Utilizing Spatio-Temporal Text Information for Cyclone Eye Annotation in Satellite Data

Large amount of archived unannotated satellite data is publicly available. The automated retrieval of satellite data from these public domains based on ad-hoc user request is extremely useful to scientists for analysis and as evidence to support scientific hypotheses on weather phenomenon. One efficient approach for such data/information retrieval is to use the publicly available information about weather events (presented as text) to assist in better and more accurate identification of the relevant satellite data. Furthermore, by identifying the weather event in the satellite datasets that is available at finer temporal scales, one can fill in the significant “information gaps” in the text data. In this paper, we describe an approach for crossmedia cyclone track summarization and cyclone eye automated annotation using publicly available satellite data and cyclone information on the World Wide Web. Using a hurricane event as an example, we show (i) automated cyclone eye annotation for satellite data using spatio-temporal text information about the hurricane, and (ii) hurricane track summarization by processing and presenting the multiple media sources. Finally, we discuss some open problems and future work related to cross-media information retrieval and mining for weather events with an emphasis on tropical cyclones.