[an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive]
We deployed a wireless sensor network to monitor eruptions at Volcán Tungurahua, an active volcano in central Ecuador. This network consisted of five tiny, low-power wireless sensor nodes, three equipped with a specially-constructed microphone to monitor infrasonic (low-frequency acoustic) signals emanating from the volcanic vent during eruptions. We gathered over 54 hours of continuous infrasound data, transmitting signals over a 9 km wireless link back to a base station at the volcano observatory.
Composed of small, low-power wireless devices, sensor networks constitute a new kind of computing platform. A typical wireless sensor network platform, or "mote", integrates a computationally-limited processor and limited storage capacity with an flexible interface allowing various sensors to be attached. The energy, memory, computation, and bandwidth limitations of sensor network devices pose system design questions that require continued research if robust deployments are desired.
To date, habitat, medical, and structural monitoring applications have been deployed on sensor networks. To our knowledge this is the first dense wireless array deployed for volcanic monitoring.
Volcanologists collect seismic and infrasonic signals to monitor and study volcanic activity. Volcanos emit powerful seismic waves while erupting; but tectonic earthquakes, mining operations, and ambulating quadripeds can also induce seismic energy, complicating eruption detection based on seismic events alone. Because volcanos also emit powerful pulses of infrasonic energy near the moment of eruption, in cases where visual monitoring is impossible, inconvenient, or dangerous, correllating infrasonic and seismic events provides better eruption detection than either taken alone. Additionally infrasonic signals provide additional informational content not found in the seismic data.
Wireless sensor networks consist of small, low-power devices equipped with a radio, a variety of sensors, and a modest amount of computational power and local storage. A typical sensor "mote" is powered by 2 AA batteries and includes an 8-bit microcontroller, 4 KB of memory, a low-power radio with a range of approximately 100 meters and a bitrate of about 38 Kbit/sec. The low cost, size, and energy requirements of sensor networks makes them very attractive for volcanic monitoring. Our small array of infrasonic sensors monitored volcanic eruptions, reporting real-time data over a wireless link to a base station. A larger infrasonic array consisting of dozens of motes could be used to eliminate sources of noise as well as to triangulate the source of an eruption event.
Tungurahua (map) is an active volcano in central Ecuador, near the town of Baños. The Instituto Geofísico of the Escuela Politecnica Nacional (IG-EPN) in Quito maintains an observatory near the volcano that is responsible for monitoring eruptions and apprising the government and media of changes in eruptive activity. Dormant for over 80 years, in 1999 Tungurahua began showing increased seismic activity which led researchers to believe that the volcano was awakening. Baños was evacuated by the Ecuadorian military in anticipation of a large event, which did not in fact occur. After several months the populace of Baños was allowed to return to their homes. Since then the volcano has been experiencing a period of increased eruptive activity, making it an ideal candidate for volcanic research.
This collaboration between researchers at Harvard, UNH, IGEPN, and UNC resulted in the deployment of a wireless infrasonic sensor network at Volcán Tungurahua from July 19-23, 2004. During this time, the volcano was erupting at the rate of several small or moderate explosions an hour, though the rate and energy of eruptions varied considerably. Our sensor network recorded over 54 hours of data from three wireless infrasonic microphone nodes, relaying the data back to the observatory over a 9 km radio link. A laptop at the observatory recorded the complete data traces and visualized the signals in real time. The wireless sensor array was colocated with a wired monitoring station recording infrasound and seismic signals, allowing us to verify our signals against an trusted monitoring platform.
Five Mica2 sensor network devices performing three different tasks composed our volcano monitoring network: three data-collection motes fitted with custom-built infrasonic sensors, one receiver mote forwarding data over a long-range serial point-to-point link, and one time synchronization mote interfaced to a GPS unit providing a common time base for the data collection elements.
The three data-collection motes integrated the Mica2 with an infrasonic microphone (Panasonic BM-034Y) and custom amplification and filtering circuitry. Sampling at approximately 102 Hz, they transmitted data packets containing multiple readings to the receiver mote at approximately 4Hz. A Mica2 receiver mote attached to a MIB600 interface board forwarded data packets along a long-range serial point-to-point link, provided by a pair of FreeWave modems fitted with 9dBI directional Yagi antennas, back to the observatory 9 km away. A laptop connected to the FreeWave modem at the observatory logged the data and provided real-time monitoring capabilities.
To provide the required common time base for the data-collection motes, an additional Mica2 mote was interfaced with a off-the-shelf Garmin GPS receiver. The time synchronization mote receives a time pulse every second from the GPS unit and relays the pulse to the infrasound motes via radio. Each mote marks the infrasound sample taken when each GPS timepulse is received, allowing the signals from each mote to be synchronized across time.
The sensor node software was implemented in TinyOS, a tiny operating system for sensor networks. Laptop software was implemented in Perl and Java running on the Linux operating system.
Through the deployment the Freewave Modems and GPS receiver were powered by standard 12 Volt car batteries, a readily available energy source in Ecuador. All other devices were powered by 2 AA batteries. No power sources required replacement over the duration of the deployment.
Data Collection and Analysis
Our wireless sensor network stored 54 continuous hours of infrasonic data into approximately 1.7 GBytes of uncompressed log files. During that time several sets of eruptive pulses were recorded. We are continuing to analyze the data we recorded, comparing it against signals recorded by several different wired monitoring stations. Displayed below are one set of traces from our wireless sensor network compared to two different wired stations: one co-located with our deployment and the other at a different site on the volcano. Note that the wired Larson-Davis microphone signals shown have inverted polarity.
This deployment was a proof of concept, allowing us to collect initial data from a small number of nodes while addressing the problems associated with data collection, time synchronization, packaging, and signal validation. Over the next six months we plan to design a much larger infrasonic array, on the order of 20 to 50 nodes, for deployment on Tungurahua or elsewhere. The volcanologists who partnered with us on this initial deployment are very excited about the capabilities such a large, spatially-seperated array would provide. Using more nodes increases the aperture of the microphone 'antenna', facilitating much more advanced analysis of collected data.
Such a large deployment also introduces challenges common to the field of sensor network research. Once fifty nodes are in place bandwidth limitations make it infeasible to sample continuously, power limitations make it necessary to do intelligent duty cycling, and logistical limitations require sophisticated hardware and software solutions for retasking the array. Therefore we expect that our next deployment will introduce a triggering model to conserve bandwidth, intelligent hardware and software support for long-term lower-power operation, and remote-reprogramming and macroprogramming concepts to allow intelligent retasking. We view this project as an exciting opportunity to take ideas and solutions developed for the sensor networking community and test their mettle outside of the artifical labratory environments in which they were created. For as we were constantly reminded by this small, simple, short deployment: it's a long way from the lab to the field.
Professor Welsh put together some slides from a short trip report to Tungurahua, containing more information and pictures. These slides were first given to a meeting of the Harvard Motes group on July 26th, 2004.
Galleries of pictures taken by Professor Welsh and Geoffrey Werner-Allen are available on Geoff's website. Geoff's photos are here and Matt's are here.
For more information on this project, please contact either Geoffrey Werner-Allen (werner AT eecs) or Professor Matt Welsh (email, phone: 617-495-3311).