Quake-Catcher Aims to be Largest Distributed Seismometer Network

Nature is reporting that a new distributed computing application is looking to monitor earthquake data using the accelerometer in many computing devices. In the long run, "Quake-Catcher" will hopefully be fast enough to give warning before major earthquakes. "If it works, it will be the cheapest seismic network on the planet and could operate in any country. It wouldn't be as sensitive as traditional networks of seismometers, but Lawrence says that's not the point. 'If you have only two sensors in an area, you have to have a perfect system. If you have 15 sensors in a system it [can] be less perfect. One hundred, one thousand, ten thousand -- your need for the system to be perfect becomes much smaller,' he says. 'That's really our approach -- just to have massive numbers.'"

Comments

Re:Accelerometers

[zappepcs]

Every piece of data that can be corroborated will help. Sure, wiimotes shake normally, but if all of them in L.A. start shaking.. well that is something to look at. That is what the summary alludes to with the statement that with more than 2 sensors the system can be less precise.

The fact that you could have corroboration from 1500 points in a 75 square mile area is quite an improvement on what they have now, and at a much cheaper price.

If you spend time analyzing data, it's amazing what you can find. That is one of the reasons that the US government wants to monitor everyone's communications... to spot small trends... and of course to gather evidence to use against political rivals thus ensuring their unending reign of ... what did the French call them ? oh yeah, terrorists

Back on track. The sensitivity of things like the wiimote add huge potential to such an endeavor. Just through sheer numbers, the size of the area shaking makes a big difference on the impact or relevance of the seismic event. It's physics, and if you are trying to see the true graph of something, the more data points you have to plot, the more informative it is. Even if some of the sensors are unreliable, they have the ability to ignore anomalous readings and use those that match others. Since you can be certain that there is an event happening (old system still in place) you can ignore or throw out data from sensors that are TOO active or not active at all, then sift through what is left to see what you find.

I'm reasonably certain that they will see a lot when they learn the true extent of the area affected by any particular event. For example, if the event stays limited to only the fault area it would be much different than if an entire area were affected outside the fault line area. Having thousands of sensors will help show that. Perhaps through this they will learn that certain geologic structures actually do redirect the energy to other areas, allowing predictions of damage to match what before were unpredictable events thus adding perhaps minutes to the warning times. That would save lives and that is what they want to do. Mapping effects through an area will help. Thousands of sensors will help achieve that despite the seemingly unreliability of the sensors themselves.

There are millions of ants in an ant hill, kill a couple hundred and they carry on. This is the same sort of idea.

Re:Hope is not a plan

[Scrameustache]

In the long run, "Quake-Catcher" will hopefully be fast enough to give warning before major earthquakes.

and the scientific basis for prediction is what, exactly?

Precursor tremors.

Re:Great vaporware application

[Eivind Eklund]

There are techniques for extracting higher quality data from overlapping low-resolution data sets. In the visual space, it's obvious that this is possible: If you have a single low-res camera, a static photographic subject, and full control of movement you can move a camera less than a pixel for reach picture taken (in a controlled way). Then you get sub-pixel-resolution data plus noise in the resulting difference set between different pictures. If you have ENOUGH difference sets, you can cancel out the noise. You then get sub-pixel resolution.

To extend this to a domain where you don't have the effective control, you have to automatically detect where different pictures fit. I remember having seen somebody that did this; I can't remember where, though.

Eivind.