Run and evaluate the scientific performance of single cycle 4dvar experiments

[kbhargava]

Description

• Rerun the 4dvar single cycle experiments after the inclusion of mask in RMSbyLevel and evaluate the impact with the soca ctest
• Repeat the above with 1/4 degree
• Test the impact of using regularization when computing the HTLM coefficients (reefer section 3a and 4b of https://doi.org/10.1175/MWR-D-20-0088.1 )
• Run 4dvar with HTLM created by using climatology as the ensemble
• Compare the linearization errors and 4dvar results from the above mentioned experiments

[kbhargava]

The impact of using regularization when computing the HTLM coefficients

ctest: no regularization leads to least linearization errors

For task 3 in the list, below is a plot showing the impact of regularization, which penalizes the impact of the ensemble part of HTLM, on the linearization errors of salinity. Similar results are observed for temperature with no changes in u, v, and h. In this small ensemble, low-resolution ocean, it looks like the information that the ensemble brings is crucial and doesn't need to be penalized.

[kbhargava]

The impact of including masks in RMSbyLevel

ctest : No significant difference at all

The difference in linearization errors
After 6 hours:
salinity: -8.84e-08 to 4.35e-7
temperature: -8.18e-6 to 1.72e-6
After 12 hours:
salinity: -4.28e-5 to 0.0004 (only visible at lev >=15)
temperature: -6.4e-4 to 7.98e-4 (only visible at lev>=15)
As expected no changes to other variables.

1/4 degree: No change at all in linearization errors 😞

[kbhargava]

Linearization errors when using the monthly climatology as the ensemble

Below are the figures comparing the linearization errors after 12 hours in the temperature (top) and salinity (bottom) with the 4dvar-htlm using August climatology and 4dvar-identity