Development and evaluation of a hydrologic data-assimilation scheme for short-range flow and inflow forecasts in a data-sparse high-latitude region using a distributed model and ensemble Kalman filtering

A forecasting system combining a physically-based distributed hydrological model (HYDROTEL), an Ensemble Kalman Filtering (EnKF) Data Assimilation (DA), and forecasted meteorological data (obtained from the North American Ensemble Forecast System; NAEFS) is developed to forecast short-range (0-14 days lead) flows and inflows in the Aishihik and Mayo basins in Yukon Territory, Canada. The system was assessed at three sites, including at the outlet of the Sekulmun River subbasin of the Aishihik basin for river flow forecasting, as well as at Aishihik Lake and Mayo Lake for reservoir inflow forecasting. To ensure accuracy of forecasting outputs, model development and evaluation was performed systematically by: (i) investigating the use of coupled EnKF and HYDROTEL models for improved flow and inflow estimations, (H) evaluating NAEFS data for short-range flow and inflow forecasts, and (Hi) using probabilistic and deterministic criteria to evaluate the forecast performance of the HYDROTEL-EnKF-NAEFS model at each site. Results illustrate that the DA framework significantly improves flow and inflow forecasts, and raw NAEFS data need to be spatially and temporally corrected to be used for hydrological forecasts. Based on probabilistic and deterministic scores, it was found that the developed forecasting system can provide flow and inflow forecasts at the Sekulmun River subbasin, Mayo Lake, Aishihik Lake sites with high, medium, and low accuracies, respectively. Differences in forecast accuracies at each site are possibly associated with: (i) uncertainties of forecasted meteorological data, (H) ability of HYDROTEL to capture daily flow and inflow variations, (Hi) DA algorithm used, (iv) heterogeneity in basin attributes, and (v) limited data availability particularly in the lake areas.