Tonight I looked at the signal from a geophone and accelerometer side by side, in order to see if they show the same ground motion and if the sensitivity factor I am using to convert from V to m/s is right. This is plotted below, along with the current estimates for accelerometer and geophone noise:
From this it is pretty clear that at least one of the sensitivity factors (V/m/s) I am using is wrong (the noise levels are much lower than the ground motion, so they can't account for the difference). I suspect it is the geophone one, because Wilcoxon provided these sensitivities for each individual accelerometer, but I was just using the number I found in online specs for the geophones.
The reason the online value is wrong is probably because of the value of the shunt resistor, a resistor that just goes across the top of the geophone (its purpose is to provide damping, by Lenz's Law). The specs assume a value of 380 Ohm, but I measured the one in the STACIS to be about 1.85 kOhm.
Assuming the accelerometer signal is correct, I multiplied the geophone signal by different factors to try to get an idea of what the true calibration factor is, and found that a value of 0.25 (m/s)/V gives decent agreement at higher frequencies (below 10 Hz the sensitivity drops off, according to the online specs). This is shown below:
Above, the geophone noise was recalculated with the new sensitivity and assuming that both geophones in the noise measurement had the same sensitivity. I took the transfer function of two geophones side by side to see if their gains were dramatically different; this plot is shown below. The coherence is only good for a small band, but looking at that band the gain is approximately unity, implying very roughly that the sensitivity of each is approximately the same. The lack of coherence is strange, and I'm not sure what the cause is. Even using the shaker near the geophones only improved the coherence slightly.