Utilizing Bivariate Climate Forecasts to Update the Probabilities of Ensemble Streamflow Prediction

Abstract

In order to enhance the streamflow forecast skill, seasonal/sub-seasonal streamflow forecasts can be post-processed by incorporating new information, such as climate signals. This study proposed a simple yet efficient approach, the “Bivar_update” model that utilizes bivariate climate forecast to update individual probabilities of the ensemble streamflow prediction. The Bayesian updating scheme is used to update the joint probability mass function derived from historic precipitation and temperature data sets. Thirty-five dam basins were used for the case study, and the modified Tank model was embedded into the ensemble streamflow prediction framework. The performance of the proposed approach was evaluated through a comparison with a reference streamflow forecast model, the “Univar_update” model, that reflects only precipitation forecast, in terms of deterministic and categorical streamflow forecast accuracy. For this purpose, multiple cases of probabilistic precipitation and temperature forecasts were synthetically generated. As a result, the Bivar_update model was able to decrease the errors in forecast under below-normal conditions. The improvements in forecasting skills were found for both measures; deterministic and categorical streamflow forecasts. Since the proposed Bivar_update model reflects both precipitation and temperature information, it can compensate low predictability especially under dry conditions in which the streamflow’s dependency on temperature increases.