Vertical coordinate conversions about hn2016_falwa

[csyhuang]

Some references from Sandro:
https://www.ncl.ucar.edu/Applications/vert_interp.shtml

[Pragallva]

Thanks Clare for the discussion for your suggested solutions! :)
Just some continuing thoughts here..
I think a better work around might be to interpolate variables from model levels (sigma or hybrid coordinates) to plevs directly using ncl or other python functions and then use the QGField class as it is since the falwa library transforms from pressure to pseudo-height coordinate.

One caution is that in this case we will be interpolating two times, model levels --> pressure --> pseudo-height (log-pressure). This could be a potential source of residual errors in the LWA budget since during the first step (model levels --> pressure) one has to make some choices about how to fill the extrapolated values below topography.

In future we should aim for having a function that directly transforms from model levels to pseudo-height coordinates and then calculates qg, LWA etc.

• One way to go about it would be to use the pyngl library (https://www.pyngl.ucar.edu/) to use the ncl functionality. But I had some issues installing it on the falwa python environment since it has some old dependencies.
• Other option is to use a plain vanilla python code (I have one using the scipy interp function). But more work needed to incorporate into the falwa code base. Also this could be bit slow compared to other methods.
• Other option is to use a fortran method if someone has one!

[Pragallva]

Some updates about pre-processing model data when the variables are on hybrid vertical coordinates. The following method works for me now.

1 - First I do a coordinate transformation from hybrid vertical levels to pressure levels using scipy interp function.
2 - There is missing data for regions below topography (or lat-lon grids where no values exist at the required pressure levels). I fill them with nans.
3 - Finally I replace the nans by solving Poisson's equation using a relaxation scheme. This is done using the gridfill python library available here - https://github.com/ajdawson/gridfill . The gridfill function is similar to NCL's poisson_grid_fill. I use itermax = 10000 to give as smooth interpolated values as possible. This seems to be necessary for getting reasonable values when using falwa library on this dataset.

[csyhuang]

Hi @Pragallva ,
I'm looking into your code to interpolate from hybrid coordinate to pressure levels. Yeah - it is pathetically slow 😢 took me 4 seconds to interpolate via a single (32, 192, 288) time slice... 🤦 The NGL package seems no longer maintained... and I cannot find anywhere official that does such preprocessing...

QGField.interpolate_fields only do the interpolation once per variable, but the hybrid coordinate conversion is doing it 192 x 288 times, so the latter is incredibly slower. But still, let me try modifying the QGField such that it you can skip interpolation if the input is on regular psudoheight grid. I'll keep you updated when this is done.

I created #115 to cover this