The seismic aircraft footprint: probing near surface and tracking aircraftGang Fang, Yunyue Elita Li (National University of Singapore), and Ohad Barak (Mentor Graphics)

SUMMARY

We report on using aircraft-induced ground motions to esti-mate properties of the near-surface medium and also to calcu-late flight parameters of aircraft. We analyze the characteristicsof aircraft-induced signals recorded with a joint 3-C geophoneand smartphone array, deployed in the National University ofSingapore (NUS) campus for traffic monitoring purposes. Us-ing time-frequency analysis, we identify clear ground motionsignal induced by a nearby commercial aircraft in the spec-trogram of the seismic data in the 50-300 Hz frequency band.The aircraft-induced ground motions are transmitted from thesound wave generated by aircraft engines and can be used forestimating elastic properties of the near-surface. We proposeusing the Doppler effect to obtain aircraft flight parameters in-cluding speed, direction and distance from seismic recordings.This experiment shows that low-flying aircraft provide a po-tential source for high spatial resolution near-surface seismicinvestigation, as well as that the geophones can be used to de-tect aircraft and monitor flight parameters.

INTRODUCTION

In urban areas, where the use of active seismic sources is usu-ally prohibited, passive seismic acquisition is a common toolfor conducting urban seismic surveys. Passive seismic en-ergy in urban areas, linked to anthropogenic activities such asquarry blasts (Fang et al., 2020) and urban traffic (Riahi andGerstoft, 2015), has received more and more interest both forits natural characteristics but also as a probing signal for sub-surface seismic investigation. Recently, both geophone andoptical fiber sensing arrays (Chang et al., 2016; Dou et al.,2017; Zhang et al., 2019) have been used to record the ambientnoise field dominated by urban traffic-induced surface seismicwaves for site investigation. At the same time, moving groundsources such as subways and vehicles are studied and trackedthrough the use of spatiotemporal analysis (Riahi and Gerstoft,2015).

Seismologists have shown evidence of observing aircraft-inducedground motions with seismic instruments. Previous researchfocuses on the seismic effects of sonic booms to study po-tential damage to ground facilities and buildings (Oliver andIsacks, 1962; McDonald and Goforth, 1969). These studiesreported that the ground motions are linearly related to themaximum pressure of the boom and to the geological prop-erties of the ground. Sound radiated by low-flying aircraft andrecorded with acoustic sensors has been used for aircraft de-tection, tracking and classification (Sutin et al., 2013; Sedunovet al., 2016). With the increasing number of high-density seis-mic stations in cities and the corresponding increase in aircraft-induced signals observed by these stations, two questions areraised: Can ground motions generated by aircraft be used tounderstand the near-surface material properties? Can we use

the aircraft-induced seismic energy for aircraft detection andflight parameter estimation?

In this work we study the aircraft-induced seismic and soundenergy recorded with a joint 3-C geophone and smartphone ar-ray. We first introduce how the data are acquired. Next, wecharacterize the signals generated by the aircraft using time-frequency analysis. Additionally, we analyze the feasibility ofusing aircraft-induced sound and seismic signals to estimatethe transmission coefficient of the air-ground interface. Fi-nally, we present a method to calculate flight parameters fromseismic recordings using the Doppler effect.

JOINT GEOPHONE AND SMARTPHONE EXPERIMENTAT NUS CAMPUS

The data are acquired by a joint geophone and smartphone ar-ray on the National University of Singapore (NUS) campus onDecember 4, 2019. The array comprises 12 three-componentgeophones (SmartSolo IGU-16HR 3C with 5Hz corner fre-quency) and 7 smartphones of various models (iPhone 6S, 7,7Plus, MI MaxII, Redmi Note 4X). Figure 1 shows the arraylayout. The geophones (denoted by green dots) were planted insoil along Kent Ridge Cres with a 10 m spacing and recordedwith a sampling frequency of 1000 Hz. The smartphones (de-noted by orange arrows) are held by students standing next tothe corresponding geophones and record both video and au-dio signals. The audios are recorded at 44.1 kHz or 48 kHzdepending on the smartphone model.

1

12810

6 4 2

Smart Phone

3C Geophone

20m

NUSAirport

Figure 1: Layout of the joint geophone and smartphone exper-iment at NUS campus. Green dots denote the three-componentgeophones labeled by the channel number. The geophone Z-component is vertically down, the N-component is parallel tothe road and the E-component is perpendicular to it. The ar-rows denotes the position of the person holding the smartphoneand standing next to the corresponding geophones. The bluedotted arrow denotes the civil airliner’s flight trajectory as de-scribed by witnesses on site, and is not a precise descriptionthe path the airplane took.

