Objectives Net ecosystem productivity (NEP) is a key indicator of carbon sinks in terrestrial ecosystem. To accurately estimate NEP, it is essential to use high-resolution remote sensing images and account for topographic effects, especially in mountainous areas, which have important implications for regional carbon budget and land use structure.

Methods We first calculate NEP based on Carnegie-Ames-Stanford approach (CASA) and geostatistical model of soil respiration (GSMSR) for Hubei Province, then optimize the NEP results by applying two types of topographic correction, including meteorological factor correction and surface area correction. We compare the NEP values before and after correction and analyze their spatiotemporal pattern.

Results The results show that topographic correction can effectively improve the accuracy of NEP calculation. Compared with the data of national flux stations, the mean absolute error decreases from 153.69 gC·m^-2·a^-1 before correction to 150.73 gC·m^-2·a^-1 after correction, and the total NEP of Hubei Province is 21.12 million tons. Furthermore, the previous studies underestimated the carbon sequestration of forest land. After topographic correction, the NEP of forest land increases by about 22% compared with that before correction, followed by grassland, while other land use types showed less increase after topographic correction. Finally, the topographic correction significantly improves the accuracy of NEP in the areas with high altitude and large terrain variations. The spatial autocorrelation trend also decreases after correction, as indicated by the reduction of Moran's I index from 1.24 to 1.18. Moreover, the topographic correction causes a noticeable west-to-east shift of the center of gravity in spring and autumn.

Conclusions Compared with NEP before topographic correction, NEP results after correction have higher accuracy, and the carbon sink distribution of different land use types also differs. But because the influence of terrain on carbon sink estimation is reflected in multiple aspects. Hence, in future work, we will continue to explore the influence of topography on vegetation factors, meteorology, and remote sensing images, to improve the simulation accuracy of the model.