University of Twente Student Theses
Post-Disaster Damage and Response Assessment with PlanetScope Satellite Imagery : Case Study February 2023 Türkiye-Syria Earthquakes
Bansod, Shreya Deep (2024) Post-Disaster Damage and Response Assessment with PlanetScope Satellite Imagery : Case Study February 2023 Türkiye-Syria Earthquakes.
|
PDF
4MB |
| Abstract: | Every year, major disasters occur worldwide, requiring substantial response and recovery efforts. These efforts are often hampered or delayed due to incomplete information from all affected areas. High-spatial resolution remote sensing data, such as optical imagery from satellites and 3D point clouds generated from sources like LiDAR etc. are widely used to support post-disaster phases, focusing mainly on damage assessment and long-term recovery. However, the immediate and short-term response often goes unmonitored once the event fades from the spotlight. Additionally, most efforts tend to concentrate on the major urban areas, making them highly location-specific and often neglecting rural and other hard-to-reach human settlements. This also leads to limited understanding of the complex response phase due to constraints like limited coverage, infrequent data, and high costs associated with very-high resolution remote sensing data. These limitations make such datasets unsuitable for large-scale damage assessments and frequent monitoring of dynamic activities after a disaster event. This research therefore explores the utility of PlanetScope for (1) large-scale damage assessment covering both urban and rural areas, (2) detection and monitoring of response activities, and (3) integration with higher resolution images (WorldView and Pleiades) to understand their synergistic potential. PlanetScope is a constellation of 130+ satellite doves capturing almost the entire planet (except the poles) daily. The spatial resolution of PlanetScope is 3-4 meters, lower than WorldView and Pleiades. However, the spatial coverage and high temporal resolution provides the avenue to acquire images for any area of interest and date, except when hindered by high cloud cover or extreme events. Using the February 2023 Türkiye-Syria earthquakes as a case study, we employed texture feature extraction using GLCM for selected locations from Türkiye with a supervised machine learning algorithm, Random Forest, for classification. The potential of higher-resolution datasets was evaluated by using them as ground truth, confirming presence of activities in focus, and for enhancing PlanetScope image details through sharpening. Due to the limited availability of very-high-resolution images and ground truth data from Syria, the analysis primarily focused on locations in Türkiye. However, a location from Syria was selected to assess the generalizability of model trained and tested on data from Türkiye. Our findings show that PlanetScope effectively supports grid-based damage detection, identifying zones with collapsed buildings and calculating damage densities to highlight hotspots in both densely (urban) and sparsely (rural) populated areas. PlanetScope imagery is also useful for monitoring response activities, such as the evolution of temporary shelters and debris detection. However, damages not showing debris or missing roofs and small isolated temporary shelters are challenging to detect due to PlanetScope’s nadir acquisition and limited resolution. High temporal resolution of PlanetScope aids debris detection but is less effective for tracking removal due to the limitation posed by its spatial resolution and unavailability of related ground truth data. Integration of PlanetScope with higher-resolution data in multiple ways shows their synergistic potential to produce a detailed and comprehensive damage assessment. However, large-scale assessment and frequent monitoring can be hindered by the limited coverage and frequency of the higher-resolution satellite images. Lastly, the generalizability test conducted on a location in Syria showed that direct application of the model resulted in poor accuracy due to differences in settlement pattern and texture properties between the two regions. However, when the model was trained on data from Syria, it achieved high accuracy. In conclusion, the results prove that PlanetScope can effectively support rapid, large-scale damage and response assessment that can enable efficient allocation of resources and support for emergency responders. Advance feature extraction and classification algorithms can further help in improving the accuracy of results. Additionally, advance integration methods can help in enhancing details of PS images which can also lead to direct quantification of changes observed resulting in more meaningful insights. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Subject: | 38 earth sciences |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/105366 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page
