Comparison of Sequential and Variational Streamflow Assimilation Techniques for Short-Term Hydrological Forecasting

Abstract

This study compares sequential and variational streamflow assimilation techniques for short-term hydrological forecasting based on a lumped conceptual rainfall-runoff model and two dissimilar watersheds (Canada and Germany). The assessment targets the Ensemble Kalman filter (EnKF) and variational data assimilation (VDA). Deterministic streamflow forecasts are computed on a daily time step over a 10-day forecast horizon, using meteorological observations as inputs to the model. Results show that the EnKF leads to the highest performance for all forecast horizons while the optimal set-up for the VDA, which often competes with the EnKF, varies from one watershed to the other. EnKF surpasses forecasts without assimilation for all horizons and for both watersheds where the NSE varies between 0.88 and 0.79 on the au Saumon watershed in Canada and between 0.92 and 0.87 on the Schlehdorf watershed in Germany on a 10-day horizon, which is not always true for the VDA. The naïve output assimilation is also assessed and is only helpful for the first two days of the forecasts.