Estimating lake water storage and water level using Sentinel-1 C-band SAR

Intensification of climate change and extreme climate conditions around the world caused an increase in the frequency and intensity of water disasters (e.g., drought and flood). Particularly, the Korean Peninsula underwent a significant amount of rainfall during the summer monsoon, along with the increased number of typhoons passing through. In early August 2020, heavy rainfall occurred across the southern part of the Korean Peninsula (e.g., Jeolla and Chungcheong provinces), which resulted in the loss of life and properties. Accordingly, there is a continuous need to establish flood system monitoring over a wide region.

Accordingly, various studies have utilized different types of satellite imagery (e.g., optical, synthetic aperture radar [SAR], LiDAR) for flood inundation mapping. For example, optical satellite imagery (e.g., MODIS, Landsat, Sentinel-2/3) has been widely utilized for flood mapping, while it has limitations with regard to weather conditions. Synthetic Aperture Radar (SAR) imagery has been brought as an alternative as it is not hindered by weather conditions and has relatively high spatial resolution. Therefore, this study utilizes Sentinel-1 C-band backscatter (from 01/2016 to 12/2022) provided by the European Space Agency (ESA) to estimate the inland water body storage as well as water level at Naju Lake located in the Yeongsan River basin, South Korea.

 Prior to estimating the water body storage and water level, two threshold-based methods (i.e., Otsu threshold method, k-mean clustering) were used to distinguish water and no-water pixels based on the bimodal histogram of Sentinel-1 C-band backscatter. The validation of the water body area is conducted by comparing against optical image-based modified normalized difference water index calculated from the harmonized Landsat sentinel-2 (HLS) imagery. The overall evaluation confirmed that the accuracy of the water body area with k-mean clustering (0.8) showed better performance than that from the Otsu threshold method. Especially, the water body area from the Otsu threshold method showed a clear overestimation during the monsoon period. Afterwards, we established support vector regression (SVR) with the number of water pixels and ground-based water storage datasets. Estimation of water storage with SVR showed similar trend with observed water storage with the coefficient of determination (R²) of 0.92, while estimated water storage showed slight underestimation (bias = -899 m³).

Overall, Sentinel-1 C-band backscatter showed the capability to capture the inland water body as well as the volume of the inland lake. Even though there are several limitations (e.g., sensitivity toward vegetation, coarse revisit frequency) in the context of near real-time flood monitoring, it still has value in monitoring the spatio-temporal behavior of inland water body.

Acknowledgement: This work was supported by Korea Environment Industry & Technology Institute(KEITI) through R&D Program for Innovative Flood Protection Technologies against Climate Crisis Program(or Project), funded by Korea Ministry of Environment(MOE)(RS-2023-00218873)