Ocean Carbon Uptake and Productivity in The CMIP6 Ensemble
Scientific Motivation
The invasion of anthropogenic carbon and heat into the upper ocean is expected to continue into the 21st century, raising major concerns for the health of marine ecosystems and sustainability of global fisheries. Past CMIP5 assessments (e.g. Bopp et al. 2013) identified clear trends in upper ocean warming and acidification across models and scenarios, but outlined divergent responses for net primary productivity (NPP) and thermocline oxygen in several productive and biodiverse regions of the world ocean.
This project will examine historical simulations and future projections of ocean carbon and heat uptake and storage in the CMIP6 ensemble, and explore their impacts on marine ecosystems drivers, with a focus on primary productivity. Hacking will explore:
How do models simulate carbon and heat uptake compared to available observations? What are future projections under different ssp’s? what regions dominate carbon and heat uptake vs. storage? What are differences across models (uptake rates, regional differences, etc.) and what processes might explain these differences?
How well do models simulate the mean distribution and long-term trends of ecosystem drivers (SST, pH/aragonite saturation depth, [O2]/hypoxic depth, NPP) compared to available observations? Can we identify persistent or new biases since CMIP5? What are CMIP6 projections for marine ecosystem drivers in a warming climate? What are spatial characteristics, timescales of emergence, and differences across models? Is there a relationship between models’ carbon and heat uptake efficiency and the severity of their projected impacts on these ecosystem drivers? How did increased climate sensitivity since CMIP5 in certain models (e.g. CESM2) influence these impacts in these models?
Projections of NPP especially showcase major differences across CMIP5 models in sign and amplitude over important ocean regions (Eastern/central tropical Pacific, Subpolar gyres, etc. Fig 5. Bopp et al 2013). Does this spread still exist in CMIP6? Do projected changes in carbon export at depth mirror the changes in NPP? What physical and biogeochemical processes explain NPP and carbon export changes across regions and models (e.g. changes in easterlies, upwelling, nutrient transport by the EUC or the overturning circulation, stratification and ventilation, remineralization rates, etc.)?
We will prioritize developing efficient workflows for loading, analysis, and plotting of model outputs and comparison to observations, and aim for process-based examination of model projections, inter-model differences, and model biases.
Anticipated Data Needs
Monthly (for season-sensitive processes/variables) and/or annual ocean physical and biogeochemical variables from the piControl, Historical, and main ssp experiments.
Anticipated Software Tools
We will primarily rely on Xarray/Dask for analysis, matplotlib/cartopy for visualization, and explore additional tools (ESMintake, ESMlab, xgcm) as needed.
Desired Collaborators
Ocean BGC enthusiasts; Python beginners and experts are all welcome. The proposed hacking and guiding questions (1-3) are made broad enough to serve as starting points for team members to focus on or spawn off/lead as a sub-project with more specific (or different?) questions that hopefully will still fall under the broad theme of carbon cycling and ecosystem drivers in the CMIP6. Inter-projects synergies and exchange of expertise are encouraged.
Thanks for putting this together Yassir! I think these are wonderful questions. I am particularly interested in contributing to 2 & 3, as I think there is opportunity for overlap between the two.
Nicely outlined project. Regarding the software tools, I just commented over in the oxygen project, regarding the role of xgcm. I would appreciate any comments on whether to split out/centralize the development of tools which might be used over several of the science objectives (e.g. vector calculus operators like grad, div, curl).
That would be extremely helpful @jbusecke! Also super interested in density coordinate interpolation, and we (and other projects) might think of a couple other needs as projects progress, so perhaps a centralized technical project would be more efficient? This is already great text for a proposal:
Thanks for writing up these nice ideas! I think it would be helpful to split this proposal into a few concrete projects for the hackathon. From a technical perspective, the questions laid out here suggest the need to develop tools to analyze spatial variability (but I’m less familiar with the existing packages!)
Here are some ideas:
1.a. What are the spatial differences between ocean heat uptake and ocean carbon uptake in the CMIP6 models? While there are areas of the ocean that will be important for both heat uptake and carbon uptake, given the distinct sources, spatial patterns of the forcing and sources and sinks of heat and carbon in the oceans each will also be dominated by different regions. Distinguishing the patterns of both heat and carbon uptake (and the differences between them) could aid understanding the of biotic v abiotic controls on ocean carbon uptake in the models and the sources of uncertainty in trends.
2.a. Given the disagreement in sign of the changes in marine ecosystem variables across CMIP5 models, it will be an interesting challenge to find a time of emergence of the forced signal. It would be interesting to work with the Can We Better Quantify the “Time of Emergence” of Regional Climate Change Signals using CMIP6? project to adapt the approaches to ocean and biogeochemical variables. One question that I have is how regions defined by machine learning may depend on the mean state of the models.
2.b. A specific example of examining an ecosystem driver is laid out in Ocean oxygen in a warming world
I am also interested in this project. Paul Lerner will probably be the point person from our group. He will explain what we want to get out of this. Thanks for posting this project!
These are some great project ideas, which I think are quite useful for those of us who are working on the ocean biogeochemistry fields of the various CMIP6 models. Here are some follow-up questions I think are worth considering, under two broad categories:
In addition to, as Yassir suggests, comparing simulated ecosystem drivers to available observations, can we also identify the drivers of the biases themselves? For example, are biases in NPP and nutrients related to biases in physical properties (temperature, mixed layer depth), parameterization of processes such as growth rate and remineralization, or structural uncertainty? Can we determine causal relationships between biases of different biogeochemical properties which are interactive? For example, biases in surface nutrient fields will lead to biases in NPP, but NPP, along with advection and mixing, affects the nutrient fields themselves. And how do the relationships between biases change in different oceanic environments? These questions are perhaps even more important for ocean carbon uptake, which is affected by NPP but also wind speed, T/S, and the air-sea pCO2 gradient.
What does the seasonal cycle of productivity and carbon uptake look like across the CMIP6 models. Can we identify ocean biogeochemical provinces with unique seasonal cycles (e.g., Henson et al., 2010;2013, Biogeosciences)? In each of these regions, what are the driving factors of the seasonal cycles in uptake and productivity? How and what differences are there in the predicted changes of the seasonal cycle (changes in amplitude + phase) of the CMIP6 models? Moreover, alluding to point (3) of Yassir, are changes in annual mean NPP systematic across the seasonal cycle, or are there seasonal variations in the sign and amplitude of the long-term changes?
I am interested in contributing to this project, particularly Question 1: spatial patterns of carbon and heat storage in historical and SSP simulations. I have been analyzing carbon surface fluxes / transport / storage in the Southern Ocean in a suite of CMIP5 models and have used Pyferret for all of my analysis but I would like to learn other python tools and this seems like a good opportunity to branch out beyond Pyferret.
I am currently trying to implement some of this functionality ahead of the hackathon (so that I can spend more time testing the tools with some real research
The major PR needed for this will hopefully be merged soon and then I will add some more examples for everyone here (or where appropriate). Let me see if I can throw a proposal together tonight to centralize these efforts.
