| dc.description.abstract | The goal of this research is to review the current state of long-term, multi-decadal lake dynamic monitoring and develop novel techniques for scalable analysis at local, regional, and global levels. This dissertation is comprised of three chapters formatted as journal manuscripts with each chapter progressively addressing some key limitation in current lake dynamic monitoring methodologies. Chapter 1 tracks lake dynamics (surface elevation, surface area, volume, and volume change) for a single water body, Lake McConaughy, which is the largest lake and reservoir in the state of Nebraska, using the cloud-based geospatial analysis platform Google Earth Engine. Lake dynamics were estimated using bathymetric survey data, the Shuttle Radar Topography Mission 30-meter digital elevation model, and Landsat 5 image composites for 100 time periods between 1984 and 2009. Water surface elevation was estimated and assessed for 5,994 different combinations of water indices, segmentation thresholds, water boundaries, and statistics and produced elevations as accurate as 0.768 m CI95% [0.657, 0.885] root-mean-square-error. The method also detected seasonal and long-term trends which would have major implications for regional agriculture, recreation, and water quality. Chapter 1 was published as an article in the peer-reviewed journal Water Resources Research in October 2019. Chapter 2 expands and improves upon the techniques explored in Chapter 1 in multiple ways. First, the techniques were improved to remove image contamination sources such as snow, ice, cloud cover, shadow, and sensor error for individual images using the Pixel Quality Assurance (QA) band available as a part of the Landsat 4, 5, 7, and 8 Top-of-Atmosphere Tier-1 Collection-1 archives. Using the Pixel QA band information, image contamination was removed from each image between August 1982 and December 2017 and water surface elevation was estimated with the remaining visible water boundary extents overlaying merged National Elevation Dataset digital elevation model and bathymetric survey data resampled to 30-meters which resulted in enhanced temporal resolution compared to the techniques used in Chapter 1. Second, the analysis was expanded from a single water body to fifty-two lakes/reservoirs to provide a better understanding of how the techniques generalize to imagery and water bodies encompassing a wide range of ecotypes, geologies, climates, and management strategies. A variety of common water indices, such as the Modified Normalized Difference Water Index, naïve and dynamic water indices, water boundary types, and filtering strategies were tested and individual lake accuracies are as low as 0.191m RMSE CI95%[0.129, 0.243], and 45 of the 52 lakes produced sub-meter root-mean-squared-error accuracies. Furthermore, accuracy of surface elevation estimates is highly correlated with the mean slope of surrounding terrain with low-slope shorelines having greater accuracy than high-slope shorelines such as those in canyon-filled reservoirs or in mountainous regions. Overall, the improved techniques extend our ability to track long-term lake dynamics to lakes with bathymetric datasets while lacking in-situ hydrological stations, provide a framework for scale-able analysis in Google Earth Engine, and balance a need between high-accuracy estimates and maximum temporal resolution. Bathymetric survey data, such as that used in Chapters 1 and 2 is, unfortunately, not available for most water bodies at regional and global scales. Chapter 3 introduces a method of tracking long-term lake dynamics without bathymetry data and only using available digital elevation models such as Shuttle Radar Topography Mission, the National Elevation Dataset, and Advanced Land Observing Satellite. In digital elevation models, the water surface is often, but not always, hydroflattened producing a flat surface approximating the surface of the water at the time of the data capture which precludes using water boundaries like those in Chapter 1 and Chapter 2 to estimate water level when it is lower than the hydroflattened surface in the digital elevation model. However, using hypsometric relationships developed from the digital elevation models, subsurface water dynamics can still be estimated by extrapolating the low water levels using regression, albeit with increased uncertainty compared to levels above the hydroflattened surface. Using multiple digital elevation models, the lowest hydroflattened surface can be identified for each water body which reduces uncertainty for low water levels by reducing the extrapolation distance to those values while simultaneously increasing the number of above hydroflattened surface estimates. In addition to low-level uncertainty, hypsometric techniques are highly impacted by image contamination such as cloud, cloud shadow, snow, ice, and sensor error which reduces the observable water surface area resulting in erroneous surface elevation, volume, and volume change estimates. To help alleviate this issue, a technique of using proportional hypsometry was developed to remove contamination effects. Together, using the lowest hydroflattened surface and proportional hypsometry, this research produced 12,680 additional water surface elevation estimates for 46 lakes in comparison to traditional hypsometric techniques, reduced the number of sub-hydroflattened water surface estimates by 549 or more compared to individually using any of the three digital elevation models assessed, and lays the groundwork for regional and global scale surface water dynamic research without bathymetric survey data. | |
