Application of Wireless Sensor Network to Global Environmental Monitoring

1
Application of Wireless Sensor Network to Global
Environmental Monitoring
Presented at 2nd International Conference on
Sensor Networks and Applications (SNA06), October
19, 2006

• Yu Hen Hu
• University of Wisconsin Madison
• Dept. Electrical and Computer Engineering
• Madison, WI 53706
• Hu_at_engr.wisc.edu
• Collaborators Tim Kratz, Barbara Bensen, Paul
  Hanson, Laurence Choi and GLEON team

Research sponsored by US National Science
Foundation, Moore Foundation and others
2
Agenda

• Sensor network
• Multi-disciplinary technologies
• Multi-domain applications
• Sensor network for environmental monitoring
• Great duck island project
• GLEON project
• Other related project
• GLEON
• Objectives, status
• Technology challenges

3
Wireless Sensor Network

• A computer network consisting of spatially
  distributed autonomous devices using sensors to
  cooperatively monitor physical or environmental
  conditions, such as temperature, sound,
  vibration, pressure, motion or pollutants, at
  different locations.
• The development of wireless sensor networks was
  originally motivated by military applications
  such as battlefield surveillance. However,
  wireless sensor networks are now used in many
  civilian application areas, including
• environment and habitat monitoring,
• healthcare applications,
• home automation, and
• traffic control

Wikipedia
4
Characteristics of a Sensor Network

• Small-scale sensor nodes
• Limited power they can harvest or store
• Harsh environmental conditions
• Node failures
• Mobility of nodes
• Dynamic network topology
• Communication failures
• Heterogeneity of nodes
• Large scale of deployment
• Unattended operation

Wikipedia
5
Sensor Network Design

• Sensor node
• Hardware
• Sensors, wireless transceivers, actuators
• Software
• Operating system, network stacks, sensor
  middleware
• Ad Hoc Network
• Self configuration, re-configuration
• Collaborative (in-network) processing
• Data aggregation

6
Sensor Network Applications

• Environmental monitoring
• Habitat monitoring
• Acoustic detection
• Seismic Detection
• Military surveillance
• Inventory tracking
• Medical monitoring
• Smart spaces
• Process Monitoring

Wikipedia
7
A Typical SNA Scenario

• Sensor nodes are deployed over a sensing field
• Sensing
• To measure physical quantities
• Communicating
• To communicate with neighboring nodes or
  collector nodes to
• Report (and aggregate) measurements
• Relay data and command within the network
• Actuating
• To take proper actions/decisions based on
  commands and measurements

8
Examples of Sensor Nodes
Accsense sensor pod
Atlas ZigBee modules
CrossBow Mica Motes
EYES sensor module (EU project)
9
Smart Dust

• a self-contained, millimeter-scale sensing and
  communication platform for a massively
  distributed sensor network. 
• device will be around the size of a grain of sand
  and will contain sensors, being inexpensive
  enough to deploy by the hundreds. 

• computational ability, bi-directional wireless
  communications, and a power supply

10
Smart dust
http//www-bsac.eecs.berkeley.edu/warneke/SmartDu
st/index.html
11
Sensor Network for Environmental Monitoring

• Sensor networks will produce a revolution in our
  understanding of the environment by providing
  observations at temporal and spatial scales that
  are not currently possible.
• Recommendations
• Development of models, algorithms,
• Process automation (scaling)
• Better sensing technologies
• Infrastructure support for sensor network
  development and deployment

US NSF workshop on Sensors for environmental
Observations Report
http//www.wtec.org/seo/final/Sensors_for_Environm
ental_Observatories.pdf
12
Grand Environmental Challenges

• Biological diversity and ecosystems function,
• Invasive species,
• Climate variability and ecological responses to
  climate change,
• Hydrologic forecasting to predict changes in
  fresh water sources,
• Biogeochemical cycles and their impacts on
  ecosystems,
• Infectious diseases and their interactions with
  the environment,
• Land-use changes as they impact ecosystems
  services and human welfare, and
• Materials uses in relation to environmental
  impacts of their residuals.

