Here is an excellent USGS document on earthquake lights. Sky- or ground-associated earthquake lights have been reported preceeding earthquakes by an hour or so. A man reported the sand at Ocean Beach in San Francisco glowing in the pre-dawn darkness an hour before the 1906 quake struck (and of course I can't find the reference). One of the earliest good documentations of earthquake lights were the Matsushiro earthquakes in central Japan in the 1960s (right over the North American-Asian-Philippine Plate Triple Junction). Want to see for yourself? These weird rainbow clouds appeared over Sichuan in the hour before the quake there in May 2008.
People have reported weird lights preceding the Chile quake, although the only footage I've seen so far is here, and I'm betting it's just transformers blowing up, like these in Peru from 2007. (Notably, they're during the quake.)
(Note that these observations in no way support the tinfoil-hat-wearers who blame the Chilean earthquake and everything else on "HAARP". Were the earthquake lights in Japan HAARP too? What about in San Francisco a century ago? I guess those earthquake lights were from aliens trying to blow up the Earth.)
The best explanation for earthquake lights so far (but it's still not great) is ultra-low frequency (ULF) electromagnetic radiation generated by plate motion. That I know of no one has a coherent model to explain how it works or originates, especially since even empirically a) lights and large earthquakes don't always occur together, and b) electromagnetic anomalies don't always occur in high activity areas. Parkfield, California (where all the parallel Northern California faults split off from the San Andreas) is probably the most seismically active place in the Lower 48, and when an acquaintance went there to record with his field rig, he got nothing.
Empirically, the ULF connection has arguably been reinforced by the Vogel Study, a set of observations in 2001 of what was almost certainly a man-made phenomenon near Toppenish on the Yakama reservation in central Washington State. The report is worth reading. Essentially, the reservation fire department kept getting calls about fires from people at who saw lights at night. These lights turned out to be lights in the sky rather than burning hillsides. This was costing the reservation money, so they called in the team that wrote this report to figure out what was going on. What these observers picked up were ULF signals, almost certainly of artificial origin (look at how square the wave forms are), modulated both by amplitude and frequency. Entirely on a lark because I had read this study, a few summers ago I drove through the area and until I actually went there I didn't look at the map too closely. It turns out Hanford Nuclear Facility and Yakima Firing Center are nearby, in one of the possible directions from which the signal was coming. Without getting all tinfoil-hatty, the Federation of American Scientists maintains a document supposedly from the EU Parliament about a U.S. surveillance network called ECHELON which supposedly has operations at Yakima Firing Center (see page 52). One problem is that while these ULF emissions were detected, there were no lights detected during the observation period. It's still interesting that the lights were seen at this location, and that there were ULF signals detected.
Since the lights sometimes precede earthquakes and could conceivably help us avoid future loss of life, this phenomenon requires closer study.