This is an introduction to how this metadata is stored.

http://wiki.gbif.org/gbif/wikka.php?wakka=MMMetaData

XMP Sidecar Files
A sidecar file is an alternative to storing the metadata directly in the image file itself by instead storing the data in a separate .xmp file with the same base name as the photo. Sidecars are typically used in cases where the file format of the photo doesn't directly support embedding metadata or in cases when the image file should not be edited directly. It should be noted that very few programs support the reading the xmp sidecar files, most will default to reading and writing to the photo directly.

Gracefully copied from http://www.earlyinnovations.com/photolinker/xmp-sidecar-files.html

Note: that side car files are separate files from the photos so if that photo were to be archived it would need to form a package of two files a sidecar file and the main photo image.

Embedded Metadata
Exif
http://www.digicamhelp.com/glossary/exif-data/
http://www.opanda.com/en/iexif/
http://www.sno.phy.queensu.ca/~phil/exiftool/TagNames/EXIF.html
http://www.stuffware.co.uk/cheese/

IPTC
http://www.iptc.org/std/photometadata/2008/specification/IPTC-PhotoMetadata-2008.pdf

The funky Dublin Core Metadata Schema
http://www.iptc.org/std/Iptc4xmpCore/1.0/specification/Iptc4xmpCore_1.0-spec-XMPSchema_8.pdf
http://www.prismstandard.org/about/relationships.asp

Ok, so the items which can be recorded in Exif data spots, might not be able to be recorded in ITCP spots and vise versa. That is that some elements like copyright holder or photographer can be recorded in the ITCP data but not in the Exif data. This means that there is not 100% correspondence between the two sets. We can not choose to use one and ignore the other. When we throw XMP into the mix there are attentional things which can be recorded in XMP but not in either of the other sets. Additionally, XMP is in its own file, not embedded. Dublin Core (DC) is also a set of options for metadata. They are not embedded in the photo itself, rather in a way it is a way to organize a database of metadata about objects. REAP uses DC. DC is extensible, that is we can move embedded metadata (or sidecars) from photos into the REAP database's metadata structure. But then what happens when the photo is removed from the REAP container. Does the metadata travel with it?

Here is a clip gracefully copied from http://www.earlyinnovations.com/photolinker/annotation-philosophy.html

Many popular websites and applications allow you to annotate your photos by adding keywords, a description, a title, a location, a list of the people in the photograph and many other tags. These websites and applications generally suffer from two major deficiencies:

Annotations are often exclusively added to a propriety database, and not written back to the photo. This means that unless the software or website is still available in, say, 50 years, the annotations will be completely lost.
Programs that do write the annotations directly to the image file usually corrupt existing tags or write partial information.
PhotoLinker solves both of these issues.

1. PhotoLinker write the annotations directly to the photo so that your annotations stay with the photo forever. After annotating with PhotoLinker you can use other programs or upload to popular websites with the knowledge that your annotations will stay with your copy of the photos.
2. PhotoLinker is one of first application to adhere to Metadata Working Group Guidelines for Handling Image Metadata. These guidelines ensure that annotations are handled correctly. In addition, PhotoLinker maintains transparency about how it handles the metadata by using open source tool ExifTool and showing exactly which tags are between read and written.