基于偏最小二乘回归的冬季北太平洋风暴轴指数的特征诊断

Diagnosic prediction of winter storm track indices in the North Pacific based on partial least square regression

针对现有风暴轴指数分析大多采用相关分析等较为简单方法,难以对风暴轴指数变化有效诊断分析的问题,引入偏最小二乘回归(Partial Least Square Regression,PLS)的线性方法和核偏最小二乘回归方法(Kernel Partial Least Square Regression,KPLS),对冬季北太平洋风暴轴指数变化进行了特征诊断研究,并与传统的线性无偏最小二乘回归结果进行了试验比对。结果表明:偏最小二乘回归方法的诊断结果能够更好地反映风暴轴内部变化规律,并有效降低诊断误差。对于PNYI(北太平洋风暴轴纬度指数),采用r>0. 2的因子筛选方案(r为因子与风暴轴指数的相关系数)并应用KPLS算法时,预测效果最佳;对于PNXI(北太平洋风暴轴经度指数)和PNII(北太平洋风暴轴强度指数),采用全因子方案并应用KPLS算法时,预测效果最佳。

With most of the existing analyses about storm track indices using such simple methods as correlation analysis and doing a bad job in diagnostic prediction,the Partial Least Square regression(PLS)containing the linear method and the Kernel Partial Least Square regression(KPLS)were introduced to study the variation of winter storm track indices in the North Pacific.The experiment results,comparing with those of conventional least square regression,indicates a better performance of PLS in revealing the inherent variation rules and reducing the diagnostic error.In terms of the winter North Pacific storm track latitude index(PNYI),the combination of KPLS and the factorfiltering scheme of r>0.2(r is the correlation coefficient between the factors and the storm track indices)performs best.As for the longitude index(PNYI)and the intensity index(PNII),the combination of KPLS and the allfactor scheme performs best.