Talk:Attribute Convention for Data Discovery 1-2 Working
- 1 List of Open Issues
- 2 -- Graybeal (talk) 16:44, 3 May 2013 (MDT)
- 2.1 Summary of Changes re Publisher/Creator
- 2.2 Re: -- Graybeal (talk) 16:48, 3 May 2013 (MDT)
- 2.3 Re: -- Graybeal (talk) 16:48, 3 May 2013 (MDT)
- 2.3.1 Re: Re: -- Graybeal (talk) 16:49, 3 May 2013 (MDT)
- 184.108.40.206 Re: Re: Re: -- Graybeal (talk) 16:50, 3 May 2013 (MDT)
- 220.127.116.11.1 Re: Re: Re: Re: -- Graybeal (talk) 16:53, 3 May 2013 (MDT)
- 18.104.22.168.2 =Re: Re: Re: Re: Re: -- Graybeal (talk) 17:00, 3 May 2013 (MDT)=
- 22.214.171.124.3 =Re: Re: Re: Re: Re: -- Graybeal (talk) 16:58, 3 May 2013 (MDT)=
- 126.96.36.199.4 ==Re: Re: Re: Re: Re: Re: -- Graybeal (talk) 18:03, 3 May 2013 (MDT)==
- 188.8.131.52.5 ==Re: Re: Re: Re: Re: Re: -- Graybeal (talk) 17:10, 3 May 2013 (MDT)==
- 184.108.40.206 Re: Re: Re: -- Graybeal (talk) 16:50, 3 May 2013 (MDT)
- 2.3.1 Re: Re: -- Graybeal (talk) 16:49, 3 May 2013 (MDT)
- 2.4 Re: -- Graybeal (talk) 17:09, 3 May 2013 (MDT)
- 3 -- Graybeal (talk) 16:44, 3 May 2013 (MDT)
- 4 -- Graybeal (talk) 18:31, 3 May 2013 (MDT)
- 5 Adding Guidance
- 6 Computability
- 7 Cross-Referencing
- 8 Roles-by-Position vs. Roles-by-Code -- Ted.Habermann (talk) 09:16, 5 May 2013 (MDT)
- 9 creator_name and institution definitons. -- Dpsnowden (talk) 13:05, 9 May 2013 (MDT)
- 10 Feature Types (cdm and otherwise) -- Graybeal (talk) 17:40, 20 May 2013 (MDT)
- 11 Depth (!) -- Graybeal (talk) 19:17, 20 May 2013 (MDT)
- 12 People and Institutions -- Ted.Habermann (talk) 13:55, 4 June 2013 (MDT)
- 13 Conventions or Metadata_Convention -- NanGalbraith (talk) 09:40, 19 November 2013 (MST)
List of Open Issues
You may add to this list (each issue gets a row).
Also, soon someone will read all the comments below, and consolidate the open issues into this list.
|Issue number||Issue name||Description||Reference below||Recommended fix|
|1||Roles in Suggested section||Cleanup requested; current selection of role_entity not satisfactory||Ted, John et al create more precise definition of <role>_<entity> substitutable pattern|
|2||Attributes that are part of NUG or CF||Identify which, if any, terms on our list are actually defined by another standard||Nan has added NUG or CF to the attributes that are part of those standards. (was:John is working on a blended list of all attributes)|
|3||Guidance||How do we include/reference guidance?||Discussion||Every section that needs/includes guidance should have a Guidance link, that refers to the appropriate Guidance text on a separate page|
|4||Undecided Terms||Resolve open issues with terms||Resolved (review), Open (discuss)||Pick an answer and keep moving|
|5||Purpose of document||Is this document just for discovery? a LOT of terms are clearly not discovery|
|6||Internally complete||Does this document need to be internally complete per CF philosophy?|
|7||Conventions attribute||NUG recommends putting all conventions into this single attribute; ACDD originally used Metadata_Conventions attribute. Do we suggest ACDD_1.2||Use ACDD-1.2.2|
|8||Data type||Is the term cdm_data_type appropriate? We mean the THREDDS scientific layer; what terms are allowed?|
|9||creator_institution duplicate||does creator_institution duplicate CF's 'institution'?||delete?|
It might be a good idea to cross check against the definitions that NODC has added - as part of their NetCDF template project they wrote some better descriptions. They're at http://www.nodc.noaa.gov/data/formats/netcdf/
There are a few categories of terms that need better definitions, IMHO.
