http://news.sciencemag.org/earth/2015/04/smart-phones-could-...
I would guess that there are enough phones plugged in at any given moment, even during the day, to provide a large enough sample size. Especially if that were paired with other data sources.
The CORS GPS network is used for precise positioning using position averaging and correction for any atmospheric interference.
Armchair engineering, I would assume the cost for a solid state seismometer (3 axis accelerometer) would be trivial enough to add onto existing reference stations to provide precise positioning data, as well as comprehensive movement data.
These usually end up detecting earthquakes though http://www.ga.gov.au/scientific-topics/hazards/nuclear-monit...
Not necessarily. They could also follow the SETI approach by providing an app for volunteers who collect the data. The app evaluates the data, filters the noise and sends the abstract data results to a server. There are already many data collections of past earthquakes which could be used as patterns for a neural network in the app.
Battery life impact could be reduced by creating an automatic timeshare where only a small fraction of those millions of people are collecting data at any given time; the data would probably still be sufficient. The percentage of users actively collecting data could also be dynamically increased to improve accuracy when a possible seismic event is detected anywhere in the system.