We propose a next-generation cloud-based social media visual analytics disaster response system that will enable decision-makers and first-responders to obtain real-time analytics (both descriptive and predictive) for situational awareness and early-warning. Social media sites like Twitter and Facebook provide rapid "sensory" feedback from "human sensors" the millions of people on the ground including those affected by the disaster as well as hundreds of first-responders and relief organizations. The problem is that this social media data can be overwhelming and noisy (lots of irrelevant stuff). Our proposed innovation is to combine and analyze this social media unstructured "big-data", together with other structured data from a variety of sources including NASA. The emphasis is on visual analytics which will allow all these diverse geospatial data to be integrated and analyzed on a real-time basis in an intuitive manner. We utilize geobrowsers such as COAST, Google Earth, and OpenStreetMap for display of the output of the visual analytics prediction system. This is an interactive system which enables the relevant users and decision-makers to provide input to the analytics process as the solution evolves.