ISSNIP

 Anomaly Detection in Wireless Sensor Networks

 

The Great Barrier Reef (GBR): 

     The Great Barrier Reef (GBR) consists of 3200 coral reefs extended over 280,000km2 [3] (see Figure 1). Understanding the patterns of thermal stress and other environmental parameters is essential for monitoring the health of the coral reefs (eg., refer [10] for environmental change and impacts on GBR). The health of the coral reefs can be affected by cold water intrusions, hot water intrusions, coral calcification, ocean acidification, coral-algae phase shifts and the spread of coral diseases due to temperature increases. For example, a temperature rise of about 2-3 degrees Celsius over the normal maximum summer temperature can kill corals [12]. In order to monitor the health of the coral reefs, sea temperatures need to be measured in fine spatial resolutions at various depths. Traditional methods of using satellite images can only reveal water surface temperature distributions at coarse spatial resolutions such as 1km2. This resolution does not provide sufficient detail to investigate the cause of coral bleaching or coral growth events in the GBR, such as 2002 coral bleaching event [3, 11]. Measurements at small spatial scales and at various depths are required in order to enable an in-depth analysis of the implications and causes of these bleaching events.

gbr-reefs-sitewsn-mgtisland    

Figure 1: Great Barrier Reef [3].                                          Figure 2: Sensor nodes used in Nelly Bay, Magnetic Island [9]. 

Another important factor in the analysis of the reef is the abundance of a sea water organism called plankton, which plays an important role in the GBR food chain [1]. Understanding plankton production recently became popular due to its ability to recycle CO2 and therefore its potential role in global climate change. The productivity of plankton in the GBR is influenced by the nutrient rich cold water intrusions that originate in the Coral Sea and upwell on the reef. Monitoring high frequency sea temperature changes due to daily tides and upwelling in near real-time enables the study of how the changing sea temperature affects the abundance of plankton [1]. Further more, nutrient concentrations like nitrate, phosphate and silicate are strongly correlated with water temperatures [2].

The coral reef monitoring system using wireless sensor networks in the Davis Reef, North Queensland involves the placement of a number of environmental sensors that measure temperature, salinity, light and oxygen [3]. Current infrastructure installed in this reef site includes a sensor gateway, which provides the aggregation point for sensor data, a hybrid power supply utilising solar cells and a battery, a high speed microwave link, which operates using a 'humidity duct' at a data rate of 10Mbps, and cameras [4, 7, 8].

The first experimental wireless sensor network was implemented in Nelly Bay, Magnetic Island during 2007 [9] (refer to Figure 2). The sensor network consisted of two sensor arrays that comprise four moorings, each having seven temperature sensors vertically positioned below the ocean surface 2m apart [5]. An accelerometer sensor is also being deployed at each node to measure the wave tidal frequency.

 HeronIslandSN   HI-gbr-hs

Figure 3: Sensor network deployment in Heron Island, Qeensland. Hierarchical (or layered or tiered) topology of the network.

 HI-GBR-Nodes

Figure 4: Heron Island deployment: Poles, buoys, base station and electronics inside the boxes.  

Currently, a sensor network is being implemented in Heron Island, Qeensland as part of the Great Barrier Reef Ocean Observing System (GBROOS) project [14]. This sensor network is a two tiered, hierarchical topological network with heterogeneous sensor nodes in each level as shown in Figure 3. The nodes in the first tier are called poles and the nodes in the second tier are called buoys. There are 5 buoys and 6 poles used in the deployment. Buoys and poles are deployed in the lagoon area of the GBR approximately 2 km apart. The buoys communicate with the poles via single hop. The poles communicate to the base station via multiple hop. The base station is located in the Heron Island, which transmits the collected data to the mainland 75 km away.

Each pole and buoy collects temperature at a depth from the sea surface. The temperature is samples every 10 minutes and saved in memory before transmitted to the base station periodically. Electronics at each buoy and pole consist of memory, temperature sensor interface, antenna for wireless communication between poles, buoys and base station, solar panels and a battery. Figure 4 shows the poles, buoys, base station and a box of electronic devices in one of the buoys. A third tier of sensor consisting of small sensor nodes is planned for future deployment. These sensor nodes will communuicate their data to the buoys. Figure 5 shows the hierarchical organisation with the third tier of sensor nodes. 

