Optimum Usage of Prior Forecast Information for Bias Correction

A bias correction is most effective way to reduce model systematic errors for NWP forecast. The one of important NAEFS Statistical Post Process (SPP) steps is bias correction which is using Kalman Filter method (or decaying average method) with 2% (or w=0.02) weight coefficient. In fact, this decaying weight coefficient is the functions of forecast lead-time, forecast variables, seasonal and geographic areas. For example, an accumulated bias of surface wind is not sensitive to the length of prior forecast information accumulation, but surface temperature is. Meanwhile, many other methods for bias accumulation, such as equal weights, non-equal weights, diagonal (triangle) weights, and other designed methods, have been tested and compared for selected variables. The preliminary results indicate that the forecast errors are strongly associated to forecast time scales (and forecast lead-time). For short-time forecast, the systematic error from large sample average is not necessary to improve short-term forecast, but for long-term forecast or extended-range forecast, the more historical information is, the better result is.