摘要
为探究Landsat8 OLI反演蓄积量的潜力,研究不同特征选择方法对蓄积量反演精度及不同蓄积量反演模型对反演精度的影响。以湖南省怀化市排牙山国有林场作为研究区,森林资源二类调查数据作为样地地面实测数据,选用Landsat8 OLI作为遥感数据源,将传统的Pearson相关系数法及主成分分析法2种方法结合,得到一种顾及变量相关性的主成分分析法(PCA-P)对遥感变量进行降维。使用3种变量选择方法构建了随机森林(RF)、K最近邻(KNN)、支持向量机(SVR)、多元线性回归(MLR)模型进行森林蓄积量的估测,使用决定系数(R^(2))、均方根误差(R_(MSE))、相对均方根误差(R_(RMSE))对蓄积量估测模型进行精度评价。结果表明:通过Pearson相关系数结合方差膨胀因子得到I_(B2)、I_(ND25)、I_(MSR)3个遥感变量,其与蓄积量相关性分别为0.716、0.623、0.597。使用主成分分析法得到前3个主成分,累计贡献率为93.42%。通过PCA-P得到前2个主成分,累计贡献率为89.99%。使用PCA-P筛选变量并构建的随机森林模型取得了最佳效果,其决定系数为0.59,精度达到77.9%。
In order to explore the potential of Landsat8 OLI accumulation inversion,the influences of different feature selection methods on the accuracy of accumulation inversion and different accumulation inversion models on the accuracy of inversion were studied.Taking Paiyashan National Forest Farm in Huaihua City of Hunan Province as the study area,the forest resources secondary survey data as the ground measurement data of the sample site,and Landsat8 OLI as the remote sensing data source.Combining Pearson correlation coefficient method and principal component analysis method,a principal component analysis method(PCA-P)is proposed to reduce dimension of remote sensing variables considering correlation of variables.Three variable selection methods were used to construct random forest(RF),K-nearest Neighbor(KNN),support vector machine(SVR)and multiple linear regression(MLR)models to estimate forest stock.The determination coefficient(R^(2)),root mean square error(R_(MSE))and relative root mean square error(R_(RMSE))were used to evaluate the accuracy of the estimation model.The results show that three remote sensing variables,B2,ND25 and MSR are obtained by Pearson correlation coefficient combined with variance inflation factor,and their correlations with stock are 0.716,0.623 and 0.597,respectively.Principal component analysis was used to obtain the first three principal components,with a cumulative contribution rate of 93.42%.The first two principal components were obtained by PCA-P,with a cumulative contribution rate of 89.99%.PCA-P was used to screen variables and construct the random forest model,which achieved the best effect with a determination coefficient of 0.59 and an accuracy of 77.9%.
作者
崔博文
佘济云
张廷琛
刘兆华
王潇
Cui Bowen;She Jiyun;Zhang Tingchen;Liu Zhaohua;Wang Xiao(Central South University of Forestry&Technology,Changsha 410004,P.R.China)
出处
《东北林业大学学报》
CAS
CSCD
北大核心
2022年第2期29-34,69,共7页
Journal of Northeast Forestry University
基金
国家林业公益性行业专项(201504301)。
关键词
主成分分析
Pearson相关系数
机器学习
森林蓄积量
Principal component analysis(PCA)
Pearson correlation coefficient
Machine learning
Forest stock