基于ECC方法的青藏高原冬季气温可预报性研究

Predictability to the Winter Temperature over the Qinghai-Xizang Plateau based on ECC

选取前期9、6和3个月欧亚大陆地表温度、东北半球500 h Pa高度场、热带印度洋海表面温度和西太平洋海表面温度作为预报因子,使用变形的典型相关分析(BP-CCA)方法,并选取各因子预报效果最好的时期作为关键时期,建立起各因子和青藏高原冬季气温之间的统计降尺度模型。之后用交叉验证和集合典型相关分析(ECC)方法评估模型实际预报能力。进一步用独立样本检验来评估模型更长时间尺度的年际变化预测效果。结果表明,BP-CCA方法能很好地识别出不同因子影响青藏高原的空间模态。其中,温度积雪反照率的正反馈机制体现了欧亚大陆地表温度的可预报性;东北半球500 h Pa高度场环流型不利于高纬的冷空气入侵高原地区;热带印度洋海表面温度反映出典型的印度洋偶极子对高原气温的调控作用;西太平洋海表面温度通过控制副热带高压的位置,从而影响高原冬季气温。各因子预报场和观测场的相关系数在交叉检验和独立样本检验中分别约为0.5和0.3,均有一定的预报技巧。而利用ECC方法能综合各因子所提供的预报信息,从而得出更为可信和稳定的预报。

Eurasian continent surface temperature（EA）,Northeastern hemisphere 500 h Pa geopotential height（NH）,sea surface temperature over the tropical Indian Ocean（TI）and sea surface temperature over the Western Pacific（WP）were considered as predictors at 9,6 and 3months lead times.The Barnett-Preisendorfer canonical correlation analysis（BP-CCA）was used to establish statistical downscaling models between four predictors and the winter temperature over the Qinghai-Xizang Plateau（QXP）,and the most skillful period of each predictor was selected as the critical period.Then one year out cross validation and ensemble canonical correlation（ECC）was applied to evaluate the practical model prediction ability.The independent sample test was used to estimate the predictability of interannual variation with a longer time scale.The main resultsare as follows：（1）The spatial patterns of how different predictors influence the QXP are well recognized by BP-CCA.The predictability of EA was reflected by the positive feedback of temperature-snow-albedo.The NH pattern shows a cyclonic anomaly and the East Asia trough is relative weak,which prevents the cold air from high latitudes.TI reveals the impact of canonical Indian Ocean Dipole（IOD）on temperature over the QXP,more moisture was brought to the QXP region.WP controls the winter temperature over the QXP by changing the position of subtropical high.（2）Correlation coefficient between predictand and observation,in the cross validation and independent sample test,are about 0.5 and 0.3 respectively.This indicates that all the four predictors have a good predictable skill to the winter temperature over the QXP.（3）The ECC method synthesizes different predictable information from all the four predictors,which increase the correlation coefficient to 0.7 in the eastern QXP.It means that ECC can give a more reliable and stable prediction.