Real-Time Correction Method for Numerically Modeled Typhoon Tracks and Its Integrated Application

Using the short-lead-time forecast bias and target-lead-time latitude forecast (i.e., the lead time required to be corrected) of numerically forecasted typhoon tracks as the predictors, a track forecast bias prediction equation is established by multiple linear regression. The equation permits typhoon track forecast correction in real time. In this paper, 12 h is taken as the short lead time, and the European Center for Medium-Range Weather Forecasts deterministic prediction system (ECMWF-IFS) and ensemble prediction system (ECMWF-EPS) for typhoon track forecasts are applied. The modeled forecast results from 2018 show that the mean track error of the corrected typhoon tracks forecasted by ECMWF-IFS at 24 h, 36 h, 48 h, 60 h, 72 h, and 84 h is reduced by 7.3 km, 9.3 km, 8.9 km, 6.5 km, 6.9 km, and 2.6 km, respectively, compared with those of uncorrected typhoon tracks. In general, the corrective effect is better for strong typhoons (observed intensity at 12 h≥32.7 m s⁻¹). First, the typhoon track forecast of each ensemble forecast member from ECMWF-EPS is corrected; the integrated forecast is then obtained. The typhoon track forecasts obtained by the following five methods are compared: “corrected deterministic prediction,” “ensemble mean of all ensemble forecast members,” “ensemble mean of selective ensemble forecast members,” “ensemble mean of all corrected ensemble forecast members,” and “ensemble mean of corrected selective ensemble forecast members.” The modeled forecast results from 2018 show that “ensemble mean of corrected selective ensemble forecast members,” “ensemble mean of all corrected ensemble forecast members,” and “ensemble mean of selective ensemble forecast members” yield the lowest track error at 24 h and 36 h, 48 h and 60 h, and 72 h and 84 h, respectively. When applied in a targeted manner in operational application, an objective comprehensive forecast result for typhoon tracks with excellent performance for each lead time is expected. At 24 h, 36 h, 48 h, and 72 h, the mean track error of “ensemble mean of corrected selective ensemble forecast members” is reduced by 13.3 km, 11.7 km, 10.0 km and 7.6 km, respectively, compared with those of “ensemble mean of all ensemble forecast members” and by 0.7 km, 2.0 km, 3.9 km, and 2.4 km, respectively, compared with those of the Central Meteorological Office official track forecast (with corresponding lead times of 12 h, 24 h, 36 h, and 48 h, respectively).