The Xiamen Symposium on Marine Environmental Sciences (XMAS 2027)

Biological carbon pump (BCP) is critical for ocean carbon storage, yet its efficiency and future trajectory remain highly uncertain. This uncertainty stems from two primary sources: physics-driven nutrient biases and ecosystem model complexity. Using the South China Sea as a testbed, we optimize four NPZD-type models (1P1Z/2P2Z structures and two grazing formulations) under prescribed physics and nutrient nudging to establish an optimized physical and nutrient baseline. We then drive these optimized NPZD-type models with physical and nutrient forcings from multiple CMIP6 models to assess how much BCP uncertainty is attributable to physics-driven nutrient biases versus ecosystem model complexity. The original CMIP6 ensemble mean underestimates primary production (PP) and export production (EP) in the South China Sea, with substantial inter-model spread. Forcing-replacement experiments show that CMIP6-derived differences in forcing, particularly nitrate, reproduce most of this spread in PP and EP. This demonstrates that the physically controlled nutrient supply is the primary driver of PP and EP uncertainty in oligotrophic systems like the South China Sea, and that improving physical fields is a necessary first step toward narrowing BCP uncertainty. In contrast, while the native CMIP6 ensemble shows a narrow spread in export ratio (e-ratio), forcing-replacement reveals a much larger spread, comparable to that from parameter perturbation. This indicates that forcing and model parameters contribute equally to e-ratio uncertainty. The apparent CMIP6 consensus on e-ratio therefore masks substantial underlying uncertainty arising from both forcing and model diversity. Furthermore, optimized NPZD-type models degrade significantly when driven by CMIP6 forcing, revealing that BGC calibration is tightly coupled to the parent physical model and not portable across forcings. The apparent skill of original CMIP6 models may therefore reflect compensating physics-BGC errors. Reducing BCP uncertainty requires more realistic physical fields (controlling nutrient supply), not just more complex BGC. More broadly, BGC optimization must be evaluated within the physical context for which it was optimized.