Draft for FESOM recipe

[roxyboy]

I tried to build upon the INALT60 PR #26 but I couldn't quite figure out the correct syntax...

[roxyboy]

Also, currently only the surface data is available to flux to the cloud. The interior should be ready soonish.

[cisaacstern]

Thanks @roxyboy. I'll take a look today.

[cisaacstern]

@roxyboy, your draft was super helpful. I'm now caching these inputs to GCS.

[cisaacstern]

@roxyboy, FESOM inputs are cached. Before building the Zarr, I wanted to confirm your target_chunks preference.

Currently, target_chunks is set to {"time": 15} in the recipe here. By my calculation (see code, below), that puts each chunk at about 120 MBs in size. I don't know much about chunking, but I've noted that pangeo-forge-recipes/docs/tutorials/cmip6-recipe.ipynb, e.g., makes a general recommendation that chunk sizes be between 50-100 MBs.

Thoughts? If you don't have another preference, I'll build the Zarr with target_chunks={"time": 6}, which is the integer chunk size which gets us closest to 50 MBs per chunk.

cache_path = rec.input_cache.root_path  # `rec` is the `"FESOM/surf/fma"` recipe
first_input_url = fs_gcs.ls(cache_path)[0]  # `fs_gcs` is the endpoint where inputs are cached
first_input_file = fs_gcs.open(first_input_url, mode='rb')
ds = xr.open_dataset(first_input_file)

ntime = len(ds.time)  # the number of time slices
ncfile_size = ds.nbytes  # the netcdf file size

target_chunk_options = (6, 15)
for option in target_chunk_options:
    chunksize = int((ncfile_size/ntime) * option)
    print(f"With {option} time slices per chunk, chunksize is {chunksize/1e6} MBs")

With 6 time slices per chunk, chunksize is 48.000177 MBs
With 15 time slices per chunk, chunksize is 120.000442 MBs

[roxyboy]

Thoughts? If you don't have another preference, I'll build the Zarr with target_chunks={"time": 6}, which is the integer chunk size which gets us closest to 50 MBs per chunk.

target_chunks={"time": 6} is fine with me :)

[cisaacstern]

Noting that, along with #24 (comment), this build is also blocked by pangeo-forge/pangeo-forge-recipes#164. The surface data for this recipe contain single-variable arrays of shape (744, 1000, 1000) (for dims time, lat, lon) and size ~6 GBs.

[cisaacstern]

I was able to build the FESOM surface data using an un-merged draft of pangeo-forge/pangeo-forge-recipes#166.

It can be accessed via the project catalog, as demonstrated here: https://github.com/pangeo-data/swot_adac_ogcms/blob/main/intake_demo.ipynb

[roxyboy]

@cisaacstern Are the interior data for FESOM on OSN...?

[cisaacstern]

@cisaacstern Are the interior data for FESOM on OSN...?

Not yet. Is it on the ftp server? Last I was aware it wasn't available over ftp yet.

[roxyboy]

@cisaacstern The FESOM group notified me that their data was corrupted so they re-ran their simulation. They've finished remaking the winter data. Could we swap the data on OSN??
The surface and 3D winter data are here: https://swiftbrowser.dkrz.de/public/dkrz_035d8f6ff058403bb42f8302e6badfbc/SWOT_intercomparison2/

[cisaacstern]

Sure thing, @roxyboy. Do you have a deadline (either specific or general) for this?

[roxyboy]

It'd be nice if I could include it in the talk I'll give at the high-res ocean modelling workshop that will take place from Sep. 29 so... the sooner the better but if you're occupied with other projects, it's ok if the FESOM data doesn't make it to it.

[cisaacstern]

That's very doable, probably won't get to it this week, but will get it done by end of next week.

[cisaacstern]

@roxyboy, apologies that I won't be able to complete this by the end of this week as I'd initially planned. I will make sure to revisit this on Monday and should still be able to get it to you in advance of your deadline. I'll follow-up with a better time estimate next week once I've taken a closer look.

[cisaacstern]

@roxyboy, I'm removing the existing FESOM surface winter Zarr store from OSN now, to replace it with the corrected data you've linked in #52 (comment).

I'll be adding the winter 3D data at the same time and will ping you again here when both new Zarr stores are complete.

[roxyboy]

@roxyboy, I'm removing the existing FESOM surface winter Zarr store from OSN now, to replace it with the corrected data you've linked in #52 (comment).

I'll be adding the winter 3D data at the same time and will ping you again here when both new Zarr stores are complete.

Ok, thanks a lot! :)

[cisaacstern]

@roxyboy, the updated FESOM data (surface and interior) is now on OSN and I've added FESOM interior data as an option in the project catalog pangeo-data/swot_adac_ogcms#6

Here's a quick summary of how you'd use the workflow described in pangeo-data/swot_adac_ogcms/intake_demo.ipynb to open the FESOM interior data:

Note this would need to be run from within theswot_adac_ogcms project repo to have access to the local imports

# load the catalog and determine the allowable params for the selected item
from validate_catalog import all_params
params_dict, cat = all_params()
item = "FESOM"
params_dict[item]

[{'datatype': 'surf', 'season': 'fma'},
 {'datatype': 'surf', 'season': 'aso'},
 {'datatype': 'int', 'season': 'fma'},
 {'datatype': 'int', 'season': 'aso'}]

# select a param set from the list and confirm its the one you want
params = params_dict[item][2]
print(item, params)

FESOM {'datatype': 'int', 'season': 'fma'}

# load item with the selected params
ds = cat[item](**params).to_dask()
print(ds)

<xarray.Dataset>
Dimensions:  (depth: 48, lat: 1000, lon: 1000, time: 90)
Coordinates:
  * depth    (depth) float64 0.0 -5.0 -10.0 -15.0 ... -760.0 -860.0 -1.04e+03
  * lat      (lat) float64 30.0 30.01 30.02 30.03 ... 39.97 39.98 39.99 40.0
  * lon      (lon) float64 -78.0 -77.99 -77.98 -77.97 ... -68.02 -68.01 -68.0
  * time     (time) datetime64[ns] 2012-02-01 2012-02-02 ... 2012-04-30
Data variables:
    salt     (time, depth, lat, lon) float32 dask.array<chunksize=(10, 48, 1000, 1000), meta=np.ndarray>
    temp     (time, depth, lat, lon) float32 dask.array<chunksize=(10, 48, 1000, 1000), meta=np.ndarray>
    u        (time, depth, lat, lon) float32 dask.array<chunksize=(10, 48, 1000, 1000), meta=np.ndarray>
    v        (time, depth, lat, lon) float32 dask.array<chunksize=(10, 48, 1000, 1000), meta=np.ndarray>
    w        (time, depth, lat, lon) float32 dask.array<chunksize=(10, 48, 1000, 1000), meta=np.ndarray>
Attributes:
    CDI:          Climate Data Interface version 1.9.6 (http://mpimet.mpg.de/...
    CDO:          Climate Data Operators version 1.9.6 (http://mpimet.mpg.de/...
    Conventions:  CF-1.6
    history:      Thu Sep 09 15:44:25 2021: cdo mergetime 2012-02_01_w_cubic....

[roxyboy]

@cisaacstern FESOM data appears to working fine :) Thanks again!