Updating inflow forecasts using empirical statistical matching for real-time prediction of daily net inflows to Okanagan Lake

Abstract

Accurate predictions of flow periods are important for decision-making within the Okanagan Lake System. A nonparametric method to predict the hydrograph to achieve a closer match with the timing and volume of reservoir inflows during the dominant flow period (February1 to July 31) in Okanagan Lake was developed in this study. The method employed, Real-Time Statistical Matching (RTSM), uses a combination of information from a changing suite of best-fit historical years, existing forecasts, and recent inflow trends. This included a comparison between the current hydrograph against hydrographs derived from historical inflows based on the predicted volume and pattern of the hydrograph. The RTSM-based approach is hypothesized to improve the ability of hydrological models to predict shifts in the general timing of peak net inflows. This makes the RTSM model more robust to both historic and non-historic conditions. The performance of the RTSM-based predictions was compared to the legacy hydrology model based on average timing of historic flows. Results indicate an improvement in predictive accuracy of 10%, 6%, and 80% for Nash-Sutcliffe Efficiency (NSE), root mean squared error to standard deviation ratio (RSR), and percent bias (PBIAS) respectively, which are three different measures of the accuracy of predictions. Further, the success of the Okanagan Fish/Water Management Tool (FWMT) relies on the water and fish managers that use the tool, which extends beyond the quantitative metrics in this study. The authors’ also discussed how the tool’s utility has changed over time from when it was put into practice. It was learned that in practice, the best use of the model was based on the volume-based prediction with the real-time adjustment.