Quantifying water volumes of ungauged lakes using optical satellite imagery and high-resolution LiDAR DEMs

Abstract

Surface water bodies such as lakes, rivers and wetlands are critical water resources to both human and ecological systems, but are under increasing pressure from competing users. Monitoring of the location, quantity and movement of water is crucial in effectively managing these resources and implementing a sustainable water management strategy for the future. However, continuous information on the quantity and distribution of water across the landscape is limited in some regions because of the high cost of traditional in-situ monitoring. As such, global remote sensing datasets are being used more frequently to complement these sparse networks. This paper aims to develop new methods to estimate the quantity of water (volume) in open water storages such as lakes, using remote sensing data. Lake Menindee, part of the greater Menindee Lakes complex in the Murray-Darling Basin, was selected as the case study of this research because of its geographic location and data availability. Water management in the Murray-Darling Basin has been under increasing scrutiny partly due to exposure of water theft by irrigators. As such, there is a pressing need for large scale monitoring of water resources in the region using novel data and methods. This paper developed three methods to estimate water volumes in a lake, all of which only used a high-resolution (5m) LiDAR DEM in conjunction with optical imagery. As an initial preprocessing step, the water observations from space (WOfS) algorithm (Mueller et al. 2016) was applied to Landsat optical imagery to detect areas of surface water in the lake which was used as an input to all the methods. The first method derived a relationship between lake inundated surface area and volume using the DEM. Subsequently, this relationship was used to convert WOfS-derived surface areas to volumes. The second method evaluated the quality of match between the WOfS spatial inundation pattern and DEM-modelled inundation patterns at 0.1m water level increments, from which an optimal match and the respective DEM-derived volume was picked. Quality of match was quantified with three metrics commonly used in weather forecasting. In the third method, the elevation of the WOfS water body edge was derived from the DEM, and a volume was estimated by filling the lake DEM to this height. Water volumes by all three methods were estimated using 19 years of high-quality Landsat data equivalent to 209 scenes, and daily gauged measurements were used for validation. A combination of scatterplots and statistical metrics were used for evaluation. Initial findings show that all methods have reasonable skill in estimating water volumes with high Pearson correlation coefficients, and estimates from methods 2 and 3 have relative biases of less than 10 percent. No single method performed consistently better across all ranges of volumes, with method 3 having poorest performance for low volumes while method 1 substantially overestimated high volumes. Additionally, estimation errors were volume-dependent, with medium-range of volumes having highest accuracy estimates while prediction skill consistently worsened at higher volumes across all methods. Future research should further investigate drivers of the volume-dependent errors, expand the evaluations to multiple case studies, including the large on-farm water storages across the Murray-Darling Basin, and test other remote sensing data sources such as radar altimetry. These results clearly demonstrate the potential of remote sensing based methods for lake volume estimation.