210 likes | 229 Views
This study aims to build and learn from a deployed sensor network system for long-term soil monitoring. It addresses the challenges of capturing heterogeneity at the mesoscale using wireless sensor networks and provides valuable input for terrestrial hydrology models.
E N D
Building and End-to-end System for Long Term Soil Monitoring Katalin Szlávecz, Alex Szalay, Andreas Terzis, Razvan Musaloiu-E., Sam Small, Josh Cogan, Randal Burns The Johns Hopkins University Jim Gray, Stuart Ozer Microsoft Research
Motivation for Building a Sensor Network Monitoring: background data, trends => • Soil animal activity/metabolic processes depend on moisture, temperature • Frequent visits disturb the sites • Soil respiration, trace gas fluxes • Better input for terrestrial hydrology models • CS: Build and learn from a deployed system
Heterogeneity • Sampling problem • Scaling problem • Large scale estimates?
Capturing Heterogeneity at Mesoscale: Wireless Sensor Networks • Small computers with radio transmitter • Each connected to multiple sensors (moisture, air and soil temperature, light) • Automatic data upload
2m 2m 8m Network Design • Ten mote network • Each mote • samples every min • data stored in FLASH • status every 2 min, wait for data request • Single hop network • Gateway connected to campus network
Temperature SensorCalibration Calibrationsin the Lab Mote Resistor Calibration Temperature sensorA/D units SoilTemperature Resistance Reference voltageA/D units Voltage Moisture SensorCalibration Moisture sensorA/D units Voltage Resistance Water Potential Light IntensityA/D units Voltage Air TemperatureA/D units CPU clock TemperatureConversion Soil Water Potential->Volumetric Conversion UTC DateTime Air TemperatureCelsius Water ContentVolumetric From Raw Data to Useful Quantities
Current Status Olin Deployment • Operating since Sep 2005 • Over 8M data points • Winding down
Database/Datacube • SQL Server 2005 database • Rich metadata stored in DB • Adopted from astronomy: NVO • Data access through web services • Graphical interface • DataCube under construction(multidimensional summary of data)
all year Season of Year season all Week of Season week Site day Patch Day of Season Hour of Day hour all sensor tenMinute all category Measurement all depth Sensor Datacube Dimension Model
Lessons Learned: Wireless Sensor Networks • Network lifetime is predictable • Nodes continue operate despite large environmental fluctuations • Waterproofing is still an issue Bathtub test
Lessons Learned: Wireless Sensor Networks II • Single-hop network: transmission range is considerably shorter than in lab due to foliage • Relay node helps • Low level programming is still required • Importance of sensor uniformity is essential • Switch to Echo sensors
Lessons Learned: Data Systems • We got real data, end-to-end ! • Sensors respond to environmental changes • Database from off-the-shelf components • Getting high level summaries : DataCube • We need a fully automated pipeline: the current two manual steps are still too labor intensive
Additional Deployments I • Leakin Park • Urban forest, BES permanent plot • Since March 06
Additional Deployments II • Baltimore Polytechnic High School • Two days ago
Integration of Sensor Data into Baltimore Ecosystem Study Projects • Urban-rural gradient studies • Water and Carbon Cycling • 200 node network at Cub Hill • Ecology of invasive species • Less fluctuating? More refuges? • Light composition – onset of reproduction • Spatio-temporal patterns of soil C and N cycling • Attachment of additional gas sensors
Many different land use /land management types Different soil conditions, soil communities Plan: to deploy 200 motes in summer 06 Neighborhood Scale Heterogeneity: Cub Hill CO2 Flux tower Maps by E. Ellis and D. Cilento, Dept. of Geography, UMBC
Acknowledgement • Microsoft Research • The Gordon and Betty Moore Foundation • Seaver Foundation • Gordon Bell • Allison Smykel, Katy Juhaszova