1. people: creator_name (recommended) publisher_name (suggested)
In a 'normal' research/observing/modeling situation, who are these people?
I think there are 2 necessary points of contact, the person who 'owns' the research and gives you the go-ahead to use/publish the data, and the person who put the data into the file and/or on line. You don't really need to know how to contact the other contributors, even if they had equally or more important roles.
I believe that NODC recommends naming the principal investigator as the 'creator' - although in some circumstances there is no single PI, so maybe we should say this is the person who grants the use of the data.
I'm using the publisher as the person who wrote the actual file that contains the terms, and I'm listing co-PIs and data processors as contributors.
Other comments are moved below. jbg
Summary of Changes re Publisher/Creator
I went with Publisher Name, Creator Name, Publisher Info (rich metadata), Creator Info (rich metadata), and Contributor Info (rich metadata); the latter could include owner or any other person/role. All of the 'rich metadata' items could include s role explicitly, presumably from a controlled vocabulary; either the same role or (if you want to create havoc) a different one.
I deleted creator_email and creator_url; if you want to add those, do it in the Info field.
Re: Summary of Changes re Publisher/Creator -- NanGalbraith (talk) 08:40, 30 July 2013 (MDT)
- I noticed that there was no publisher, just publisher institution etc, so I added publisher with a definition of The person responsible for the data file, its metadata and format.
Is that the definition we're using?
I think we have reached consensus that the _info fields are too difficult to parse (Ted's comment); should we go back to _email and _url?
Also, I moved a lot of these out of the 'recommended' category: creator_institution_info, publisher_institution, publisher_institution_info, publisher_project*
One last pitch: with thanks for reminding me, to Mike McCann:
These terms exist in ISO CI_RoleCodes, so why are we not using them?
publisher - The person responsible for the data file, its metadata and format.
principalInvestigator - The person who is responsible for the science content and intellectual property of the dataset
originator - (alternate for principalInvestigator) the person or institution responsible for the science and intellectual property in the dataset, when there is no principalInvestigator
--Graybeal (talk) 14:40, 17 September 2013 (MDT) I'm in favor of using the ISO terms. Note: The definitions above are not the ones I found in ISO; the definitions in ISO are a little bit further down, and are hopelessly self-referencing.
- Ted 4/22/2013: Most of these concerns are discussed at http://wiki.esipfed.org/index.php/NetCDF,_HDF,_and_ISO_Metadata along with more general solutions.
- Nan 4/26/2013: One other item that I think we might need to have - beyond better definitions for some of the existing terms - is a CV for contributor roles. I think one exists, somewhere, but I'm not sure where. BODC, maybe? MMI? Or should this really be free text?
- John 4/26/2013: Should be from a controlled vocabulary IMHO. BODC has (for SeaDataNet) an extension of ISO role terms, if I recall correctly. I think it isn't just for contributor roles, it's for all roles that this is needed—ISO wasn't very thorough in the first place, but there will always be new ways for people to be connected to a data set.
- I don't think we have to be restrictive (in what roles are allowed) but I think we should try to be explicit (about what a role means).
- Ted 4/26/2013: I agree completely that a shared vocabulary with definition is critical. The old ISO vocab is at https://geo-ide.noaa.gov/wiki/index.php?title=ISO_19115_and_19115-2_CodeList_Dictionaries#CI_RoleCode. Many new roles were added in the most recent revision. There is also a brief discussion at http://wiki.esipfed.org/index.php/ISO_People (I will update that list to include revisions)...
- What is really important is that the representation allow specification of the source of the code along with the code itself. This is possible in THREDDS, but not ACDD. The job of the standard is to say we use a codelist for this item and that codelist has a location. It is the communities job to say: this is the codelist that our community uses.
- Derrick S: Codelists can be seen as antithetical to the CF goal of creating self describing files. Can we figure out a way to encode ISO objects with the need for references to other objects while still staying true to our goal of remaining aligned with CF? The last thing I'd want us to recommend is to open a door down a pathway back to Grib and BUFR.
- Edward A: Regarding CF, in some ways they already use "codelists", e.g., standard names, so it is not entirely new. Its just their standard names are very human readable at the same time.
- Nan 4/26/2013: I think we can use terms from a CV, but they should be meaningful, not URLs or those lovely 5 character codes that hark back to languages we've forgotten we ever knew.