National Research Council (NRC). 2001. Grand
Challenges in the Environmental Sciences.
National Academy Press, Washington DC, USA
13
Great Duck Island Monitoring Project

• Mission
• monitor the microclimates in and around nesting
  burrows used by the Leach's Storm Petrel.
• Goal
• to develop a habitat monitoring kit that enables
  researchers worldwide to engage in the
  non-intrusive and non-disruptive monitoring of
  sensitive wildlife and habitats

• Starting time Spring 2002,
• Participants
• Intel Research Laboratory at Berkeley
• the College of the Atlantic in Bar Harbor
• University of California at Berkeley
• Task
• deploy wireless sensor networks on Great Duck
  Island, Maine.

http//www.greatduckisland.net/
14
Habitat and the Bird
Habitat to be monitored (up, yellow
microphone Red camera) and the Leachs petrel ??
(right)
15
GDI Sensor Network

• Autonomous sensor nodes "motesplaced in areas
  of scientific interest, form a multihop network
  (sensor patch)
• Each patch network gateway mote has an external
  directional antenna forward data to base station
• Base station a laptop in the light house (350ft
  away) stores the data in a database and connect
  to Internet.

Mainwaring, et. Al, wireless sensor network for
habitat monitoring, ACM workshop on wireless
Sensor networks and applications, sept. 2002,
Atlanta, GA
16
Mica Sensor Node

• Single channel, 916 Mhz radio for bi-directional
  radio _at_40kps
• 4MHz micro-controller
• 512KB flash RAM
• 2 AA batteries (2.5Ah), DC boost converter
  (maintain voltage)
• Sensors are pre-calibrated (1-3) and
  interchangeable

• Left Mica II sensor node 2.0x1.5x0.5 cu. In.
• Right weather board with temperature, thermopile
  (passive IR), humidity, light, accelerometer
  sensors, connected to Mica II node

17
(No Transcript)
18
Background

• GOAL
• building a scalable, persistent, international
  network of lake ecology observatories.
• Network
• instrumented platforms capable of
• sensing key limnological variables and
• moving the data in near real time, often through
  wireless networks, to web-accessible databases.
• A web portal with a series of web services
• to allow automation of processes associated with
  instrument management and data quality
  assurance/quality control, and
• to allow computation of metrics based on the high
  frequency data.

19
Scientific Questions

• How do nutrient loading, hydrology, geologic
  setting, and climate regime influence the
  metabolic balance in lakes?
• What roles do large-scale disturbances, such as
  typhoons, drought, and seismic activity play in
  defining lake biological communities and their
  dynamics?
• How do lake morphometry, hydrology, and climatic
  setting modulate dissolved gas and nutrient
  fluxes across the sediment and water column
  boundary, thermal strata, and the lake surface
  and atmosphere interface?

20
GLEON International Participants

• 1. Trout Lake Stn, U Wisconsin, NTL-LTER, US 2.
  Academia Sinica Natl Center of HP Computing,
  Taiwan Forestry Research Inst, TWN 3. Centre for
  Biodiversity Ecology Research, U Waikato, NZ
  4. Center for Lake Mgmt Research, Kangwon U, KOR
  5. Centre for Ecology Hydrology, UK 6. Centre
  for Water Research, U of Western Australia, AUS
  7. Dorset Environmental Science Centre, Inland
  Lakes, Ontario Ministry of Environment, CAN 8.
  Lammi Field Stn, U Helsinki, FIN 9. Kinneret
  Limnological Laboratory, Israel Oceanographic
  Limnological Research Ltd, STAV-GIS, ISR 10.
  Nanjing Inst of Geography Limnology, CHN 11.
  Archbold Biological Stn, USA

21

• WISCONSIN Northern Temperate Lake Long Term
  Ecology Research
• Backbone of communication and monitoring
• Augmented with short-term deployments
• Technological and analytical challenges

Tim Kratz
22
Yuan-Yang Lake
23
Lake Taihu
Guangwei ZHU, Nanjing Inst. of Geography and
Limnology, Chinese Academy of Sciences
24
YSI sensor record DO, T and pH every 10 min from
31 July to 1 Sept.
Guangwei ZHU
25
GLEON Vision

