From Data Reverence to Data Relevance: Model-Mediated Wireless Sensing of the Physical Environment

Wireless sensor networks can be viewed as the integration of three subsystems: a low-impact in situdata acquisition and collection system, a system for inference of process models from observed data and a prioriinformation, and a system that controls the observation and collection. Each of these systems is connected by feedforward and feedback signals from the others; moreover, each subsystem is formed from behavioral components that are distributed among the sensors and out-of-network computational resources. Crucially, the overall performance of the system is constrained by the costs of energy, time, and computational complexity. We are addressing these design issues in the context of monitoring forest environments with the objective of inferring ecosystem process models. We describe here our framework of treating data and models jointly, and its application to soil moisture processes.