Failure Is Inevitable: The Trade-off Between Missing Data and Maintenance

Abstract

Fixing sensor faults is intuitively important, but in a class of applications where the environment is easily disturbed, there is a trade-off: maintenance requires disruption of the phenomena, so that for a period after maintenance the measurements made are not meaningful. An instance of such an application is soil monitoring. Because of its spatial heterogeneity, in-situ sensing via wireless sensor networks is a natural fit to monitoring soils. However, deploying a sensor below ground requires disturbing the soil. After a settling time, which can be weeks or more, measurements can then be trusted to reflect the state of the soil prior to excavation. Maintenance requires a similar operation, and affects all nearby sensors, not simply the one being fixed. In the San Jacinto Mountains, soil monitoring sensors were deployed to measure the flux of CO₂ from the ground to the atmosphere. The deployment consisted of several sites, where each site monitored soil temperature, moisture, and CO₂ concentration at multiple depths. The sensors within a site were within a few centimeters of each other. This chapter addresses the trade-off between fixing a broken sensor, and leaving it broken, by designing and implementing a system to both fill in missing data and keep the scientist informed about the need for maintenance. The result is potentially half as much maintenance, which translates into far more reliable data being collected.