DR v1.2.2 sea ice cell_methods issue 1

[taylor13]

sea ice cell_methods issue 1

The correct cell_methods should be: "area: mean where sea_ice over all_area_types" (in SImon.siflsaltbot and in 7 other variables; see column DQ of https://docs.google.com/spreadsheets/d/1biV1mSsAMDIWO7lGC_kYCDC-MYS0qRqi2U2KYlFynJU ).

Justification:
The current cell_methods is inconsistent with Notz et al., (2016) and with the data request's "description" which indicate that this is the total of something in the sea ice portion (area) of the cell divided by the cell area. If the current cell_methods ("area: mean where sea_ice") is correct, then the description should read: "Total [of something] .... divided by sea ice area."

This may impact what data gets reported because clearly different groups will follow conflicting guidance provided by the descriptions and metadata.

When this issues was raised earlier (#66), it did not lead to any clarification as to what was being requested, and no technical objections were raised concerning the change.

[bethdingley]

Pinging @pderepen @martinvancop

[pderepen]

The correct cell_methods should be: "area: mean where sea_ice over all_area_types" (in SImon.siflsaltbot and in 7 other variables; see column DQ of https://docs.google.com/spreadsheets/d/1biV1mSsAMDIWO7lGC_kYCDC-MYS0qRqi2U2KYlFynJU ).

Can you clarify what over all_area_types mean? I haven't seen this cell_methods before. Isn't sea ice always over ocean? I don't understand what more it adds compared to area: mean where sea_ice.

Justification: The current cell_methods is inconsistent with Notz et al., (2016) and with the data request's "description" which indicate that this is the total of something in the sea ice portion (area) of the cell divided by the cell area. If the current cell_methods ("area: mean where sea_ice") is correct, then the description should read: "Total [of something] .... divided by sea ice area."

This may impact what data gets reported because clearly different groups will follow conflicting guidance provided by the descriptions and metadata.

Actually, my sense is that those two things are not inconsistent with each other. The cell_methods describes the method to use when calculating variable (in the case of SImon.siflsaltbot we are looking for the flux of salt between water and sea ice, which obviously can only be calculated over the portion of the grid cell covered by sea ice otherwise this flux is irrelevant), but then the description indicates how to report it (i.e., if the units of m-2 refer to the area of sea ice or the area of the grid cell). In the case of SImon.siflsaltbot and other freshwater fluxes you mentioned in https://docs.google.com/spreadsheets/d/1biV1mSsAMDIWO7lGC_kYCDC-MYS0qRqi2U2KYlFynJU, the reason we want them to be reported by grid-cell area is to be able to easily sum them over grid cells to calculate budgets.

When this issues was raised earlier (#66), it did not lead to any clarification as to what was being requested, and no technical objections were raised concerning the change.

Apologies for not responding then, I was not tagged on issue #66.

[taylor13]

Admittedly, we're getting into CF convention "weeds" here. The construct "where X over Y" in cell_methods is described here. There can also be some confusion about terminology. In the climate sciences, the term "flux" (with units W m-2) is substituted for "flux density", which in other scientific disciplines identifies the same property. We sometimes use the term "total flux" to what we get when multiplying the flux by the area of the region where the flux has been evaluated; in other disciplines, they call this simply the "flux" so that in their terminology flux = (flux density)*area. In the data request (and below), I'll refer to "flux" and "total flux" as normally used in our field and as defined above.

The current cell_methods = "area: time: mean where sea_ice (mask=siconc)". The stuff is parentheses is allowed by CF but not standardized, and I'll omit it subsequently for simplicity.

Following the current "where directive", the fluxes would be averaged over the sea_ice portion of the grid cell (i.e., the total flux at the interface between the sea ice and adjacent medium would be calculated and then divided by the area of sea ice).

Notz et al. (2016) define the variables as the "total flux in the sea ice portion (area) of the cell divided by the cell area. " Since cell area may be larger than sea ice area, this is generally different from what the current cell_methods calls for. If we want to follow the Notz prescription, then cell_methods="area: time: mean where sea_ice over all_area_types". Under the CF conventions, this means take the total flux over the sea ice portion of the cell and divide it by the area of the cell, consistent with Notz et al.

[martinvancop]

There are 8 variables concerned by this comment.

Simon.siflsaltbot
Simon.simass
Simon.sisaltmass
Simon.sivol
Simon.sisnmass
Simon.sidmassevapsubl
Simon.siflwbot
Simon.siflwdrain

All these are "per grid cell area", as stated by Notz et al (2016), which is correct (not the current averaging method)

Therefore we believe the averaging method should be "mean_where_seaice over all_area_types" for these 8, provided that this "all_area_types" includes open water

Are we on the same line ?

