融合海温偶极因子的长期月径流预报研究  

Long-term Streamflow Prediction with Sea Surface Temperature Dipoles Incorporated

作　　者：刘建华 徐文馨 石昕颜 陈杰[2,3] 胡召根[1] 陈华 LIU Jian-hua;XU Wen-xin;SHI Xin-yan;CHEN Jie;HU Zhao-gen;CHEN Hua(TSQ-1 Hydropower Plant of TSQ-1 Hydro Development Co.,Ltd,Guangzhou 510600,Guangdong Province,China;State Key Labo-ratory of Water Resources and Hydropower Engineering Science,Wuhan University,Wuhan 430072,Hubei Province,China;Hubei Provincial Key Lab of Water System Science for Sponge City Construction,Wuhan University,Wuhan 430072,Hubei Province,China)

机构地区：[1]天生桥一级水电开发有限责任公司水力发电厂,广东广州510600 [2]水资源与水电工程科学国家重点实验室武汉大学,湖北武汉430072 [3]海绵城市建设水系统科学湖北省重点实验室(武汉大学),湖北武汉430072

出　　处：《中国农村水利水电》2023年第9期46-53,共8页China Rural Water and Hydropower

基　　金：国家自然科学基金项目(52079093);湖北省自然科学基金资助项目(2020CFA100)。

摘　　要：可靠的月径流预报可以为水库科学运行与管理提供依据。通常而言,月径流预报模型可分为过程驱动和数据驱动两类。对于数据驱动模型而言,预报模型的选取和其对应的输入数据共同决定了预报的效果。然而,现有研究多集中于模型结构的对比分析,就模型输入数据对预报效果的影响的讨论较少。尽管海温与区域降水、径流的遥相关关系已被广泛证实,当前基于数据驱动模型的月径流预报在选择海温预报因子时通常仅考虑固定海域的海洋表面温度场作为遥相关因子,而忽视了海温场的空间分布特征和关联性。研究以天生桥一级(天一)水库入库径流为例,在考虑水库前期径流和大气环流因子的基础上,将海温偶极因子纳入待选预报因子集,在年内各月分别构建预见期为1~12个月的多元线性回归模型,探索各因子组合方式对预报效果的影响。结果表明:①只使用前期径流因子开展预报时效果较差,但在预见期为1~3个月时,将其与大气环流因子或海温偶极因子结合使用能有效提高两种因子单独使用时的预报精度;②含有海温偶极因子的预报因子组合在预见期较长时的预报效果优于只考虑大气环流因子和考虑径流和大气环流因子的模型,其中,提升效果最为显著的月份为9月和11月,以径流和海温偶极为预报因子的模型对这两个月份在预见期1~12个月的平均精度较以径流和大气环流为预报因子的模型分别提升了7.1%和9.3%。Accurate predictions of monthly streamflow are of great importance for reservoir operation management.The existing approaches for the monthly streamflow predictions can be broadly classified into data-driven and process-based models.Predictive performance of datadriven models is highly sensitive to model structures and input variables.However,most studies focus on comparing model structures,which ignore the relationship between model inputs and monthly streamflows.Despite the growing interest in using sea surface temperature(SST)anomalies over specific areas as predictors in data-driven models for predicting monthly streamflows,spatial distribution patterns and corre⁃lation of sea surface temperature have rarely been considered in previous studies.Therefore,this study investigates the role of SST dipole on monthly streamflow prediction and the influence of different combinations of historical streamflow observations,atmospheric indices and SST dipoles on predictive performance.Different multiple linear regression(MLR)models are developed for the twelve months of the year to pre⁃dict monthly inflows for the Tianyi Reservoir at lead times of 1 to 12 months.Results show that although historical streamflow observations have little value by themselves,they generally improve the predictive performance when used in combination with historical atmospheric indi⁃ces or with SST dipoles for 1-to 3-month-ahead streamflow predictions.MLR models generally perform better when SST dipoles are includ⁃ed as predictors for monthly streamflow prediction at longer lead times.The largest improvements are found for September and November.For example,compared with using historical streamflow observations and atmospheric indices as predictors,streamflow observations and SST dipoles used as predictors have improved the averaged prediction accuracy by 7.1%and 9.3%for the two months with lead times of 1 to 12 months,respectively.

关 键 词：月径流预见期 多元线性回归 大气环流因子 海温偶极模型 因子组合 

分 类 号：P338[天文地球—水文科学]