Abstract: Using Landsat 8 OLI remote sensing image as data source, combined with the second-class survey data of forest resources in Shangri-La City, stratified sampling was used to estimate the carbon storage of Pinus densata based on remote sensing characteristic variables under 3 conditions of 95% reliability and 95%, 90% and 85% sampling accuracy respectively. The results are compared with those of general stratified sampling and systematic sampling. The results show as follows: In the screening of remote sensing feature variables, 4 with the strongest correlation are successively the information entropy of the fourth band in the 11th window(R11B4EN), the mean of the 4^th band in the 11^th window(R11B4ME), the synergy of the 7^th band in the 11^th window(R11B7HO), and the 2^nd moment in the 2^nd band in the 11^th window(R11B2SM). Under the same sampling design accuracy, the number of sampling units presents the rule of system sampling > general stratified sampling > remote sensing stratified sampling. Under the sampling design accuracy of 95% and 85%, the actual sampling accuracy is the highest and the error is the smallest when using the 11^th window second-band 2^nd moment(R11B2SM). Therefore, stratified sampling based on remote sensing characteristic variables can provide reference for forest carbon storage investigation.