[CF-metadata] cellmethods="lat: lon: mean over land"

From: Jonathan Gregory <j.m.gregory>
Date: Fri, 10 Feb 2006 10:20:53 +0000

Dear Beate

Thanks for your comments on this. It is a tricky issue, and I didn't explain
fully the conclusion of the previous discussion (which ought to be added to
the standard, as soon as someone has time).

> We have a similar example: two variables with standard name
> surface_snow_melt_amount
>
> SFSSMLT is value local to snow covered area
> SSSMLT is value is gridbox mean (multiplied with snow fraction)
>
> as you proposed for 'fractional land coverage' (cell_methods="lat: lon:
> mean over land" or "lat: lon: mean") we may use
> cell_methods="lat: lon: mean over snow" and cell_methods="lat: lon: mean"
> The second case is OK for me

Yes, I think the second choice would be right:
mean over snow => total mass of snow melt (kg) divided by area of snow (m2)
mean => total mass of snow melt divided by area of gridbox
It is possible that you might need
mean over land => total mass of snow melt divided by land area in gridbox
because snow can also exist on top of sea ice i.e. in the sea area. As you say,
that involves the land fraction as well.

> I think it would be better to define all variables that way that they are
> valid where really existent (snow melt valid where snow, forest fraction
> valid where land ...).

The conclusion we reached before was that this either requires too much
meaning to be "built in" to the standard names i.e. defining snow to exist
only on land, or the standard names become too awkward, because we have to
add a _where to all of them. We decided that a simple solution is this:

* Remember that the default interpretation of an intensive quantity is that
it's a *point* measurement; that's what should be assumed if there is no
cell_methods. A point measurement of snow_amount could be either missing_data
or zero if there is no snow, but it doesn't really matter. If there is a
valid non-zero value, it is well defined; a local measurement has an
unambiguous meaning.

* Use the _where distinctions for locally defined quantities in different
parts of a gridbox. Thus, we might have surface_temperature_where_sea and
surface_temperature_where_land. These are by default local measurements. If you
don't need to distinguish them as point measurements, surface_temperature is
OK without qualification.

* Recognise that this _where distinction is not the same as the area over
which averaging is done if it's a mean quantity rather than a point quantity.
We use the cell_methods to define means. Thus we could have
surface_upward_latent_heat_flux_where_snow with
cell_methods="lat: lon: mean over snow" or "mean over land" or "mean". It's
the same quantity in W in each case, divided by different areas.

Does that make sense?

> The name
> 'water_evaporation_flux_from_canopy_where_land' should not be valid any
> more if one accept my definition above...
I agree, this name does not need _where_land since there is only a canopy over
land, so I don't think the distinction is ever needed for a local quantity of
this name. We should change it (and use an alias to keep the old name valid).
As you say, the cell_methods will indicate, when it's a mean, whether is a
mean over land or over all gridbox area.

Best wishes

Jonathan
Received on Fri Feb 10 2006 - 03:20:53 GMT