基于条件扩散模型的未测量流域径流预测方法

打开文本图片集

中图分类号：TP391.4；TP183 文献标识码：A 文章编号：2096-4706（2025）08-0071-07

Abstract： Deep Learning becomes apowerful tool for runoff prediction，but in ungauged basins，the lack of flow observation data makes model trainingand prediction usuallyrequiretheapproachof Transfer Learning.However，thetarget basinoftendoesnothaveenoughdataforfie-tuning，whichmakes itdiffculttocalibratethemodelparameters.Therefore， this paper proposes anungauged basins runof prediction methodbasedonconditional diffusion model.The method includesa forwardnoisingprocessandareversedenoisingprocessThedenoisingmodelis trainedinthesourebasinandthenthedatais recoveredfromthenoiseintetargetbasinasthepredictionresultInadition，thedenoisingprocesisguidedbytheconditional datancludingmeteorologicaldriversandhistoricalrunoffndtheTrasformerlayerisintroducedintothedenoisingmodelto capture the dependenceof ime andfeatures.Throughthecross-validation experimentontheCAMELS-US dataset，theresults show that the method has superiority.

Keywords： runoff prediction; ungauged basins; Transfer Learning; conditional diffusion model; CAMELS-US

0 引言

径流量能够反映特定流域内水文、王壤和地质特征，是综合反映流域内自然条件和人类活动的重要指标。（剩余7609字）