Seismic aircraft footprint

We recorded 15 minutes of data. Figures 2a and 2b show se-lected 800 seconds of the geophone Z-component and audiorecording respectively, where the audio recording is extractedfrom the smartphone video and downsampled to the same sam-pling rate as the seismic recording. Based on the video in-formation, we know that the large amplitude events that oc-cur simultaneously in the audio and seismic recordings aremainly caused by moving ground-based vehicles (includingcars, busses, vans, trucks and bikes). The seismic and soundrecordings exhibit different frequency and amplitude charac-teristics for the same traffic events. In general, the seismicrecording shows higher signal-to-noise ratio (SNR) and betterspatial coherence than the smartphone audio data. We attributethis difference to the following two factors: (1) The audio sig-nal is strongly contaminated by wind and human speech noise,both of which do not generate strong ground motions. (2) Theinconsistency in the commodity MEMS microphones and cor-responding audio processing in different smartphones. There-fore, only the geophone data are used for array processing inthe subsequent data analysis.

(a)

(b)

Car Truck Aircraft Car

E

N

Z

A

(c)

Figure 2: Comparison between the waveforms of (a) audiorecorded by smartphone with (b) the geophone Z-component.(c) The waveforms of audio and three-component geophonerecordings at channel 6 between 300 s and 340 s, labeled bythe related events.

Figure 2c compares the recordings of audio and the three com-ponent displacements between 300 s and 340 s at channel 6.The data are labeled according to the video recordings. Thoughthis experiment is primarily designed to record ground-basedtraffic data, we happen to pick up the roar from an aircraft pass-

ing overhead. The aircraft sound is identified while listeningto the recorded audio data. Although the video footage of theaircraft is not recorded, according to witnesses’ recollectionthe general direction of the aircraft’s flight is marked with theblue dashed arrow in Figure 1.

From Figure 2c we see that the aircraft signal is much strongerthan the road traffic signal in the audio recordings. However,the geophone data are overwhelmed with ground-based trafficsignal, and one can barely identify the airplane signal from thedisplacements in any of the three directions. In the subsequentsections, we uncover the aircraft signal in the geophone dataand demonstrate that it can be used to provide information toboth the near surface soil property and the aircraft itself.

CHARACTERIZING AIRCRAFT-INDUCED GROUND MO-TIONS

Figures 3a-3d are the power spectrograms of the data compo-nents shown in Figure 2c, obtained by applying short-time FFT(STFT) to the data. On the power spectrograms of the audiorecordings, the strongest energy corresponds to the fundamen-tal harmonic of the aircraft signal, whose frequency reducesfrom 95 Hz to 55 Hz as time goes from 314 s to 326 s. An-other two higher order harmonics are also visible, albeit withmuch lower energy. The ground-based traffic generates broad-band audio signals that are slightly stronger than the ambientaudio noise. As already shown in the time domain represen-

Car Truck Aircraft Car

A

Car Truck Aircraft Car

Z

Car Truck Aircraft Car

E

Car Truck Aircraft Car

N

(a) (b)

(c) (d)

Figure 3: Comparison of the spectrograms of (a) audio and(b) geophone Z-component, (c) E-component and (d) N-component recorded at channel 6. The spectrograms are ob-tained by applying STFT with a Tukey window (0.256 s win-dow length, 8% overlap). The amplitudes indicate PowerSpectral Density (PSD) in dB.

tations, the spectrograms of the geophone recordings providebetter temporal resolution and SNR for the ground-based traf-fic signals.

The aircraft signal, which seems to be undistinguishable in thetime domain presentation, is clearly visible on the power spec-trograms of all three components. Since the aircraft was cruis-ing at a high altitude of several hundred meters prior to land-ing, the sound wave mostly impacted vertically. Consequently,the ground motion is stronger on the vertical component thanin the two horizontal components.

Seismic aircraft footprint

Figure 4a shows the waveforms of aircraft-induced ground mo-tions in Z component, after applying a 50 - 300 Hz band-passfilter and zooming into a smaller time range from t = 314 s tot = 326 s to remove interference from the ground-based traf-fic signal. Taking channel 1 as the master channel and cross-correlating it with all other channels, we obtain the cross-cor-relogram shown in Figure 4b. There is a slight shift of thewaveform arrival time along the array. We observe that thereis no significant waveform dispersion in the cross-correlogram.This indicates that the observed aircraft seismic signals arecoming from sound waves generated by the aircraft’s enginesand transmitted at the air-ground interface. This observationalso provides a reasonable explanation why aircraft signals aremore clearly recorded by the seismic vertical component thanthe horizontal components. By applying a linear least-squarefitting on the time shifts in cross-correlogram, we estimate theapparent velocity of the aircraft sound wave to be approxi-mately 1308 m/s.

