DAP4: Inclusion of response metadata in the DMR

From OPeNDAP Documentation

<-- back to OPULS Development

James Gallagher

1 Background

At the recent ESIP meeting, as well as the DataOne meeting that immediately preceded it, many people mentioned that responses from services would be more useful if they contained some meta information about the response. This information would help users understand how the response was made; it is essentially provenance information without the heavyweight burden that it contain everything known about those data.

2 Problem addressed

When users are confronted with web services (or feel confronted) they often wonder about the origin of the data they are accessing. Local data does not typically present this problem because there are often people close by who can answer questions about the data's origin and subsequent processing. Binding information to the data response that holds this kind of information is one way to address a fundamental issue with remote data access systems. NB: These same issues (will) arise with file transfer systems, but users seem more comfortable (or less uncomfortable...) with using data from them. Also, DataOne in particular, addresses a user population that is less likely to have used remote data archives in the past and so, as a group, are less accepting of the (remote) data's validity. These new users are raising valid questions about how they can tell if data are suitable for their needs.

3 Proposed solution

Bundle information in the DMR. The information should include:

  1. where it came from;
  2. the source data set(s);
  3. processing that happened;
  4. the software version;
  5. how to cite the dataset in a publication;
  6. licensing information/restrictions.

Based on a cursory reading of the PROV-O specification, I think that notation could be used for all of this.

The information will be encoded in a new XML element to be added to the DMR called DAPResponseProvenance. The information will be encoded in a DAP4 attribute called DAPResponseProvenance that will use the type OtherXML. The contained XML will use PROV-O (or, as Patrick West points out, the related PROV or PROV-XML) standards from the W3C. That change to a DAP Attribute of type OtherXML from a new XML element is from Dennis and Ethan.

While this could get very detailed, I propose the following for the six kinds of information listed:

where it came from
The URL used to access the resource.
the source data set(s)
Filename (not pathname, though). For an aggregation, the name of the NCML file. Extension: for aggregations, list all files touched. Because this might be complicated, it might not make an initial version.
processing that happened
This would be the name of the access (DMR, Data, ...). Extension: any server functions that were used.
the software version
This will vary for different servers (it might be one number or a list or numbers), but in the end it is one or more of the tried and true x.y.z version numbers along with names.
how to cite the dataset in a publication
Nominally a DOI or instructions. This would hopefully be boilerplate for a given server, although having it vary will be what users want since there's a move to have dataset citations list people who did the work and an institutional server will provide access to datasets with several different authors for different datasets.
licensing information/restrictions
URL reference to a license.

4 Rationale for the solution

Adding the information to the response is important - an alternative is to provide a URL that has some or all of it, but that seems too easy to break (and depends on the network always being present).

Whether the information is encoded in DAP attributes or using non-DAP XML is a bit harder. Using plain XML seems better because to encode the information as DAP attributes means developing our own set of conventions or using PROV-O (or the equivalent) but jammed into DAP attributes. No software will understand this out of the box. Using PROV-O in plain XML should make it accessible to generic XML tools as long as they can be configured to get the element that contains the information.

5 Discussion

Jimg 11:50, 30 July 2013 (PDT) Adopted the suggestion from Dennis and Ethan to use OtherXML for the provenance information.