Implement dewpoint consistent with Ambaum (2020) formula

[kwodzicki]

What went wrong?

I noticed some of my unit test from checking/convert between RH and dewpoint were failing in v1.71. Upon investigation, I noticed that the saturation_vapor_pressure function now implements the Ambaum formula; however, the dewpoint function "inverts the [Bolton1980]_ formula for saturation vapor pressure", causing inconsistencies.

The "code to reproduce" passes using v1.6.3 but fails for 1.7.1

Operating System

Linux

Version

1.7.1

Python Version

3.12

Code to Reproduce

import numpy as np
import xarray as xr
from metpy.calc import (
    relative_humidity_from_dewpoint,
    dewpoint_from_relative_humidity,
)


dim = 'dim_0'                                                              
temp = xr.DataArray(
    data=[20.0, 25.0, 30.0],
    dims=dim,
    attrs={'units': 'degC'},
)

dew = xr.DataArray(
    data=[12.0, 16.0, 20.0],
    dims=dim,
    attrs={'units': 'degC'},
)

rh = (
    relative_humidity_from_dewpoint(temp, dew)
    .metpy                                                                 
    .dequantify()                                                          
)

dew_test = dewpoint_from_relative_humidity(temp, rh)
np.testing.assert_almost_equal(dew_test.values, dew.values)

Errors, Traceback, and Logs

AssertionError: 
Arrays are not almost equal to 7 decimals

Mismatched elements: 3 / 3 (100%)
Max absolute difference among violations: 0.01519082
Max relative difference among violations: 0.00075954
 ACTUAL: array([11.9932291, 15.9904193, 19.9848092])
 DESIRED: array([12., 16., 20.])

[dopplershift]

The summary of the state of things from @dcamron in #3950:

#3950 (comment):

Hi, thanks for the issue! I understand the confusion, though this is not a bug. Check out #3726 and related issues/PRs for scattered details. Here's the short version:

dewpoint and saturation_vapor_pressure were the inverse of each other, based on Bolton 1980. This is a simplified empirical fit of the Clausius-Clapeyron relationship for equilibrium of water vapor and liquid water based on past experiments and measurements. This loses validity and accuracy near and outside warm, Earth-atmospheric conditions.

We updated Saturation Vapor Pressure to a newer calculation, based on Ambaum 2020. Using some clever assumptions, they justified the value in using a physically-based direct calculation for saturation vapor pressure. We made this update to be internally consistent with #3728, a fast, direct LCL calculation based on the same assumptions. By tweaking these assumptions, we can make our formulation valid across phases and equilibrium regimes, eg calculating the saturation pressure of water vapor over ice below 0 degrees C. Neat! This is also required for that new LCL calculation. Romps 2021a is a great overview of this change.

MetPy's functions to calculate dewpoint still depend on that Bolton 1980 formulation, and so (a) does not round-trip with the new saturation vapor pressure formula and (b) does not support wide temperature ranges and multiple phase regimes. However, we can (and will!) update these functions to use these new assumptions, as seen in Romps 2021b. We just haven't gotten there yet, because it's going to break our tests and we haven't prioritized the work. Updating dewpoint was not required to get this fast LCL implemented.

When we make that update, these two will have the same interface and be able to round-trip. We aren't planning on removing any of this functionality for our users in the mean time.

[dopplershift]

I'll leave this open to track, though I consider this a "nice to have" and something we really do want, rather than a bug.