GEOPOTENTIAL_HCONV: error covariance modelling in the slope region

[stef]

This is a more general question about modelling background error covariance close to the slope.

Looking at the figures below, I'm guessing that the diffusion approach for correlation modeling diffuses along s-coordinates. Do you find this problematic in realistic applications?

Now I discovered GEOPOTENTIAL_HCONV and wondered if I should try to use it. I have defined

Code: Select all

** GEOPOTENTIAL_HCONV      if horizontal convolutions along geopotentials    **

, commented out the safeguard in the code and it blows up, which wasn't surprising because it's marked as experimental.

Are there any new developments regarding GEOPOTENTIAL_HCONV?

Shoud I improve the s-coordinate parameters instead?

The convolution parameters for my toy model are as in the tutorial, they may not make much sense for my domain:

50km for HdecayI and 30m for VdecayI. Hgamma for IC is 0.5.

I'm adjusting initial conditions only. The innovation is large.

Figure 1: Temperature increment in the Bay of Bengal for one single ARGO profile. Note the elevated correction patch in the shallower region towards the shelf. The horizontal line indicates the transect shown below.


Figure 2: Same as 1 but for the transect. It doesn't look as bad here because the increment at depth dominates the visual impression.

increments_330_temp.png (52.28 KiB) Viewed 109 times

increments_transect_330.png (23.07 KiB) Viewed 109 times