[taylor13]

Yes, I think we are. Those are the affected variables and "all_area_types" indeed includes open water, sea ice, (and land for data that has been regridded away from the native ocean/sea ice common grid).

[pderepen]

I'm not sure I understand why, out of the 8 variables concerned by this comment:

Simon.siflsaltbot
Simon.simass
Simon.sisaltmass
Simon.sivol
Simon.sisnmass
Simon.sidmassevapsubl
Simon.siflwbot
Simon.siflwdrain

there is only one of the sea ice mass tendency terms (sidmassevapsubl). Shouldn't all mass tendency terms have the same cell_methods? Can you clarify?

[taylor13]

I'm not absolutely positive about this, but I think the explanation is that mass is an extensive quantity, so if you measure a mass tendency, that too is extensive. A "flux" (i.e., flux density) is an intensive quantity, the flow of energy per unit area. So in regions of at least partial sea ice coverage, variables like tendency_of_sea_ice_amount_due_to_basal_melting will be reported the same whether cell_methods is "where sea", "where sea_ice", or with no "where" directive included.

I'm not sure how "where sea_ice over all_area_types" would get interpreted in this case.

[bethdingley]

I'm happy to update the cell methods to include mean_where_seaice over all_area_types for

• Simon.siflsaltbot
• Simon.simass
• Simon.sisaltmass
• Simon.sivol
• Simon.sisnmass
• Simon.sidmassevapsubl

Just noting in the earlier list there were two typos

• Simon.siflwbot --> SImon.siflfwbot
• Simon.siflwdrain --> SImon.siflfwdrain

There is also a daily variable I can update:

• SIday.sisaltmass

@pderepen I just want to check in - does @taylor13's comment address your concern about sea ice mass tendancy terms or do we need to update the cell methods of more variables? We're hoping to do another DR release next week so quick engagement would be appreciated if possible.

I'll implement the change as outlined here on Monday/Tuesday next week to allow time for any additional comments

[taylor13]

Thanks, @bethdingley for pursuing this issue (and thanks to @pderepen for pointing out the inconsistency). I have reviewed my comments above, and I think I misinterpreted the term "amount", which turns out to be an intensive quantity (kg/m^2), not extensive (kg). As described in Notz et al. (2016), all of these should also include "mean_where_seaice over all_area_types" in the cell_methods. So, they (as well as SIday.sisaltmass, which you already noticed) need corrected cell_methods.

Bottom line is the following also need cell_methods set to area: mean where sea_ice over all_area_types time: mean:
SImon.sidmassmeltbot
SImon.sidmassgrowthbot
SImon.sidmassgrowthsi
SImon.sidmassgrowthwat
SImon.sidmassmeltlat
SImon.sidmassdyn
SImon.sidmassth
SImon.sidmassmelttop

[pderepen]

I want to take a step back here to clarify things as I feel some details have gotten mixed up in this discussion (myself included!).

According to Notz et al. (2016), the difference between an extensive and intensive variable is:

• An extensive variable is proportional to area fraction, such that it naturally approaches zero as area fraction approaches zero. As such, zeros should be averaged in where no sea ice is present. Good examples of extensive variables are sivol and simass. The cell_methods for these variables should be the newly proposed area: mean where sea_ice over all_area_types by @taylor13.
• An intensive variable on the other hand is not proportional to area fraction, and as such does not necessarily approaches zero as area fraction approaches zero. In this case, only time samples with non-zero sea ice fraction are multiplied by area fraction, then summed and then divided by the sum of the sea ice area. Good examples of intensive variables are sithick and sitemptop. The cell_methods for these variables should be something like area: time: mean where sea_ice (I might not be using the correct CF nomenclature here, but basically this means summing over the sea ice portion of the grid cell and dividing by the sea ice area).

In the spreadsheet first provided by @taylor13 (https://docs.google.com/spreadsheets/d/1biV1mSsAMDIWO7lGC_kYCDC-MYS0qRqi2U2KYlFynJU), there were 8 sea ice variables initially concerned by this comment:

