A Systematic Probabilistic Approach for Estimation in Dense Wireless Sensor Networks

In this paper we present a novel systematic approach for inference of missing observations in environmental monitoring wireless sensor networks (WSNs). This capability is important in that it enables sensor networks to effectively estimate the missing data when observation is incomplete to address two critical challenges in WSNs: (1) how to achieve sensor node failure tolerant and robust data collection in harsh environments, and (2) how to minimize energy consumptions through putting some sensor nodes into sleep alternatively. In our approach, we use Markov random fields (MRFs) to model the spatial correlation in sensor networks. Our MRF model is constructed by using iterative proportional fitting (IPF), and then Loopy belief propagation (LBP) is employed to perform the approximate inference given an incomplete network observation. We demonstrate our approach using real-world sensed soil moisture data in environmental monitoring. Our preliminary results show the great promise of the proposed approach.