• International collaborative network for
  sensor-based research into lake ecology
• End-to-end integrated CI solution for data
  collection, analysis, and collaboration
• Site level
• Data acquisition, instrument deployment and
  management
• Data management, curation, and publication
• Site-level analysis and visualization
• Network level
• Resource discovery, access and utilization
• Cross-site analysis and visualization
• Collaboration, sharing of techniques,
  experiences, and best practices
• Priority 1 timely (near-real-time) data sharing

Tony Fountain, UCSD
26
Yuan-Yang Lake and Typhoons
Source http//sensor.nchc.org.tw/ecogrid/typhoon_
idx.php
Tim Kratz
27
The Old Model
Manual recording at weather station D2 on Niwot
Ridge, Colorado, USA Photo circa 1953, courtesy
of Niwot Ridge LTER web site
Tim Kratz
28
The Current Model
Portable Lake Metabolism Buoy North Temperate
Lakes LTER Wisconsin

• Instrumented Platforms
• make high frequency observations of key variables
• send data to web-accessible database in near real
  time

Tim Kratz
29
The Future Model
Yuan-Yang Lake, Taiwan
Oracle Server NCHC Taiwan
Australia

• Web Services
• metabolism models
• intelligent agents
• data retrieval

Finland
New Zealand
Application Client
Korea
United Kingdom
etc.
Oracle Server Wisconsin

• Requires significant partnerships between
• lake scientists
• information managers
• middleware developers

Trout Bog Lake, Wisconsin
Tim Kratz
30
Tim Kratz
31
Technical Challenges

• Dense spatial-temporal sampling
• Large amount of data
• Large number of sensors, and types of sensors
• Long term observations
• Observation protocol, instrument change over time
• Irregular observation periods, missing data
• Global participations
• Different observation protocols, variables, data
  format, and instruments
• Scaling!

32
Cyber-Infrastructure Issues for Field-Deployed
Sensors

• Hardware Issues
• Connectivity to field sensor
• Move IP close to sensor
• Need better penetration of signal
• Storage of raw data in field (redundancy)
• Self-calibrating sensors
• Software Issues
• Automated screening for quality of
  data/troubleshooting
• Automated data reduction, including intelligent
  agents
• Flexibility to handle changes in sensor
  configuration
• Detect events and trigger adaptive sampling

Tim Kratz, UW-Madison
33
Sensor Calibration Automation

• Example from current practice
• Sensor calibration
• Due to sensor drift, the Greenspan dissolved
  oxygen (DO) sensor requires frequent (once every
  two weeks) calibration services.
• Sensor data correction
• After the DO sensor is calibrated, the observed
  data since last calibration need to be adjusted
  accordingly.
• Calibration Agent
• Will detect calibration event,
• Retrieve or capture calibration data
• Calculate the correction factor
• Retrieve data to be corrected
• Correct data and load corrected data to the
  database
• Write event data to calibration log in the
  database

34
Staffs manually calibrate the DO sensor.
35
(No Transcript)
36
Sensor QA/QC

• Developing algorithms that detect data problems
  undetectable by range checks
• Explore how generic such checks can be
• Managing the volume of data exceptions
• Clear definition of responsibility for various
  screening
• More automated process for handling
• Automating data calibration
• Developing QA/QC checks for new instruments
• Multi-mode QA/QC checks

37
Developing algorithms that detect data problems
undetectable by range checks

• malfunctioning anemometer detected as an abnormal
  occurrence of zero wind speed values

frequency of zero hourly average wind speed
values per month
38
Another Example

• a data quality problem when the instrumented buoy
  was pulled down in the water by the ice

water temperature (deg C)
normal winter
sensors displaced
39
Current Practice

• A hard-limit QC check system. When data value
  exceeds the set range, a H flag is generated
  indicating the data is suspicious.
• The present range check is manually determined
  based on historical data (e.g. same period last
  few years)
• Manually examining H flags in incoming data and
  edit out those likely to be normal data.
• ? A laborious process that does NOT SCALE UP.

