摘要
森林生物量能直接反映森林质量,遥感技术结合地面样地能实现林分或区域范围森林生物量的反演,反演结果对制定森林资源合理利用、生态环境改善等方面的政策具有重要的指导意义。论文以旺业甸林场Landsat 8 OLI影像为数据源,从影像中提取161个植被指数,对比Pearson相关系数法和随机森林法进行特征变量选择,分别筛选出合适的因子作为模型自变量,结合实地调查数据,建立多元线性逐步回归、地理加权回归、kNN回归和随机森林等4种生物量反演模型,并对模型结果进行精度验证。研究结果表明:1)利用Pearson相关系数法进行特征变量选择要优于随机森林法。2)短波红外光和近红外区间波段组合得到的植被指数与生物量的相关性显著,相关性系数最高的前五个因子为SR627、SR637、SR647、SR64、SR213,分别达到了0.776、0.761、0.730、0.702和0.657;3)4种生物量反演模型中,随机森林模型效果最好,决定系数R2为0.72,RMSE=8.12,EA=76.54%;线性逐步回归模型次之,R2为0.65,RMSE=9.01,EA=72.88%;其次是kNN回归模型,R2为0.59,RMSE=9.75,EA=74.89%;地理加权回归模型效果最差,R2为0.58,RMSE=13.75,EA=53.95%;4)利用随机森林模型对研究区进行生物量反演,反演结果生物量空间分布与实际情况基本一致,反演效果较好。
Forest biomass can directly reflect the quality of forests.Remote sensing technology combined with ground sample plots can realize the inversion of forest biomass in forest stands or regions.The inversion results have important guiding significance for formulating policies on rational use of forest resources and improvement of ecological environment.The paper uses the Landsat 8 OLI image of Wangyedian forest as the data source,extracts 161 vegetation indices from the images,compares the Pearson correlation coefficient method and the random forest method to select the characteristic variables,and selects the appropriate factors as the model independent variables.Combined with field data,four kinds of biomass inversion models were established,such as multiple linear stepwise regression,geographically weighted regression,kNN regression and random forests regression which were used to verify the accuracy of the model results.The results showed that:1)Pearson correlation coefficient method is superior to random forest method in feature variable selection.2)The correlation coefficients between biomass and vegetation indexes obtained by combination of shortwave infrared and near-infrared are significant,and the top five factors with the highest correlation coefficient are SR627,SR637,SR647,SR64,SR213,which are reached 0.776,0.761,0.730,0.702 and 0.657.3)Among the four biomass inversion models,the random forest model works best,whose coefficient of determination is 0.72,RMSE=8.12,EA=76.54%.The multiple linear stepwise regression model is the second,whose coefficient of determination is 0.65,RMSE=9.01,EA=72.88%.The third model is the kNN model,whose coefficient of determination is 0.59,RMSE=9.75,EA=74.89%.The geographically weighted regression model is the worst,whose coefficient of determination is 0.58,RMSE=13.75,EA=53.95%;4)The inversion of biomass using random forest model in the study area shows that the spatial distribution of biomass is basically consistent with the actual situation,and the inversion effect is good.
作者
蒋馥根
孙华
林辉
龙江平
蒋治浩
雷思君
JIANG Fugen;SUN Hua;LIN Hui;LONG Jiangping;JIANG Zhihao;LEI Sijun(Research Center of Forestry Remote Sensing&Information Engineering Central South University&Technology,Changsha 410004,Hunan,China;Key Laboratory of Forestry Remote Sensing Based Big Data&Ecological Security for Hunan Province,Changsha 410004,Hunan,China;Key Laboratory of State Forestry&Grassland Administration on Forest Resources Management and Monitoring in Southern Area,Changsha 410004,Hunan,China)
出处
《中南林业科技大学学报》
CAS
CSCD
北大核心
2019年第10期88-94,共7页
Journal of Central South University of Forestry & Technology
基金
“十三五”国家重点研发计划项目“人工林资源监测关键技术研究”(2017YFD0600900)
湖南省教育厅科学研究重点项目(17A225)
湖南省普通高校青年骨干教师培养对象项目(7070220190001)
中南林业科技大学研究生科技创新基金(CX20192025)
