Reventador, located in northern Ecuador, is an active volcano. In 2002 it erupted with such massive force that it blanketed Quito, Ecuador's capital city 60 miles to the west, with a layer of ash so thick the airport had to shut down.
Today it blows its stack with lesser force but with great frequency. And just as storm chasers want to be where the cows are flying around, vulcanologists want to be as close as they can get to 11,500-foot peaks that spew grapefruit-sized molten rocks several times a day.
What scientists want even more than taking in the sights and sounds, however, is gathering precise records of seismic activity and studying the data, which is where Matt Welsh, assistant professor of computer science at Harvard University, and his wireless sensor mesh network come in.
Previously, seismologists from the University of New Hampshire (UNH) and University of North Carolina (UNC) collected data by lugging bulky, heavy sensor stations up the mountain - each one containing a car battery for power plus sensors, cables, and data logging and data storage equipment.
Every few days someone had to trek several hours back up the mountain to collect the microdrives. This put severe limits on the number of sensor stations deployed and consequently on the amount of data collected. The set-up also had other limitations - for example, there was no way to tell if one of the stations had stopped functioning.
Welsh had a better idea. He had a 190-node wireless sensor mesh running at Harvard that was designed to detect changes in a building's temperature and humidity. Now it was time to bring his mesh network out of the lab and into the line of fire.
It's a jungle out there
For the Reventador expedition, Welsh created a tiny, low-power sensor station that runs on ordinary D-cell batteries and fits into a plastic case the size of a lunchbox. It sends seismic and sonic data around the clock to a base station that collects data in real time. Total cost: $400 per node.
In August 2005, the seismologists from UNH and UNC and the computer scientists from Harvard flew to Quito, drove three hours, then bushwhacked through the jungle for several hours to the base of the mountain, where they camped. At 5 am, Reventador issued a wake-up call.
"The top blew," Welsh says, "and I freaked out."
Welsh was far enough away from the top of the volcano so that he wasn't in danger, but just to be on the safe side he sent his grad students up the mountain while he manned the base station. His team deployed a total of 16 nodes 650 to 1,300 feet apart, in an ad hoc mesh network that spanned almost 2 miles up one side of the volcano.
Each node, which consisted of a Tmote Sky sensor from Moteiv, an interface board and a battery holder, was covered with rocks to anchor it and protect it from the elements - the area had been defoliated in the 2002 eruption.
The seismic sensors were buried nearby, connected to the nodes by USB, and the sensors that measure infrasonic waves were mounted on PVC tubing Welsh used to elevate the high-gain antennas off the ground.
The network used IEEE 802.15.4 low data-rate radios and connected to a FreeWave radio modem at the base of the mountain that backhauled the data to a laptop about 2.5 miles away in the tiny hotel where the team was staying. Welsh says he decided against using 802.11 because it would have required too much power.
As one might expect, all kinds of problems arose. For example, each node had enough storage capacity to record 20 minutes worth of data, then it simply overwrote the disk. Because the scientists obviously wanted data only when the volcano was erupting, Welsh set up the system so that when any five nodes reported seismic activity to the base station, all the nodes were pinged and told to capture the last 60 seconds' worth of data.
That worked well until the time they were eating dinner in the hotel and a giant explosion was heard. Welsh raced over to check the laptop and realised that a smaller event had occurred within a minute of the giant event, so the network was busy capturing data for the small event and missed the big one.
There were other minor glitches that one probably would not run into at Harvard. For example, the hotel didn't run electricity overnight so Welsh had to hire somebody to keep filling the generator with diesel fuel to keep the laptop running. That person apparently dozed off every so often, so occasionally the laptop died.
LAN vs lava
As far as data collection goes, the experiment was a qualified success. It took around six days to get everything up and running. Then the system crashed on Day 8. The team got everything rebooted by Day 13 and from that point through Day 19, the network ran like a dream - except for the time that a giant ball of molten rock blew the top off the PVC pipe connected to the sensor node closest to the top of the volcano.
Another major problem was the fact that although ordinary data networks simply retransmit dropped packets, the scientists wanted a datastream with no dropped packets and they wanted all the data time-stamped to within 10 milliseconds.
In other words, they wanted to be able to follow an event as it cascaded down the mountain from one sensor node to the next. It took Welsh and his team more than six months to get the data to that point, but the result was useable information about 230 events.
The venture has yielded some interesting data. For example, one would expect volcanic activity to start at the top of a mountain, but some events came from deep inside Reventador in locations that the seismologists didn't anticipate.
So what's next for Welsh and his crew? The plan is to keep working toward a sensor mesh of hundreds of nodes, augmented by a middle tier of Linux-based 802.11 devices that would sit between the remote sensors and the base station.
Welsh hopes to deploy enough sensors to create a 3D view of the volcano. That would require devices with better power consumption, more powerful data collection capabilities and, of course, more grad students.