集合样本数影响月平均500 hPa高度场预测技巧分析

AN ANALYSIS OF THE INFLUENCE OF ENSEMBLE SIZE ON MONTHLY MEAN OF 500-HPA GEOPOTENTIAL HEIGHT PREDICTION

关键词:  模式评估  500 hPa高度场  集合预报  相关系数

Key words:  model evaluation  500 hPa geopotential height  ensemble forecasting  correlation coefficient

摘要:

气候系统模式输出结果是当前开展气候预测业务的重要参考依据之一，如何提高气候系统模式输出结果的可信度是改进气候业务预测能力的关键之一。利用1999—2010年NCEP CFSv2模式每日四次预测未来45天的回算数据，分析了集合样本数对模式预测能力的影响。分析结果表明，模式对月平均500 hPa位势高度的预测技巧在热带地区较高，而中高纬度地区较低；模式对500 hPa位势高度时间异常的预测能力优于空间异常。无论是空间异常还是时间异常，随着模式超前时间的增加，预测技巧均逐渐降低，但是在不同区域和不同月份，模式预测技巧随超前时间的变化存在差异。此外，模式预测技巧存在非常大的年际变率。增加集合样本数，对不同月份和不同起报时间预测技巧的稳定度和预测技巧值均有明显正效果，特别是对亚洲中纬度地区改善度较大。增加集合样本数也可以在一定程度上降低模式预测技巧年际变率。集合样本数增加对于500 hPa位势高度空间异常的改进优于时间异常。

Abstract:

The output of climate system models is one of the important references for climate prediction. Enhancing the credibility of climate system models’ output is one of the keys to improve climate forecasting. In this paper, the 45 days regression data of NCEP CFSv2 model predicted four times a day from 1999 to 2010 were used to analyze the influence of ensemble size on model prediction ability. The results showed that there were higher prediction skills on monthly mean of 500-hPa geopotential height in the tropics, but lower prediction skills in the middle and high latitudes. The temporal anomaly prediction skills of 500-hPa geopotential height were better than spatial anomaly prediction skills. The prediction technique of both spatial anomaly and temporal anomaly decreased gradually with the increase of the model lead time. However, the prediction skill varied with different lead time in different regions and months. In addition, there was large interannual variability in prediction skill of CFSv2 model. Increasing the numbers of ensemble size has a significant positive effect on the stability and the value of prediction skills in different months and at different starting times, especially on the improvement in prediction for the mid-latitude region of Asia. Increasing ensemble size can also reduce the inter-annual variability of model prediction skills to some extent. With the increase of ensemble size, the improvement of spatial anomaly prediction skills on 500-hPa geopotential height was better than the improvement of temporal anomaly prediction skills.