Forest carbon storage is an important reference index of forest carbon fixation capacity.It is of great signif-icance to estimate forest carbon storage accurately for terrestrial carbon cycle.Based on the Landsat TM/OLI images and the continuous inventory sample plots during 1987 to 2017 of Shangri-La and the terrain data,the Pearson coef-ficient,Spearman coefficient,Kendall's τ coefficient,distance correlation coefficient and decision tree methods were used to extract the predictive variables,and the environmental variables were combined with random forest(RF)to estimate the carbon storage of Pinus densata in Shangri-La.The results show that:(1)Among the predictive varia-bles selected by different methods,the texture variables such as Skewness and Second Moment are highly correlated with carbon storage of P.densata.(2)The decision tree method is superior to other methods,with the fitted results R2=0.845,RMSE=10.076 t/hm2,rRMSE=29.254％,and P=0.747;(3)The modeling accuracy is improved to some extent after introducing environmental variables.Among all environmental variables,the result of RF model with land surface temperature was best.R2 is increased by 4.80％,RMSE is decreased by 1.71 t/hm2,rRMSE is de-creased by 5.391％,and P is increased by 6.60％.(4)The carbon storage of P.densata in Shangri-La changed obviously in time and space from 1987 to 2017.The carbon storage of P.densata has increased 6.51 million t.Different variables selection methods of predictive variables can affect the accuracy of carbon storage estimation and adding environmental variables into the RF model can improve the accuracy,which can provide reference for the subsequent carbon storage estimation based on remote sensing.
维普
万方数据
