수치모델의 태풍 강도 예측성능 평가 및 컨센서스 기법 연구

Abstract

For improvement of TC (Tropical Cyclone) intensity predictability, we have investigated the performance of sixteen TC intensity guidance models in the western North Pacific during 2008-2014 and tested various consensus techniques on the basis of the results. From the evaluation using total errors up to 72 hours, we found that HWRF (Hurricane Weather Research and Forecasting model) outperforms all others overall although GFS (Global Forecast System), KWRF (Korea Meteorological Administration WRF) and JGSM (Japan meteorological agency Global Spectral Model) also shows a good performance in some forecast times. HWRF, in particular, shows the highest performance in predicting the intensity of strong TCs above Category 3, which may be attributed to its highest spatial resolution (∼3 km). GFS was the most improved model from 2008 to 2014. For initial intensity error, two Japanese models, JGSM and TEPS (Typhoon Ensemble Prediction System), had the smallest error. In track forecast, the ECMWF (European Centre for Medium-Range Weather Forecasts), including recent GFS model, outperformed others. On the basis of evaluation of 16 numerical models, we have tested two consensus techniques according to the number of model used, training period, rank criteria (correlation or absolute error), and averaging methods (equally or weighted mean). Prediction tests for 2012-2014 reveals that (i) the CON2 (using one consensus method showing the smallest error up to 72 hours during training period) generally outperformed CON1 (using the best performance consensus methods at individual forecast time during training period), (ii) 1-year training period showed lower errors than 2-year, (iii) the consensus using 4 or 5 models showed the best performance, (iv) weighted averaging matched correlation criteria while equally averaging matched absolute error criteria in terms of absolute error.