Using satellite observations of river height and vegetation to improve National Water Model initialization and streamflow prediction
Date of Award
Doctor of Philosophy (PhD)
Stream measurements., Rivers--Regulation., Scientific satellites.
The National Oceanic and Atmospheric Administration (NOAA) Office of Water Prediction (OWP) implemented an operational, high-resolution National Water Model (NWM) in August 2016 to provide unprecedented hydrologic forecast capabilities on a national scale. The NWM, which will become the standard for operational hydrologic forecasting in the United States in a few years, provides streamflow forecasts for 2.7 million channel reaches within the continental United States. However, only 0.25% of these channel reaches are currently being initialized with actual observations gathered by approximately 7000 in-situ U. S. Geological Survey (USGS) stream gauges. Furthermore, no other satellite datasets are being assimilated or ingested to improve the analysis of land surface state variables. Consequently, NWM streamflow forecast capabilities are drastically hindered due to the lack of observations for model initialization. The National Aeronautics and Space Administration (NASA) Surface Water Ocean Topography (SWOT) mission will be launched in 2021 to provide unique observations of global rivers with widths greater than 50–100 meters. The SWOT mission will complement current in situ stream gauges by increasing the spatial coverage of streamflow observations and providing interferometric measurements from a uniform observational platform, albeit with altimetric errors greater than those of in situ gauges. However, no steps have been made to prepare SWOT measurements for assimilation into the operational NWM or quantify the assimilation impacts of SWOT WSE. This dissertation develops the capabilities and methodology to assimilate synthetic SWOT water surface elevations (WSE) into the operational NWM using the Data Assimilation Research Testbed (DART) Ensemble Adjustment Kalman Filter (EAKF) so that SWOT data can be used operationally shortly after launch. As a result, this project is the first to use unique SWOT measurements in an operational model, rather than simply in a research setting. Additionally, this project will assess the impact of ingesting real-time Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (VIIRS) green vegetation fraction (GVF) into the NWM. To accomplish these objectives, this project will investigate several case studies corresponding to droughts and rain-generated flood events in Alaska and North Alabama, and answer the following research questions: (1) how does ingesting real-time VIIRS GVF into the WRF-Hydro system affect runoff and streamflow prediction at the watershed scale and (2) does the assimilation of SWOT WSE improve streamflow forecast accuracy in gauged and ungauged basins. First, ingesting real-time VIIRS GVF into the NWM resulted in improved streamflow for the three watersheds examined with improvements in correlation, RMSE, and bias, but with varying levels of significance. P-values between the model simulation using real-time GVF and the simulation using climatological GVF were 0.03, 0.41, and 0.26 for the Limestone Creek, upper Flint River, and Paint Rock River watersheds, respectively. Furthermore, the model runs calibrated with VIIRS GVF and ingesting VIIRS GVF during forecasting were found to be superior to the current NWM configuration which does not ingest real-time GVF. GVF was found to have the largest impact on streamflow following river crest. GVF impacts on the NWM also vary by watershed and its terrain, with streamflow in watersheds with steeper terrain being less sensitive to changes in GVF. GVF was also found to be strongly proportional to ET even during winter months when incoming solar radiation is minimal, emphasizing the importance of ingesting GVF into the NWM year-round for the accurate modeling of streamflow. Secondly, this research was also able to successfully develop a SWOT-specific observation operator and modules within DART to assimilate synthetic SWOT WSE derived from an Observation System Simulation Experiment. The results were promising, showing that the assimilation of synthetic SWOT WSE with errors with a standard deviation of 25 cm resulted in consistently improved correlation and biases with respect to the truth during analysis. These results suggest that assimilating real SWOT WSE into WRF-Hydro and the NWM will improve streamflow prediction. However, one point of concern is the uncertainty in channel bathymetry, which is strongly related to channel head and thus influences the impact of assimilation. This aspect of assimilating SWOT WSE, along with attempts to limit the number of artificial flood waves generated when WSE observations are assimilated, will be examined in future work.
Elmer, Nicholas J., "Using satellite observations of river height and vegetation to improve National Water Model initialization and streamflow prediction" (2019). Dissertations. 166.