CloudyDay correlates the mood of the Twittersphere with precipitation patterns. It can be used to generate pretty graphs that demonstrate, quantitatively, that a little bit of rain makes people happy and prone to saying nicer things, but a large amount of precipitation makes them sad, and therefore more likely to announce their displeasure to the world via twitter.
To get this data, I used the Twitter Stream to gather a data set of tweets from 270 of the most populous US cities. After gathering that data, I used AlchemyAPI's sentiment and concept analysis endpoints to gather data about each tweet. After taking the average for each city, I looked up the city's current weather conditions using WeatherUnderground's wonderfully easy to use API. Placing that data into a MySQL table database gives me an easy data set from which to draw correlations.
That said, it is a really small data set, mainly because nobody in the US is tweeting at 4 in the morning. Watching drunk tweets fly past on your terminal is really fun.