Abstract
Abstract. The potential of data assimilation for hydrologic predictions has been demonstrated in many research studies. Watersheds over which multiple observation types are available can potentially further benefit from data assimilation by having multiple updated states from which hydrologic predictions can be generated. However, the magnitude and time span of the impact of the assimilation of an observation varies according not only to its type, but also to the variables included in the state vector. This study examines the impact of multivariate synthetic data assimilation using the ensemble Kalman filter (EnKF) into the spatially distributed hydrologic model CEQUEAU for the mountainous Nechako River located in British Columbia, Canada. Synthetic data include daily snow cover area (SCA), daily measurements of snow water equivalent (SWE) at three different locations and daily streamflow data at the watershed outlet. Results show a large variability of the continuous rank probability skill score over a wide range of prediction horizons (days to weeks) depending on the state vector configuration and the type of observations assimilated. Overall, the variables most closely linearly linked to the observations are the ones worth considering adding to the state vector due to the limitations imposed by the EnKF. The performance of the assimilation of basin-wide SCA, which does not have a decent proxy among potential state variables, does not surpass the open loop for any of the simulated variables. However, the assimilation of streamflow offers major improvements steadily throughout the year, but mainly over the short-term (up to 5 days) forecast horizons, while the impact of the assimilation of SWE gains more importance during the snowmelt period over the mid-term (up to 50 days) forecast horizon compared with open loop. The combined assimilation of streamflow and SWE performs better than their individual counterparts, offering improvements over all forecast horizons considered and throughout the whole year, including the critical period of snowmelt. This highlights the potential benefit of using multivariate data assimilation for streamflow predictions in snow-dominated regions.
Highlights
Water resource management for reservoirs located in snowdominated regions relies on an accurate portrayal of the snow water equivalent (SWE) spatial and temporal distribution in order to make accurate streamflow predictions
Before investigating the effect of data assimilation on streamflow forecasts, a state vector configuration analysis was conducted. This was done in order to find out which variables, among the seven listed previously (VOL, SWE, snow ripening index (SRI), snow temperature index (STI), soil moisture level (SML), groundwater level (GWL), lake water level (LWL)), should be included in the state vector for each type of data assimilated in order to reduce the number of comparisons to make
This study investigated the impact that multivariate data assimilation can have on streamflow forecasts using the CEQUEAU hydrologic model applied over the Nechako watershed in a synthetic experiment
Summary
Water resource management for reservoirs located in snowdominated regions relies on an accurate portrayal of the snow water equivalent (SWE) spatial and temporal distribution in order to make accurate streamflow predictions. Some water resources managers make use of ensemble streamflow prediction (ESP) to plan reservoir operations over various lengths of time. ESPs have the benefit of integrating weather forecast uncertainty, either by making use of weather ensemble predictions (de Roo et al, 2003) or by using historical weather data (Day, 1985) as input in a hydrologic model. Many water resources managers still use a manual approach to adjust the initial state of the watershed based on available observations and the user’s experience (Liu et al, 2012). Data assimilation (DA) methods, such as the ensemble Kalman filter (EnKF; Evensen, 2003) can improve the estimation of the initial state of the watershed while providing an uncertainty on this initial state (Liu and Gupta, 2007).
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
