基于机器学习的黄河上游未来不同气候情景下径流预估  

Machine learning-based runoff prediction under future climate scenarios in upper reaches of Yellow River

作　　者：刘军 赵坤 LIU Jun;ZHAO Kun(Ji’nan Water Supply and Drainage Monitoring Center,Ji’nan 250014,China;Ji’nan Shizhong District Urban Investment Group Urban Construction Co.,Ltd.,Ji’nan 250004,China)

机构地区：[1]济南市供排水监测中心,山东济南250014 [2]济南市中城市投资集团城市建设有限公司,山东济南250004

出　　处：《水利信息化》2025年第2期39-44,共6页Water Resources Informatization

摘　　要：随着气候变化与人类活动的加剧,径流序列稳定性遭到破坏,为解决径流准确预估难的问题,收集黄河上游10个气象站的月气象、唐乃亥水文站月径流量、7个CMIP6的GCM输出等数据,利用统计降尺度方法对GCM数据进行降尺度处理,采用支持向量机、多元线性回归和随机森林3种机器学习模型模拟历史径流,经对比分析后选择随机森林模型预估未来不同气候情景下的黄河上游径流变化。结果表明:3种机器学习算法都能够较好地模拟月径流量,但对几处径流峰值的模拟效果较差;在未来远期(2070年后)SSP5-8.5情景下的径流量明显高于SSP2-4.5气候情景。年径流量呈现明显的增加趋势,且SSP5-8.5情景下增加趋势更为显著,增加速率约为1.92m^(3)/(s·a),高于SSP2-4.5情景下的0.94m^(3)/(s·a)。气温升高致使冰雪融化提前发生,黄河流域未来降水增加趋势明显。研究成果可为不同气候情景下径流预估工作提供参考。With the intensification of climate change and human activities,the stability of runoffsequences has been disrupted.To address the challenge of accurately estimating runoff,this study collected monthly meteorological data from ten meteorological stations in the upper reaches of the Yellow River,monthly runoffdata from the Tangnaihai hydrological station,and GCM outputs from seven CMIP6 models.Statistical downscaling methods were applied to process the GCM data,and three machine learning models including support vector machine,multiple linear regression and random forest were used to simulate historical runoff.After comparative analysis,the random forest model was selected to estimate runoffchanges in the upper reaches of the Yellow River under different future climate scenarios.The results showed that all three machine learning algorithms effectively simulate monthly runoffbut perform poorly in capturing several runoff peaks.In the long-term future(after 2070),runoff under the SSP5-8.5 scenario will be significantly higher than under the SSP2-4.5 climate scenario.Annual runoffexhibits a clear increasing trend,which is more pronounced under the SSP5-8.5 scenario,with an increase rate of approximately 1.92 m^(3)/s per year,surpassing the 0.94 m^(3)/s per year observed under the SSP2-4.5 scenario.Rising temperatures lead to earlier melting of ice and snow,while precipitation in the Yellow River Basin is projected to increase significantly.Thesefindings provide a valuable reference for runoffprediction under different climate scenarios.

关 键 词：径流预估 机器学习 气候变化 降水量 黄河上游 

分 类 号：TV121.2[水利工程—水文学及水资源] P338[天文地球—水文科学]