Large-scale estimation and uncertainty analysis of gross primary production in Tibetan alpine grasslands

Gross primary production (GPP) is an important parameter for carbon cycle and climate change research. Previous estimations of GPP on the Tibetan Plateau were usually reported without quantitative uncertainty analyses. This study sought to quantify the uncertainty and its partitioning in GPP estimation across Tibetan alpine grasslands during 2003-2008 with the modified Vegetation Photosynthesis Model (VPM). Monte Carlo analysis was used to provide a quantitative assessment of the uncertainty in model simulations, and Sobol' variance decomposition method was applied to determine the relative contribution of each source of uncertainty to the total uncertainty. The results showed that the modified VPM successfully reproduced the seasonal dynamics and magnitude of GPP of 10 flux tower sites on the plateau (R-2=0.77-0.95, p < 0.001). The 6 year mean GPP in Tibetan alpine grasslands was estimated at 223.3 Tg C yr(-1) (312.3g C m(-2)yr(-1)). The mean annual GPP increased from western to eastern plateau, with the increase of annual temperature and precipitation and the decrease of elevation, while the decrease of GPP from southern to northern plateau was primarily driven by air temperature. Furthermore, the mean relative uncertainty of the annual GPP was 18.30%, with larger uncertainty occurring in regions with lower GPP. Photosynthetic active radiation, enhanced vegetation index, and the maximum light use efficiency (LUE) are the primary sources of uncertainty in GPP estimation, contributing 36.84%, 26.86%, and 21.99%, respectively. This emphasizes the importance of uncertainty in driving variables as well as that of maximum LUE in LUE model simulation.