I propose creating a web based ocean atlas (similar to the WOCE Atlas, shown above) of CMIP6 models sampled as WOCE/GO-SHIP cruises. From the website, one will be able select an ocean transect (A16, P16,S4, etc) and then compare the CMIP6 model mean, single model ensembles, and individual realizations to observations. The high quality bottle samples will come from the GLODAPv2 bottle data product. When the hackathon is over, we will have a functioning website and begin a draft of the publication that will accompany the dataset of remapped files. Much of the data processing workflow has been mapped out, therefore I expect much of our collaboration to be focused on developing the web based atlas. Questions to consider will be:
How should we compare observations to single model ensembles?
What skill metrics should we use?
What variables should we include?
With this atlas we can make the CMIP6 data more accessible to observationalists, and modelers will spend much less time creating comparisons of simulated tracer fields and observed tracer fields from WOCE/GO-SHIP cruises.
Outcomes
Dataset of CMIP6 models remapped to ocean observations
A functional first-draft of the web based CMIP6 ocean atlas.
Post hackathon, we will submit the publication associated with the dataset to Earth System Science Data or as a preprint via EarthArXiv
A python package for applying this to any ocean model (such as regional models)
Feasibility
The creation of the dataset can be completed over the few days of the CMIP6 Hackathon by leveraging existing python packages. We will utilize easy_coloc to map simulations to observational space. For storage, Pangeo seems like a perfect home for the remapped files. Finally, for quantitative comparisons between models and observations, we will use skill metrics from the xskillscore package.
Data Requirements
3d fields of:
Circulation tracers
Salinity
Temperature
CFCs
SF6
Biogeochemical Tracers
Oxygen
DIC
Alkalinity
AOU
Nitrate
Phosphate
pH
Desired Collaborators
Anyone with experience in skill metrics, basic web development, and/or ocean model analysis
I think this is a great idea and something I was hoping we could work on during the hackathon. So important to make CMIP6 data more accessible. Perhaps OCB may be able to host the website? Or do you already have someone in mind who could? It might be worth adding primary production to the list of biogeochemical tracers. As for how I can input, I’ve not worked with the packages you’ve described before, I have done similar ocean model analysis but using CDO and R. But interested to contribute if I can!
I can offer guidance on the skill metrics, but won’t have the bandwidth to work primarily on this project. We talked about this in La Jolla, but a few of us have collaborated over at xskillscore to have some standard metrics available that wrap xarray and are vectorized (https://github.com/raybellwaves/xskillscore/). I think we have corr, mse, rmse, mae, and some probabilistic ones over there. We’ve also added a ton at climpred (https://climpred.readthedocs.io/en/latest/metrics.html) that could just be lifted directly from our codebase.
This is a great idea, and definitely something that would be useful to the observational community, outside those with expertise in model analysis. I am still project shopping, but would definitely be interested in contributing to this.
One suggestion I’d have would be to expand as much as possible the set of tracer variables Selfishly, I’d especially be interested in adding fluxes as a category of biogeochemical tracers, since especially with GEOTRACES, there are now more derived export flux flux observational data that would be interesting to compare with the model output. I’d be up for helping brainstorm other variables that might be of interest to the community as well so that we make sure we get in the data requests in time to have it for the hackathon.
I think this is a fantastic idea and something the community needs. I am thinking of proposing a “hack a basin” project looking at changes in metrics related to transport / biogeochemistry / physical properties under a middle of the road and high-end future scenario. Wondering if it might be useful to add a future component to this CMIP6 Ocean Atlas where a user can choose one or two scenarios to look at the projected change in the field at each cross section? Just a difference in a 2081 to 2100 average (end of century) minus 1986 to 2005 average (or whatever time period you are using for your historical comparison against obs.).
I don’t have much experience in the packages it sounds like you will be using but have a lot of experience working with ocean data from CMIP5/6 models.
