Data characterized in frequency domain
Figure 6a shows typical band-averaged amplitude spectra of background geomagnetic activity recorded on the QuakeFinder ANT4 north–south sensor at various times throughout the day. The background signals we measured were largely “external signals,” originating from natural sources in the magnetosphere and ionosphere. At our highest recorded frequencies, the first Schumann resonance (Nickolaenko and Hayakawa 2014) produced an increase in geomagnetic activity in the range 5–10 Hz. At lower frequencies, there was a monotonic increase in activity with decreasing frequency. Amplitude spectra clearly showed higher power during the day and the lowest power at night, during the “quiet time” (02:00–03:00 local time) when the electric trains were inactive. At 0.03–0.05 Hz, around which the electric-train signals are centered, the background noise was 10 times greater during the day than during the quiet time (Karakelian et al. 2000).
The amplitude spectra also showed good coherence between different sensor models (Fig. 6b). The first Schumann resonance peak (8 Hz) had an amplitude of around 0.45 pT/sqrt(Hz) as recorded by all three north–south sensors and consistent with the expected values (e.g., Heckman et al. 1998). Over most of the frequency band of interest, the QFido3 was, as expected, the noisiest system.
Spectrograms of the different coils also showed good correlation (Fig. 7). Several broadband signals appeared as anomalously large amplitudes over short time periods; these generally occurred on days of anomalously high global geomagnetic activity (geomagnetic storms are marked as black triangles in Fig. 7). An interesting phenomenon, which was present in all datasets but was not obvious from the time series or spectra, was a banded signature with harmonics at multiples of 0.5 Hz, occurring at the same time each day during April 12–18 and May 5–10. These signals were probably caused by the SLAC National Accelerator Laboratory, which has its injector ~ 1 km distant from our magnetometers. During the period in question, SLAC was using parts of the 3 km linear accelerator comprising 65 MW pulsed klystrons driving accelerator structures with beam rates at, variably, 1, 10, and 120 Hz.
Data characterized in time domain
We compared north–south component magnetic field data recorded on all three coil types at JRSC with each other and with data from a remote reference site, FRN, ~ 225 km from JRSC (Fig. 1). Day-long voltage and nanotesla time series with a 1000-s high-pass filter (Fig. 8) demonstrated the importance of the instrumental transfer functions (Fig. 4): the night-time period that is evident as a quiet time in the voltage time series was enhanced in the nanotesla time series, in which the lowest decade of frequencies (0.001–0.01 Hz) had been boosted to its correct level (Fig. 6a), dominating the 0.01–0.1 Hz bandwidth of electric-train noise. When studying transient magnetic pulses, we commonly reviewed our time-domain data after a 200-s high-pass filter (Fig. 9a), allowing us to recognize anomalies, particularly during the quiet overnight period. One such significant disturbance during the quiet period (Fig. 8) could be clearly observed in the 200-s high-pass nanotesla time series and was also seen in the data from our remote reference site, FRN (Fig. 9a). Focusing on this disturbance (Fig. 9b), we found overall good coherence between FRN and all three sensors at JRSC in the 30–300 s period range, confirming that this is an external signal almost equally affecting all sensors across a region > 200 km. Based on the period, we identified this signal as a Pi2 irregular geomagnetic pulsation (e.g., Baumjohann and Nakamura 2009).
Not all signals observed at JRSC are atmospheric in origin, as attested by examples (Fig. 10) in which signals recorded at JRSC were not observed at remote reference FRN. Eight examples of similar signals were observed during a visual inspection of a 24 h segment of data, and six examples of such signals were observed when inspecting the quiet periods of each day (2 h, while BART is non-operational). These signals were absent from the FRN records but were recorded on all three separate sensor types at JRSC (Fig. 10a), demonstrating a local origin for these single-sided pulses. Although some signals with this shape (cf. Dunson et al. 2011) are known to be associated with lightning strikes (cf. Bleier et al. 2009), no strikes were recorded by a global lightning database (https://www.earthnetworks.com) within 500 km of the Bay Area corresponding with the time of this pulse. Instrumental noise could also be ruled out as the source, because the pulse was detected on all three coils and digitized on two separate systems. An assessment of records from nine nearby QuakeFinder stations (Fig. 11) revealed that the pulse was detected at all of the sites, but with varying amplitude and polarity. This indicates that the source was large enough to register on a regional-scale network; however, we could not determine whether this signal was cultural noise or a signal generated internally within the earth, without a more detailed analysis that would have been beyond the scope of this work. To make such a distinction would require a more careful cataloging of all nearby anthropogenic sources, as well as perhaps a temporal coincidence with a possible causative tectonic event. Alternatively, because the signal was recorded at a wide range of sites, the source of the signal could be deduced by inverse-modeling of current source shape and location (cf. Minamoto et al. 2011; Nagamachi et al. 2013). The sensor-spacing of such an array would need to be compatible with the nature of the sources. Although similar pulses have been suggested to have tectonic causes prior to earthquakes (Bleier et al. 2009; Dunson et al. 2011), in this case we suggest that a cultural noise source is a more likely origin. Irrespective of the cause of this pulse, the presence of the same signal on all three north–south sensors at JRSC is a positive demonstration that we can expect to successfully integrate data from different ULF networks despite their different equipment types.
In contrast to the desired—and expected—concurrence of the data from different sensors, in some instances signals were observed only on one system or one sensor, and were absent on the others, indicating either instrument noise (perhaps related to the power supply or digitizer) or extremely local phenomena (Fig. 10b–d). Figure 10b shows a complex 4-s signal occurring on the two QuakeFinder north–south sensors but not present (or highly attenuated) on the QuakeFinder east–west sensor, nor on any USGS-Stanford sensor. This two-sided pulse must have originated locally to the QuakeFinder instruments, since it was not visible 50 m distant on the USGS-Stanford sensors; moreover, it must also have been strongly directional and likely represented an electromagnetic source, since it was at least 20× larger on the north–south than on the east–west components. The time of day (09:04 local time) was consistent with an anthropogenic source. By contrast, a signal observed on the USGS-Stanford north–south sensor but not on any other sensors or systems (Fig. 10c), or on the QuakeFinder QFido3 but not on other sensors or systems (Fig. 10d), may represent system noise or perhaps physical disturbance to a single sensor (e.g., ground squirrel activity). Whereas the example in Fig. 10a is encouraging in demonstrating signal fidelity across independent systems, the examples in Fig. 10b–d are cautionary in demonstrating that there remain many individual signals that are clearly neither atmospheric nor tectonic in origin, which defy easy explanation.