Improvement of detritus description in the BFM biogeochemical model and exploitation of the improved model, jointly with observations from ARGO floats to characterise the carbon pump in the Mediterranean.
Ocean Sciences, Climate
Research area
The dynamics of the ocean carbon pump are crucial for understanding the marine carbon cycle, and yet they remain poorly constrained in marine biogeochemistry models. These models represent an essential tool to both understand carbon cycle dynamics and to assess how will it change in the future under climate change. At the same time observation infrastructures such as the BGC-Argo fleet are providing unprecedented amounts of observational data on marine biogeochemistry cycles. Yet these data are still not fully exploited in marine biogeochemistry models.
My research project aims at finding novel ways to integrate observational data in marine biogeochemistry modelling frameworks.
Project goals
The goal of my TeRABIT project was to improve carbon cycling and plankton phenology in the Mediterranean implementation of the BFM biogeochemical model, coupled to ocean physics through the OGSTM transport model. Carbon cycling and sequestration (carbon pump) were improved by splitting BFM particulate organic into two classes: “small” (slow-sinking) and “large” (fast-sinking), each with distinct remineralisation kinetics. Plankton phenology and model skill against observations were improved by tuning selected parameters. Because the full-resolution model (1/24°) is computationally expensive, a degraded-resolution version (1/4°) was developed to test parameter sets faster. Multiple parameter sets and formulations were tested to improve key behaviors: eastern Mediterranean spring bloom peak (overestimated), summer surface chlorophyll (underestimated), deep chlorophyll maximum depth (configuration-sensitive), and the east–west production gradient (underestimated).
Computational approach
For my research I use 3D coupled physics-biogeochemistry ocean models. These are typically computationally intensive models, requiring HPC infrastructures to run (on several nodes) and large storage space for their outputs (typically 100s of TB). Due to their computationally intensive nature, fine tuning of these models is impractical.
To overcome this I am employing lighter 1D ocean column models, which can be effectively constrained by observations and/or by selected outputs of the 3D model.
These light-weight models allow to test different configuration more efficiently, and can be used to inform the parameterization of the larger 3D model.
Image
Simplified scheme of the state variables (P: phytoplankton, Z: zooplankton, D: detritus, B: bacteria) and fluxes between them that contribute to detritus carbon export (JDsink). The plots on the right represent the shape of the functional responses between model fluxes in input and output of control boxes (dashed rectangles).
Key results
The BFM model with two particulate organic groups has been successfully implemented and tested. Tests demonstrate this has a significant impact on the description of the gravitational carbon pump (more carbon and nutrients are exported, in agreement with observational data). The optimisation of plankton phenology has yielded positive results, key performance metrics were ameliorated and the phytoplankton chlorophyll production formulation was revised based on the optimisation findings.
Resource usage
For my TeRABIT project (and use case) I mainly used the Galileo100 and Leonardo clusters for running the OGSTM-BFM coupled physics-biogeochemistry modelling suite. The development of the degraded resolution (1/4 deg) model allowed for a significant reduction of computational requirements and time, for model fine-tuning. This model now runs on two (almost) fully populated Leonardo nodes (220 cores), and takes about one hour to complete a three-year run, which is sufficient for gauging performance. At present I completed 30+ model runs, and consumed circa 55000 core-hours. More will be used in the coming weeks, till project completion.
What's next
Mediterranean 3D Model results with two particulate organic classes are to be soon delivered as a model product within the NECCTON European project and will be distributed through the project online platform. Results from the optimisation of plankton phenology will serve as a basis to be implemented in the operational OGSTM-BFM that will be deployed, within the framework of Copernicus Marine products, for near-real time forecasts, for the planned reanalysis product (2027) and for climate projections within the recently started RIVIERADE European project.
Image
Mean modelled vs observed (satellite) surface chlorophyll in the North Western Mediterranean. The modelled values are after the optimisation of plankton phenology.
Giovanni Galli
Istituto Nazionale di Oceanografia e di Geofisica Sperimentale
After graduating in Environmental Engineering, at the University of Padova, I completed a PhD with OGS and the University of Trieste. My thesis was on modelling Mediterranean Benthic assemblages and their response to climate change. I was for five years employed as research scientist at the Plymouth Marine Laboratory (PML) in the UK, where I worked on coupled physics-biogeochemistry regional ocean models. Recently I re-joined OGS, Where I work on the marine carbon cycle in the Mediterranean.