第５２卷第４期２０２３年８月湖北林业科技HubeiForestryScienceandTechnologyVol．５２,No．４Aug．,２０２３∗收稿日期:２０２２Ｇ１２Ｇ１３作者简介:邹泽林(１９９１~),男,工程师,主要从事森林资源监测和林业信息化应用等工作.Stacking集成学习模型在森林蓄积量估测中的应用邹泽林程霞刘紫薇黄鑫(国家林业和草原局中南调查规划院长沙４１００１４)摘要:为了探索SentinelＧ２遥感数据在估测森林蓄积量中的适用性,以及开发一种提高蓄积量估测精度的集成学习算法,选择江西省兴国县为研究区,以SentinelＧ２为遥感数据源,利用Boruta算法进行特征选择后开发了一种Stacking集成学习模型,并且与MLR、KNN、SVR和RF四种基础模型进行对比.结果表明,相比于MLR模型,机器学习模型具有更强的蓄积量估测能力,利用机器学习模型估计森林蓄积量的RMSE降低了１８．０２~２２．５０m３hm－２,rRMSE降低了９．０１％~１１．２５％.并且相比于基础模型,利用Stacking算法将４种模型进行集成后,模型的均方根误差进一步降低了１１．９５~７．４７m３hm－２,说明Stacking集成学习算法可以有效的提高森林蓄积量的估测性能.关键词:森林蓄积量;SentinelＧ２;集成学习;遥感估测中图分类号:S７７１．８文献标识码:A文章编号:１００４－３０２０(２０２３)０４－００５２－０６ApplicationofStackingEnsembleLearningModelinForestVolumeEstimationZouZelinChengXiaLiuZiweiHuangXin(CentralSouthInvestigationandPlanningInstituteofNationalForestryandGrasslandAdministrationChangsha４１００１４)Abstract:InordertoexploretheapplicabilityofSentinelＧ２remotesensingdatainestimatingforeststockandtodevelopanintegratedlearningalgorithmtoimprovetheaccuracyofstockestimation,XingguoCounty,JiangxiProvincewasseＧlectedstudyarea,usesSentinelＧ２astheremotesensingdatasource,developsastackingintegratedlearningmodelusingBorutaalgorithmforfeatureselection,andcomparesitwithfourbasicmodels,namelyMLR,KNN,SVRandRF．TheresultsshowedthatcomparedwiththeMLR,KNN,SVRandRFmodels,thestackingintegratedlearningmodelwasmoreefficientthantheMLRmodel．ThemachinelearningmodelhasstrongeraccumulationestimationabilitythantheMLRmodel,andtheRMSEofforestaccumulationestimationusingthemachinelearningmodelwasreducedby１８．０２~２２．５０m３h...
