Use this URL to cite or link to this record in EThOS:
Title: Improving flood prediction using data assimilation
Author: Cooper, Elizabeth S.
ISNI:       0000 0004 7971 9228
Awarding Body: University of Reading
Current Institution: University of Reading
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
River flooding is a costly problem worldwide. Timely, accurate prediction of the behaviour of flood water is vital in helping people make preparations. Mathematical hydrodynamic models can predict the behaviour of flood water given information about inputs such as river bathymetry, local topography, inflows, and values for model parameters. Uncertainty in these inputs leads to inaccuracies in model predictions; data assimilation can be used to improve forecasts by combining model predictions with observational information, taking into account uncertainties in both. In this thesis we investigate ways to maximize the impact of observational information from satellite-based synthetic aperture (SAR) instruments in data assimilation for inundation forecasting. We show in synthetic twin experiments using an ensemble transform Kalman filter that using joint state-parameter data assimilation techniques to correct the model channel friction parameter as well as water levels provides a significant, long lasting benefit to the model forecast. We show that errors in the channel friction parameter and inflow are interdependent. We propose a novel observation operator that allows direct use of measured SAR backscatter values, potentially allowing the use of many more observations per SAR image. We test our new observation operator in synthetic experiments, showing that we can successfully update inundation forecasts and the value of the model channel friction parameter using our new approach. We show that different observation operator approaches can generate significantly different updates to model forecasts and illustrate the physical mechanisms responsible. Lastly, we use our new observation operator to assimilate backscatter values from real SAR images, showing that our new approach can be used to improve inundation forecasts in a real case study. Improved understanding of the physical mechanisms by which updates are generated by different observation operators provides insights into improving the observation impact of SAR data.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral