Water Level Estimation Using Sentinel-1 Synthetic Aperture Radar Imagery And Digital Elevation Models (2012.07627v2)

Abstract: Hydropower dams and reservoirs have been identified as the main factors redefining natural hydrological cycles. Therefore, monitoring water status in reservoirs plays a crucial role in planning and managing water resources, as well as forecasting drought and flood. This task has been traditionally done by installing sensor stations on the ground nearby water bodies, which has multiple disadvantages in maintenance cost, accessibility, and global coverage. And to cope with these problems, Remote Sensing, which is known as the science of obtaining information about objects or areas without making contact with them, has been actively studied for many applications. In this paper, we propose a novel water level extracting approach, which employs Sentinel-1 Synthetic Aperture Radar imagery and Digital Elevation Model data sets. Experiments show that the algorithm achieved a low average error of 0.93 meters over three reservoirs globally, proving its potential to be widely applied and furthermore studied.

• The paper presents a novel algorithm that uses gradient changes in Sentinel-1 SAR data to accurately determine water boundaries.
• It integrates DEM data to simulate water rise, avoiding the pitfalls of traditional water body extraction techniques.
• Experimental results across three dams showed an R² of 0.96 and an average error of 0.93 meters, highlighting its precise performance.

Enhancing Water Level Estimation through Sentinel-1 SAR Imagery and DEM Data

Introduction

Monitoring water levels in reservoirs harnesses critical information for resource management, forecasting droughts, and mitigating flood risks. Traditional in-situ measurement methods, although effective, suffer from limitations regarding cost, accessibility, and global coverage. This paper introduces an innovative approach to water level estimation that leverages Sentinel-1 Synthetic Aperture Radar (SAR) imagery and Digital Elevation Model (DEM) datasets, presenting it as a viable alternative for global water level monitoring applications.

Current Water Level Estimating Techniques

Existing methodologies primarily utilize satellite altimetry for large reservoir water level estimation. However, the challenge with altimetry data lies in its limited global coverage, rendering it ineffective for small to medium-sized reservoirs, which are predominant in regions such as the Greater Mekong Subregion. Optical imagery has also been explored, with several studies using it in conjunction with accurate reference DEMs for mapping water surfaces to water levels. Yet, the susceptibility of optical images to cloud cover limits their reliability. Synthetic Aperture Radar (SAR) imagery emerges as a promising alternative due to its cloud penetration capability, though existing methods relying on water body extraction through SAR image thresholding face limitations in varying hydrological and geographical contexts.

Proposed Methodology

The paper presents a novel methodology for water level estimation that abandons the traditional dependence on water extent extraction, due to its unreliability. Instead, the proposed algorithm is designed around the principle that amidst the dynamic range of backscatter coefficients in SAR images, the gradient magnitude change along the actual water boundary is distinctly higher. Employing this key observation, the algorithm endeavors to simulate the water rising process on the DEM image to identify the level that optimally aligns with this principle. Implemented on the Google Earth Engine platform, the method incorporates preprocessing steps such as band combination, speckle filtering, and edge detection to enhance the reliability and accuracy of the estimation process.

Experimental Validation

The proposed algorithm's performance was rigorously evaluated across three dams (Burrendong, Hume, and Mosul) contrasting with their respective reference data. The results revealed an impressive average $R^2$ score of 0.96 and an average error margin of just 0.93 meters, showcasing the algorithm's potential for accurate and reliable water level monitoring. These findings were substantiated by comparing estimated water levels against reference data over a series of selected dates, further affirming the algorithm's applicability and effectiveness in diverse hydrological and geographical scenarios.

Conclusion and Looking Forward

The paper effectively addresses the limitations of existing water level estimation methodologies by introducing a robust, reliable, and highly accurate approach that harnesses Sentinel-1 SAR imagery and DEM datasets. With an average error rate of 0.93 meters, this method sets a new benchmark in the field. Looking ahead, the paper suggests several avenues for enhancing the algorithm's efficiency and accuracy, including experimentation with alternative optimization methods and the potential integration of deep learning segmentation networks for precise water body segmentation. Through these improvements, the proposed methodology holds the promise of revolutionizing water level monitoring, enabling better management and response frameworks for dealing with extreme hydrological events.

The implications of this research are vast, opening doors to more accurate and globally applicable water level monitoring techniques. Future work could further refine this approach, potentially incorporating it into integrated systems for comprehensive environmental monitoring and management. As the global community grapples with the challenges of climate change and water resource management, innovations like these provide valuable tools in our arsenal.