• simass, sivol, and sisaltmass: These variables are all extensive variables and had the cell_methods area: mean where sea time: mean, which was correct as it is averaging the zeros where no sea ice present (i.e., where sea, not where sea ice). This is equivalent to the newly proposed cell_methods area: mean where sea_ice over all_area_types, thus that change can be implemented.
• siflsaltbot, siflfwbot, and siflfwdrain: According to Notz et al. (2016), these are intensive variables and had the cell_methods area: time: mean where sea_ice (mask=siconc), which is consistent. However, the description of those variables states that those fluxes are divided by grid-cell area, which contradicts the cell_methods. Interestingly, out of all the sea ice flux terms (sifl***: energy, sensible, latent, conductive, freshwater/salt), these are the only ones that mention "divided by grid-cell area" in their description and contradicts the fact that they are all be intensive variables. Note that this is not something we changed for this phase of CMIP, the descriptions were the same in CMIP6. @martinvancop do you see a reason why the freshwater/salt fluxes are divided by grid-cell area and not any of the other fluxes? Could it be to make it easier to close the salt budget with the ocean?
• sisnmass: According to Notz et al. (2016), this is an extensive variable but it had the cell_methods area: time: mean where sea_ice (mask=siconc), which is inconsistent with the definition of an extensive variable and with its description, which states that this is the total mass of snow on sea ice divided by grid-cell area. Therefore, the cell_methods needs to be updated to area: mean where sea_ice over all_area_types.
• sidmassevapsubl: This variable and all other sea ice mass tendency terms (sidmassth, sidmassdyn, sidmassgrowthwat, sidmassgrowthbot, sidmassgrowthsi, sidmassmelttop, sidmassmeltbot, sidmassmeltlat) are extensive variables and had the cell_methods area: mean where sea time: mean, which is consistent with the definition of an extensive variable and with their description. Again, this is equivalent to the newly proposed cell_methods area: mean where sea_ice over all_area_types, thus that change can be implemented.

Hopefully, this makes things a bit clearer.

A few notes:

• @bethdingley: I did not add SImon or SIday when discussing these variables, but any change should apply to any compound name that includes that variable, regardless of if it's requested at the daily and/or monthly time frequency.
• @taylor13: The newly proposed cell_methods area: mean where sea_ice over all_area_types is missing the time dimension, so there probably a time: mean that needs to be added somewhere. Since I'm less familiar with CF convention I'll leave that up to you.

[taylor13]

Thanks, @pderepen , for reviewing carefully all of this.

To answer your last question, yes all the variables under discussion should precede the "where" directive with area: time: mean (so, area: time: mean where sea_ice over all_area_types).

Also note that area: mean where sea is not always equivalent to area: mean where sea_ice over all_area_types. Consider the case when data on a native ocean grid is mapped to some more regular grid and we find a target grid cell that happens to fully include only 3 native grid cells, with each native cell occupied entirely by a single area type: one cell being ice-free sea, one cell being land, and one cell being sea ice). Then for a variable "v", the value in the target (regridded) cell is computed from the values of v found in the original cells as follows, assuming the 3 cells are of equal area:

for area: mean where sea:
v(target) = [v(ice-free sea) + v(sea_ice)]/2.
Now under the presumption that v(ice-free sea) is zero, then
v(target)= 0.5*v(sea_ice)

for area: mean where sea_ice over all_area_types:
(target) = (1/3)*v(sea_ice)
In this case the mean is computed only over sea_ice (yielding "v(sea_ice)) and then scaled by (area of sea ice)/(area of all area types), i.e., by 1/3.

Only when there is no land contribution to the value will the two cell_methods yield the same value.

[bethdingley]

• @bethdingley: I did not add SImon or SIday when discussing these variables, but any change should apply to any compound name that includes that variable, regardless of if it's requested at the daily and/or monthly time frequency.

I can confirm changes will be made across SIday and SImon variables consistently. There was a previous issue of the daily and monthly variables being inconsistent with each other but we have fixed that now.

Based on your comments I have implemented the below. If Karl's comments change anything (or if I've misinterpreted your comments), please let me know. We are planning to create the release today, but will likely be able to incorporate small updates like these tomorrow if needed!

