Flow-Dependent Bias correction to the NCEP Short-Range Ensemble Forecasting system

Model bias could at least be classified into two kinds: long-term (climate or theoretical or true) and short-term (practical) biases. The former might be constant and needs long training period of historical data (therefore more expensive), while the latter should be flow dependent and needs relatively short training period. For example, a model might show very different biases between an El Nino year and a La Nino year but show almost no bias if averaged over those two years. Similarly in daily weather prediction, a model could show warm bias in one type of weather regime such as high pressure system in some days and show cold bias in low pressure system in other days. Therefore, it's necessary and important to remove those short-term or practical portion of model bias in order to steadily reduce forecast error in daily NWP operation. In another word, there are rooms to improve in traditional statistically-based bias correction approaches which are usually a running-mean or a decaying average or a similar type over a period of past forecasts without distinguishing weather regimes. Recently, a flow regime dependent bias correction method was proposed and tested with the NCEP Short Range Ensemble Forecasting (SREF) system. As a first step, the method was applied only to the first moment (ensemble mean) and showed promising results. How to apply the method to the 2nd (ensemble spread) and higher moments (such as probability) is still under invesitigation.