Abstract
The billions of public photos on online social media sites contain a vast amount of latent visual information about the world. In this paper, we study the feasibility of observing the state of the natural world by recognizing specific types of scenes and objects in large-scale social image collections. More specifically, we study whether we can recreate satellite maps of snowfall by automatically recognizing snowy scenes in geo-tagged, time stamped images from Flickr. Snow recognition turns out to be a surprisingly doff cult and under-studied problem, so we test a variety of modern scene recognition techniques on this problem and introduce a large-scale, realistic dataset of images with ground truth annotations. As an additional proof-of-concept, we test the ability of recognition algorithms to detect a particular species of flower, the California Poppy, which could be used to give biologists a new source of data on its geospatial distribution over time.