Taylor Swift’s Eras Tour began in the summer of 2023. Attendance at the two Seattle concerts averaged over 70,000 each night, which at the time set the stadium’s attendance record. Fans were allowed to enter the stadium at 16:30 PDT, and the concert was slated to begin at 18:00 PDT. The concert began with two warm-up acts: Singer Gracie Adams played six songs, after which the band HAIM played an additional six songs. Taylor Swift then played for approximately 3.5 hours (Figure 2).
As implied by the tour title, the setlist was broken into segments, or “eras”, each relating to a musical period in the artist’s history. The character of the music changed during some eras, with some eras inspiring more dancing, and others having a slower beat and more subdued nature. Other forms of media and entertainment, including video, dancing, and changes in set design, accompanied each of the eras as well: the stage was backed by a large video screen on which images were shown both between and during songs. Sets and costumes were changed between Eras, and this process took up to several minutes, resulting in times in which sound production was greatly reduced.
The setlist for all concerts in the tour was largely identical, with both the order and choreography of songs scripted. Each concert, however, included two “surprise songs” which were different on the two nights at Lumen Field. This allowed us to compare time periods in which songs were repeated, and in which they were distinct between the two concerts.
STADIUM:
Lumen Field (previously known as CenturyLink Field, Qwest Stadium, and Seahawks Stadium), is oval-shaped, with its long axis oriented north-south. It is reputed to be among the loudest in the National Football League: during a 2013 football game fans broke a Guinness World Record for sound volume (https://www.foxnews.com/sports/seattle-seahawks-fans-set-stadium-noise-record-sunday). The stadium is located on the site of the former Kingdome, which was destroyed in a controlled demolition in 2000. Seismic analysis of that implosion, as well as drill holes in the region confirm that the stadium is underlain by thick alluvial sediments, including sands and muds [1,12]. The stadium’s maximum capacity is ~69k for events in which spectators are restricted to permanent seating but can be increased if attendees are permitted on the field. The crowd at the Taylor Swift concerts was estimated to be ~72k for each of the two nights.
CITIZEN SCIENCE:
Taylor Swift has a large number of dedicated fans who refer to themselves as “Swifties”. When a local Seattle news station reported that the concert had been recorded seismically, dozens of Swifties reached out to the lead author to offer their help. We created a Google Drive to which Swifties could upload concert videos as well as a spreadsheet on which they could log song start times and personal observations of the concerts. In total we received ~80 videos and dozens of comments and observations (Supplementary Table S1). Time stamps on videos and photos allowed us to confirm song and event timing, although most were precise only to the minute. The second author attended the second concert.
We received videos from attendees seated in a variety of locations in the stadium. Videos taken from high in the bleachers allowed us to observe crowd behavior across the field, and images from near the stage provided us with a view of the two bands (located on either side of the main stage). Some Swifties sent video of the substantial crowd located outside of the stadium, and we were able to use visible landmarks to identify their location.
SEISMIC DATA:
Seismograms were recorded during the concerts on station UW.KDK [13], located ~150 m due west of Lumen Field. KDK is a 3-component TITAN accelerometer, with flat response at frequencies < ~30 Hz. Data from station KDK were downloaded from the EarthScope Data Management Center and plotted in both the time and frequency domains. We selected a time period for analysis that spanned the expected time of the concert, including several hours prior to the venue opening to capture sound checks, and several hours after the concerts were expected to end.
Waveforms for the two concerts show a prolonged (hours) series of regular, high-amplitude pulses (Figure 2). These can broadly be separated into three segments: The first significant period of increased signal strength lasts 23 minutes, the second lasts 30 minutes, and the final sequence lasts ~3.5 hours. Each of these time periods is composed of shorter duration (3-6 minute) bursts, separated by short periods of quiescence (Figure 2, inset). We propose that the first two periods represent performances by the two opening acts, followed by the headline event. The duration of the shorter segments is consistent with the length of many pop songs. Further analysis confirming that these pulses represent discrete songs is described below.
