Prediction of early summer rainfall over South China by a physical-empirical model
Prediction of early summer rainfall over South China by a physical-empirical model
复制标题
DOI：
10.1007/s00382-013-2014-3
复制
发表时间：
2014-10
期刊：
影响因子：
4.6
通讯作者：
S. Yim;Bin Wang;W. Xing
中科院分区：
文献类型：
--
作者：
S. Yim;Bin Wang;W. Xing
初夏（5 月至 6 月，MJ）北半球最强的降雨带出现在东亚（EA）副热带锋面。在此期间，华南 (SC) 降雨量达到年度峰值，代表东亚地区降雨量最大变率。因此，我们建立了 SC 降雨指数，它是 SC 上空 72 个站点（北纬 28°以南，东经 110°以东）的 MJ 平均降水量平均值，极好地代表了 EA 上 MJ 降水变率的主要经验正交函数模式。为了预测SC降雨，我们建立了物理经验模型。对 34 年观察（1979-2012）的分析揭示了三个物理后果预测因素。去年冬天，在出现丰富的超临界降雨之前，（a）印度太平洋暖池的偶极海表温度（SST）趋势，（b）北大西洋的三极海表温度趋势，以及（c）亚洲北部的变暖趋势。这些前兆预示着初夏菲律宾海副热带高压和鄂霍次克高压增强，这是副热带锋面降雨增强的控制因素。基于这些预测变量建立的物理经验模型在 1979-2012 年实现了 0.75 的交叉验证预测相关性技能。令人惊讶的是，这项技能远远高于 1979-2010 年期间四动力模型的集合预测（0.15）。结果表明，当前动力学模型的预测能力较低，很大程度上是由于模型的缺陷，动力学预测还有很大的改进空间。
In early summer (May–June, MJ) the strongest rainfall belt of the northern hemisphere occurs over the East Asian (EA) subtropical front. During this period the South China (SC) rainfall reaches its annual peak and represents the maximum rainfall variability over EA. Hence we establish an SC rainfall index, which is the MJ mean precipitation averaged over 72 stations over SC (south of 28°N and east of 110°E) and represents superbly the leading empirical orthogonal function mode of MJ precipitation variability over EA. In order to predict SC rainfall, we established a physical-empirical model. Analysis of 34-year observations (1979–2012) reveals three physically consequential predictors. A plentiful SC rainfall is preceded in the previous winter by (a) a dipole sea surface temperature (SST) tendency in the Indo-Pacific warm pool, (b) a tripolar SST tendency in North Atlantic Ocean, and (c) a warming tendency in northern Asia. These precursors foreshadow enhanced Philippine Sea subtropical High and Okhotsk High in early summer, which are controlling factors for enhanced subtropical frontal rainfall. The physical empirical model built on these predictors achieves a cross-validated forecast correlation skill of 0.75 for 1979–2012. Surprisingly, this skill is substantially higher than four-dynamical models’ ensemble prediction for 1979–2010 period (0.15). The results here suggest that the low prediction skill of current dynamical models is largely due to models’ deficiency and the dynamical prediction has large room to improve.