Abstract: [Objective] Due to the inconsistency of spatial resolutions between sample plots and image pixels, in this study the estimation of forest carbon density was conducted by using geostatistics method and multi-resource remote sensing data, which aims to provide reference for the application of MODIS data in the regional estimation of forest carbon density.[Method] Firstly, the spatial block co-simulation algorithm was employed to scale up the sample plots of forest carbon density in You county of Hunan Province from the spatial resolution of 25.8 m×25.8 m to the spatial resolutions of 250 m×250 m, 500 m×500 m and 1 000 m×1 000 m respectively. Then, MODIS images with three spatial resolutions corresponding to those mentioned above, were applied to map forest carbon density for this county using sequential Gaussian co-simulation algorithm. Finally, the best model was applied in the estimation of forest carbon density for Hunan Province.[Result] There were highly similarities for sensitive factors of forest carbon density between Landsat5 and MOD09A1 data, according to theresult of Pearson product moment correlations, the top three sensitive factors were 1/TM3, 1/TM2, 1/TM1 for Landsat5 and 1/Band1, 1/Band4, 1/Band3 for MOD09A1 respectively; compared to the vegetation product of MOD13Q1 and MOD15A2, multi-spectral data such as Landsat5 and MOD09A1 showed great potential in the simulation of forest carbon density with the accuracy of 82.02% and 75.64%, respectively; there is good application for the sequential Gaussian co-simulation algorithms based on the image of MOD09A1, which can be used in the spatial simulation of forest carbon density for Hunan Province with the accuracy of 74.07%.[Conclusion] The spatial block co-simulation algorithm could be used to realize the conversion of spatial resolutions from sample plots to the pixels of MODIS images. It was also found that the MODIS derived maps were more smoothed than those from Landsat5 due to the limitation of spatial resolutions, especially in the terms of capturing the spatial variability of forest carbon density. The adopted method was well suited for regional estimation of forest carbon density based on the combination of forest inventory sample plot data and remotely sensed images, especially for the areas that the plot sizes and images pixels were inconsistent.

Key words: forestry remote sensing, forest resource inventory, multi resource remote sensing, sequential Gaussian block co-simulation, forest carbon density