Seismic signals recorded during the concerts may be separated into two frequency bands: low frequency signals are between 1-8 Hz, and high frequency signals are broadband between 30-80 Hz (Figure 3). Signals in the high frequency band include a variety of patterns: some have short pulses at narrow frequencies while others are more smoothly broadband (Figure 3). In contrast, the low frequencies are extremely narrow-band, and they exhibit harmonics at frequencies that vary between songs (Figures 3 and 4).
Two prolonged diffuse low frequency (-5-20 Hz) signals are visible in the seismic record, beginning at ~16:30 and ~23:30 (Figure 3). The venue was opened to the public at 16:30, and both seismic and video data indicate that the concert ended at ~23:30. We attribute these signals to fans arriving and departing. It is unclear whether the signal reflects vehicular or foot traffic.
Even when observed by eye it is apparent that data recorded over the two nights were highly similar (Figures 2 and 5). This is consistent with seismic energy generated during a repeated set list. Cross-correlation of five hours of data between the two nights reveals a maximum correlation when the data are offset by 26 minutes; subsequent conversations with concert attendees confirmed that the second night of the concert was delayed by an estimated half hour. This provides additional evidence that the seismic signals were generated during the two concerts.
To test that the waveforms represented a predictable set list played on both nights, we cross correlated signals from the shorter-duration pulses (interpreted as, and hereafter referred to as songs). Correlations of a song waveform over both nights were generally high (>0.75), with some correlations exceeding 0.95 (Figure 5a).
In each concert of the Era tour, Taylor Swift played two “surprise songs” (Figure 6). These differed each night of the concert, and thus represent an excellent opportunity to test concert correlations. Indeed, the surprise songs represent the only part of the concert that was not highly similar in time series. Because the two nights’ surprise songs differ in length, the last era of the concerts were also offset relative to one another. Unsurprisingly, the time period of the surprise songs shows the lowest correlations, with a maximum correlation coefficient of 0.03 (Figure 5b).
SONG IDENTIFICATION:
That the short-duration pulses were in fact different songs was confirmed by (a) signal duration, (b) song rhythm, and (c) sonification of the seismic signal. Concert setlists were published online (e.g. https://www.setlist.fm/setlist/taylor-swift/2023/lumen-field-seattle-wa-3ba4a0dc.html) and confirmed by videos submitted by attendees.
To first order, waveforms were correlated with songs by comparing their onset times with photos, comments, and videos sent in by attendees (Supplementary Table S1). Most videos shared by Swifties were time stamped only to the minute, which resulted in some uncertainty in the precise start times of the songs. In many cases, a clear increase in amplitude was visible in the time series near the approximated start time, but in other cases, the onset time was less certain. We were able to refine start times by viewing a complete video of the concert posted online. In many cases, we were able to compare song durations to published versions, but many songs were performed with extended intros or shortened for the concert.
Station KDK has a sample rate of 200 Hz and an anti-aliasing filter at 80 Hz. Thus the highest frequencies available in the seismic record would, at best, capture the bass portion of each song; much of the vocals, guitars, and piano music fall outside of the instrument’s detection capacity.
In nearly all cases, the seismic data have distinct, narrow-band low frequency (1-8 Hz) signals that exhibit harmonics (Figure 4). The spectral content of these signals is well below what would be expected of music, but precisely matches the rhythms of specific songs. Figure 4 shows examples in which the fundamental frequency or first harmonic of each song correlates with the published rhythm of the song. For example, Ready For It (Figure 4) has a published rhythm of 160 beats per minute (BPM), or 2.67 beats per second (Hz). The next three songs exhibit strong shaking at 1.6 Hz (96 BPM), 2.2 Hz (132 BPM), and 2.1 Hz (126 BPM) consistent with published values for Delicate (95 BPM), Don’t Blame Me (136 BPM) and Look What You Made Me Do (128 BPM).
