To isolate this time variant signal, P waves from each timeslice were correlated with the arbitrary reference record for that station. To accomplish this, the first arrival P wave had to first be extracted from the record. All signals were bandpass filtered between 200 and 1000 Hz and then tapered to isolate the first P arrival.
Figure 1 This example shows the preparation of a record from station 18 recorded at hour 5.5 (file 12) from source B. The top record is the raw data. The middle record has had a 200-1000 Hz bandpass filter applied. The bottom record has been tapered to isolate the P wave signal of interest. All signals were prepaed in this manner before being correlated.
Figure 2 Two P waves from source B recorded at station 18. The violet and blue signals were recorded at hours 0 and 7.5, respectively (files 1 and 16).
Figure 3 The traveltime curves vary systematically with time. This diagram shows the evolution of the tt curve associated with shots A and B. The residuals are relative to the mean tt at at each station. It is clear that traveltimes are least at the start of the experiment, timeslice 0, and again 12 hours later. The traveltimes vary on two scales. There are variations of 0.2 ms on the scale of 1 hour. This undesireable signal may be the result of poor origin times resulting from imprecise triggering. It could also be the result of slight variations in the waveform which bias the correlation procedure. On the scale of 10 hours however, the signal variation exceeds 1 ms making it clear despite the low amplitude noise.
Traveltimes calculated before ah_timediff revision
Figure 4 These diagrams compare the fluctuations of the water table, as measured in the wells, with the traveltimes from source A and B. Each line on the lower two plots shows the evolution of the traveltime to a particular station. The mean tt to a station has been removed to isolate the time variant signal. It is clear that the changes between adjacent timeslices are consistent across the array. These diagrams show a clear dependence of the P wave traveltime on the water table level.
The correlation is different than one might expect however. From the reference model, the wells and intuition, we expect a layer of slow drier sand to lie on top of a fast saturated zone. When the water table rises, the thickness of the top layer decreases. Seismic energy can reach the bottom layer more quickly. This should cause traveltimes to decrease. The figures above imply that traveltimes increase as the water table rises.
The energy reflected off the water table offers a possible explanation. This PwP phase follows a similar raypath as the direct arrival but reflects from the free surface and then off the water table before arriving at the station. Since the velocity contrast between the two layers is a factor of four, the impedence contrast, and hence the strength of the reflection, is large.
Figure 5 A typical P phase was convolved with a simple filter to simulate the effects of the PwP reflection. The filter used the velocities and densities of the layers to calculate the relection coefficient. The angle of the raypaths through the top layer was calculated assuming the ray had traveled nearly parallel to the boundary in the bottom layer.