[vikp]

I've been bouncing the idea around in my head for a while to predict the weather in an area from a photostream (ie flickr, twitter, etc). More brightness -> hotter is a simple correlation, but you could extract a lot of interesting image and meta features, I am sure (for instance, maybe people take less indoor photos in the summer).

This is more of a "fun" project in realtime, but historical weather data is actually very valuable for things like energy price prediction, and is pretty hard to get. I don't know what the market is like, but there is almost definitely one (I would be a customer if the data is accurate).