Forest Biomass Model Of Deciduous Broad-leaved Forest In Mazongling Forest Farm Based On Remote Sensing

As the main body of the terrestrial ecosystems,forest plays a very important role in the biogeochemical process.Deciduous broad-leaved forests as one of the important components of the forest,conducting biomass research to measure forest carbon sequestration capacity and evaluate forest carbon balance is critical.This paper takes the natural secondary deciduous broad-leaved forest in Mazongling Forest Farm of Jinzhai County as the research object and use the typical sampling method to set up plots with different ages,different site conditions and 20 * 20 m in size.The biomass of the plots was calculated by the general hardwood biomass formula.Then,using World View-2 as the remote sensing data source to extract the vegetation index,texture feature and other factors.Combined with the terrain factor as the candidate modeling factor,three machine learning algorithms(RF,k-NN,ANN)were used to construct the remote sensing quantitative inversion model of natural secondary deciduous broad-leaved biomass in Mazongling Forest Farm,and using the determination coefficient R2 and the root mean square error RMSE to measure the prediction ability of the model and select the optimal remote sensing quantitative inversion model.The forest types of Mazongling forest farm were classified by using and comparing three supervised classifiers,and draw the distribution map of natural secondary deciduous broad-leaved biomass in Mazongling Forest Farm and make correlation analysis with altitude,slope and slope position.The main research results of this paper are as follows:(1)The calculation results showed that the total aboveground biomass was between789.03 t / hm2 and 1552.15 t / hm2,and the biomass was in the order of medium diameter >small diameter > large diameter.Among them,the biomass of trunk accounted for 57 % of the total biomass,the biomass of branch accounted for more than 25 % of the total biomass,the biomass of leaf accounted for more than 8 %,and the biomass of bark accounted for more than 4 % of the total biomass.(2)Compare the performance of RF,k-NN and ANN machine learning models by leave-one-out cross validation,ANN model has maximum R2 and minimum RMSE.In ANN model,when decay=0.1,size=2,the model accuracy reached the highest,which is R2=0.69,RMSE=31.53 t / hm2.Finally,ANN model was selected as the remote sensing inversion model of natural secondary deciduous broadleaf forest biomass in Mazongling forest farm.(3)In this paper,by comparing the classification accuracy of random forest method,maximum likelihood method and Mahalanobis distance method,it is concluded that the effect of random forest classification is the best,and the Kappa coefficient is 0.97.Based on ENVI5.3 to achieve random forest classification,and Mazongling forest farm was divided into deciduous broad-leaved forest,coniferous forest,coniferous and broad-leaved mixed forest and non-forest land.(4)The AGB inversion of World View-2 images covered by natural secondary deciduous broad-leaved forest in Mazongling Forest Farm was conducted by using the ANN optimal model.The estimated biomass was 90.34 t / hm2 in average,the standard deviation was 47.96 t / hm2,and the total biomass was 336215.57 t / hm2.Overall,the aboveground biomass of natural deciduous broad-leaved forest in Mazongling forest farm is mainly distributed in Lingtou area and Heshangping area,followed by Dacaoping area and Donggaoshan area,and the least distributed area is Qianping village.(5)Based on DEM data,the biomass of deciduous broad-leaved forest in Mazongling Forest Farm accounts for 72.37% in the area of 1450-1660 m above sea level,22.63% in other areas;biomass with a slope of 26-35° accounts for 34.57%,others It is 65.43%;the biomass of the southwest slope is 78.70%,and the others are 21.3%.