- We can select one CV, or we can add a term 'rolecode_vocabulary' (that would be fairly reasonable, since we're already using 'keyword_vocabulary').
- The SDN roles below are new, but the ISO roles are from a slightly outdated page at NODC. I just find this format easier to look at than the full xml and csv formats that are available on line.
- Personally, neither of these is very appealing - I hope the new ISO codes will be better.
- metadata collator: Responsible for the compilation of metadata for one or more datasets and submission of that metadata to the appropriate SeaDataNet metadata repository.
- programme operation responsibility: Responsible for the operation of a data collecting programme.
- programme archive responsibility: Responsible for the archive centre handling distribution of delayed mode data from a collecting programme and the long term stewardship of its data.
- programme realtime responsibility: Responsible for the centre handling distribution of true and near real time data from a collecting programme.
- contact point: Person responsible for the provision of information in response to queries concerning the metadata or underlying data.
- principal funder: Person or organisation that funds the majority of an activity. contributing funder: Person or organisation that contributes to the funding of an activity.
- principal investigator: Scientific lead of data collection within a programme
- resourceProvider: party that supplies the resource
- custodian: party that accepts accountability and responsability for the data and ensures appropriate care and maintenance of the resource
- owner: party that owns the resource
- sponsor: party that sponsors the resource
- user: party who uses the resource
- distributor: party who distributes the resource
- originator: party who created the resource
- pointOfContact: party who can be contacted for acquiring knowledge about or acquisition of the resource
- principalInvestigator: key party responsible for gathering information and conducting research
- processor: party who has processed the data in a manner such that the resource has been modified
- publisher: party who published the resource
- author: party who authored the resource
- collaborator: party who conducted or contributed to the research
- Thanks for aggregating these terms. I agree none of these role vocabularies are very appealing, but I suspect that's because the world they describe is messy. I do not see how a single vocabulary can satisfy everyone's needs, especially for keywords; nor how naming the vocabulary title creates an unambiguous reference that everyone can use to look up terms from it. I guess I'm just stuck on the lack of provided functionality in this respect.
- Ted H 4/27/2013: The suggestion to add an attribute called rolecode_vocabulary demonstrates very well the problem with this approach - a community has a documentation need and, in order to address that need, we need to add a new concept into the convention. Do we end up with a *_vocabulary attribute for every attribute that can benefit from a shared vocabulary? I think this would be difficult to maintain.
- As an alternative, we create a responsibleParty type group that includes a role from a shared vocabulary and information that describes people or organizations. The role has a value and a source which is the shared vocabulary that it comes from.
- Are we a community of convention users or convention developers? When we say we need a mechanism for describing responsibleParties that includes a role from a shared vocabulary and descriptive information, we are convention developers. When we say we need a vocabulary to describe roles like principleInvenstigator or instrumentDeveloper, we are acting as a community using a convention.
- What I am trying to do is separate these two roles so that when a community says "we need a shared vocabulary for x", we do not have to add a new attribute called x_vocabulary to the convention.
- Ken C 4/27/2013: All we say at NODC in our netCDF templates for the creator_ attributes is copied below… we discussed attributes like this a lot when documenting our templates and finally "settled" on the idea of creator being associated with "collector" of the data. Of course even that is not perfect. We don't say anything about PIs, since as Nan points out there is often no single PI. I would add that there is often no PI at all… many, many, datasets come to us now as a result of sustained and operational observing programs and systems, where the idea of a "PI" itself doesn't even apply.
- * creator_email: Email address of the person or institution that collected the data. -- The email of the person or institution may be found in the NODC tables for persons (http://www.nodc.noaa.gov/cgi-bin/OAS/prd/person) and institutions(http://www.nodc.noaa.gov/cgi-bin/OAS/prd/institution). Use the short name of the institution if available.
- * creator_name: Name of the person who collected the data. -- Use the name from the NODC persons(http://www.nodc.noaa.gov/cgi-bin/OAS/prd/person) table when applicable.
- * creator_url: The URL of the institution that collected the data. -- The url of the institution can be found in the NODC institutions (http://www.nodc.noaa.gov/cgi-bin/OAS/prd/institution) table
Nan 4/22/2013: There are a few categories of terms that need better definitions, IMHO. (continued)
2. file times
- date_created (recommended)
- date_modified (suggested)
- date_issued (suggested)
These could well have different meanings for model data; for my in situ data, I have 2 (or, for real time data, possibly 3) useful file times; the time the last edit or processing occurred, which is the version information and could be useful if the underlying data has been changed, and the time the file was written, which could provide information about translation errors being corrected. (We don't update files, we overwrite them; some people might need to describe the time the original file was written and time of last update?) For real time data it could also be interesting to know the last time new data arrived, which could be asynchronous.
