Statical prediction system of the typhoon intensity using Numerical Weather Prediction model for correction

Abstract

The accuracy of tropical cyclone (TC) forecasts from NWP models have been improved especially for the track. Relatively, TC intensity forecasts still include huge uncertainties though the dynamics, physics processes, and resolutions of NWP systems become higher in both horizontal and vertical. For this reason, many operational centers and academia for TC forecasts implemented statistical prediction systems and Artificial Intelligence (AI) algorithms based on long-term dynamic model forecasts for better predictions of typhoon intensity.The National Hurricane Center (NHC) developed the Statistical Hurricane Intensity Prediction Scheme (SHIPS) which is a statistical model based on NWP forecasts (parameters from atmosphere and ocean). Also, infrared imagery from geostationary satellite is used as predictors for the regression. SHIPS is implemented for the North Atlantic and East Pacific regions. Otherwise, the Joint Typhoon Warning Center (JTWC) implemented this model for the Northwest Pacific region. Also, Korea Meteorological Administration (KMA) and Japan Meteorological Administration (JMA) developed the statistical based typhoon prediction systems (called STIPS and TIFS, respectively). However, the accuracy of these systems is not stable because it is not easy to define the tendency of NWP forecasts for TC intensity. The National Typhoon Center of KMA developed a new statistical model (Statistical Prediction Intensity of Korea mEteorological administrator, SPIKE) for typhoon intensity prediction based on ECMWF forecast. While the ECMWF Integrated Forecast System (IFS) has an excellent performance in forecasting track of typhoons, the intensity tends to be underestimated compared to typhoons analysis information. SPIKE is basically developed as a multi-linear regression model, and its predictors are extracted from the IFS forecast. The average prediction error of typhoon intensity of SPIKE in 2022 decreased by about 30% compared to the ECMWF forecasts. However, there was still a limitation, especially for cases of rapid intensification (RI). More studies to reflect real-time intensity, cloud development, center location, and prediction errors of the model are conducted. Then, the second multi-linear regression model to account for these parameters is developed. Finally, an additional improvement of about 30% was achieved. Also, the performance for RI cases developing more than 35 knots within 24 hours was greatly improved.