EOF和CCA方法在台风路径预报试验中的比较

COMPARISIONS OF EOF AND CCA METHODS IN TYPHOON TRACK FORECAST TEST

利用气候持续性因子，并分别考虑５００ｈＰａ高度场的ＥＯＦ因子和ＣＣＡ因子，分别建立了基于ＥＯＦ因子的和基于ＣＣＡ因子的南海台风２４－１２０小时路径预报模式。经过非独立样本和独立样本的预报检验、比较。将两种方法运用于５００ｈＰａ高度场的分析表明，ＣＣＡ方法得出的典型变量权重系数分布虽不能象ＥＯＦ方法的特征向量一样表示出明显的而且平滑的空间场形式，但ＣＣＡ方法浓缩了更多的与台风路径整体相关性最好的原变量场信息，因而更能有效地减小预报误差。预报检验表明，无论是基于ＥＯＦ因子的路径预报模式，还是基于ＣＣＡ因子的路径预报模式，都具有一定的预报能力，其中以后者的预报效果为更好，从５年预报来看，２４、４８、７２、９６和１２０小时的平均矢量误差分别为１５９．２７、３１４．８４、５２４．１２、８１３．０３和９８７．１２ｋｍ。该路径预报模式为南海台风短。

Based on both the CLIPER predictors and the EOF predictors,CCA predictors of 500 hPa height fields respectively, two sets of track forecast models for typhoons over the South China Sea at intervals from 24h to 120 h were proposed in this paper. The models were verified and compared using the dependent and the independent samples. The applied analyses of the two methods performed on the height fields or allohypsic fields at 500 hPa have been shown that the canonical weighted coefficient distribution derived by the CCA method did not exhibit the apparant and smooth space patterns of the eigenvectors derived by the EOF method. However, the ensemble information of original variable fields correlative to the tracks of typhoon were much more enriched by the CCA method. Consequently, the forecast errors were reduced effectively. The verifications indicated that the track forecast models either based on EOF predictors or based on CCA predictors show skill over CLIPER models, whereas the latter forecast performance is better. The 5 year averaged vector errors are 86, 170, 283, 436 and 533 nautical miles at 24,48, 72, 96 and 120 h respectively. The track forecast models can be taken as effective objective prediction techniques for predicting the short and mid range tracks of typhoon over the South China Sea.