Rapid detection of rare geospatial events
Michael Olson, Annie Liu, Matthew J. Faulkner, K. Mani Chandy · 2011
The paper presents theory, algorithms, measurements of experiments, and simulations for detecting rare geospatial events by analyzing streams of data from large numbers of heterogeneous sensors. The class of applications are rare events - such as events that occur at most once a month - and that have very high costs for tardy detection and for false positives. The theory is applied to an application that warns about the onset of shaking from earthquakes based on real-time data gathered from different types of sensors with varying sensitivities located at different points in a region. We present algorithms for detecting events in Cloud computing servers by exploiting the scalability of Cloud computers while working within the limits of state synchronization across different servers in the Cloud. Ordinary citizens manage sensors in the form of mobile phones and tablets as well as special-purpose stationary sensors; thus the geospatial distribution of sensors depends on population densities. The distribution of the locations of events may, however, be different from population distributions. We analyze the impact of population distributions (and hence sensor distributions as well) on the efficacy of event detection. Data from sensor measurements and from simulations of earthquakes validate the theory.