Based on Landsat 8 multispectral imagery and ground survey data, taking the aboveground carbon density of arbor forest as the research object, the field survey data of aboveground carbon density of arbor forest, Landsat 8 multispectral image and DEM data were used to extract vegetation indices, texture features, principal component transformation factors, cap transformation factors and topographic factors as modeling variables. Pearson correlation coefficient method combined with residual mean square criterion method was used to screen variables. CoKriging interpolation and geographic weighted regression method were used to construct aboveground carbon density of arbor forest. And the estimated effect of the two methods were compared and analyzed. The results showed that the accuracy of the estimated model constructed by the geographic weighted regression method (R2 was 0.74, RMSE was 6.84t/hm2, MAE was 5.13t/hm2, RE was 0.74%), which was superior to the CoKriging interpolation method (R2 was 0.47, RMSE was 9.72t/hm2, MAE was 7.41t/hm2, RE was 012%), and the spatial heterogeneity of the estimated variables was well preserved (CVGWR=0.5372, CVCOK=04968), the geographic weighted regression method can obtain higher estimation accuracy. The research can provide a reference for estimating the aboveground carbon density of arbor forest and other forest parameters of forest at regional or large scale.