Wireless sensors promise to provide researchers a flexible new tool for studying biogeochemical processes in heterogeneous environments where line power may not be available and ad-hoc deployments are desirable. Because sensor battery power is limited, algorithms for adaptive measurement and communication that optimize power usage while ensuring that phenomena of interest are adequately captured could significantly enhance such deployments. Additionally, a method that allows researchers to detect areas where additional sensors would be useful, or where deployed sensors are redundant, would allow the sensor distribution to be optimized.

This paper describes a novel strategy based on random forests and reinforcement learning for improving sensor placement and organizing an optimal network topology for capturing carbon flux and related measurements in complex sub-alpine terrain at the Niwot Ridge site near Boulder, CO. The approach consists of two facets. First, a machine learning algorithm is used to learn statistical relationships between the various sensors as data are collected. The strengths of these relationships are then used to diagnose regions of unpredictability where additional sensors should be deployed or current sensors should report more frequently to better measure the processes of interest. Second, reinforcement learning techniques are used to optimize the sensor reporting and network routing strategy based on recorded network activity, battery usage, and an assessment of each sensor's value. These operational parameters are re-learned nightly and transmitted to the sensor network nodes to improve future network reporting performance and power usage balance. The proposed technique is demonstrated using simulated data, and may be utilized in a planned wireless sensor array deployment in the summer of 2007.