Abstract: Accurate estimate of gross primary productivity (GPP) is the basis for the modeling of terrestrial ecosystem carbon cycle and climate change projection. We evaluated the performance of CLM4.5 simulated GPP and analyzed the causes of errors using observed GPP data from FLUXNET, combined with  leaf area index (LAI） data. We conducted a sensitivity analysis to examine the key parameters in simulating GPP in CLM4.5 and discussed potential ways to improve the accuracy of photosynthesis simulation. Our results showed that CLM4.5 was better than CLM4 in simulating monthly and annual GPP. The mean absolute error (MAE) for the monthly and annual simulated GPP in CLM4.5 across plant functional types was reduced by 15% and 29%, respectively. However, CLM4.5 still presented a large bias in annual GPP with a MAE of 366.06 g C·m^-2·a^-1. The annual GPP bias had different seasonal variation for different plant functional types, with the bias mainly occurring in spring and summer. The GPP simulated by CLM4.5 was most sensitive to three parameters, i.e. specific leaf area, leaf carbontonitrogen ratio, and the fraction of nitrogen in Rubisco. The improvement in GPP simulation requires better modeling in phenology and leaf area index, P dynamics and C N P interactions, and a ecosystem-level photosynthetic parameter dataset.

Key words: Fanhe River watershed, boosted regression tree, soil and water assessment tool, quantitative analysis.