Recently, light use efficiency (LUE) models have been widely used for calculating gross primary productivity (GPP). Unfortunately, these models are only suitable for flat areas without considering the topographic effects on plant photosynthesis. In this study, considering the topographic controls on direct radiation, diffuse radiation, and sunlit canopy area, an adjusted model called mountainous two‐leaf (MTL)‐LUE is developed from the original two‐leaf LUE model (TL‐LUE) to improve the GPP simulations in mountainous areas. The MTL‐LUE model was validated at an eddy covariance (EC) site with apparent topography in the carbon footprint area. Results showed that the daily GPP simulations from MTL‐LUE (root mean square error = 1.29 gC m⁻² d⁻¹) matched better with EC GPP than those from the TL‐LUE model (root mean square error = 1.84 gC m⁻² d⁻¹), confirming the improvement of the MTL‐LUE model in mountainous areas. Moreover, results also indicated that the MTL‐LUE‐simulated GPP showed obvious spatial variations in the study area, suggesting that the MTL‐LUE model could characterize the effects of topography on plant photosynthesis. More specifically, canopy GPP was found to decrease when slope increased in shaded terrains and has no obvious variations in different slope ranges when exposed to the sun. Furthermore, results also indicated that a moderate increase in canopy GPP would be expected when the sky clearness decreased, which depended on the gap between the GPP increase of shaded leaves and the GPP decrease of sunlit leaves. This study highlights the potential of the MTL‐LUE model in improving GPP simulations in mountainous areas.

Land surface models intended for large‐scale applications are often executed at coarse resolutions, and the sub‐grid heterogeneity is usually ignored. Here, a spatial scaling algorithm that integrates the information of vegetation heterogeneity (land cover type and leaf area index) and surface topography (elevation, slope, relative azimuth (Raz) between the sun and the slope background, sky‐view factor, and topographic wetness index), was proposed to correct errors in gross primary productivity (GPP) estimates at a coarse spatial resolution. An eco‐hydrological model named BEPS‐TerrainLab was used to simulate GPP at 30 and 480 m resolutions for 16 mountainous watersheds selected globally. Results showed that an obvious improvement on GPP estimates at 480 m resolution was achieved after the correction in comparison with GPP modeled at 30 m resolution, with the determination coefficient increased by 0.38 and mean bias error reduced by 203gCm⁻² yr⁻¹. The combination of all the seven factors made the largest improvement for GPP estimation at 480 m resolution, suggesting that a larger improvement would be achieved when more factors of surface heterogeneity are considered. More specifically, our results indicated that five factors, including land cover type and leaf area index regarded as integrated outcomes of all the environmental conditions, Raz and sky‐view factor associated with radiation redistribution, and slope related to soil water redistribution, were especially important in the spatial scaling procedure. This study suggests that incorporating the information of surface heterogeneity into the spatial scaling algorithm is useful for improving coarse resolution GPP estimates over mountainous areas.

Due to the spatial heterogeneity of land surfaces, downscaling is an important issue in the development of carbon cycle models when evaluating the role of ecosystems in the global carbon cycle. In this study, a downscaling algorithm was developed to model gross primary productivity (GPP) at 500 m in a time series over rugged terrain, which considered the effects of spatial heterogeneity on carbon flux simulations. This work was carried out for a mountainous area with an altitude ranging from 2606 to 4744 m over the Gongga Mountain (Sichuan Province, China). In addition, the Moderate Resolution Imaging Spectroradiometer (MODIS) GPP product at 1 km served as the primary dataset for the downscaling algorithm, and the 500 m MODIS GPP product was used as the reference dataset to evaluate the downscaled GPP results. Moreover, in order to illustrate the advantages and benefits of the proposed downscaling method, the downscaled results in this work, along with ordinary kriging downscaled results, spline downscaled results and inverse distance weighted (IDW) downscaled results, were compared to the MODIS GPP at 500 m. The results showed that (1) the GPP difference between the 500 m MODIS GPP and the proposed downscaled GPP results was primarily in the range of [−1, 1], showing that both vegetation heterogeneity factors (i.e., LAI) and topographic factors (i.e., altitude, slope and aspect) were useful for GPP downscaling; (2) the proposed downscaled results (R² = 0.89, RMSE = 1.03) had a stronger consistency with the 500 m MODIS GPP than those of the ordinary kriging downscaled results (R² = 0.43, RMSE = 1.36), the spline downscaled results (R² = 0.40, RMSE = 1.50) and the IDW downscaled results (R² = 0.42, RMSE = 1.10) for all Julian days; and (3) the inconsistency between MODIS GPP at 500 m and 1 km increased with the increase in altitude and slope. The proposed downscaling algorithm could provide a reference when considering the effects of spatial heterogeneity on carbon flux simulations and retrieving other fine resolution ecological-physiology parameters (e.g., net primary productivity and evaporation) over topographically complex terrains.

Vegetation biophysical products offer unique opportunities to examine long-term vegetation dynamics and land surface phenology (LSP). It is important to understand the time-series performances of various global biophysical products for global change research. However, few endeavors have been dedicated to assessing the performances of long-term change characteristics or LSP extraction derived from different satellite products, especially in mountainous areas with highly fragmented and rugged surfaces. In this paper, we assessed the time-series characteristics and LSP detections of Global LAnd Surface Satellite (GLASS) leaf area index (LAI), fractional vegetation cover (FVC), and gross primary production (GPP) products across the Three-River Source Region (TRSR). The performances of products’ temporal agreements and their statistical relationship as a function of topographic indices and heterogeneous pixels, respectively, were investigated through intercomparison among three products during the period 2000 to 2018. The results show that the phenological differences between FVC and two other products are beyond 10 days over more than 35% of the pixels in TRSR. The long-term trend of FVC diverges significantly from GPP and LAI for 13.96% of the total pixels, and the percentages of mismatched pixels between FVC and two other products are 33.24% in the correlation comparison. Moreover, good agreements are observed between GPP and LAI, both in terms of LSP and interannual variations. Finally, the LSP and long-term dynamics of the three products exhibit poor performances on heterogeneous surfaces and complex topographic areas, which reflects the potential impacts of environmental factors and algorithmic imperfections on the quality and performances of different products. Our study highlights the spatiotemporal disparities in detections of surface vegetation activity in mountainous areas by using different biophysical products. Future global change studies may require multiple high-quality satellite products with long-term stability as data support.