本文利用神经网络模型、多元线性回归模型和马尔科夫模型分别建立了统计预报模型,对热带印度洋海表温度异常(SSTA)和印度洋偶极子(IOD)指数进行了63 a的长时间回报实验,并详细比较了线性和非线性统计预报模型的差异.结果表明:统计模型对IOD指数的预报技巧和现有动力模式预报技巧相差不大,对偶极子指数(DMI)有效预报时效为3个月,东极子指数(EIO)为5~6个月,西极子指数(WIO)达到8~9个月.IOD事件强烈的季节锁相特性使得对秋季的DMI指数可以提前4个月做出有效预报.加入同期的ENSO指数来预报IOD指数,能有效地提高IOD预报技巧,特别是对IOD峰值的预报.复杂的神经网络模型和简单的多元线性回归模型在对SSTA 和IOD指数的预报具有同等的效果.%The tropical Indian Ocean Sea Surface Temperature Anomaly(SSTA)and the Indian Ocean Dipole (IOD)indices are predicted,using the multiple linear regression model,the Markov model and the neural network model respectively.63 years'hindcast experiments are set up to compare the differences between linear and nonlinear statistical models in detail.And the results reveal that the statistical models are little different from the complicated dynamic model.Their skillful prediction(correlation coefficients above 0.5) could reach 3 months for DMI,about 5-6 months for EIO index and 8-9 months for WIO.Since the IOD event has a strong seasonal phase lock,the DMI can be predicted previously for 4 months in fall.When the synchronistic ENSO index is added as a predictor,the prediction skill,especially the IOD peak,will be improved.The complicated neural network and the simple regression model are proved to be with a similar prediction skill.
展开▼