UQ in Ocean Biogeochemical Models

UQ in Ocean Biogeochemical Models#

The ocean biogeochemical (OBGC) processes play a central role in shaping the earth’s climate by absorbing and sequestering atmospheric carbon dioxide (CO\(_2\)). To improve our ability to predict the climate, it is very important to understand the biogeochemistry of the ocean. Numerical OBGC models are a powerful tool for investigating ocean biogeochemistry and its effects on the carbon cycle. Therefore, including an OBGC component in the Earth System Modeling efforts is essential for climate simulation and predictions. OBCG model outputs are increasingly used by scientists and policymakers when assessing the impacts of climate change. In addition, OBCG models, along with reanalysis datasets, are used for different purposes such as in the development of marine environmental applications and services [Jones et al., 2016] monitoring algal blooms and the movement of fish populations [Brasseur et al., 2009]. However, current OBGC models used to simulate and thus better understand the marine ecosystem processes are associated with large undefined uncertainties. The uncertainties are unavoidable, but UQ of the models can make the models more useful and mitigate the discrepancies in the model outputs.

Parameter Uncertainty#

OBGC models are generally a set of nonlinear equations of marine physical, biogeochemical, and ecological processes [Fennel et al., 2022, Franks, 2002] that are translated into computer code, with each equation expressing how each component of the model (e.g., the biomass of phytoplankton) changes with time due to the hydrodynamical effects (e.g., ocean circulation and mixing) and to fluxes between the various components of the marine ecosystem. The evaluation of the fluxes involves a wide variety of complex biological and chemical processes, which are converted to simplified schemes in models, a methodology known as parameterization. Hence, OBGC models have numerous parameters whose values are poorly known. The available reference values were usually estimated only from a limited field or laboratory experiments but not in the ocean basin of interest. The uncertainty of these parameter values is large [Schartau et al., 2017], translating into possibly significant uncertainty in the model outputs.


We use the OBGC model Regulated Ecosystem Model 2 (REcoM2) [Hauck et al., 2013] to assess the uncertainty of the model fields and the parameters. REcoM2 describes two phytoplankton classes, diatoms, and nanophytoplankton, with an implicit representation of calcifiers and a generic heterotrophic zooplankton class (Figure Fig. 83). It has one class of organic sinking particles whose sinking speed increases with depth. REcoM2 simulates 22 passive tracers (see Figure Fig. 83).

Schematic diagram of the BGC model REcoM2

Fig. 83 Schematic diagram of the BGC model REcoM2. The abbreviations are for the 22 passive tracers – dissolved inorganic carbon (DIC) and alkalinity (ALK) for the carbonate system; the macro-nutrients dissolved inorganic nitrogen (DIN) and silicic acid (DSi); the trace metal dissolved iron (DFe), nanophytoplankton biomass content of carbon (NanoC), nitrogen (NanoN), calcium carbonate (NanoCaCO\(_3\)) and chlorophyll (NanoChl); diatoms biomass content of carbon (DiaC), nitrogen (DiaN), silica (DSi) and chlorophyll (DiaChl); zooplankton biomass content of carbon (ZooC), nitrogen (ZooN); detritus content of carbon (DetC), nitrogen (DetN), silicate (DetSi) and calcium carbonate (DetCaCO\(_3\)); extra-cellular dissolved organic carbon (DOC) and nitrogen (DON); and dissolved oxygen (DO\textsubscript{2}). Arrows depict source and sink terms. The elements (e.g., N) flow from the inorganic nutrients to phytoplankton, to zooplankton, produce nonliving organic matter detritus, and remineralize back to the inorganic pool.#

We use ensemble data assimilation (EnDA) (see [Schartau et al., 2017] for a review) methods to estimate model parameters and state variables. In the ensemble technique, the inputs parameters are varied to get an ensemble of simulations which is used to represent the most likely outcome with a quantification of its uncertainty. In addition, the correlation of the different variables can be estimated from the distribution of the ensemble members. We implement the EnDA in the Parallel Data Assimilation Framework - PDAF [Nerger and Hiller, 2013], an open-source software package developed to implement parallelized data assimilation applications. To identify the parameters whose uncertainty has the largest impact on the variability of prediction, thus to better understand the model itself, we implement a global sensitivity analysis (GSA). In this project we consider Sobol sensitivity indices based GSA. The parameter estimation and sensitivity analysis are interrelated but can be done independently, so is in this projects.

