Geospatial Analysis for Identifying Blocked Roads after Earthquakes: A Case Study from the Kahramanmaras Earthquake
Journal of Performance of Constructed Facilities, vol.40, no.3, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 40 Issue: 3
- Publication Date: 2026
- Doi Number: 10.1061/jpcfev.cfeng-5371
- Journal Name: Journal of Performance of Constructed Facilities
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, Criminal Justice Abstracts, ICONDA Bibliographic, INSPEC, Criminal Justice Periodical Index, The International Construction Database (ICONDA), Academic Search Ultimate (EBSCO), Engineering Source (EBSCO), Sociology Source Ultimate (EBSCO)
- Keywords: Disaster management, Geospatial analysis, Search-And-rescue optimization
- Hacettepe University Affiliated: Yes
Abstract
Rapid identification of road blockages is essential for effective emergency response in the aftermath of large-scale urban disasters. This study introduces a geospatial methodology for detecting roads obstructed by earthquake-induced building collapses, utilizing high-resolution unmanned aerial vehicles (UAV) imagery collected after the February 6, 2023, Kahramanmaras Earthquake. A deep learning-based object detection model (YOLOv9-C) was employed to detect collapsed buildings, and subsequent intersection analysis with road network data enabled automated blocked road identification. The analysis was based on a geospatial data set comprising 4,188 UAV images systematically captured across Nurdagi, Kirikhan, Iskenderun, and Onikisubat, representing some of the most heavily impacted regions. To evaluate the framework's robustness, two complementary strategies were adopted. The first was randomly split analysis (RSA), which randomly partitioned the data set into training, validation, and testing subsets. The second was Nurdagi test analysis (NTA), which excluded all Nurdagi data during training and validation and used it exclusively for testing to assess generalization in unseen areas. The blocked road detection yielded precision values of 0.963 (RSA) and 0.961 (NTA), closely aligning with a manually labeled ground truth data set. Overall, 21.55 km of blocked roads were detected across the study area. By integrating UAV-based deep learning detection with GIS, the proposed framework provides near real-Time, scalable insights to support search-And-rescue operations and postdisaster logistics.