Snow cover percentage for albedo diagnostics

[tammasloughran]

I'm interested in the diagnostics in recipe_albedolandcover.yml

I tried adding ACCESS-ESM1-5 to the yaml file. Running the recipe tells me that it needs ACCESS-ESM1-5 variable snc, which is the percentage of snow cover. This does not exist for ACCESS CMORIzed output on fs38. How can I get this variable?

[tammasloughran]

On closer inspection of raw ACCESS-ESM1-5 output, I cant even find a variable for snow cover fraction at all...

[tammasloughran]

After reading the docs for this recipe, the snc variable is used to do the diagnostics for snow free regions and snow covered regions.
Should I modify the script to ignore the snow-free and snow covered parts of the diagnostics for ACCESS?

[tammasloughran]

I've tried removing references to snc, but I'm getting some other error now. Something to do with stacking arrays. After some investigation I suspect that this has something to do with how ACCESS has no representation of pasture, while MPI/Hadgem does. However, for whatever reason, the script executes despite pastureFrac not being available.

At this point I would be inclined to throw out ESMValTool altogether and write a custom one for ACCESS myself. The diagnostics are too complicated and the data aren't available for ACCESS. i.e. I just wanted a simple linear regression for albedo against treeFrac.

[flicj191]

True, might be best to make a new script and recipe suitable for ACCESS that is simpler. Using just the variables albedo and treeFrac? and output a map with values?
fyi this is how esmvalcore derives alb: https://github.com/ESMValGroup/ESMValCore/blob/main/esmvalcore/preprocessor/_derive/alb.py
Happy to try to help

[har917]

@tammasloughran CABLE doesn't carry a snow cover fraction variable - all land in one tile will have snow or non-snow on a particular time step. In grid cells with mixed land cover you may be able to get an estimate of spatial snow cover for those few time steps when one tile has snow and the others don't - but that would require tiled snow output (i.e. again not usually available).

So in practice albedo will be best calculated via the script @flicj191 has linked (or something very similar). I would have gone with the non-clear sky variables rsds and rsus for CABLE/ACCESS since these include diffuse/direct radiation differentiation and fractionation as they impact the albedo.

More generally there are other science elements to snow cover percentage that CABLE does not consider, or that we will never be able to assess from monthly output. First (data availability) is variation within the month - i.e. how many days snow was present. Albedo is nonlinear to snow so this can be important. Second (parameterization) is sub-tile level variation in snow - meaning an ACCESS tile with 1cm snow on average will, in practice, have some area with no snow and some area with >1cm snow. This can be particularly important for energy considerations during the snow melt periods.