Proposal: Modify Cell Methods Guidelines to Accomodate Empirical Sea Ice Datasets

[Bruce-T-T-Calvert]

Topic for discussion

The current definition of the sea_ice_area_fraction standard name does not well include how it is currently being used in empirical sea ice datasets. I suggest that the CF Convention guidelines be modified to provide advice on how to better indicate the precise definition of sea_ice_area_fraction used by modifying the cell methods.

Background:
As discussed in this recent thread, the definition of sea_ice_area_fraction was clarified to include the following: "an area fraction is evaluated as the area of interest in the cell divided by the grid cell area... "Sea ice" means all ice floating in the sea which has formed from freezing sea water, rather than by other processes such as calving of land ice to form icebergs." Also, a different thread suggests that the Caspian Sea and the Great Lakes are lake regions rather than sea regions. I understand that these definitions were constructed primarily with CMIP climate modelling in mind, but their application to empirical datasets can be problematic for a few reasons:

1. Many empirical datasets, especially pre-satellite datasets, do not have information on the origin of sea ice, be it from the freezing of sea water or the calving of land ice.
2. Many empirical datasets (e.g., HadISST2, COBE-SST3, NSIDC datasets of G02202v4 G10010v2, and OSI SAF datasets of OSI-430-a and OSI-450-a) provide netcdf files with sea_ice_area_fraction values for the Caspian Sea and the Great Lakes.
3. When calculating sea ice concentrations, some datasets apply thresholds below which sea ice concentrations were set to zero. For example, HadISST2 applies a 15% threshold while G02202v4 applies a 10% threshold.

Two alternative approaches were mentioned in the first thread: changing the definition of sea_ice_area_fraction such as to include calved ice or creating a new standard name to better accommodate empirical datasets. The first alternative approach isn't desirable due to the use of sea_ice_area_fraction by the climate modelling community, while the second approach isn't desirable as it could create confusion and inconsistencies due to the existing prevalence of "sea_ice_area_fraction" and "sea ice concentration" by the empirical sea ice dataset community.

Proposal:
I propose to indicate in the CF Convention guidelines ways to clarify the sea_ice_area_fraction in the cell methods. If no cell methods are listed, then the default assumption should be that sea_ice_area_fraction aligns with its current definition (which is oriented towards the climate modelling community).

I am not sure of the best way to indicate these nuances using cell methods, but below are some ideas:

• "cell_methods="area:mean origin: all where sea"" to indicate that sea ice concentrations include calved ice, or "cell_methods="area:mean origin: sea_water where sea"" to indicate that sea ice concentrations include frozen sea water but exclude calved ice.
• "cell_methods="area:mean where sea or lake"" to indicate the inclusion of the Great Lakes or Caspian Sea.
• "cell_methods="area: mean*(ice_area_fraction > 0.15) where sea"" to indicate that a cutoff threshold of 15% was used when calculating sea ice concentrations.

This approach would involve a slight broadening of the definition of the sea_ice_area_fraction standard name. For example, the text ""Sea ice" means all ice floating in the sea which has formed from freezing sea water, rather than by other processes such as calving of land ice to form icebergs." would have to be changed, such as to "If cell_methods are not provided, then "sea ice" means all ice floating in the sea which has formed from freezing sea water, rather than by other processes such as calving of land ice to form icebergs."

[JonathanGregory]

Dear @Bruce-T-T-Calvert

Thanks for setting out these issues. It seems to me that three specific use-cases you give are of different kinds.

• The different interpretations of sea_ice in model and observational datasets is the most difficult issue. I don't think it's appropriate to use cell_methods to disambiguate the meaning of a standard name, because cell methods describe statistics of variation over coordinates. I am a modeller rather than an observationalist, but I don't think I'm prejudiced in saying that the model definition is correct. Googling "what is sea ice" returns a pageful of sources that say sea ice is frozen sea water. Floating ice that comes from land ice is not sea ice. Of course I understand that it may not be possible to distinguish them in remote sensing. For existing data, we could make a note in the description of the sea_ice standard name that in some observational datasets it might include floating ice which is not sea ice. For new data, I think we ought to recommend different and possibly new standard names, such as floating_ice_area_fraction.