All of the low frequency signals exhibit at least one harmonic. It is possible that part of the audience moved with a primary rhythm where others moved in double-time. Studies have shown that crowds jumping as a group are most likely to bounce at frequencies between 1-8-2.3 Hz [14]. Thus if a song has a tempo significantly outside of this range the audience may be most comfortable moving at twice or half of the song’s primary rhythm.
It is most likely, however, that the harmonics result from a Dirac comb effect, in which regularly spaced pulses exhibit a comb-shaped spectrum. This effect has been invoked for a variety of natural signals that appear harmonic but are not associated with resonance, such as regularly-spaced earthquakes [15,16] and ocean swell [17]. Diaz et al. (2017) invoked the same process for harmonic signals recorded during a Bruce Springsteen concert [4]. It is unlikely that the crowd moved at frequencies much higher than 3 Hz, lending support to a model in which the harmonics are a consequence of the Fourier transform rather than audience motion.
DATA INTERPRETATION:
Concert attendees reported observations that allowed us to interpret seismic signals throughout the concert and correlate ground shaking with specific crowd and band behaviors. We begin by describing in detail the observations and seismic data observed at the beginning of the concert, and then present data for several other periods of the concert in which signals were significant.
At approximately 19:49 PDT, a cart in which Taylor Swift was hidden was brought on stage, generating an enthusiastic audience response (Figure 7). During this time, a recording of the song Applause by Lady Gaga was broadcast across the sound system. At 19:54:46, a large digital countdown clock was revealed on the main screen, marking the final two minutes and 23 seconds before the main event. The recorded music now switched to You Don’t Own Me as performed by Dusty Springfield. At 19:56:12, the countdown reached zero, and performers entered the stage area. Between the countdown clock and Taylor Swift’s first appearance at 19:58:15, several events inspired a loud crowd response: (a) the song Miss Americana & the Heartbreak Prince began at 19:56:38, (b) a door opened in the main screen, from which performers emerged (19:56:50), and (c) the first performer entered the stage (19:57:07).
Each of these events is visible in the seismic data (Figure 7). The moment of the cart’s arrival is poorly known: an attendee noted the time to the minute only. The song Applause is visible in the seismic data as a period of higher amplitude lasting 3:30, with sustained low frequencies of ~2.3 and ~4.6 Hz; these values are consistent with the song’s published duration and tempo of 140 BPM. You Don’t Own Me is a ballad with a more mellow cadence than that heard in Applause, and indeed the seismic data are lower in amplitude and low frequencies are only weakly visible.
Once the countdown clock reaches zero, video data show the crowd loudly cheering at events a-c above. This is a period in which the seismic data have no low frequency harmonics but contain brief (seconds) pulses of energy between 40-80 Hz. We propose that these reflect the screams and cheers of an enthusiastic crowd that is not moving in a synchronous manner. In contrast, once Taylor Swift begins singing, a strong low-frequency signal initiates, as do high frequency broadband signals (Figure 7). As noted above, the low frequency signals are consistent with song rhythms, and the broadband signals likely reflect the lowest frequencies in the music.
Seismic amplitudes vary over the course of a given song, and crowd-sourced video and observational data confirm that periods with different amplitude correspond to different parts of the song, including verses, choruses and bridges. In Figure 8 we present low-frequency (1-8 Hz bandpassed) waveform data for the song Love Story. Text labels along the top denote changes from verse to chorus, or to instrumental sections of the song, with representative lyrics. Lower text labels describe observations, from video data, of crowd motion or singer/crowd energy levels. The data show a strong correlation between song structure, crowd activity, and amplitude of ground shaking. Video data show that during the chorus (“Romeo take/save me…”) the audience begins to jump synchronously; these periods correspond to sharp increases in ground acceleration at KDK (Figure 8). On a qualitative scale, no significant change in the volume of the music was observed at this time. Further, seismic amplitudes remain low at the end of chorus 3, while video data show an audible increase in the energy and amplitude of the drums. This provides strong evidence that changes in seismic amplitude are associated with crowd behavior rather than the music or sound system.
