Mississippi River Climate and Hydrology Conference

Friday, 17 May 2002: 2:30 PM
Assessment of Bias-Correction Methods for Probabilistic Forecasts of Monthly Streamflow Volumes
Tempei Hashino, Iowa Institute of Hydraulic Research and Department of Civil and Environmental Engineering, University of Iowa, Iowa City, IA; and A. A. Bradley and S. S. Schwartz
Ensemble streamflow prediction systems are currently being used to make probabilistic streamflow forecasts up to seasonal time scales. With this approach, an ensemble of streamflow traces is generated, starting from the forecast date initial conditions, using a hydrologic model. However, the ensemble traces produced by a hydrologic simulation model can have biases. These biases can adversely impact forecast quality and the operational usefulness of the forecasts. How best to remove biases is an important science question for GAPP research.

This study assesses three bias-correction methods. The first method uses the expected value of streamflow conditioned on the model-simulated flow for the ensemble (conditional expected value approach). The second method assigns a multiplicative bias correction to each simulated ensemble trace based on the ratio of model-simulated and observed flows for the same weather sequence multiplicative approach). The third method uses the observed flow that has the same nonexceedance probability as the simulated flow for the ensemble (quantile-mapping approach). A distributions-oriented approach, developed for the verification of probabilistic streamflow forecasts, is used to assess the forecast quality corresponding to the three bias-correction methods. Comparisons are made of experimental forecasts of monthly flow volumes for the Des Moines River, issued sequentially for each month over a 49-year record. The results shows that all three bias-correction methods significantly improve forecast quality. Still, the multiplicative approach yields forecasts with the highest skill for a lead time of one-month, while the quantile-mapping approach has the highest skill for longer lead times. The experimental results demonstrate the importance of bias-correction, and the need to examine various approaches for the range of forecast variables of interest in water resources operations.

Supplementary URL: