Uncertainty quantification for ocean biogeochemical models

Predicting climate change necessitates a thorough understanding of marine biogeochemical (BGC) processes and the coupling between marine ecosystems and the global carbon cycle. Ocean BGC models are tools employed for this purpose. However, current ocean models used to simulate and thus better understand the ocean BGC processes are highly uncertain in their parameterization. This work delves into research to quantify uncertainties that arise in ocean BGC models and obtain improved parameters to reduce those uncertainties utilizing the BGC ocean model Regulated Ecosystem Model Version 2.
A Global Sensitivity Analysis (GSA) is performed to identify which parameters most influence the uncertainty of model outputs in a one-dimensional (1-D) configuration at two ocean sites in the North Atlantic (BATS) and the Mediterranean Sea (DYFAMED). This work finds that the grazing parameter, the maximum chlorophyll-to-nitrogen ratio, the photosynthesis parameters, and the chlorophyll degradation rate are significant for BGC simulation.
This dissertation uses ensemble data assimilation to estimate the most important BGC process parameters. First, data assimilation experiments are carried out in a 1-D model using an ensemble Kalman Filter to estimate preselected BGC parameters at BATS and DYFAMED stations. Subsequently, the scope and application of experiments are broadened to a global scale 3-D model by incorporating spatial variations in parameter values. Replacing the default parameter values with the optimal values obtained in this work improves the model outcomes in both 1-D and 3-D configurations. This work underscores the importance of spatially varying parameter optimization and highlights the potential benefits of incorporating spatially varying BGC parameters in regional and global 3-D BGC models. Through such rigorous scientific endeavors, we inch closer to a more coherent understanding of the complex interplay between the ocean BGC processes and the carbon cycle.