Enhancement in long-term hydrologic forecasting accuracy using the Bayes’ theorem |
---|
학술지명 Asia Oceania Geosciences Society
저자 강신욱,전근일,남우성,서승범,김영오
발표일 2019-07-31
|
The National Drought Information-Analysis Center of Korea that operates all the multi-purpose dams currently runs its Ensemble Streamflow Prediction (ESP) system in the beginning of week and month, which employs the probabilistic precipitation forecast of the KMA (Korea Meteorological Administration) to assign weights of the ESP traces. The first part of this study evaluates the current skill of the K-water ESP system; the overall skills in terms of N-RMSE (Normalized Root Mean Squared Error) and POD (Probability of Detection) are 1.06 and 37.6%, respectively but the skill in the below-normal category of streamflow is significantly low (18.2% of POD). Secondly, this study proposes the Bayesian ESP (B-ESP); the observed (i.e., the prior) streamflow distribution is updated with the ESP (i.e., the posterior) information through the Bayes’ theorem with a likelihood function. Note that the role of the likelihood is correcting systematic errors in the hydrologic forecasting model. Testing of B-ESP for the thirty-five Korean watersheds shows that overall skills in the below normal category are improved (23.2% of POD) although it is still lower than those from climatology. |