Ensemble Data Assimilation for Flood Forecasting in Operational Settings: from Noah-MP to WRF-Hydro and the National Water Model

The National Water Center (NWC) started using the National Water Model (NWM) in 2016. The NWM delivers state-of-the-science hydrologic forecasts in the nation. The NWM aims at operationally forecasting streamflow in more than 2,000,000 river reaches while currently river forecasts are issued for 4,000. The NWM is a specific configuration of the community WRF-Hydro Land Surface Model (LSM) which has recently been introduced to the hydrologic community. The WRF-Hydro model, itself, uses another newly-developed LSM called Noah-MP as the core hydrologic model. In WRF-Hydro, Noah-MP results (such as soil moisture and runoff) are passed to routing modules. Riverine water level and discharge, among other variables, are outputted by WRF-Hydro. The NWM, WRF-Hydro, and Noah-MP have recently been developed and more research for operational accuracy is required on these models. The overarching goal in this dissertation is improving the ability of these three models in simulating and forecasting hydrological variables such as streamflow and soil moisture. Therefore, data assimilation (DA) is implemented on these models throughout this dissertation. State-of-the art DA is a procedure to integrate observations obtained from in situ gages or remotely sensed products with model output in order to improve the model forecast. In the first chapter, remotely sensed satellite soil moisture data are assimilated into the Noah-MP model in order to improve the model simulations. The performances of two DA techniques are evaluated and compared in this chapter. To tackle the computational burden of DA, Massage Passing Interface protocols are used to augment the computational power. Successful implementation of this algorithm is demonstrated to simulate soil moisture during the Colorado flood of 2013. In the second chapter, the focus is on the WRF-Hydro model. Similarly, the ability of DA techniques in improving the performance of WRF-Hydro in simulating soil moisture and streamflow is investigated. The results of chapter 2 show that the assimilation of soil moisture can significantly improve the performance of WRF-Hydro. The improvement can reach 58% depending on the study location. Also, assimilation of USGS streamflow observations can improve the performance up to 25%. It was also observed that soil moisture assimilation does not affect streamflow. Similarly, streamflow assimilation does not improve soil moisture. Therefore, joint assimilation of soil moisture and streamflow using multivariate DA is suggested. Finally, in chapter 3, the uncertainties associated with flood forecasting are studied. Currently, the only uncertainty source that is taken into account is the meteorological forcings uncertainty. However, the results of the third chapter show that the initial condition uncertainty associated with the land state at the time of forecast is an important factor that has been overlooked in practice. The initial condition uncertainty is quantified using the DA. USGS streamflow observations are assimilated into the WRF-Hydro model for the past ten days before the forecasting date. The results show that short-range forecasts are significantly sensitive to the initial condition and its associated uncertainty. It is shown that quantification of this uncertainty can improve the forecasts by approximately 80%. The findings of this dissertation highlight the importance of DA to extract the information content from the observations and then incorporate this information into the land surface models. The findings could be beneficial for flood forecasting in research and operation