Date of Completion
Thomas M. Hopson
Emmanouil N. Anagnostou
Amvrossios C. Bagtzoglou
Field of Study
Doctor of Philosophy
Data assimilation (DA) is a method that optimally combines imperfect models and uncertain observations to correct model states using new information acquired from the incoming observations. In recent years, DA has been extensively used for improving the uncertainty of hydrologic prediction, largely due to the emergence of advanced remote sensing tools for observations of soil moisture, river discharge and precipitation. Several DA methods have been explored in hydrology; however the choice and the effectiveness of a specific DA method may vary depending on the model and the observation. The goal of this dissertation study was reducing streamflow forecast uncertainty, and was carried out in three parts.
First, the effectiveness of four different DA methods (ensemble Kalman filter (EnKF), particle filter (PF), Maximum Likelihood Ensemble Filter (MLEF) and variational method (VAR)) for improving streamflow forecasting were evaluated. In-situ discharge was assimilated into The United States National Weather Service (NWS) river forecasting model (Sacramento Soil Moisture Accounting model (SAC-SMA)) for Greens Bayou basin (with area of 178km2), in eastern Texas. The results indicate that all the four DA methods enhanced the short lead time forecast when compared to the model without the data assimilation; however the performances of each method vary with flow magnitude and longer lead time forecasts. Overall, the PF and MLEF performed superior to other DA algorithms across all flow regimes.
In the second part of this thesis, the value of satellite-based soil moisture retrievals for enhancing river discharge was assessed. Surface and root zone satellite-based soil moisture retrievals from AMSR-E (passive microwave) and ASCAT (active microwave) sensors were separately assimilated into the SAC-SMA model in Greens bayou using ensemble Kalman filter. Two different data assimilation experiments were carried out over a period of four years (2007 to 2010): updating the soil moisture state of the SAC-SMA model and combined correcting of soil moisture and total channel inflow (TCI) of the model. It was found that the remotely-sensed soil moisture assimilation reduced the discharge RMSE compared to the open loop for both assimilation schemes, and there was no appreciable difference between surface and root zone soil moisture results, as well as between the AMSR-E and ASCAT results. Furthermore, the dual correcting of soil moisture and TCI produced lower river discharge RMSE.
In the third part, the utility of passive microwave-based river width estimates for river discharge nowcasting and forecasting were assessed for two major rivers, the Ganges and Brahmaputra, in south Asia. Multiple upstream satellite observations of river and flood plains were used to track downstream flood wave propagation, and using a cross-validation regression model, the downstream river discharge was forecasted for lead times up to 15 days. The results showed that satellite derived flow signals were able to detect the propagation of a river flow wave along both river channels. And the approach also provided better discharge forecasts at downstream location compared to a purely persistence forecast, especially for high flows when the water spills out of the river bank. Overall, it was concluded that satellitebased flow estimates are a useful source of dynamical surface water information in regions where there is a lack of ground discharge data.
Hirpa, Feyera Aga, "Hydrologic Data Assimilation for Operational Streamflow Forecasting" (2013). Doctoral Dissertations. 152.