气候预测对象与影响因子关系不稳定性及其统计集合预测模型的改进

Unstability of Relationship between Climate Prediction Objects and Affecting Factors and Improvement of the Ensemble Statistical Prediction Model

通过对"滑动相关——逐步回归——集合分析"短期气候预测方法的研究及其业务应用进行了综述。从深入分析气候预测对象与影响因子关系不稳定性问题入手,提出物理统计与集合分析相结合的气候预测建模理念,设计了"滑动相关——逐步回归——集合分析"气候预测模型。在改进该预测方法的过程中,先后提出了"交叉建模"集合预测思路,建立了"持续异常"与"转折影响"不同因子信息拓展方案,构建了"原始序列"与"去线性趋势"不同建模方案,再对最优模型进行集成分析。近8 a该方法参与国家气候中心全国汛期预测、全国秋季初霜冻初日预测,试用效果良好。在新疆地方气候预测服务中广泛应用,包括降水、气温、积温、热量指数、界限温度初终日期、沙尘日数的预测等。该方法在极端气候事件指数、应用气候指数的预测领域具有应用潜力。

In this paper, the study and application of prediction methods of seasonal precipitation were reviewed based on the moving correlation, stepwise regression and ensemble analysis. The instability between the climate pre- diction objects and the affecting factors was analyzed, which brought us an idea of combination of the physical-statis- tical model and ensemble analysis. A series of schemes were employed to improve the prediction methods including the cross-modelling, treating of continuity and transition features of the prediction factors, preprocessing of both orig- inal and detrended time series of variables, and ensemble of optimal prediction models. In the recent eight years, the outputs of this method have been used as reference to predict the summer precipitation and first frost date at Nation- al Climate Center, and it was widely used in climate prediction and services in Xinjiang, such as predicting precipi- tation, temperature, cumulative temperature, heat indices, start and end dates of given temperature, dust days, et al. The method will be useful in predicting the extreme climate events.