Suggestion from Duncan Dickinson:

The least controversial form may appear as per the code below. This would require no work from implementers unless they wanted to add the format
element (which I think would be optional):

<relatedInfo type=”metadata”>

<title>EML</title>
<identifier="http://data.example.com/meta-eml.xml" type=”uri”/>

<format>

<identifier type="uri">http://some.schema.location/</identifier>
<identifier type="internet-media">application/blah+xml</identifier>
</format>
<notes>This is the EML version of the metadata</notes>

</relatedInfo>

The other version I suggested is to treat the relatedInfo as a location rather than an identifier. This has a larger impact on the relatedInfo element but I would suggest it better indicates intent - the relatedInfo is a location to which the user can go and get something rather than an identifier. The controlled vocab doesn't allow for a URL type on identifier (http://services.ands.org.au/documentati ... aries.html) but the example in the CPG shows one in use (http://ands.org.au/guides/cpguide/cpgrelatedinfo.html):

<relatedInfo type=”metadata”>

<title>EML</title>
<location>

<address>

<electronic>
<value type="url">http://data.example.com/meta-eml.xml</value>
</electronic>
<format>

<identifier type="uri">http://some.schema.location/</identifier>
<identifier type="internet-media">application/blah+xml</identifier>
</format>
</address>
</location>
<notes>This is the EML version of the metadata</notes>

</relatedInfo>

I tend to favour the first option I think. It's simpler, to the same effect. One could argue that the <location> element provides clearer semantics, but URLs are just a particular subset of URIs, and I tend to think that using "http" as your URI scheme actually implies that it's a locator (a URL) anyway, rather than just an identifier.