Discrepancies shown in the single observation assimilation from Rawinsonds.

[gmao-wgu]

A single u wind observation from Rawindsondes was assimilated in both GSI and JEDI. Below is the basic information from this single Rawinsondes observation:

• Locations: lat=33.1662, lon=273.3192, Pressure=700mb

• Observation Error: u_wind: GSI 1.7, JEDI 1.7

• OMF: u wind: GSI 6.7582. JEDI. 7.476

The increments from the single Rawinsondes u wind at lat= 33.1662, lon=273.3192 and pressure=700mb are shown below. It is noted that u increments from GSI are scaled by 7.476/6.7582 to account for the difference in OMF between GSI and JEDI. So the u increment differences are either from H^TBH or from the minimization. I did another test by reducing the norm reduction required in GSI from 10-3 to 10-6 and got nearly identical plots.

[Dooruk]

@rtodling mentioned during JEDI Wednesday meeting that this difference is expected, is this still an issue?

[gmao-wgu]

I understand that we have to accept this kind of difference. Now I realize that although the background error inputs provided are identical between GSI and JEDI, the background error covariance matrix or its inverse projected onto the observation locations are different due to additional operations in JEDI, such as transforming from lat-lon grids to cubic sphere grids, then to observation locations, whereas in GSI, it goes directly from lat-lon grids to observation locations. This might explain the larger differences in increments near the surface and at certain high levels where background errors are relative larger, as well as the larger increments and greater gradient norm reduction observed always in JEDI.