Reconsider the use of NetCDF-4 groups for instrument metadata

[kerfoot]

Proponent(s): @kerfoot

Moderator: @OceanGlidersCommunity/format-maintainers

Describe the error

Filing this issue to reconsider the use of NetCDF-4 groups to store instrument metadata for the following reasons:

1. It is unclear how/if ERDDAP will handle multiple groups during aggregation of 2 or more data sets. I've submitted this question to the ERDDAP Google Groups and strongly recommend reading this thread for background and the current thinking and advice.
2. The current strategy is not a CF-approved use of groups. See CF guidance on use of groups for more information.
3. To my knowledge, the current proposed NetCDF-4 groups strategy has not been tested by either simply serving a test data set and/or aggregating the CDL example to ensure that it addresses the original question of how to handle instrument metadata. Please comment with examples if this has been done and I have missed it.

4. Members of the community expressing reservations against the use of groups in this way, largely for the reasons mentioned above, and there likely are others.

   Example

   I took the CDL example, created a NetCDF file and served it up via ERDDAP:

   http://slocum-test.marine.rutgers.edu/erddap/tabledap/sp041_20191205T1757_n_measurements.html

   As you can see, the data set does not include any of the group meta data information. I have used multiple EDDTableFromFiles ERDDAP data types, but cannot get the groups included in the ERDDAP data set. I filed and additional set of questions on this topic and Bob Simons responded on September 23, 2022 with comments and suggestions. I would also suggest giving this thread a thorough reading before our next DMTT meeting.

   Potential solution

   Use empty scalar variables to attach instrument meta data. Here is an example in which I modified the CDL template from above to remove the groups and replace them with scalar variables:

   http://slocum-test.marine.rutgers.edu/erddap/tabledap/sp041_20191205T1757_n_measurements_instrument_scalars.html

   There is an additional option, though I'm less enthusiastic about implementing it in this spec as I don't believe it provides as elegant of a solution. However, we should discuss both options before proceeding.

   Platforms affected

   None

   Additional context

   As @vturpin mentioned in an email sent out September 23, 2022, it would be good to get Kevin O'Brien on the next DMTT call for some additional thoughts on this subject.

[castelao]

Thanks for opening this issue @kerfoot .

Great idea! It would be great to have @kevin-obrien 's expertise to help us.

About question 2, what is the exact issue with CF Conventions?

Could we list here what are all the reservations? It is difficult to weigh in without knowing what are the issues.

[JuangaSocib]

Hi all, very interesting issue here, thanks @kerfoot. I read carefully all the links and these are my conclusions so far:

• For now, get rid of groups: as they are it is clear that don't follow any known standard. We should re-define them to include data as well, but it seems that this not handle yet by most relevant tools (e.g. ERDDAP).
• Use scalar ancillary variables to store sensor metadata: this CF compliant and can be handled by ERDDAP as indicated above.

[justinbuck]

@kerfoot useful thread on the google ERDDAP group is here: https://groups.google.com/g/erddap/c/7g6ecOZZNNU Includes feedback on the sensor metadata.

The discussion is ERDDAP focused but eh key thing is that we are declaring the file to be CF complaint so we should be reaching a level of compliance that enable the files to work in common tools that depend on CF e.g. ERDDAP, Panoply, etc