NODC doesn't seem to use date_issued, but they have defs for created and modified.
- date_created: "The date or date and time when the file was created.... This time stamp will never change, even when modifying the file."
- date_modified: This time stamp will change any time information is changed in this file.
Summary of Changes re File Times
If there is the concept of date_modified, it has to be the last time the data changed (as the public sees it). That's the most important metadata of all, so now it's in the Recommended section.
If that is date_modified, then date_created has to be the original creation date, when information was first available on this file.
I could not think of a non-bizarre use case for date_issued, so I deleted it.
Since iso uses keyword type codes instead of cramming all the possible keywords (theme, place, etc) into one structure, I don't see why we don't do something similar. We could use our pseudo-groups syntax; keywords_theme, keywords_dataCenter ...etc.
Summary of Changes re Keywords
I created an arcane way to specify multiple keyword vocabularies, and implicitly allowed it to specify prefixes for the keyword field (e.g., "CF:air_temperature, IOOS_Key:Nutrients, My Favorite Keyword, AirTemperature"). I opened up the format (it's free text, why not), which leaves the battle to be fought over best practices.
- Not sure how the type codes are being considered in this context, as additional attributes or as an organizing technique inside the keywords attribute?
- I consider it a fail that there is no agreed way to support two keyword vocabularies. I therefore propose the following: If a keyword is a URI, it does not have to be a member of the Keyword Vocabulary (because its vocabulary can be derived through other means).
- I wish there were a way that Keywords and Keyword Vocabulary could have a default treatment that makes these two fields fully computer-friendly. Could we permit the Keyword Vocabulary format to be a URI, or to be specified as Name|URI, wiki-like.
4. coordinate 'resolution' terms
The word resolution is a poor choice, and if it's going to be kept, it needs to be defined as meaning 'spacing' or 'shape' and not an indication of the precision of the coordinate. For measurements that are irregularly spaced along a mooring line, it's fairly useless - unless we come up with a vocabulary describing this and other possible values.
For my data, the term might be more useful with the other definition; our depths are approximate 'target depths', and, while we may know the lat/long of an anchor and of a buoy (the latter being a time series, the former being a single point) we don't actually know the lat/long of any given instrument on a mooring line. The watch circle of the buoy is really the 'resolution' we need to supply here.
- Ooh, good point. I think in context of geospatiotemporal coverage, 'resolution' is a meaningful word, but without a definition it's wide open to misinterpretation.
- Your need is in regard to the measurements/locations provided for the data, right? The three terms that often get used to satisfy your need are precision, accuracy, and error. Can they be specified by the corresponding variable attributes?
Do we want to provide any guidance, in addition to the definition?
Re: Adding Guidance -- Ted.Habermann (talk) 09:36, 5 May 2013 (MDT)
- Guidance is incredibly important on many levels. I think it is really important to integrate the guidance into the conformance tool. We have done this more in the ISO rubric then in the ACDD rubric. The rubric results include the links to the guidance and examples... This ends up providing an integrated evaluate / improve environment...
I often try to make the definition of a parameter clear enough that a computer could recognize and do something with the answer. Is that strongly desirable, weakly desirable, or not of interest?
Summary of Approach re Computability
Some of us find it strongly desirable, but not enough to enforce it throughout. So I added it as an option in a number of places, and tried to encourage it with some of the definitions.
There are other pages with guidance and discussion about these terms. Do we want to refer the user explicitly to them, either in the document as a whole or in specific terms?
Re: Cross-Referencing -- Ted.Habermann (talk) 09:37, 5 May 2013 (MDT)
- See Guidance discussion above
Roles-by-Position vs. Roles-by-Code -- Ted.Habermann (talk) 09:16, 5 May 2013 (MDT)
Organizations and people play many roles in the scientific data life-cycle. There are two ways that those roles can be reflected in a metadata record: by position and by code. Many metadata managers are familiar with the roles by position approach because it is used in the FGDC CSDGM. The person referenced from the metadata section is the metadata contact, the person referenced from the distribution section is the distributor, and so on. Using this approach means that the object that holds information about organizations/people does not need a role indicator. That information is inferred by the position in the structure.
