Surface water extent monitoring using the Global WaterPack product: automated extraction, refinement, and analysis

Abstract

Monitoring and analyzing surface water dynamics is critical for understanding hydrological variations, climate change impacts, and water resource management. Traditional methods of surface water storage monitoring rely on in-situ measurements, which are often spatially and temporally limited. Remote sensing has revolutionized this field by enabling large-scale, consistent, and continuous observations of area and height water storages. This study presents a methodology for generating accurate water area time series using the Global WaterPack (GWP), a monthly satellite-derived dataset. Developed by the German Aerospace Center (DLR), the GWP dataset is specifically designed for monitoring surface water dynamics on a global scale. A Python-based processing tool is developed to systematically extract and analyze lake and river water extents, mitigating key challenges such as cloud contamination, defining proper threshold, and classification inaccuracies. By integrating high-frequency surface water observations from GWP with the Prior Lake Database PLD (a static dataset for extracting the initial search area), and the SWOT (The Surface Water and Ocean Topography) prior River Database (SWORD), which provides a standardized framework of high-resolution river nodes and reaches, this tool enhances the reliability of time-series analysis. This framework improves surface water change detection, reduces computational complexity, and refines water occurrence assessments under diverse hydroclimatic conditions. The workflow automates the entire process, allowing users to select lakes interactively via a geospatial interface or upload coordinate lists for batch processing. Key steps include (1) downloading images, (2) defining the search area, (3) normalization, (4) generating a water occurrence map, (5) defining constant water and land masks, (6) residual analysis, (7) deriving and applying thresholds, (8) generating time series plots, and (9) correlation analysis. Advanced filtering methods, such as threshold-based classification and residual analysis, refine water occurrence measurements, while an adaptive thresholding approach using the Cumulative Distribution Function (CDF) enhances water body delineation accuracy. To evaluate the reliability of the extracted data, the resulting surface area time series are compared against altimetry-derived water height records using correlation analysis. The analysis revealed clear seasonal and interannual variations in lake water areas, aligning well with natural hydrological patterns. Many lakes showed strong positive correlations between satellite-derived surface area and altimetry-based water levels, validating the method’s effectiveness. However, weaker correlations in some cases were attributed to issues like cloud cover, sensor limitations, and complex hydrodynamics. The study emphasized that a fixed threshold is insufficient for all systems, whereas the corrected method provided more reliable results across diverse conditions. Although river analysis showed varied hydraulic responses, the tool proved useful for monitoring floods, seasonal changes, and long-term water trends, especially with proper calibration. By providing an automated, scalable, and accurate tool for water body monitoring, this thesis contributes to advancing hydrological analysis using remote sensing and geospatial processing techniques. The developed tool can aid in climate studies, water resource management, and flood risk assessment, offering a valuable framework for long-term surface water monitoring.