基于不同特征选择方法的森林地上生物量估测研究——以平江县芦头林场为例  

Estimation of Forest Above-Ground Biomass Based on Different Feature Selection Methods——Taking Lutou Forest Farm in Pingjiang County,as an Example

作　　者：李富宇 龙江平 Li Fuyu;Long Jiangping(College of Forestry,Central South University of Forestry and Technology,Changsha 410004,China,Hunan)

机构地区：[1]中南林业科技大学林学院,湖南长沙410004

出　　处：《绿色科技》2024年第16期94-100,共7页Journal of Green Science and Technology

基　　金：湖南省科技厅重点领域研发计划(编号:2020NK2051)。

摘　　要：森林地上生物量(forest aboveground biomass,AGB)是森林生物量的主要组成部分,是森林资源监测的重要指标之一。以湖南省平江县的中南林业科技大学芦头实验林场为研究区(以下简称芦头林场),结合2021年采集的50个样地调查数据,基于Sentinel-1 SAR数据与DEM数据,提取后向散射系数、纹理与地形因子作为自变量,结合前向特征筛选,使用随机森林(random forest,RF)算法和极致梯度提升树(XGBoost,XGB)算法建立研究区森林AGB估测模型。结果表明:整体上XGB模型优于RF模型。使用全部特征的RF和XGB模型的R^(2)分别为0.17和0.25,RMSE分别为49.50 t/hm^(2)和51.91 t/hm^(2)。在特征筛选后,模型的估测精度均有提升:使用RF特征重要性的特征筛选的模型R^(2)为0.26~0.31,RMSE为47.37~49.10 t/hm^(2);使用XGB特征重要性的方法更优,R^(2)为0.34~0.42,RMSE为43.48~46.19 t/hm^(2)。研究区的最优模型为基于XGB特征重要性的XGB模型,R^(2)为0.42,RMSE为43.48 t/hm^(2),研究区的森林AGB总量为8.51×10~5 t。特征筛选后的森林AGB估测模型精度有所提升,可为林业部门在监测森林资源等方面提供重要支持。Forest Aboveground Biomass(AGB) is one of the main components of forest biomass and an important index of forest resource monitoring.This study took the Lutou Experimental Forest Farm of Central South University of Forestry and Technology in Pingjiang County,Hunan Province as the research area(hereinafter referred to as Lutou Forest Farm).Combined with the sample survey data,based on Sentinel-1 SAR data and DEM data,the backscattering coefficient,texture and terrain factors were extracted as independent variables.Combined with the forward feature screening,Random Forest(RF) algorithm and XGBoost(XGB) algorithm were used to establish the forest AGB estimation model in the study area.The results show that the XGB model is better than the RF model.The R~2 and RMSE of RF and XGB models using all features were 0.17 and 0.25,respectively,and 49.50 and 51.91 t/hm~2,respectively.After feature screening,the estimation accuracy of the models was improved.R~2 and RMSE were 0.26~0.31 and 47.37~49.10 t/hm~2 for the models with RF feature importance screening.The method using XGB feature importance is better,with R~2 of 0.34~0.42 and RMSE of 43.48~46.19 t/hm~2.The optimal model in the study area was XGB model based on the importance of XGB features.R~2 was 0.34.RMSE was 43.48 t/hm~2,and the total forest AGB in the study area was 8.51×10~5 t.The accuracy of forest AGB estimation model after feature screening has been improved,which can provide important support for forestry departments in monitoring forest resources.

关 键 词：森林地上生物量 Sentinel-1 随机森林 XGB 特征选择 

分 类 号：S718[农业科学—林学]