Clarify that `sftlf` includes lakes

[jesusff]

See euro-cordex/jsc-compliance-check#19 (comment)

There are already quite likely inconsistencies across models for this key variable

Should we modify the long_name in the CMOR tables to read as in CMIP6?
https://github.com/PCMDI/cmip6-cmor-tables/blob/6a30acaea9b63fdecbe08894d7e50ef36879a993/Tables/CMIP6_fx.json#L125-L132

[gnikulin]

I don't know about all CORDEX RCMs but we never included lakes to to sftlf. In general, I would say it's very strange to include lakes in land_area_fraction.

[larsbuntemeyer]

This is an overview from jsc-cordex. Seems that there is indeed a disagreement in the interpretation of sftlf:

Some models seem to count, e.g., small lakes in Norway into the land fraction while preserving, e.g., Lake Ladoga (RegCM5) while others even include that one into the land fraction (ALADIN64E1). All models, at least, seem to agree on Black Sea to be excluded from the land fraction.

EDIT: This might also depend on whether models define fractional masks or or boolean.

[larsbuntemeyer]

I now discussed this quickly with @jppietikainen. Our conclusion is (or how we do it with REMO) that sftlf is basically a land water mask, it excludes lakes from the land fraction and includes them in the water fraction. If you want a pure land sea mask without rivers and lakes, you add sftlf and sftlaf.

Can we clarify what "Including Lakes" in "Percentage of the Grid Cell Occupied by Land (Including Lakes)" mean, e.g.

• lakes are included in the water fraction? -> if there is a lake, sftlf < 1
• lakes are included in the land fraction? -> even if there is a lake sftlf == 1

I quickly checked sftlf for MPI-ESM and it seems to be the first case. But again, this could depend on whether the model allows a fractional mask at all.

[larsbuntemeyer]

Just for convenience, this is CMIP6 GCMs (also seems to be some different interpretation):

[jesusff]

Thank you, Lars. I had this check on CMIP6 models pending :-) It seems that in CMIP the definition is not clear as well.

My understanding of the CMIP "Percentage of the Grid Cell Occupied by Land (Including Lakes)" was your second option: that lakes are part of the land fraction. So, on a lake, sftlf would be 100% (land). This was definitely not standard in CORDEX up to now and came as a surprise when I learnt about it. I always expected that sftlf is a land/water mask (or fraction), regardless of whether the water surface is an open sea, a lake or a river.

From the figures above, the problem goes beyond this interpretation, as there are also water surfaces which are considered lakes by some models but not by others. E.g. the Black or Caspian seas do not appear consistently in the different models. Could this be related to the fact that these models have the ocean model handling the dynamics of the sea water (sftlf==0), but a lake model on the other water surfaces (sftlf==100 and sftlaf locating the areas handled by a lake model)?

I ping @taylor13, just in case he can shed some light on this or ping other relevant people

[taylor13]

A quick response, which I hope is right. As @jesusff suggested, the sftof (sea_area_fraction) cells > 0 are those included in dynamical ocean model calculations, and sftlf (land_area_fraction) cells = 1 - sftof.

Most ocean models have binary native grids (ocean cells and .not. ocean cells), but when the data are regridded, sftlf and sftof can be fractional.

[jesusff]

Thank you for the quick reply. This was clearly not applied in the case of RCMs. We have mainly prescribed SSTs and the lakes can have a different treatment either by being handled by a proper lake model or by setting the water surface temperature to respond with a lag to the air temperatures. On the other hand, river water is clearly visible as well in the RCM sftlf examples above with a couple of exceptions.

@pierrenabat @graziano-giuliani how did you compute the sftlf variable?

To account for all water surfaces, we are currently missing the river area fractions (sftrif?). The CF area type for this already exists (cf-convention/vocabularies#65), so it would be straightforward to define.

[jppietikainen]

Just a comment on the different Black and Caspian sea handling. One possibility is the underlying surface map. With REMO, for example, our surface is based on GLCC data, which for some reason defines Black and Caspian seas as lakes. This is not what we want when we activate our lake model and thus, we have a modified GLCC dataset in which B&C seas are sea points. Anyhow, I wanted to point this out because features like this can potentially make differences, even if we strictly define sftlf.

[gnikulin]

With regard to CMIP6, perhaps only a couple of the CMIP6 models include a lake model and in this case it's logical to provide sftlf as land+lake in CMIP6

[gnikulin]

The Caspian Sea is a very specific case. Most CMIP6 models don't provide SST for the Caspian sea even if it's presented as a sea in sftlf. Some CMIP6 models provide SST for the Caspian Sea taking values from the nearest point where SST is available (Black Sea or Persian Gulf). When we run RCM simulations for this region, we consider the Caspian Sea as a lake in the FLake model.

[pierrenabat]

Hello, for CNRM-ALADIN64E1, indeed we follow the CMIP6 definition, considering that lakes are included in the land fraction (so we can have sftlf=1 in grid points with lakes). I agree this needs to be clarified for CORDEX output, but I think it could be dangerous to have the same variable name as CMIP models with a different definition

[larsbuntemeyer]

I think one conclusion to draw from this is that sftlf is not as strictly definded as one might think but rather reflects the handling of model specific internals with regard to land atmosphere interaction, ocean coupling, and lake models. One consequence is that, e.g., if we use it for masking data in the comparison with EOBS. For example, ALADIN64E1 includes lake ladoga in the evaluation while, e.g., ICON masks it out:

So in the future, we should probably use a combination of sfltf and sftlaf in the case of EOBS.