40
Research Directions

• Long Term Vision Develop QA/QC agents capable of
• Establishing/updating data quality criteria
• Monitoring the sensor data quality,
• Annotating data quality assessment, and
• Tipping-off human operator to handle special
  situations and providing useful information to
  assist the operators decision.
• Medium Term Goals
• Data-mining based data quality cross-checking
  among different
• Sensing modalities
• Sensing times
• Sensing locations
• Short Term Objectives
• H flag screening agent
• Adaptive, soft-limit data quality assessment
  flagging system

41
Adaptive data quality assessment

• Estimating the probability that the observed data
  is within the normal range.
• Record data quality as the estimated probability
• Flag data when the quality drops below a
  hard-limit in terms of probability
• Update periodically the probability distribution
  of normal range data adaptively based on
• historical data
• Operator decisions
• Correlated sensor data from sensors of different
  locations, modalities

42
An Example
43
Correlation Coefficients
44
Joint PDF Estimation

• Use lattice vector quantization (high dimensional
  histogram) to estimate coarse joint pdf for
  efficiency.
• Other methods including clustering, kernel
  estimation, mixture of Gaussian model, etc.

45
Conditional Probability
46
Comparing Dissolved Oxygen Measurements over
GLEON Lakes
47
The Lake Metabolism Project Toward a Global
Network of Lake Observatories

• High Frequency Raw Data
• Water temperature
• Dissolved Oxygen
• Wind Speed/Direction
• Chlorophyll a
• Barometric Pressure
• etc.

• Reduced Data
• Gross Primary Production
• Respiration
• Net Ecosystem Production

Instrumented Buoy
Publicly available in near real time
Status, Trends, Mechanisms
Tim Kratz
48
Objectives

• Exploiting DO variation patterns over medium
  (months) intervals at high frequency time scale
  (10 minutes)
• Develop a basic taxonomy to describe DO variation
  patterns symbolically.
• Develop an abstract representation and
  description (e.g. vocabulary, ontology, or even
  grammar) of DO variations
• To relate the vocabulary phenomenon to
  ecological, biological, hydraulic events and
  explanations
• Use the vocabulary as a tool to describe and
  compare DO variations at different GLEON lakes

49
Approach

• Collect DO data from GLEON lake projects
• Develop data processing tools to
• Pre-process and normalize raw data to a standard
  format
• Develop data transformation tools to represent
  data in different domains (spatial, frequency,
  time-space, multi-scale, etc.)
• Use both manual and automated method to spot
  recurring features from the data
• Develop automated algorithms to detect and
  annotate identified features, and label each type
  of features
• Develop a multi-channel, multi-scale, symbolic
  representation of data from multiple sensors

50
Day
Night
O2 Concentration
Respiration (12 hr)
NEP (12 hr)
Time
GPP NEP (12hr) R (12 hr) R (24hr) R (12hr)
x 2

NEP (24hr) GPP R (24hr)
Tim Kratz
51
Daily Cycle Segmentation

• Basic Idea each day is roughly trapezoid in
  shape.
• Identify the morning rise and evening decline
  based on the curve gradients
• The area between morning rise and evening decline
  represents the mid-day DO saturation

52
Daily Cycle Segmentation

• Assuming the daily cycle is of trapezoidal shape,
  the longest positive consecutive streak
  corresponds to the morning rise, and vice versa
  for the evening decline.

53
Midnight Surge Detection

• In some of the data, it was noticed that there
  was a secondary bump in the DO levels occurring
  approximately at midnight.
• The detection procedure
• Identify bumps
• Extract features (bump volume, bump height etc.)
• Make decision on whether the bump is a Midnight
  Surge or not

54
Midnight Surge Detection

• First, use a gradient search method to find local
  minima in our midnight window

55
Midnight Surge Detection

• Then, we form an envelope bottom and subtract
  it from the curve to yield a base-line surface to
  measure volume and height
• Finally, we extract features from the largest
  bump, and select a classification method to
  determine which days exhibit the midnight surge
  behavior.

56
Concluding Remarks

• Sensor network applications to environmental
  monitoring is an emerging application of great
  importance.
• It provides an opportunity for inter-disciplinary,
  international collaboration to advance science
  and technology
• Real world application stimulate new technical
  challenges that demands efficient, cost-effective
  engineering solutions.