• The problem of seas and lakes is a long-standing one. In a vocabularies issue we have a potential solution to allow names that could be used without making the distinction. That would help with ambiguous areas like the Great Lakes and the Caspian Sea.

• The modification of sea_ice_area_fraction to exclude values below some threshold could perhaps be done with cell_methods, but it feels a bit awkward to me, because sea_ice_area_fraction is not a coordinate variable of sea_ice_area_fraction. Maybe we could do this with a new standard name as well, such as sea_ice_area_fraction_where_above_threshold, with a default value of 0.15 for the threshold (that's commonly used, isn't it) and the possibility of providing a scalar coordinate variable to specify it otherwise. Do you think that would be acceptable?

Best wishes

Jonathan

[Bruce-T-T-Calvert]

@JonathanGregory Thanks for your thoughts and suggestions.

Rereading the previous thread, concerns with using a different standard_name were generally related to the use of marine_ice, rather than floating_ice. Thomas Lavergne mentioned that floating ice means "all ice floating on water including lake-ice, sea-ice, ice-shelves and icebergs", so adopting the floating_ice standard name could deal with both the lake issue and the calved ice issue. The current standard names include some names that have floating_ice as a prefix or suffix, but additional standard names such as floating_ice_area or floating_ice_area_fraction would need to be added. Ideally, the guidelines would be modified to suggest the use of the floating_ice standard name instead of sea_ice when dealing with observational datasets, such as the following sentence "Observational ice datasets that include ice from frozen seawater, frozen lake water, and calved land ice should use the standard name floating_ice_area instead of sea_ice_area".

That leaves the issue of the threshold. I don't have any strong views on using an alternative standard name with a "where_above_threshold" suffix. One thing to note is that the existing standard names only include "where" as aliases, so this chance could create a new precedent.

Does anyone know if thresholds are applied to other variables outside of the context of sea ice? For example, could thresholds be applied to some rainfall measurements? If there are other cases of thresholds being applied, then modifying the cell_methods conventions might be a better approach than potentially multiplying the number of standard names for a large number of variables.

[JonathanGregory]

Dear Bruce

Yes, I agree with you that if we provide floating_ice standard names, their description should indicate that they're appropriate for datasets where calved land ice can't be distinguished from frozen sea or lake water. Also, they should be advertised in the descriptions of corresponding sea_ice standard names.

There are 15 standard names which use thresholds and contain threshold e.g. number_of_days_with_air_temperature_above_threshold and fraction_of_time_with_sea_ice_area_fraction_above_threshold. I suggested using a threshold in this case as well by analogy. As you say, it would be possible to devise a cell method to describe the relationship to the threshold, but I'm not convinced that we need to do that for the moment, since the number of cases is quite small, compared with the total of 5000 standard names. Also, the case we're considering with sea ice is a different from the others in that it's self-referential; the threshold is wrt the data variable itself, not to some other quantity. There could be other uses for this idea as well, but they haven't been requested as far as I recall. I agree that there might be better words we could use for where in the standard name.

Best wishes

[Bruce-T-T-Calvert]

There are 15 standard names which use thresholds and contain threshold e.g. number_of_days_with_air_temperature_above_threshold and fraction_of_time_with_sea_ice_area_fraction_above_threshold. I suggested using a threshold in this case as well by analogy.

My apologies. Earlier, I must have been either looking at an outdated standard name table, or had text in the search bar. You are correct regarding the prevalence of the use of "above_threshold" in standard names, so it makes sense to use that approach.

There is also the question as to how to indicate the threshold amount. In the existing CF Convention guidelines, there is a single example (7.12) where a temperature threshold was indicated as a coordinate. It seems very odd to define thresholds as coordinates rather than in cell methods. An alternative approach could be to include text like "threshold: 0.15" in the cell_methods.

So, overall, I suggest the following:

1. Add the following standard names: floating_ice_area, floating_ice_area_fraction, floating_ice_area_fraction_above_threshold, sea_ice_area_fraction_above_threshold
2. The description of sea_ice_area_fraction should include the following sentence: "Observational ice datasets that include ice from frozen seawater, frozen lake water, and calved land ice should instead use the standard name floating_ice_area_fraction."
3. The CF Convention guidelines subsection on cell methods (currently 7.3) should include a brief subsubsection on thresholds, which indicates that threshold values should be indicated in cell_methods such as by using "threshold: 0.15" to indicate that a 15% cut-off was applied to an observational floating ice concentration dataset.
4. Modify example 7.12 to use cell_methods rather than coordinates to define the threshold.

