Optimizing PISCES Parameters for North Atlantic Seasonal pCO2 Predictions: An Ensemble-Based Approach Using BGC-Argo Data

Numerical models of ocean biogeochemistry serve as critical tools for detecting and predicting the impacts of climate change on marine resources, and for monitoring ocean health. Recent research by Rodgers et al. (2023) has identified significant limitations in current CMIP6 class models, particularly in their representation of seasonal partial pressure of CO2 (pCO2) temporal phasing and magnitude in key ocean regions such as the North Atlantic. These limitations arise primarily from parameter uncertainty, as model parameters are typically derived from laboratory experiments using a limited range of species that do not represent the diversity of marine organisms. In addition, certain parameters remain experimentally indeterminate, resulting in wide plausible ranges that introduce considerable uncertainty into model predictions.

Our research addresses these challenges through a comprehensive approach to parameter optimisation using ensemble-based data assimilation techniques. In particular, we focus on reducing the systematic bias in the PISCES marine biogeochemical model distributed with NEMO v4.2, simulation of the North Atlantic seasonal pCO2 cycle, while generating robust uncertainty estimates through ensemble methods.

The optimization process began with an extensive sensitivity analysis using SOBOL indices to identify the parameters most influential in controlling seasonal pCO2 dynamics. We then implemented a particle filter algorithm to optimize these key parameters in a NEMO-PISCES 1D configuration using data from a North Atlantic BGC Argo float. The particle filter generated an ensemble of thousands of state variable solutions, each representing a different PISCES parameterization with reference values varying between 0.01 and 2 times their nominal range. From this ensemble, we identified the ten parameter combinations that most effectively minimized the model-data discrepancy. These optimized parameter sets were then used to generate a 3D regional NEMO-PISCES ensemble in the North Atlantic - the ensemble approach providing a robust framework for uncertainty quantification.

Our results show significant improvements in model performance, with the optimized PISCES parameter set in the 1D configuration achieving a 40% reduction in RMSE for seasonal cycle predictions of surface nutrients, chlorophyll, and carbon components compared to the standard PISCES configuration. Most importantly, all ensemble members successfully reproduce seasonal pCO2 phasing and magnitude in agreement with observation-based data, addressing a critical limitation of the reference model while providing uncertainty estimates consistent with observational uncertainties.

This research demonstrates the effectiveness of ensemble-based data assimilation techniques in optimizing biogeochemical model parameters, thereby enhancing the accuracy and reliability of ocean simulations. These improvements significantly strengthen our capacity to monitor ocean health, forecast climate change impacts on marine ecosystems, and provide robust scientific guidance for marine resource management decisions.