Parameter sensitivity analysis and optimization of carbon and water fluxes in grassland ecosystems based on the Biome-BGCMuSo Model

Accurate quantification of carbon and water fluxes dynamics in arid and semi-arid ecosystems is a critical scientific challenge for regional carbon neutrality assessments and sustainable water resource management. In this study, we developed a multi-flux global sensitivity discriminant index (D_sen) by integrating the Biome-BGCMuSo model with eddy covariance flux observations. This index was combined with a Bayesian optimization algorithm to conduct parameter optimization. The results demonstrated that: (1) Sensitivity analysis identified 13 highly sensitive parameters affecting carbon and water fluxes. Among these, the canopy light extinction coefficient (k) and the fraction of leaf N in Rubisco (FLNR) exhibited significantly higher sensitivity to carbon fluxes (GPP, NEE, Reco; D_sen > 10%) compared to water flux (ET). This highlights the strong dependence of carbon cycle simulations on vegetation physiological parameters. (2) The Bayesian optimization framework efficiently converged 30 parameter spaces within 50 iterations, markedly improving carbon fluxes simulation accuracy. The Kling-Gupta efficiency (KGE) values for Gross Primary Production (GPP), Net Ecosystem Exchange (NEE), and Total Respiration (Reco) increased by 44.94%, 69.23% and 123%, respectively. The optimization prioritized highly sensitive parameters, underscoring the necessity of parameter sensitivity stratification. (3) The optimized model effectively reproduced carbon sink characteristics in mountain meadows during the growing season (cumulative NEE = −375 g C/m²). It revealed synergistic carbon-water fluxes interactions governed by coupled photosynthesis-stomatal pathways and identified substrate supply limitations on heterotrophic respiration. This study proposes a novel multi-flux sensitivity index and an efficient optimization framework, elucidating the coupling mechanisms between vegetation physiological regulation (k, FLNR) and environmental stressors (VPD, SWD) in carbon-water cycles. The methodology offers a practical approach for arid ecosystem model optimization and provides theoretical insights for grassland management through canopy structure regulation and water-use efficiency enhancement.