In this paper, we investigate the development of distributed real-time data mining algorithms for sensor network applications with a focus on air pollution monitoring applications. Our approach is based on a considering two-layer sensor network framework comprising mobile and stationary sensors. We present architectural abstractions for such a network that are suitable for conducting peer-to-peer data mining algorithms. We also develop and evaluate the performance of a real-time peer-to-peer data clustering algorithm for the identification of pollution hotspots and for analyzing the dispersion of pollution clouds. We conduct an experimental evaluation of our algorithms to compare the accuracy of our algorithms to centralized implementations and identify the associated tradeoffs.