Data gathered from multi-month to multi-year battery-powered environmental monitoring sensor networks present numerous challenges. This thesis explores three problems. First, design issues related to loading, storing and data integrity are studied in detail. An end-to-end system addressing these tasks by decoupling deployment-specific and deployment-independent phases is presented. This solution places a strong emphasis on the ability to trace the origins of every collected measurement for provenance and scientific reproducibility. Second, we explore the problem of assigning accurate global timestamps to the measurements collected using the motes’ local clocks. In deployments lacking a persistent gateway, a data-driven approach is employed to assign timestamps to within 10 parts per million. Building on these experiences, we developed a novel low-power approach to accurately timestamp measurements in the presence of random, frequent mote reboots. The system is tested in simulation and on a real deployment in a Brazilian rain forest. It is able to achieve an accuracy in the order of seconds for more than 99% of measurements even when a global clock source is missing for days to months. Lastly, this thesis explores a generic data-driven approach to reduce communication costs in order to increase network lifetime. In particular, spatial correlation among sampling stations is leveraged to adaptively retrieve data from a small subset of informative sensors rather than all instrumented locations. Soil temperature data collected every half hour for four months from 50 locations is used to evaluate this system. The method achieves a factor of two reduction in collected data with a median error of 0.06 C and 95th percentile error of 0.325 C.
This work is part of the Life Under Your Feet project developed at the Hopkins Inter- Networking Research (HiNRG) and eScience Labs at the Johns Hopkins University.
Speaker Biography
Jayant Gupchup received his Bachelors in Computer Engineering from Mumbai University in 2003. From Sep 2003 to July 2005, he worked at the Inter-University Centre for Astronomy and Astrophysics (IUCAA). In Fall of 2005, he began his Ph.D. at the Department of Computer Science at the Johns Hopkins University. His research focusses on data management in long-term environmental monitoring networks, and he is jointly advised by Dr. Andreas Terzis and Prof. Alex Szalay. In 2007, he worked at the Microsoft Bay Area Research Center as a summer intern. He received a masters in Applied Mathematics and Statistics in May 2010 under the supervision of Prof. Carey Priebe. After his Ph.D., he will join the parallel data warehousing team at Microsoft in March 2011.