Estimation of Forest Above-Ground Biomass Based on Different Feature Selection Methods——Taking Lutou Forest Farm in Pingjiang County,as an Example
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 Uni-versity 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 estab-lish the forest AGB estimation model in the study area.The results show that the XGB model is better than the RF model.The R2 and RMSE of RF and XGB models using all features were 0.17 and 0.25,respectively,and 49.50 and 51.91 t/hm2,respectively.After feature screening,the estimation accuracy of the models was improved.R2 and RMSE were 0.26～0.31 and 47.37～49.10 t/hm2 for the models with RF feature importance screening.The method using XGB feature importance is better,with R2 of 0.34～0.42 and RMSE of 43.48～46.19 t/hm2.The optimal model in the study area was XGB model based on the importance of XGB features.R2 was 0.34.RMSE was 43.48 t/hm2,and the total forest AGB in the study area was 8.5 1× 105 t.The accuracy of forest AGB estimation model after feature screening has been improved,which can provide important support for forestry departments in moni-toring forest resources.
NETL
NSTL
万方数据
