Compare interface pressure computation methods

[CoryMartin-NOAA]

Compare GeoVaLs and H(x) when computing interface pressures (prsi) using the following two methods:

• prsi(k) = ak(k) + bk(k) * p_surface (GSI way)
• prsi(k) = sum(delp(1:k)) (FV3-JEDI way)

[CoryMartin-NOAA]

A previous GSI run with H(x) and GeoVaLs differs in two way from the newer GSI run:

• prsi is computed the JEDI wa
• Lossy netCDF compression in the write-grid component of GFS was turned off, allowing for only lossless compression

The second point explains figures such as the first one: Where the differences in model temperature profiles interpolated to x,y are mostly just noise and small (~+-0.001K) Similar behavior is shown for things like profiles of q.

However, here is model pressure: Where there are noticeable differences, particularly at low model levels. Here are a few select profiles: For some locations, the differences are small throughout, but for many, the difference can be ~20Pa towards the model surface.

[CoryMartin-NOAA]

Taking a quick look at H(x) for amsu-a_n19 and conventional temperature

AMSU-A N19: Differences range from very small to ~0.15K (depending on channel/location), it is generally a positive difference when not near zero, suggesting the model derived observations are warmer than before. I believe this is consistent with the pressure differences shown above, as those mean the model is now slightly lower pressure for each layer for the same temperature, so that would tend to make things a bit warmer when interpolating in height.

Differences in H(x) for conventional temperature are much larger: Frequently the differences are +-1K and can be as high as 6K (but this is rare and seems limited to 1 profile?)

[CoryMartin-NOAA]

For good measure, here is conv_ps: Most differences are again small (within 0.25 mb) but some can be as large as 2.5 mb