Cool SSTs over Congo's submarine canyon; is this realistic?

[lmp4]

I am running a high resolution application over the Congo/Angola region (3 km resolution) and upon running for a year I have noticed some cool SST's over a certain region in my domain.

Specifically this region seems to be linked to the submarine canyon off the Congo river mouth. I have done a little reading and I now understand that it is possible that these canyon regions (when located off the continental shelf) are prone to enhanced up-welling and thus cooler SSTs temperatures however I am not sure the effect should be so noticeable (cooling of around 1K).

In my 5 km resolution model this effect is not present, leaving me to believe that the increased resolution is resolving the flow around the canyon more.

Congo-Avg-April.png (18.63 KiB) Viewed 8560 times

That is the SST cooling effect,

2014-11-27 13_57_46-regular include inferior.png (35.21 KiB) Viewed 8560 times

and there's the bathymirty.

Currently I have only been looking at 25km resolution observations but even when switching to 6km observations of SSTs this effect is still not present for the 20km wide canyon.

Anyone have any thoughts what could be going wrong? or is this actually realistic?

Also I feel as though the SSTs are a little strange one level below my surface level;

SST_49_level.ps

(1.43 MiB) Downloaded 386 times

[nacholibre]

Do you think trying different vertical advection schemes for tracers or changing the stretching near surface might help you get a feel for how much of it could be a computational artifact? Just a thought...
Zafer

[jivica]

do you interpolate atmo forcing (i.e. T2) before run or you leave that to ROMS linear interpolation?
If you use ROMS then it might use atmo values over the land and interpolate them as they were above the channel, hence used as forcing.
To test that you have to interpolate atmo forcing before onto ROMS grid but using only wet atmo forcing values (you have to use landmask from the atmo model!!).
In my case where I have many channels and islands this makes huge difference.
Ivica

[lmp4]

After looking at observations at 6 km resolution and changing my output to coarser resolution (6, 15, 30 km) I think I can conclude this is most definitely a computational artifact.

At the moment I am using ERA-Interim for my atmo forcing, indeed interpolated before the run.

Currently I am running three models with three different vertical advection schemes for tracers; originally I used the 4th ordered centered scheme but now I am trying 4th-ordered Akima, TS_MPDATA and TS_U3ADV_SPLIT.

I am also trying a few runs where I have 'filled' the Congo canyon bathymetry, just to see how closely the bathymetry is related to my problem.

[lmp4]

Quick update on my last post, so changing to TS_MPDATA and TS_U3ADV_SPLIT made the model blowup, TS_A4VADVECTION however seemed to perform much better than TS_C4VADVECTION (my original setting).

Here you can see the improvement where there is less cooling over the canyon as compared to using C4VADECTION in Figure 2.
FIGURE 1

FIGURE 2

I wonder how I can improve this further or do you think I have reached my model's limit with this bathymetry? Would changing the stretching help?