The ISO Standards combine the roles-by-position approach with the roles-by-code approach. Roles can generally be inferred from the positions of CI_ResponsibleParty objects in the structure, but flexibility is increased by adding a code for role to the each object. This is helpful when citing a dataset that involves people in multiple roles (principle investigator, publisher, author, resourceProvider) or when specifying the point of contact for a particular section.
The roles-by-position approach allows the roles of the people involved with a dataset to be known when they are accessed separately. For example, the xPath /gmi:MI_Metadata/gmd:contact can be used if one were interested in the metadata contact for a resource. A more general xPath (//gmd:CI_ResponsibleParty) can be used to answer the question “what people or organizations are associated with this dataset”. In the latter case, the role code provides information about roles even though the people are being accessed independent of the structure.
Multiple CI_ResponsibleParties can be included in almost all ISO objects that can include CI_ResponsibleParties. In those cases, roleCodes can be used to associate appropriate roles with particular organizations people if necessary. For example, the ISO CI_Citation object is used to refer to a variety of resources that are not included in a metadata record. It is modeled after a bibliographic reference and can include any number of organizations or people (CI_ResponsibleParties) in any roles. Typically a CI_Citation includes originators or authors and a publisher.
Re: Roles-by-Position vs. Roles-by-Code -- Ted.Habermann (talk) 09:45, 5 May 2013 (MDT)
- The discussion of role codes is interesting from many points of view. The lack of groups in the netCDF model essentially eliminates both of these approaches from consideration. There is no structure to attach organizations or people to and there are no objects to attach roles to. The only remaining alternative is the "named element" approach in which the name of the element includes the role. Are there advantages to that?
The definition of creator_name is now
- The data creator's name, URL, and email. The "institution" attribute will be used if the "creator_name" attribute does not exist.
The discussion about the roles for individuals is elsewhere in the document. My point here is that the second sentence of the existing definition includes a description of some action that will be taken. While many of us know that the actor in this case is ncISO, not everyone does. Further, we're conflating two concepts, the definition of a term and the use of that term in a particular use case (i.e. translation to ISO 10115* via ncISO). I propose that for this definition in particular and for the entire wiki in general, that we strive to separate these two concepts in the text. Let's first state what ACDD is, and what each term means, and then state one of the admittedly most common use cases.
Summary of Approach re Using Terms in Use Cases
Strove to separate the concept of how it is used from the concept of a term's definition. (One place you can't do that is in the cdm_feature term, which is very explicit about its connection to THREDDS features.)
The Unidata ACDD says
- The "cdm_data_type" attribute gives the THREDDS data type appropriate for this dataset. E.g., "Grid", "Image", "Station", "Trajectory", "Radial". Its use is recommended.
The NOAA ACDD says
- The THREDDS data type appropriate for this dataset
This is what ours currently says.
The NODC guidance says
- This attribute is used by THREDDS to identify the feature type, what THREDDS calls a "dataType". The current choices are: Grid, Image, Station, Swath, and Trajectory. These data types do not map equally to the CF feature types. If the CF feature type = Trajectory Time Series, use "Trajectory"; if Point, Profile, or Time Series Profile, use "Station".
The actual THREDDS list is called either dataTypes (code) or dataType Types (doc header), and has the same 5 types listed in the NODC guidance.
If you look up "netcdf feature type" the first link is http://www.unidata.ucar.edu/software/netcdf-java/reference/FeatureDatasets/Overview.html, which says the choices are ANY, NONE; GRID, RADIAL, SWATH, IMAGE; and ANY_POINT, which encompasses POINT, PROFILE, SECTION, STATION, STATION_PROFILE, and TRAJECTORY.
I went with something NODC-like, though it killed me not to include radial, station_profile, etc.
Re: Feature Types (cdm and otherwise) -- NanGalbraith (talk) 13:15, 9 September 2013 (MDT)
featureType is a special NetCDF attribute in CF; it gives the type of Discrete Sampling Geometry, and its presence indicates that the file contains DSG features. This opens a whole set of expectations fr the file contents, and some limitations on the dimensions and coordinates allowed. We should stick with cdm_data_type, in my opinion - although I have to ask if it is actually a discovery attribute.
