Quantitative assessment of parameterization sensitivity and uncertainty in Noah-MP multi-physics ensemble simulations of gross primary productivity across China’s terrestrial ecosystem

Abstract. Understanding the carbon cycle and its interactions with climate systems requires precise simulation of Gross Primary Productivity (GPP). However, achieving this remains challenging due to the inherent complexity of the models. Current research lacks quantification of how uncertainties in physical process parameterization affect GPP simulation across various ecosystems, and the dominant physical processes goberning GPP variability are pooly identified. To address these issues, this study generated a 48-member Noah Land Surface Model with multi-parameterization options (Noah-MP) ensemble by manipulating key physical parameterization schemes. The model was validated using ChinaFlux tower measurements and Penman-Monteith– Leuning Version 2 data. We employed the Sobol’ total sensitivity index to assess the influence of four key physical processes on GPP: radiation transfer, the soil moisture limitation factor for transpiration (β-factor), turbulence, and runoff generation. Results demonstrate that Noah-MP effectively captured GPP's spatiotemporal patterns in Chinese ecosystems but overestimated GPP in forest and cropland during spring and summer. Sensitivity analysis indicates that the β-factor dominates GPP simulations across most of China, while radiation transfer is the primary driver on the Tibetan Plateau. The main difference between the two radiation transfer schemes lies in whether vegetation gaps fraction are considered. On the Tibetan Plateau, where grasslands and shrublands exhibit large canopy gaps, consider it or not could lead to in substantial differences in simulated radiation and consequently in GPP, making GPP highly sensitive to the choice of radiation scheme. Across ecosystems, water-related factors (β-factor and runoff) mainly affect croplands and savannas, radiation transfer dominates grasslands and shrublands, and turbulence is most influential in forests. There are also distinct seasonal patterns: radiation and turbulence dominate in spring and summer, while radiation and β-factor prevail in autumn and winter, especially in arid regions. Based on systematic performance evaluations and sensitivity analyses, this study proposes optimized Noah-MP model configurations for China's terrestrial ecosystems. The radiation transfer scheme considering the three-dimensional canopy structure (option 1) is recommended for grasslands and shrublands. Our findings offer insights for enhancing GPP simulation accuracy in Noah-MP, thereby improving the model’s ability to represent carbon–water dynamics from regional to continental scales.