The usefulness of any model mainly depends on the uncertainties of its output data. Therefore, a rigorous skill assessment is required to understand the ability of OBGC models to represent ocean processes and distributions. Skill assessment for OBGC data assimilation is more complex than for free-run models. There are more types of information (free-run model, data, and assimilation model) that should be inter-compared. We apply graphical and statistical analysis (e.g., Continuous Ranked Probability Score (CRPS), Root Mean Square Difference (RMSD), and bias) to assess the skill of OBGC data assimilation. Finally, we will study the effect of estimated spatially and temporally varying parameters on the biogeochemical fields and dynamics by finding meaningful structure, explanatory underlying processes, generative features, and indentifying corelations between parameters and state variables for improved understading on OBGC process and their representation in the models.


The most widely available observational data for ocean biogeochemistry are ocean color observations. The color of the ocean is set by incident light interacting with constituents (both dissolved, living, and non-living particles) in the water. “Ocean color data” is the satellite remote sensing of the light reflected from the ocean. This surface reflectance can be used to determine concentrations of chlorophyll-a, a photosynthetic pigment found in phytoplankton cells. As the data is collected by satellite remote sensing technique, ocean color data have good spatial and temporal coverage for assimilation.

One limitation of ocean color assimilation is that ocean color observations are available only for the ocean’s upper layers. Simultaneously assimilating remotely sensed and in situ biogeochemical profiles can be a way forward. Biogeochemical-Argo (BGC-Argo) floats data are promising. Argo [Johnson et al., 2022] is an international program that collects information from inside the ocean using a fleet of robotic instruments (floats) with measurement sensors that drift with the ocean currents and move up and down between the surface and a mid-water level. BGC-Argo is a particular extension of the Argo program, where floats are equipped with biogeochemical sensors.



P. Brasseur, N. Gruber, R. Barciela, K. Brander, M. Doron, A. El Moussaoui, A. J. Hobday, M. Huret, A. S. Kremeur, P. Lehodey, R. Matear, C. Moulin, R. Murtugudde, I. Senina, and E. Svendsen. Integrating biogeochemistry and ecology into ocean data assimilation systems. Oceanography, 22(3):206–215, 2009. URL: <Go to ISI>://WOS:000269506500024, doi:DOI 10.5670/oceanog.2009.80.


Katja Fennel, Jann Paul Mattern, Scott C. Doney, Laurent Bopp, Andrew M. Moore, Bin Wang, and Liuqian Yu. Ocean biogeochemical modelling. Nature Reviews Methods Primers, 2022. doi:10.1038/s43586-022-00154-2.


P. J. S. Franks. Npz models of plankton dynamics: their construction, coupling to physics, and application. Journal of Oceanography, 58(2):379–387, 2002. URL: <Go to ISI>://WOS:000175729100012, doi:Doi 10.1023/A:1015874028196.


J. Hauck, C. Volker, T. Wang, M. Hoppema, M. Losch, and D. A. Wolf-Gladrow. Seasonally different carbon flux changes in the southern ocean in response to the southern annular mode. Global Biogeochem Cycles, 27(4):1236–1245, 2013. URL: https://www.ncbi.nlm.nih.gov/pubmed/26074664, doi:10.1002/2013GB004600.


G. C. Johnson, S. Hosoda, S. R. Jayne, P. R. Oke, S. C. Riser, D. Roemmich, T. Suga, V. Thierry, S. E. Wijffels, and J. Xu. Argo-two decades: global oceanography, revolutionized. Ann Rev Mar Sci, 14:379–403, 2022. URL: https://www.ncbi.nlm.nih.gov/pubmed/34102064, doi:10.1146/annurev-marine-022521-102008.


E. M. Jones, M. E. Baird, M. Mongin, J. Parslow, J. Skerratt, J. Lovell, N. Margvelashvili, R. J. Matear, K. Wild-Allen, B. Robson, F. Rizwi, P. Oke, E. King, T. Schroeder, A. Steven, and J. Taylor. Use of remote-sensing reflectance to constrain a data assimilating marine biogeochemical model of the great barrier reef. Biogeosciences, 13(23):6441–6469, 2016. URL: <Go to ISI>://WOS:000391123700001, doi:10.5194/bg-13-6441-2016.


L. Nerger and W. Hiller. Software for ensemble-based data assimilation systems-implementation strategies and scalability. Computers & Geosciences, 55:110–118, 2013. URL: <Go to ISI>://WOS:000318384600010, doi:10.1016/j.cageo.2012.03.026.


M. Schartau, P. Wallhead, J. Hemmings, U. Loptien, I. Kriest, S. Krishna, B. A. Ward, T. Slawig, and A. Oschlies. Reviews and syntheses: parameter identification in marine planktonic ecosystem modelling. Biogeosciences, 14(6):1647–1701, 2017. URL: <Go to ISI>://WOS:000398193700002, doi:10.5194/bg-14-1647-2017.


Nabir Mamnun