Environmental Monitoring Using Sensor Networks

1
Environmental Monitoring Using Sensor Networks

• Christos Panayiotou, and
• Michalis P. Michaelides
• Dept. of Electrical and Computer Eng.
• University of Cyprus

2
Motivation

• Ecosystems and habitat monitoring
• Air quality monitoring
• Water quality monitoring

3
Monitoring the Impact of Urban and Agricultural
Land Use.

• Directives
• Article 8, WFD Directive (2000/60/EC) Monitoring
  each river basin district.
• 91/676/EEC Nitrates EC Directive

4
Overview

• Source localization problem
• Wireless Sensor Networks
• Related work
• The source localization problem using sensor
  networks (a first take).
• Some simulation results
• Conclusion and future work

5
Complex System

• Human activities contaminate the environment to
  the point where entire ecosystems are destroyed
  and risking human lives.
• Diffusion of toxins and chemicals into water,
  soils and sediments.
• Industrial activities may contaminate air, soil
  and water
• Fertilizers, sewage disposal, manure storage
  (nitrates and nitrites)
• Biological indicators (macro invertebrates and
  other microbial contamination - Streptococcus
  count, E-Coli, Cryptosporidium)

6
Objective

• Can we detect the contamination?
• Can we locate the source of the contamination?
• Can we locate multiple sources of contamination?
• Can we control pollution?

7
Sensor Nodes

• Simple and fairly cheap devices
• They consist of
• Sensing unit multiple onboard sensors in
  acoustic, seismic, IR, magnetic modes, imagers,
  micro radars
• Data processing unit - storage
• Communication (transceiver) unit wireless links
  to neighboring nodes
• Power supply unit (battery)
• Optionally,
• Location Identification Unit (e.g. GPS), Mobility
  platform, Power generation unit ()
• They are capable of performing simple tasks

8
What Are Sensor Networks?
Sink
Internet or Satellite network
sensor nodes
sensor field
Task Manager Node

• A collection of Sensor Nodes collaborating to
  perform a complex task

9
Characteristics of Sensor Networks

• Limited Resources
• computational capability
• communication bandwidth
• power
• memory
• Large number of nodes densely deployed
• Sensor nodes may not have global identification
  ID because of the large amount of overhead and
  large number of sensors
• Prone to failure (power outage)
• Topology may change frequently
• Ability for self-organizing and self-configuring

10
Related Work in Sensor Networks

• Remote Underwater Sampling Stations (RUSS) at
  Syracuse lakes
• Non-intrusive breeding habitat monitoring on
  Great Duck Island
• Temperature and humidity monitoring at Pickberry
  Vineyards
• SENSPOL monitoring environmental pollutants in
  water, soil and sediments.
• Oak Ridge National Laboratory in USA
  Comprehensive incident management system that
  will rapidly respond to a chemical, biological or
  radiological event.
• Los Alamos National Laboratory Sensor Network
  that will detect a motor vehicle carrying a
  Radiological Dispersion Device.
• CSIP (Collaborative Signal Information
  Processing) deals with the energy constrained
  dynamic sensor collaboration.

11
Related work in plume tracking using unmanned
vehicles

• Source localization
• Bio-mimetic robotic plume-tracing algorithms
  based on olfactory sensing (Homing, foraging,
  mate seeking)
• Basic steps in robotic plume-tracing
• Sensing the chemical and sensing or estimating
  fluid velocity.
• Generating sequence of searcher speed and heading
  commands such that the motion of the vehicle is
  likely to locate the odor source.
• J. Farrell et al. uses an autonomous vehicle
  operating in the fluid flow capable of detecting
  above threshold chemical concentration and
  sensing fluid flow velocity

12
Vehicles vs. Sensor Networks

• Static sensor network
• If plume passes by a sensor, it is detected.
• Position of the source needs to be remotely
  estimated using fusion techniques.
• Energy constraints
• Need efficient routing techniques to relay the
  information hop by hop to the sink.
• Low cost.

• Vehicle Search
• Spend a good amount of time searching for the
  plume in reachable areas.
• Plume tracking
• Potentially it can go very close to the source.
• All necessary computation can be done on-board.
• Once source location is identified it returns to
  base to report.
• High cost

13
Sensor Network Plume Tracking
Contaminant Source
Sensor nodes
14
Sensor Network Plume Tracking with Current
Contaminant Source
Current Direction
Sensor nodes
15
Propagation model

• N sensor nodes stationary, randomly placed in a
  rectangular field R with locations known ( xi ,
  yi ).
• Contaminant source ( xs , ys ) is somewhere
  inside R.
• Propagation model fi(.) includes the
  concentration at the source, speed and direction
  of the wind or current and other environmental
  parameters.

16
Simple Simulation Model

• Propagation of contaminant transport is uniform
  in all directions.
• Additive Gaussian white noise.
• a2, c106

17
Least Squares Estimation

• Centralized approach
• Sensor nodes calculate the mean of M measurements
  and then send the computed mean to the sink.
• After the sink receives the information from all
  sensor nodes it employs the nonlinear least
  squares method to compute an estimate of the
  source location by minimizing function J.

18
Simulation Results

• MATLAB simulation package
• K100 randomly placed sources for each experiment
• Effect of varying number of sensors, noise
  variance and number of measurement samples.

19
Least Squares Start Position

• LS max start start the minimization in the
  neighbourhood of the sensor node with the highest
  measurement.
• LS random start randomly pick 10 start
  positions in the sensor field.
• LS combo choose the method that minimizes the
  squared 2-norm of the residual.
• CPA Closest point approach
• The source position is the location of the sensor
  that measured the highest concentration.

20
Random Sensor Field
21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
Preliminary Conclusions

• A large number of sensors is required to
  guarantee a good enough source position estimate

25
(No Transcript)
26
Simple Propagation Model with Current

• Only sensors in the active area A may detect the
  plume
• When a sensor node is triggered by the presence
  of the plume it wakes-up, it takes a number of
  discrete measurements and calculates the mean.
• Centralized Approach If the mean exceeds a
  predefined threshold T the sensor communicates
  this value to the sink and continues measuring
  otherwise it goes back to sleep.

27
Results with Current

• LSp Least squares estimator with initial
  concentration known.
• LSc Least squares estimator with initial
  concentration unknown.
• Use separable least squares techniques
• Further improvements
• LSu Unconstrained optimization
• LSw Constrained search based on wind direction

28
(No Transcript)
29
(No Transcript)
30
Threshold considerations

• Determines the number of sensors involved in the
  estimation.
• Needs to be large enough to minimize probability
  of false alarms.
• Needs to be small enough to ensure maximum
  detection probability.
• Needs to be appropriately chosen to minimize
  energy consumption while not compromising
  estimation accuracy.

31
(No Transcript)
32
Decentralized Estimation
SINK
33
The More Realistic Model

• Once released at its source odor is carried by
  the wind to form a plume.
• As the plume travels further away it becomes on
  average more dilute due to molecular diffusion.
• Dominant cause of diffusion is turbulence.
• Characteristics of odor plume depend on physical
  environment.

34
Single Sensor Output

• Graphical interpretation of the output of an odor
  plume with moderate turbulence.
• Sensor was stationary at 10 cm downstream of the
  odor source and at the geometric center of the
  plume.

35
Conclusion and Future Work

• Sensor network technology will allow us to
  monitor and better understand the environment
• Future work
• Estimation using the more realistic model
• Decentralized approach
• Include mobile nodes to improve the network
  coverage