Added wagner_pruss option for saturation vapour pressure.

[mariuswinkler]

With that, the calculation for the mixing ratio has changed throughout pysonde. Although Hardy1998 is used to calibrate the radiosondes from Vaisala side, we decided to go for the Wagner Pruss approach to align with the french MAESTRO Team which.

[mariuswinkler]

The last test fails because of:

"FAILED test/test_mwx_output.py::test_mwx_file_consistency - AssertionError: difference between old and new dataset is too large for mr
assert <xarray.DataArray 'mr' ()> Size: 4B\narray(0.00014251, dtype=float32)\nAttributes:\n units: dimensionless <= 0.0001"

which must be related to the changes I made in the computation of the saturation vapour pressure whiche affects the mixing ration computation. Since the threshold value is set at 0.0001 and the error is raised for a value slightly larger (0.00014251), would it make sense to raise the threshold value for this test?

[observingClouds]

The last test fails because of:

"FAILED test/test_mwx_output.py::test_mwx_file_consistency - AssertionError: difference between old and new dataset is too large for mr assert <xarray.DataArray 'mr' ()> Size: 4B\narray(0.00014251, dtype=float32)\nAttributes:\n units: dimensionless <= 0.0001"

which must be related to the changes I made in the computation of the saturation vapour pressure whiche affects the mixing ration computation. Since the threshold value is set at 0.0001 and the error is raised for a value slightly larger (0.00014251), would it make sense to raise the threshold value for this test?

The issue is probably that in metpy the molecular weight ratio is given more precisely than 0.622. Could you adjust the calc_wv_mixing_ratio function to use Md and Mv from metpy.constants? I am hopeful that the test will pass after that. Since the test is still covering the Hardy forumlation.

[mariuswinkler]

Won't quite work still... running out of ideas rn

[observingClouds]

The brackets are incorrectly set here and in the next occurrence. It is

w = w_s * RH / 100

[mariuswinkler]

Thanks for the note. Initially the RH scaling was applied after computing the saturated mixing ratio. It should go inside to compute the actual vapor pressure first. I've corrected it so that e = RH * e_s is passed directly into the function.

For the else case when self.profile.humidity.values is in percent the / 100 needs to also go inside the brackets of course.

[mariuswinkler]

hmmm...

The only explanation I can think of is that the test might use some form of checksum or internal consistency check based on previous test-datasets. By moving the relative humidity inside the brackets of the mixing ratio computation, we might now be slightly overstepping the threshold that causes the test to fail even though the result is actually correct.

[observingClouds]

So what was wrong with df559b8? You like breaking tests?! ;) Can we not do the scaling afterwards as I did before?

[mariuswinkler]

Now I've stumbled over my own leg. One more try.

[mariuswinkler]

I think the scaling by relative humidity (dividing by 100) needs to happen inside the mixing ratio function, since the computation is nonlinear and depends on the actual vapor pressure, not RH in percent, no ?

[observingClouds]

Well, don't we want to calculate the:

• saturation mixing ratio (ws) via td.calc_wv_mixing_ratio(self.profile, e_s)
  and than scale it to ambient mixing ratio (w) via saturation_mixing_ratio * RH / 100?

w = w_s * rh / 100

[mariuswinkler]

Actually, both should work no ?
What I am suggesting is:

e = e_s * RH / 100.0
w = td.calc_wv_mixing_ratio(self.profile, e)

What you are suggesting is:

w_s = td.calc_wv_mixing_ratio(self.profile, e_s)
w = w_s * RH / 100.0

... only, that my version does not pass the test.

[observingClouds]

Well, to stay simple, let's stay with what works.

[mariuswinkler]

I'll turn it around in a new push. 1 sec.

[mariuswinkler]

I don't understand why this test won't go through. I took "method=wagner_pruss" out so that the test would pass but without success. I am running out of ideas here. Maybe you can have a look at it, Hauke.