春季陆面植被对长江流域夏季降水可预报性的影响分析

利用美国全球监测与模型研究中心(GIMMS)1982—2006年逐月归一化植被指数(NDVI)、美国国家海洋和大气局(NOAA)1854—2008年海温资料以及中国国家气候中心(NCC)1951—2006年160站月降水资料,通过旋转经验正交函数分解(REOF)和相关分析获得了长江流域夏季降水预报序列和植被、海温预报因子集。基于最优子集回归方法(OSR),并借助交叉验证(CV)以及空间重建等手段,构建了单独以前期春季海温为预报因子和同时引入前期春季海温与归一化植被指数为因子的两类预报模型,对比分析引入陆面植被因子前后长江流域夏季降水预报效果改善状况,评估春季陆面植被对长江流域夏季降水可预报性的影响及预报效果的稳健性。结果表明:(1)相对于海温因子,春季陆面植被因子对长江流域夏季降水预报具有同样重要性,引入春季归一化植被指数后,长江流域夏季降水预报得到明显改善,相关系数平均由0.49提升到0.66,提高0.17左右,模型解释方差提升平均60%左右,其中单纯海温因子预报效果较差的汉江—淮河地区和淮河流域地区,相关系数更是提高了0.20—0.30,模型解释方差提升1倍左右;(2)交叉验证预报表明,相对于仅考虑海温因子模拟情形,交叉预报相关系数下降较多,模型稳健性较低,引入归一化植被指数后,长江流域夏季降水预报稳健性得到明显提升,长江中下游及其以南的长江三角洲地区、洞庭湖—鄱阳湖地区改善尤为明显;(3)长江流域降水可预报性存在明显的区域差异,嘉陵江流域地区、汉江—洞庭湖地区预报效果最好,汉江—淮河地区、淮河流域地区、长江三角洲地区预报效果最差,但引入归一化植被指数后预报效果提高最明显,而洞庭湖—鄱阳湖地区虽然模拟效果较好,但预报稳健性较低,交叉验证相关系数降幅达到0.27,这也从侧面说明了长江流域夏季降水分区预报的重要性。

The impact analysis of spring vegetation on the summer precipitation predictability over the Yangtze River Basin

The Rotated Empirical Orthogonal Functions (REOF) and correlation analysis are used to get the summer precipitation forecasting objects, the vegetation and ocean factors over the Yangtze River Basin, based on the 1982-2006 monthly Normalized Difference Vegetation Index (NDVI) from the Global Inventory Monitoring and Modeling Studies (GIMMS) in USA, the 1854-2008 sea surface temperature (SST) data supported by the American National Oceanic and Atmospheric Administration (NOAA) and the 1951-2006 monthly precipitation dataset of 160 stations from the National Climate Center (NCC) of China. The optimal subset regression models (OSR), the cross validation (CV) tests and space reconstruction methods, are respectively introduced to analyze the improvement owing to the incorporation of the vegetation factors, the spring vegetation impacts on the summer precipitation predictability and robustness over the Yangtze River Basin, under the two different situations of ocean factors alone and both ocean and land vegetation included. The results show that: (1) The spring land vegetation is at least as important as the ocean temperature indices. Compared with the pure pre-spring SST forecast models, the predictability of the Yangtze River Basin is obviously improved after the introduction of the pre-spring vegetation factors. The average forecast correlations are increased by about 0.17 from 0.49 to 0.66 (the explained variance is increased by about 60%), especially for the poor prediction areas using SST factors alone such as the Han River-Huai River subarea and the Huai River Basin subarea in which the correlations are raised by about 0.20-0.30 (the explained variance is increased by about 100%). (2) The cross validation results indicate that the pure SST forecast models of the Yangtze River Basin have the low prediction robustness in which the cross validation forecast correlations have the great drop. The introduction of the vegetation NDVI factors can receive the better performance in which the forecast correlations are significantly improved over the middle-lower reaches of the Yangtze River Basin and its south such as the Yangtze River Delta subarea and the Dongting Lake-Poyang Lake subarea. And, (3) there are the obvious predictability difference among the subareas over the Yangtze River Basin, which show the necessity of the subareas forecasts one by one. The best performance areas are the Jialing River Basin subarea and the Han River-Dongting Lake subarea, and the worst ones are the Han River-Huai River subarea, the Huai River Basin subarea and the Yangtze River Delta subarea. In spite of the well simulation results, the Dongting Lake-Poyang Lake subarea shows the weaker robustness for the cross validation tests in which the forecast correlation coefficients are reduced by 0.27.