Interesting concept that could prove to be quite useful. I am, however, concerned about mining shelter websites based on visual cues or a shelter's determination of "breed". As noted on your Methods page, your data mining will be most advantageous for those animal of "particular breed or mix". In most instances, those are the animals most easily adoptable or able to move into rescue (excluding bully types, which is a whole other kettle of fish.) Additionally, there is a movement to eliminate categorizing animals by breed because it sets unrealistic expectations for an adopter. Those visual cues are irrelevant. The information you'll gather will help determine how certain physical attributes such as size and age will move out of a shelter, but any contentious shelter is collecting that data now. So... are you re-inventing the wheel?
How does your measuring euthanasia rates relate directly to animals transported to another shelter? How do you factor in regional reasons for euthanasia rates increasing or decreasing?
You discuss moving animals from one shelter to another, and mention automated notification. What triggers that notification? Is this where selection by breed is used? What problem are you addressing that is not already being addressed by shelters who are currently in active transfer programs?
I think this idea is exciting, but I would like more clarification on how not only clearer determination on 'breed", but also factors in moving animals, and how you specifically envision notification.