Re: Re: Feature Types (cdm and otherwise) -- NanGalbraith (talk) 08:44, 30 September 2013 (MDT)
- The term cdm_datatype seems to have originated with ACDD, and it's a poor choice of terms, IMHO, since most THREDDS docs use 'data type' to mean float/int etc. Also, we might want to point to the actual unidata document that defines what we are calling cdm_data_types, at http://www.unidata.ucar.edu/software/thredds/current/netcdf-java/tutorial/PointDatatype.html That page uses the term Observation Datatypes, which is not really any more explicit than cdm_data_type. Feature type is more descriptive, but (as above) it's an overloaded CF attribute.
From the unidata page linked above, these are the definitions of the types:
"Several types of observation collections are described in the Common Data Model's Scientific Datatype layer. A Point Observation dataset contains observations which are not necessarily related in space or time. A Station Observation dataset contains time series of observations at named locations called stations. A trajectory is a collection of observations which are connected along a one dimensional track in space, with time increasing monotonically along the track. A Trajectory Observation dataset contains one or more trajectories."
Depth is fraught.
(0) Vertical positive: I almost made this required. Instead, I moved it from Suggested to Recommended. Obvious reasons.
(1) Vertical min/max: I didn't see in casual inspection a clear practice for min/max specification as a function of vertical_direction_positive = up or down. So I reused a convention established, after long thought, by OOI CI, and documented here. Trust me, there is one other option for a convention, and it is at least as confusing if not more so.
(2) Vertical units: I assume we are not going to insist on depth as the only vertical coordinate, so I explicitly mention pressure and the use of bar.
People and Institutions -- Ted.Habermann (talk) 13:55, 4 June 2013 (MDT)
The definitions that John proposed are helpful, but raise several issues. Before, we had eight attributes with roles embedded in their names (creator_name, _url, _email, publisher_name, _url, _email, contributor_name, _role) now we have twelve proposed. Many of these proposals would encourage the concatenation of multiple information elements into single fields (contributor_info, ...) with a recommendation of using vcard, ISO 19139 or free text. I am not aware of a mechanism for including ISO 19139 in netCDF attributes. Remember that NcML has the content as XML attributes which makes it fundamentally impossible to embed XML in them and very ugly to embed delimited text. This makes it likely that freetext would be the format of choice. This creates information blobs that are many times difficult to untangle and use, particularly for machines. It is also not clear how we deal with datasets that have multiple creators from multiple institutions. This is a very common circumstance these days. I am not aware of a mechanism for connecting appropriate creator_persons to appropriate creator_institutions when there are multiple occurrences of each. In fact, I do not know of an unambiguous way to include multiple creators in netCDF as it is currently implemented.
Re: People and Institutions -- NanGalbraith (talk) 13:09, 6 September 2013 (MDT)
I replaced _info fields with _url and _email for creator and publisher, because I agree that these are easier to parse. I would like to move the _url fields (along with a few others) from the Recommended section to Suggested, or possibly to add a category that isn't so much suggested as ... might be to be considered. The creator_institution_info, creator_project*, publisher_institution*, and publisher_project* fields don't aid in discovery enough to include them, in my opinion.
Conventions or Metadata_Convention -- NanGalbraith (talk) 09:40, 19 November 2013 (MST)
We need to discuss whether to remove the existing Metadata_Conventions attribute and add ACDD-1.3 (or other) to the 'Conventions' attribute, as is recommended by the unidata guidance.
From Writing NetCDF Files: Best Practices and other unidata guidance documents:
If present, Conventions is a global attribute that is a character array for the name of the conventions followed by the dataset.
The `Conventions' attribute may be a single text string containing a list of the convention names separated by blank space (recommended) or commas (if a convention name contains blanks)
Document the convention you are using by adding the global attribute "Conventions" to each netCDF file, for example:
- Conventions = "CF-1.3";
This is under discussion on the ACDD team email:
'I have always preferred the idea of using the "Conventions" attribute rather than "Metadata_Conventions". However, client support for multiple values in the "Conventions" attribute was not very good back when ACDD was originally written. And, while explicit mention of multiple values in the "Conventions" attribute have been in the NUG for some time, it is (I believe) only now slated for the next version of CF .
Does anyone have a good sense of client support for this now?
Then again, there's the chicken and egg issue. Clients will be slow to support this feature until someone starts producing data that uses this feature.' - Ethan
'We should discuss the deprecation of Metadata_Conventions more closely at the next telcon. We for one are using it currently in many, many GHRSST granules.' - Ed Armstrong