We apply a new method to obtain accurate locations of tremor sources beneath southern Vancouver Island. Unlike more standard "cross-time" methods, which compare waveforms from different time windows at the same station, this "cross-station" method compares waveforms from the same time window at widely separated stations. It performs well, relative to cross-time methods, when the response to an impulsive tremor source is dominated by the main arrival rather than coda and when multiple colocated sources are active within the time window examined. We focus on a region roughly 10 km across that was monitored by the POLARIS deployment from 2003 to 2006. Relative location errors appear to be <1 km, allowing us to image in great detail rapid and small-scale tremor migrations that arise behind the main slow slip front. The secondary fronts tend to (a) start at or within about 1 km of the main tremor front, and propagate back along strike at rates of 10-20 km/h; (b) less commonly do the reverse, ending at the main front; or (c) propagate up or downdip at or within 1-2 km of the main front. Estimated stress drops in the secondary events are comparable to that in the main event, implying that their 25-50 times greater propagation speed results from a similarly greater slip speed.