduction
Remote sensing is a useful science and technology for different applications, especially disaster management. Remote sensing can be used to produce building damage maps after the earthquake. Recently, researchers used remote sensing data for producing building damage maps [1-4]. However, the used approaches are based on training samples. Preparing training samples is a time consuming task. For this reason, scientists would like to develop rapid damage mapping. Tiede et al. proposed a method to map damage areas of the Haiti earthquake using a shadow analysis approach. The proposed approach can produce damage areas after 12 hours [5]. The main goal of this paper is to develop a rapid damage mapping approach based on pre- and post-event images in Sarpol-e Zahab. The developed method benefits from decision making approaches to make a rapid map.
Methodology
The proposed method is done in four steps according to Figure 1. In the first step, some essential pre-processing tasks including georeferencing and radiometric correction are performed. In the second step, difference image is produced and some textural features are extracted from it. In the third step, change and unchanged areas are identified using three change detection approaches. Finally, TOPSIS decision making approach is employed to make a damage map.
 
Fig. 1.  Workflow of the proposed method
Results
Since the proposed method is based on change detection, we applied it to two data sets. Results of change detection over two case studies present in Figure 2. According to validation results, the proposed approach can detect changed and unchanged areas with about 95 % accuracy.




 
 
 



Nearest neaghbour of Region 1


Nearest neaghbour of Region 2


Spectral angle mapper of Region 1



 
 
 



Spectral angle mapper of Region 2


Maximum likelihoo of Region 1


Maximum likelihoo of Region 2





Fig. 2. Results of change detection approaches over two study areas
Using pre- and post-event Sentinel-2 images and our proposed approach, damage map of Sarpol-e Zahab was produced. Figure 3 shows pre- and post-event Sentinel-2 images and damage map of the study area.




 
 




 
Fig. 3. Pre- and post-event Sentinel-2 images and damage map of the study area
The accuracy of our damage detection approach is assessed using damage map produced by European space agency (ESA). Table 1 depicts the confusion matrix regarding the accuracy of our proposed method. Based on this table, the overall accuracy of our proposed approach is about 70 %.
Table 1. the confusion matrix of our proposed approach





Overall acc.  (%)


User acc. (%)


Producer acc. (%)


Damaged


Undamged


 




68.26


43.85


68.84


18468


14442


Undamaged




85.77


680.6


39355


6527


Damaged





 
Conclusion
In this paper, a rapid damage mapping approach is proposed to detect damage areas from Sarpol-e Zahab earthquake. The proposed method is based on change detection and unsupervised. From the perspective of change detection, our proposed approach is robust. To assess the capability of the proposed method, it was applied in Sarpol-e Zahab earthquake. Using pre- and post-event Sentinel-2 images, the proposed approach can detect damaged areas with an accuracy of 80 %.