LSTM和EnKF在农业土壤降雨径流模拟中的应用

打开文本图片集

中图分类号：P333.1 文献标识码：A文章编号：0439-8114（2025）05-0070-10DOI：10.14088/j.cnki.issn0439-8114.2025.05.011开放科学（资源服务）标识码（OSID）：

Application of LSTM and EnKF methods in agricultural soil rainfall-runoff simulation

LINLin 1 ，GAO Zhao-tian1，DINGYi-jia1，HU Xiao-long1，ZHANG Zhong-bin² （1.Schoolof Water Resources and Hydropower Engineering，Wuhan University，Wuhan 43oo72，China; 2.Instituteof Soil Science，ChineseAcademyof Sciences，Nanjing211135，China）

Abstract：Terelatioshipetwenrainfallandrunoffisofgreatsgnificaceforteallcationofwaterresourcsandtheprotetionof waterandlandresousiagriculuralareas，utitisfult todealwithteainfall-unofprossunderdiferentlndueypes insmall watersheds.Thelongshort-termmemorymodel（LSTM）andtheXin'anjiangmodelcombinedwithensembleKalmanfilter （EnKF）technologywereusedtoexplorethesimulatioefectivenessofdata-drivenmachinelearning（ML）modelonrainfal-runoff processunderdiferentlandusepaterns，andthesimulationefectiveness wascomparedwiththatofSWAThydrological model.The estimationefectivenesofEnKFonhdrologicalparametersensemblesinheXin'njangmodelandthepattsoffilestiated parameters werestudied，andterunoffprocessesfordiferenagriculturallndusetypesbasedonthecalibratedparametersweresimulated.Theresultsshowedthattherunoffvaluewaseasiertolearninthecaseofhighrunoffwithaslightlysmallslopeandthelowrunoffprocess ithalargeslope.ThesimulationaccuracyandstabilityoftheSWATmodelwerenotasgodas thoseoftheLSTMmodel， butSWATmodelcouldreflctthelocalsoilhydrologicalconditionstoacertainextent，whichwasconvenientforgeneticanalyis.The EnKFtechologyhadthefunctionsofparameterupdateandparameterestimation，whchcouldoptimizetheunoffsimulatiofectiveness of the Xin'anjiang model.

KeyWords：ainfallrunofsiulation；datadriven;dataassiilation；LSTM;EnKF;Xinaangmodel;landusepae;ptiie forecasting

中国南方小流域的红壤区生态系统面临着严重的水土流失、土壤酸化、肥力退化、季节性干旱及土壤污染等问题[12]，是由于该区域降雨时空分布不均匀以及不合理的开发利用，因此研究区域内气候、植被、土壤等对降雨径流有影响的因素成为土壤生态环境等领域的热门话题

随着科技的不断发展，数据处理与机器学习等技术提高了各类模型的精度与构建效率。（剩余10061字）