CMIP6 compound name	Cell methods	Description	Possible further change needed (awaiting confirmation)
SImon.simass	area: mean where sea_ice over all_area_types time: mean	Total mass of sea ice divided by grid-cell area.
SImon.sivol	area: mean where sea_ice over all_area_types time: mean	Total volume of sea ice divided by grid-cell area, also known as the equivalent thickness of sea ice.
SIday / SImon.sisaltmass	area: mean where sea_ice over all_area_types time: mean	Total mass of all salt in sea ice divided by grid-cell area. Sometimes, models implicitly or explicitly assume a different salinity of the ice for thermodynamic considerations than they do for closing the salt budget with the ocean. In these cases, the total mass of all salt in sea ice should be calculated from the salinity value used in the calculation of the salt budget.
SImon.siflsaltbot	area: time: mean where sea_ice (mask=siconc)	Total flux of salt from water into sea ice divided by grid-cell area. This flux is upward (negative) during ice growth when salt is embedded into the ice and downward (positive) during melt when salt from sea ice is again released to the ocean.	Possibly need to remove 'divided by grid-cell area' from Description
SImon.siflfwbot	area: time: mean where sea_ice (mask=siconc)	Total flux of fresh water from water into sea ice divided by grid-cell area. This flux is positive when fresh water enters the ocean, and is therefore negative during ice growth and positive during ice melt.	Possibly need to remove 'divided by grid-cell area' from Description
SImon.siflfwdrain	area: time: mean where sea_ice (mask=siconc)	Total flux of fresh water from sea-ice surface into underlying ocean divided by grid-cell area. This combines both surface meltwater that drains directly into the ocean and the drainage of surface melt ponds. By definition, this flux is always positive.	Possibly need to remove 'divided by grid-cell area' from Description
SImon.sisnmass	area: time: mean where sea_ice over all_area_types	Total mass of snow on sea ice divided by grid-cell area. This then allows one to analyse the storage of latent heat in the snow, and to calculate the snow-water equivalent.	Is the placement of the time averaging in the cell methods correct?
SImon.sidmassevapsubl	area: mean where sea_ice over all_area_types time: mean	Rate of change of sea-ice mass change through evaporation and sublimation divided by grid-cell area. If a model does not differentiate between the sublimation of snow and sea ice, we recommend to report sidmassevapsubl as zero as long as the ice is snow covered, and to report any sublimation within the variable sisndmasssubl.
SImon.sidmassth	area: mean where sea_ice over all_area_types time: mean	Total rate of change in sea-ice mass from thermodynamic processes divided by grid-cell area.
SImon.sidmassdyn	area: mean where sea_ice over all_area_types time: mean	Total rate of change in sea-ice mass through dynamics-related processes (advection, divergence, etc.) divided by grid-cell area.
SImon.sidmassgrowthwat	area: mean where sea_ice over all_area_types time: mean	Rate of change of sea-ice mass due to sea ice formation in supercooled water (often through frazil formation) divided by grid-cell area. Always positive or zero.
SImon.sidmassgrowthbot	area: mean where sea_ice over all_area_types time: mean	Rate of change of sea-ice mass due to vertical growth of existing sea ice at its base divided by grid-cell area. Note that this number is always positive or zero since sea-ice melt is collected in sidmassmeltbot. This is to account for differential growth and melt in models with a sub-grid scale ice thickness distribution.
SImon.sidmassgrowthsi	area: mean where sea_ice over all_area_types time: mean	Rate of change of sea-ice mass due to transformation of snow to sea ice divided by grid-cell area. Always positive or zero.
SImon.sidmassmelttop	area: mean where sea_ice over all_area_types time: mean	Rate of change of sea-ice mass through melting at the ice surface divided by grid-cell area. This number is independent of the actual fate of the meltwater, and will hence include all sea-ice meltwater that drains into the ocean and all sea-ice meltwater that is collected by a melt-pond parameterisation. Always negative or zero.
SImon.sidmassmeltbot	area: mean where sea_ice over all_area_types time: mean	Rate of change of sea-ice mass through melting/dissolution at the ice bottom divided by grid-cell area. Note that this number is always zero or negative since sea-ice growth is collected in sidmassgrowthbot. This is to account for differential growth and melt in models with a sub-grid scale ice thickness distribution.
SImon.sidmassmeltlat	area: mean where sea_ice over all_area_types time: mean	Rate of change of sea-ice mass through lateral melting/dissolution divided by grid-cell area (report zero if not explicitly calculated thermodynamically). Always negative or zero.

[pderepen]

@taylor13: thank you for the explanation of when area: mean where sea would not be equivalent to area: mean where sea_ice over all_area_types, I hadn't been thinking about regridded data and land portion of grid cells being averaged in.

@bethdingley: After additional discussions with SIMIP steering group members, we have decided that it would be preferable to have the 3 freshwater terms (siflsaltbot, siflfwbot, and siflfwdrain) follow a sea-ice perspective of the freshwater budget and be considered intensive variables, such that their cell_methods should be kept as area: time: mean where sea_ice (mask=siconc). As such, to be consistent with the cell_methods, the mention of "divided by grid-cell area" needs to be removed from the description of these variables.

Finally, regarding the other comment you have in the last column of the table about the placement of the time averaging in the cell_methods, I believe how you wrote it for SImon.sisnmass is correct (i.e. area: time: mean where sea_ice over all_area_types) and should also apply to all other variables listed in the table, based on @taylor13's response (except for the 3 freshwater terms just discussed):

To answer your last question, yes all the variables under discussion should precede the "where" directive with area: time: mean (so, area: time: mean where sea_ice over all_area_types).