Before I waste any more time attempting the impossible, can someone advise me if my problem is simply unsuitable for ROMS? I want to model upwelling over a complex (coral reef) topography. The grid goes from the coast, across a coral-studded shelf about 40 m deep, out to deep water. There are numerous masked points (for islands and reefs). I tried running a smoother over the domain without making much difference (I'd actually prefer not to have to smooth since smoothing is inversely proportional to reality).
I start with the upwelling test case, add a real grid (about 240x145 cells, each about 1.5 km square), and change the bcs to radiation (have tried a few different). Made a few other changes that shouldn't harm anything (eg went from 16 to 2 layers, and running on 16 cpus using MPI). Basically did not change the forcing (ie the same moderate wind). No tides. It works if I use a simple geometric bottom with no islands or channels. But with the real grid, after less than one day energy accumulates around certain reefs until it finally blows. I wonder if one of the other bc choices is less susceptible to trapping energy. Presently my choice is M2radiation.
Has ROMS been used for this in the past? I've seen the Dinniman and Klinck paper that used ROMS in a submarine canyon geometery that was
also complex, but not so shallow.
Thanks - John
Blow-up
Why only two levels? I would have thought 16 was rather minimal compared to our current runs of 30-40 levels. I don't know if that is your problem, though.
You don't say how steep your bathymetry is. Our main metric is a delta h/ over 2 h, or |h1-h2| / (h1+h2). In the bad old days, anything over about 0.2 was considered risky and the model would blow up at about 0.4-0.5 in the barotropic mode. Now we have had successful runs up at 0.4. However, the vertical stratification is also part of the puzzle and it may be that you are under-resolving that, especially if you are looking at upwelling issues.
You don't say how steep your bathymetry is. Our main metric is a delta h/ over 2 h, or |h1-h2| / (h1+h2). In the bad old days, anything over about 0.2 was considered risky and the model would blow up at about 0.4-0.5 in the barotropic mode. Now we have had successful runs up at 0.4. However, the vertical stratification is also part of the puzzle and it may be that you are under-resolving that, especially if you are looking at upwelling issues.
-
johnluick
Blow-up
Kate,
thank you for the response to my question. I realise you have suggested this metric in a previous posting. After reading your response I went back and found that it was in fact violated (despite that I thought I'd smoothed sufficiently). So I took a different approach this time - starting from an unsmoothed bathymetry, I smoothed heavily but only around the sites which were problematic. Still blew up (despite metric < 0.3 everywhere). Next I took your other hint and increased number of levels from 2 to 5. Then, no blow-up. Hmmm. I wonder why that might be - but never mind, I'm happy. To answer your question, the reason I went to 2 levels is that normally the system is vertically homogeneous but sometimes a buoyant wedge is trapped by wind against the coast and when the wind relaxes it spreads out at the surface. I figured that 2 layers would tell me all I needed to know because I am not interested in the details of the thermocline.
John
thank you for the response to my question. I realise you have suggested this metric in a previous posting. After reading your response I went back and found that it was in fact violated (despite that I thought I'd smoothed sufficiently). So I took a different approach this time - starting from an unsmoothed bathymetry, I smoothed heavily but only around the sites which were problematic. Still blew up (despite metric < 0.3 everywhere). Next I took your other hint and increased number of levels from 2 to 5. Then, no blow-up. Hmmm. I wonder why that might be - but never mind, I'm happy. To answer your question, the reason I went to 2 levels is that normally the system is vertically homogeneous but sometimes a buoyant wedge is trapped by wind against the coast and when the wind relaxes it spreads out at the surface. I figured that 2 layers would tell me all I needed to know because I am not interested in the details of the thermocline.
John
John-
You would be able to handle a freshwater wedge with a small vertical
resolution in a layer model, like those from FSU, in which the depth
of the layers can vary in time. However, in a model like ROMS where
the level positions are fixed in time, you must have about six or
more levels IN your thermocline in order to resolve it sufficiently.
This remains true even if you don't particularly care about the
details of the thermocline, for issues with a lack of resolution can
cause spurious circulation away from the thermocline.
This is even more of a problem in terrain following coordinate
systems. For example, if you had two levels and a freshwater layer
contained in the topmost level, and the water depth varied but the
depth of your salt wedge did not, the salinity in the topmost level
would vary over the bathymetric feature as the top level contained
varying proportions of fresh water lens versus the water below it.
This would appear to the model as a horizontal density gradient, and
would tend to drive spurious currents.
Likewise, the propagation speed of an under-resolved buoyant plume
can be very incorrect. In all numerical models, you under-resolve
features at your peril.
If you want to continue with a few vertical layers/levels, you may
want to look at other model formulations -- but layer models can
have other issues with steep bathymetry which can only be resolved
by having many vertical layers -- for instance, the Miami models
tend to revert back to level models when the layers would like to
intersect bathymetry, and thus would have the same resolution
requirements in regions of steep shallow topography.
Hope this helps,
Jamie Pringle
You would be able to handle a freshwater wedge with a small vertical
resolution in a layer model, like those from FSU, in which the depth
of the layers can vary in time. However, in a model like ROMS where
the level positions are fixed in time, you must have about six or
more levels IN your thermocline in order to resolve it sufficiently.
This remains true even if you don't particularly care about the
details of the thermocline, for issues with a lack of resolution can
cause spurious circulation away from the thermocline.
This is even more of a problem in terrain following coordinate
systems. For example, if you had two levels and a freshwater layer
contained in the topmost level, and the water depth varied but the
depth of your salt wedge did not, the salinity in the topmost level
would vary over the bathymetric feature as the top level contained
varying proportions of fresh water lens versus the water below it.
This would appear to the model as a horizontal density gradient, and
would tend to drive spurious currents.
Likewise, the propagation speed of an under-resolved buoyant plume
can be very incorrect. In all numerical models, you under-resolve
features at your peril.
If you want to continue with a few vertical layers/levels, you may
want to look at other model formulations -- but layer models can
have other issues with steep bathymetry which can only be resolved
by having many vertical layers -- for instance, the Miami models
tend to revert back to level models when the layers would like to
intersect bathymetry, and thus would have the same resolution
requirements in regions of steep shallow topography.
Hope this helps,
Jamie Pringle