Facilitating driving Raven from SPS

[huard]

Addressing a Problem?

SPS is ECCC's driver for land-surface schemes. It outputs surface fields in FST format, which we flip to netCDF. These are often massive files, like 4Gb per variable per year.

To run the Raven routing from SPS outputs, we need to feed Raven

1. the sum of lateral flows and surface runoff as PRECIP;
2. the groundwater recharge as RECHARGE;
3. and surface temperature as TEMP_AVE.

There are two things that would reduce the amount of pre-processing required to run Raven:

1. Adding new FORCING variables that would map to surface runoff and lateral flows. They would be added together to generate PRECIP. This would avoid having to create a new variable on disk that is just a sum of two variables.
2. Adding support for vertical levels in GriddedForcing. We would specify :ElevationVarNameNC and a new command, e.g. :ElevationIndexNC. This would avoid having to make copies of a single level to drive Raven. In our case it's not really an elevation, but conceptually it would work the same.

Potential Solution

No response

Additional context

No response

Contribution

• I would be willing/able to open a Pull Request to contribute this feature.

[huard]

I'm investigating another option that would require only small changes to Raven to read specific vertical levels.

The netCDF ibrary has support for NCZarr. We could possibly create a virtual NCZarr file that only maps the content for the level we're interested in. Zarr files are essentially json files specifiying where the data is, so creating a subset requires only new JSON files, and no copy of the binary data.

Issues so far:

• NCZarr file names look like: "file:///tmp/salmon.zip#mode=xarray,zip", and Raven will stop the flename parser at the first comma, so the read operation fails.

Converting a netCDF to Zarr is easy with nccopy:
nccopy -u -F none Salmon-River-Near-Prince-George_meteo_daily.nc "file:///tmp/salmon.zip#mode=xarray,zip"

[analytophile]

Can you clarify the need for levels? Are you trying to read the elevation grid out of a 3D array which has several elevations (say the elevation of each climate model grid?) You said it is not really elevation, then what is it?

[huard]

SPS has different area types, e.g. ['ice_free_land', 'land_ice', 'ice_free_sea', 'lake_ice_or_sea_ice', 'all_area_types', 'urban']. So a typical surface temperature file will have dimensions (time: 2926, sfctype: 6, rlat: 800, rlon: 1050). I want to drive Raven with the 5th level, corresponding to all_area_types. What I now do is copy that unique level to a new netCDF file that Raven can parse, because it has three dimensions (time, rlat, rlon).