短期气候数值预报中的集合个数问题

利用中国科学院大气物理研究所IAP L2 AGCM1.1模式18年(1980～1997年,每年包含28个积分)的集合后报试验结果,讨论了数值预测中的最小集合个数问题.研究表明预测结果达到稳定所需的最小集合数随区域和年份而变化,对于我国汛期降水跨季度预测,最小集合数在全国范围为19～20个,对于东南区域为10个左右,东北、华北地区不应少于20个.当出现多平衡态时,最小集合数增大,这时应对各平衡态分别进行集合.研究结果同时还显示,对于有预报能力的地区,集合方法可以提高预测技巧及其可信度,而在负技巧区域,集台反使得预测效果更差,因而有必要进行系统性订正.

Estimations of Ensemble Size for Numerical Climate Predictions

Ensemble size has been estimated by using the hindcast results of 18-year(1980-1997), 28 members contained in each year. Taking extra-seasonal predictions of the summer rainfalls in China for example, the minimum ensemble size necessary to approach stable prediction result varies with area and year. The minimum ensemble size is 19 to 20 for the whole China, 10 or so for Southeast China, at least 20 for North China and Northeast China. In the year when the phenomenon of multi-equilibrium-state is found, the minimum ensemble size will be large. It is also shown that ensemble method is able to im- prove predictions for the area with positive skill, but not for the area with negative skill. In this case, correction of systematic errors is needed.