New cell_methods: mabs/mibs/mebs

[zender]

1. Title

Add absolute value statistics to cell methods

2. Moderator

David Hassell

3. Requirement

The statistics mabs/mibs/mebs stand for "Maximum absolute value", "Minimum absolute value", and "Mean absolute value", respectively. They are similar to max/min/mean statistics, and they can be useful in characterizing data when one wants positive-definite metrics. mebs (unlike mean) does not allow positive and negative values to compensate eachother. Unlike rms, mebs not does weight outliers quadratically. NCO implements mabs/mibs/mebs as fundamental statistics (like max/min/mean), and annotates the cell_methods attribute of variables reduced by these statistics with the strings "maximum_absolute_value", "minimum_absolute_value", "mean_absolute_value".

4. Initial Statement of Technical Proposal

This recommendation could be implemented in CF by inserting the following lines into Table E.1. Cell Methods

​http://cfconventions.org/Data/cf-conventions/cf-conventions-1.7/build/cf-conventions.html#appendix-cell-methods

cell_methods:		Units:	Description:
maximum_absolute_value	u	Maximum absolute value
minimum_absolute_value	u	Minimum absolute value
mean_absolute_value	u	Mean absolute value

5. Benefits

This would improve the CF-compliance of fields processed by absolute-value statistics.

6. Status Quo

There is no standard CF terminology to describe absolute-value statistics so fields analyzed with these methods may have missing, erroneous, or non-standard cell_methods for this stage of processing. Unless the statistics were computed with NCO which has implemented the proposed cell_methods since February, 2015.

[zender]

Thanks David and Jonathan for your comments thus far. Jonathan asked for a use case for tabs(). I don't have a real-world use case for that. mebs() is widely used to compute absolute mean errors for intensive variables. I expect that someone somewhere sometime will want to obtain equivalent information for an extensive variable. With extensive variables often the appropriate arithmetic is summation, and tabs() would fill that need.

[taylor13]

Thanks, Charlie, for suggesting the addition of these cell_methods. I am in favor of adding all of them except sum_absolute_value, as I can't think of anyone needing this (except as a step toward computing mean_absolute_value).

I would note that most extensive properties of interest to climate scientists (that I can think of) seem to be positive definite quantities (e.g., precipitation amount, energy content, sea ice area), so perhaps it would be useful to provide an example where that isn't the case if that is to be the rationale for including sum_absolute_value.

best regards, Karl

[zender]

Thanks, Karl, for your thoughts. As I said, I do not have a real-world example for tabs(). I added it for completeness because I think someone will eventually want it. If you and others prefer to excise it, I will acquiesce. Let's give it a day or two in case David or someone else wants to fight to save it.

[jonathan]

Dear Charlie

Thanks for making this proposal. Like Karl, I support it all except for sum_absolute_value if that doesn't have a present use-case - that is our usual requirement for additions, except for very obvious corollaries. It could be added later of course if required.

Best wishes

Jonathan

[balaji]

I agree with the current proposal as well (with the omission of sum_absolute_value).

[zender]

No one has stepped forward to defend tabs()=sum_absolute_value statistics so I deleted all reference to tabs() from the proposal. As far as I know the proposal is now acceptable to everyone who has commented. I'm not sure altering the proposal was the right step but it seemed to make more sense than stating all this only in the comments section. II've never been involved with a CF trac ticket before so I don't know what the next step is. Expect that someone will let me know via taking the next step and I'll be informed electronically. Best, Charlie

[jonathan]

Thanks, Charlie. The rules are at ​http://cfconventions.org/rules.html. If no-one raises any new concerns within three weeks, the moderator (David) will conclude the discussion. As things stand now, enough support has been expressed for the ticket to be accepted according to the rules. Jonathan

[davidhassell]

I, too, am happy with the revised proposal. Thanks.

David

[davidhassell]

Three weeks have passed without comment and all are in favour with the revised proposal, so it will be marked as accepted for the next version of CF. Thanks to everyone who took the time to discuss this ticket.

David