Investigate runoff and iceberg water balance

[blimlim]

ESM1.6 redistributes a fraction of the liquid runoff from around the ice sheets into an iceberg melt flux. For conservation, the sum of the liquid runoff and the iceberg melt around the ice sheets should equal the total ice sheet runoff sent from the UM. I think masking inconsistencies might currently be causing conservation problems here.

The total runoff from the UM is first calculated as the area integral of runoff imported from the UM.

cice5/drivers/access/cpl_forcing_handler.F90

Line 1036 in 198fae2

trunoff_s = trunoff_s + gtarea(i,j) * grunoff(i,j)

Here, grunoff has had the land mask applied to it, and so only runoff over ocean points is counted.

The liquid runoff at each point in an area around the ice sheets is then set as:

cice5/drivers/access/cpl_forcing_handler.F90

Line 1202 in 28f2788

grunoff(i,j) = grunoff(i,j) * (1.0 - iceberg_rate_s) !do deduction

A total iceberg weighting is then calculated as the integral of the iceberg file data:

cice5/drivers/access/cpl_forcing_handler.F90

Line 443 in 198fae2

ticeberg_s(im) = ticeberg_s(im) + gtarea(i,j) * gwork(i,j)

Importantly, no masking is applied to the iceberg data. Data from points that fall over land are included in this weighting.

The iceberg melt at each point is then calculated as a fraction of the total liquid runoff:

cice5/drivers/access/cpl_forcing_handler.F90

Line 1065 in 198fae2

gwork(i, j) = gicebergfw(i, j, month) * iceberg_rate_s * trunoff_s / ticeberg_s(month)

Since the iceberg data gicebergfw has not been masked and is positive over several land points (see further below), some of the runoff ends up being redistributed to land points. The integral over the ocean points of the iceberg melt plus the liquid runoff will be strictly smaller than the runoff from the UM.

Land points in the CICE kmt file which have positive FICEBERG_GC3 data:

I think masking the iceberg data after it is read in from the input file will ensure that none of the runoff is redistributed over land.

There a few things that I think would be good to test out:

1. As mentioned, CICE currently applies the land mask to the UM runoff. Is any of the UM's runoff being placed over land in the first place? If so, that could be a source of non-conservation.
2. What does runoff_in - (runoff_out + iceberg_out) look like in the current set up. Does masking the iceberg data improve any inconsistencies here? Would it let us reduce/remove add_lprec?
3. How does (runoff_out+iceberg_out) compare to the total runoff output by the UM. Are there any other conservation issues happening?

[blimlim]

I've made some changes to the ice history to output variables without masking them. The following shows the iceberg flux io_licefw on the left, and io_licefw over the land points on the right:

We are getting some iceberg flux over the land. This averages to ~2.2e6 kg/s over a 5 year run. For reference, the loss in ocean mass in the AprilSpinup-lprec0 run is ~11.0e6kg/s (though it uses CICE4).

If we apply the mask to the iceberg data when it is read from the file, io_licefw is then restricted to only the ocean points:

Note: These runs were affected by the negative runoff issue, but I think the results still point to a significant amount of water going missing over land.

[DaveBi]

Thanks @blimlim for the detailed review of the "iceberg scheme" code and digging into the runoff/watermass conservtion issue which bothered me for quite a while when I worked on the runoff redistribution as iceberg flux for ESM1.5/6. I would try to answer your (very good) questions:

1. As mentioned, CICE currently applies the land mask to the UM runoff. Is any of the UM's runoff being placed over land in the first place? If so, that could be a source of non-conservation.

Yes, UM's runoff is indeed placed over land along coastlines. I didn't pay attention to this until I started working on the runoff redistribution for ESM1.5/6. I have to "axe-off" this part of runoff (as "extra") to guarantee the same watermass (runoff + iceberg) received by ocean with different partition rate. I have left comment in cpl_arrays_setup.F90 on this action:
!202407-10: note gwet added here is landsea mask for global domain (1 = wet, 0 = dry),
!passed from kmt.nc which is read in ice_grid. It is used to axe off "extra" runoff for
!global/domain "effective" runoff amount calculation, required by the "iceberg scheme"!

Doing so is based on the assumption that the runoff (from UM) placed on land points (of ocean-sea ice grid) should not be included as effective watermass input for the ocean. ACCESS coupled model always makes use of only the runoff over wet points because the runoff delivered from UM into CICE is passed directly into ocean where only wet points get the runoff. This is why I didn't pay attention to this "issue" before.

Could this be a source of non-conservation? I would believe it is not. For OASIS remapping between the UM/MOM grids, special treatment is made at coastlines where "fractional points" are applied for conservation purpose. This part is pretty tricky and I think only @MartinDix has the expertise to explain and confirm...... I just "assume" that those runoff remapped over dry points is "ineffective" because of the "fractional point" strategy.

2. What does runoff_in - (runoff_out + iceberg_out) look like in the current set up. Does masking the iceberg data improve any inconsistencies here? Would it let us reduce/remove add_lprec?

As explained above, runoff_in - (runoff_out + iceberg_out) would NOT be zero if runoff_in (in CICE) is NOT masked using the "gwet" mask.

Iceberg here is part of the runoff (e.g. 50%) around Antarctica and Greenland which is masked as "required".

Non-zero add_lprec is used if the system does not conserve matermass, which is unfortunately the case for most of the test runs -- but I remember that there is one configuration which sets add_lprec = 0 and the waterwass looks pretty stable (I wouldn't say "conserved" because the UM family doesn't really conserve water as per my impression).

3. How does (runoff_out+iceberg_out) compare to the total runoff output by the UM. Are there any other conservation issues happening?

It is answered above. Coupling resultant non-conservation would happen only if the "fractional point" strategy does not work as we expect......