CNSR 2007 Keynote Speaker:

Sensor Networks for Environmental Monitoring

Gregory J Pottie
Deputy Director, Center for Embedded Networked Sensing
University of California, Los Angeles
Thursday, May 17, 2007 (Scheduled Presentation Date)

When sensor network research began in earnest in the 1990's it was in the context of applications such as detection of vehicles or personnel using dense networks of low-cost sensor nodes, each of which was limited in some combination of energy reserves, processing capability, and memory, and possessing quite simple sensors. This led to a large number of interesting optimizations, and much work continues in these directions today. However, the vision of large-scale deployments of dense, flat networks has not actually come to fruition in practical applications. Our experience in developing sensor networks for environmental science and monitoring applications provides some insight into why this is so, and how the challenges of practical deployments give rise to deep theory questions. In this talk, our experiences in deploying networks for terrestrial and aquatic ecology applications will be described. Some of the design consequences are outlined below.

In science applications, the goal is typically to construct better models of a physical environment. We begin with a trusted apparatus and model that is limited in some way (e.g., too costly, limited in scale) and seek to extend it, in the process generating new science questions. Ecological processes are very complicated, with typically many layers of modeling required to provide even an approximate characterization. Thus, experimentation is an iterative process in which user interactions are very important. Calibration is an ongoing problem in deployments in natural settings, which together with uncertainty in the models leads to the question of trust in the data being a paramount design consideration. Validation of instruments, procedures, and models dominates the effort, which together with the requirements for iterated design results in the requirement that nodes have much more robust communications, storage, and processing capabilities than earlier assumed. Energy and communication infrastructure can be required, and mobile elements employed to ease many of the logistical tasks and to more deeply probe regions where the observations indicate uncertainty of results. The design space is thus much richer than earlier imagined, with classic sensor networks being one tier of many required for investigating nature at dense scales.