We further examine the contribution of crowd behavior to low frequency seismic amplitudes by considering times when the crowd was absent. With this in mind, we evaluated seismic data recorded during periods interpreted as the concert’s sound check. Assuming that we can identify songs by their seismic character, there were two sound checks prior to the July 22 concert: one at ~17:25 on July 21, and one at ~12:00 PDT on July 22 (Figure 2). A similar signal appears in the seismic record at 13:15 PDT on July 23, several hours before the second concert.
Both the presumed sound check data and the signals recorded during the concerts have strong signals between 30-80 Hz, in bursts lasting several minutes, consistent with the duration of songs (Figure 2). However, only the songs played during the actual concerts exhibit strong low frequencies (1-8 Hz). Although we were unable to find a set list for the sound checks, videos posted online suggest that some of the same songs were played in both sound check and concert. Low frequency signals were thus only recorded seismically when fans were in attendance. This provides additional support for our contention that these signals were generated by crowd activity.
The two “surprise songs” played at each concert were distinct from the rest of the setlist in that (a) they were different each night, and (b) the singer introduced them with the words “welcome to the acoustic section of the evening”. During most other songs the singer was accompanied by a band with guitars, bass, and drums, but the surprise songs were performed solo, on guitar or piano. The different character of these songs is evidenced in spectrograms: the surprise songs contain significant low-frequency energy, but the 30-80 Hz shaking is absent from the record (Figure 6). The absence of drums or other low-frequency instruments (e.g. bass guitar) does not correlate with a loss of low frequency signal, again confirming that audience motion is a sufficient seismic source. High frequency energy is lacking from other periods of the concert, but video data confirm that these periods either involve Taylor Swift talking with the audience or playing solo (e.g. during the song Champagne Problems). These observations provide a window into the cause of the high frequency signals–these appear to be specifically associated with amplified music.
BEAST QUAKE (TAYLOR’S VERSION):
There is no means by which the 2011 Beast Quake and the 2023 Taylor Swift concerts can be directly compared. The Beast Quake represented less than a minute of ground shaking caused by an enthusiastic, but randomly moving crowd of ~66,000. In contrast, the Eras Tour concerts lasted ~3.5 hours and had significantly larger attendance. Further, thousands were on the field for the concert, where their energy could couple directly into the ground, whereas Seahawks fans were mostly confined to the stands. Most importantly the motion of the Swifties was in a uniform manner, as the crowd jumped and swayed to the beat of the music. This caused constructive interference and amplification of seismic energy at frequencies identical to the beat of the music. Studies of vibrations induced by crowd behavior show that the energy imparted into the ground by synchronized motion such as jumping or swaying is proportional to crowd size, whereas random motion scales with the square root of crowd size (Parkhouse and Ewins, 2006). The larger signals recorded during the concert are likely a function of the type of motion rather than the relative enthusiasm of the fan base.
If credit for the seismic signal is to be given to the fans, we must rule out the contribution to ground shaking from other sources. Studies of seismic data recorded in association with other stadium events have suggested shaking may reflect resonance of the stadium itself and/or the subsurface sediments [2]. Our data show no common frequencies observed throughout the concert, which might be expected of resonance, although we cannot rule out contributions from non-resonant stadium shaking induced by the crowd.
Green and Bowers (2008) state that seismic signals recorded during an electronic music festival were caused by the sound system coupling into the ground [6]. As with the Taylor Swift concert, they show that spectral frequencies are similar to the music tempo. While sound system coupling may be a contributor to the signals presented here, the lack of low frequency energy recorded during empty-stadium sound checks strongly suggests that the primary source of low-frequency energy was crowd motion.