基于机器学习算法的森林生物量多源遥感估测OACSTPCD

In order to further explore the estimation performance of forest biomass at 10 m×10 m geoscale and the estimation effect of collaborative machine learning algorithms(RF,SVM,DT,GBM,k-NN,Stac-king)of the images with different spatial resolutions,this study used optical remote sensing GF2(Gaofen-2 satellite),sentinel 2A,Landsat 8 OLI,SUM(integration of three remote sensing data sources)images and auxiliary variables DEM elevation data,environmental factors,forest stand factors(forest type,domi-nant tree species).Under the variable selection of Boruta algorithm,the machine learning algorithm was used to estimate the forest biomass(above ground+under ground)in Yuanmou area,and the estimation accuracy under the four images was compared.The results showed that 1)based on the Boruta algorithm,the vegetation index PEIP of sentinel2A,the texture factor of Landsat 8 OLI b2_ME_9 X 9 and the GNDVI of GF2 in a single image were the highest score variables under the three images,and the GNDVI of GF2 in forest biomass was estimated by multi-source fusion as the best score variable.2)Based on the variables selected by the Boruta algorithm,RF,SVM,DT,GBM,k-NN algorithm and stacking integration algorithm of 5 models,SUM's Stacking integration algorithm had the best estimation effect,model R2 was 0.73,RMSE was 28.46 t·hm-2,the estimation performance of SUM under the ensemble algorithm was better than sentinel2A and Landsat 8 OLI,GF2 was better than sentinel2A,sentinel2A outperformed Landsat 8 OLI,indicating that high-resolution images have good estimation effects in biomass remote sensing estima-tion,and multi-source remote sensing collaborative estimation and integrated algorithms can improve the accuracy of forest biomass remote sensing estimation,which can provide references for forest biomass re-mote sensing estimation.