 HeronIslandSNPlanned

Figure 5: Planned deployment of the third tier of sensor nodes.

Anomaly Detection: 

The sensor measurements collected by such sensor network deployments can become contaminated with errors (either fully or partially) due to either loss of calibration of the sensing elements or faulty sensor nodes (e.g., see graph in [1]). Such errors are highly prevalent in sensors and electronic equipment deployed in the harsh marine environment [1]. This contamination may gradually accumulate over a period of time (gradual drift), or occur in one-off transients. Such errors need to be detected at their source in real time and corrected, in order to collect reliable data from the sensor network deployments. Further more, there is a need to automatically detect natural events of interest in the monitored environment, such as cold water intrusions. Once these events occur, we need the ability to automatically adjust the sampling frequency and type of measurements collected in response to the event of interest.

The general problem of detecting interesting changes from the normal observed behavior in sensor measurements is known as anomaly detection. An anomaly can be caused by an unusual change in the phenomena (e.g., water temperature or nutrient concentration), or by faulty sensors: that cause incorrect measurements, or even by malicious events such as security attacks in sensor networks [6]. Important challenges for the management of sensor networks in complex environments such as the GBR are the detection, inference, reporting and correcting of anomalies. Centralised solutions to anomaly detection, which involve collection of all data from sensors to a centralised node for processing, can be communication intensive and thus very energy inefficient. An alternative approach for anomaly detection in sensor networks is to use in-network processing in order to prolong the lifetime of the resource constrained wireless sensor networks. Our research into distributed anomaly detection in wireless sensor networks addresses the above challenges in order to provide a reliable, energy efficient and self-correcting wireless sensor network for use in small to large scale deployments.

People Involved: Mr. Sutharshan Rajasegarar, Dr. Christopher Leckie, A/Prof. Marimuthu Palaniswami, Prof. James C Bezdek, Dr. Yee Wei Law and Dr. Jayavardhana Gubbi.

Related Publications:

References:

[1] Olga Bondarenko, Stuart Kininmonth and Michael Kingsford, "Underwater Sensor Networks, Oceanography and Plankton Assemblages", in the Proc. of International Conference on Intelligent Sensors, Sensor Networks and Information Processing (ISSNIP 2007), pp 657-662, (Melbourne, Australia), 2007.

[2] M.J. Furnas and A.W. Mitchell, Nutrient inputs into the central Great Barrier Reef (Australia) from subsurface intrusions of Coral Sea waters: a two-dimensional displacement model. In Continental Shelf Research, 1996. 16(9): pp. 1127-1148.

[3] Stuart Kininmonth, Scott Bainbridgea, Ian Atkinsonc, Eric Gilla, Laure Barrald and Romain Vidaude (2004), "Sensor Networking the Great Barrier Reef", Spatial Sciences Qld Journal, Spring 2004, pp34-38.

[4] Cameron Huddlestone-Holmes, Gilles Gigan, Graham Woods, Adam Ruxton, Ian Atkinson, and Stuart Kininmonth, Infrastructure for a Sensor Network on Davies Reef, Great Barrier Reef, in the Proc. of International Conference on Intelligent Sensors, Sensor Networks and Information Processing (ISSNIP 2007) , pp 657-662, (Melbourne, Australia), 2007.

[5] Olga Bondarenko, Stuart Kininmonth and Michael Kingsford, Coral Reef Sensor Network Deployment for Collecting Real Time 3-D Temperature Data with Correlation to Plankton Assemblages, in the Proc. of International Conference on Sensor Technologies and Applications (SENSORCOMM 2007), pp 204-209, (Valencia, Spain), 2007

[6] A Perrig, J Stankovic, and D Wagner, Security in wireless sensor networks, in Wireless Personal Communications, vol 37, no 3-4, 2006.

[7] http://www.reefgrid.org/sensors/

[8] http://www.qcif.edu.au/industry/ReefGrid.htm

[9] http://www.reeffutures.org/sensornet/display.cfm

[10] http://www.aims.gov.au/pages/research/research-teams/rt-environmental-change-and-impacts.html

[11] http://www.reeffutures.org/topics/bleach/event.cfm

[12] http://www.reeffutures.org/topics/bleach/temp.cfm

[13] http://www.aims.gov.au/

[14] http://imos.org.au/?id=266

Back
top of page