[JonathanGregory]

Dear @Bruce-T-T-Calvert

CF uses coordinate variables for all purposes where the data variable is a function of the numerical value of some other quantity. The coordinate variables are the independent variables on which the data variable depends. Thresholds are a particular case of this, because the data variable's values depend on the choice of threshold. There are many other examples e.g.

air_potential_temperature: To specify the standard pressure to which the quantity applies, provide a scalar coordinate variable with standard name reference_pressure.

blue_component_of_toa_outgoing_shortwave_flux_expressed_as_rgb_color: The radiation wavelength or frequency range for blue can be denoted as bounds of a scalar coordinate variable.

tidal_sea_surface_height_above_mean_higher_high_water: To specify the tidal datum epoch to which the quantity applies, provide a scalar coordinate variable with standard name reference_epoch.

For thresholds, I think this is better than putting a numerical value in cell_methods, because:

• It's harder to extract a numeric value from cell_methods, which is a complicated string attribute, than from a numerical coordinate variable identified by a particular standard name.

• You might want to compute the results for more than one choice of the threshold and store them in the same data variable, in which case a multivalued coordinate variable is very convenient.

Regarding the new standard names, I think above_threshold by itself in floating_ice_area_fraction_above_threshold is potentially confusing. It could be interpreted as meaning the amount by which the sea ice area exceeds the threshold. My suggestion of where was to avoid this ambiguity. I suggested floating_ice_area_fraction_where_above_threshold for the quantity which equals the area of floating ice in the gridcell divided by the area of the gridcell if this ratio equals or exceeds the threshold (usually 0.15), but equals zero if the ratio is less than the threshold. Is that the quantity you want? It can be regarded as floating_ice_area_fraction multiplied by a binary mask which is zero where floating_ice_area_fraction<0.15 and unity where floating_ice_area_fraction ≥ 0.15. Is that right?

I agree that we could define a new cell_methods for this kind of operation, of multiplying X by a binary mask of (X ≥ threshold), where the threshold is stored in a scalar coordinate variable or a coordinate variable, but yours this is the only use-case we have for this operation so far. Since a new cell_methods would require a change to the convention, it's easier to define a standard name to indicate this operation. Also, I think it's more convenient for a program to search a file for a particular standard name than for a particular cell_methods entry.

Others' views would be helpful. I'll label this discussion as suitable to be refered from the FAQ.

Best wishes

Jonathan

[Bruce-T-T-Calvert]

@JonathanGregory thanks for this additional context and insight. I wasn't considering these other uses of coordinates such as the use of reference_pressure. Given this context, it makes sense to have a consistent approach for floating_ice, so there is no need to modify cell methods.

I also agree that it is more sensible to use _where_above_threshold rather than _above_threshold for empirical sea ice datasets.

Perhaps an example of using floating_ice_above_threshold in the CF guidelines could be useful.

[larsbarring]

I have not followed this issue in detail, nor the one on lake or sea standard names. Hence I limit my comment to the question of a new cell method mechanism vs. threshold coordinate. I fully support @JonathanGregory's preference for the latter for exactly the same reasons as Jonathan write about.

[GeyerB]

Dear @Bruce-T-T-Calvert and @JonathanGregory's,
as we are regional climate modellers, we would like to get a new standard_name for all ice covered water surfaces (nowadays the long_name is wrongly "fraction of sea ice" in the ICON model). So, the suggestion by Jonathan
floating_ice_area_fraction would be perfect for us.
This relates to the discussion on water bodies and the separation of land and sea water variables (cf-convention/vocabularies#25 (comment))
Best regards,
Beate

[efisher008]

Dear @GeyerB,

Have you opened a new issue in the cf-vocabularies repo to propose this name? NB: I believe that, as floating_ice is an existing area type, and area_fraction is a phrase in use in many existing standard names, this could be accepted quite quickly (given the correct description).

Best wishes,
Ellie

[Bruce-T-T-Calvert]

I have now created a new vocabularies issue: cf-convention/vocabularies#271