(a) (b)

Figure 4: (a) aircraft-induced ground vibration in geophone Z-component recording (after applying a 50 - 300 Hz band-passfilter.) (b) Cross-correlogram of (a) taking channel 1 as themaster trace.

HIGH RESOLUTION NEAR SURFACE PROPERTY VARI-ATION

Considering the altitude of the aircraft and the size of the array,we assume a normal-incidence planar sound-wave impingingon the air-ground interface. The transmission coefficient ofsound is defined by the ratio of the amplitude of the trans-mitted P-wave to the amplitude of the incident sound wave,T = AT /AI . We use seismic Z component to approximate thetransmitted P wave in soil. The transmission coefficient can beapproximated by the following deconvolution,

Di(ω) =Zi(ω)S∗(ω)

S(ω)S∗(ω)+ ε 〈S(ω)S∗(ω)〉, (1)

where Zi(ω) is the observed Z-component seismic signal atchannel i, S(ω) denotes the sound signal, 〈·〉 is the mean valuefunction, and ε is a regularization parameter.

Figure 5a shows the deconvolution result obtained by decon-volving the sound wave recorded at channel 1, with ε=0.1.The audio recording with different smartphones can be seenas multiple measurements of the incident aircraft sound wave.We apply equation 1 to each of the audio recordings and pick

the maximum amplitude from corresponding deconvolution re-sults, which are plotted with the dashed lines in Figure 5b. Theaudio recordings are normalized before applying deconvolu-tion to remove the inconsistency in amplitude caused by thedifferent microphones. The solid line in Figure 5b denotes theaverage value of the dashed lines. These lines show a similartrend which indicates an amplitude anomaly around 90 me-ters. Noted that the numbers shown in the Figure 5b do notmean the actual transmission coefficients, but values normal-ized to between 0 and 1. After deconvolving the sound signalfrom seismic signal, we can obtain an approximation of thespatial variation of the transmission coefficient, which reflectsthe variation of velocity and density of the near surface soil.Since aircrafts cruise at a high altitude, we assume the incidentsound wave is a plane wave and treat each geophone individu-ally to provide point-wise relative soil property variation.

(a)

(b)

Figure 5: (a) The result of deconvolving sound wave recordedat channel 1 from seismic recording shown in Figure 4a. (b)The maximum value of deconvolution resulting from decon-volving the acoustic signals from the vertical geophone signalsat all stations. The blue solid line denotes the average value ofthe dashed lines.

FLIGHT PARAMETERS ESTIMATION

An interesting observation in this experiment is the Dopplerfrequency shift shown in both the audio and the seismic record-ings (see Figures 3a-3b). The Doppler effect shifts the truesource frequency to a higher observed frequency as the sourceapproaches an observer, and shifts the observed source fre-quency down as the source recedes from the observer. The air-ground coupling transmits the Doppler effect of the aircraft’sflight from the acoustic air signal to the seismic ground signal.Since the Doppler frequency shift is determined by the speedof the aircraft and by the orientation of the aircraft’s heading

Seismic aircraft footprint

relative to the geophone location, we propose using the ob-served Doppler effect in the seismic recording to estimate theaircraft’s flight parameters.

𝑥𝑦

𝑧

𝑣 ∗ 𝑡

𝑠

𝑟(𝑡)𝑣

𝛼(𝑡)

𝜽

𝑑(𝑡)ℎ

A

Figure 6: Schematic diagram of the aircraft passing the geo-phone array.

A schematic diagram of the aircraft passing the geophone ar-ray is illustrated in Figure 6. We assume that the aircraft moveswith a constant speed v along a straight line, emiting narrow-band sound with a constant frequency f0. Point A representsthe initial position where the time t is set to be zero. The ob-served frequency f at geophone (denoted by triangles in Fig-ure 6) can be described by the following formula (Gill, 1965),

f = f0c

c− v · cos(α(t)), (2)

where f0 is the emitted frequency from the source, c denotesspeed of sound in air (established as 340 m/s), and α standsfor the angle between the vector of aircraft’s velocity ~v andthe vector connecting the aircraft and the geophone ~d. Whiletaking into account that,

cosα(t) =r(t)d(t)

=s− vt√

h2− (s− vt)2, (3)

where s denotes the flight distance from the initial position tothe position closest to the geophone, h denotes the vector be-tween the aircraft and the point where it is closest to the geo-phone, we can obtain the observed frequency by

f = f0c√

h2 +(s− vt)2

c√

h2 +(s− vt)2− v(s− vt). (4)

We estimate the unknowns of f0, v, s and h by using equa-tion 4 to fit the the Doppler effect observed in the spectro-gram of the seismic recording. In Figure 7 the red dots aremanually picked observed frequencies of the aircraft at differ-ent times. The blue dashed line denotes the nonlinear fittingcurve to the dots. Considering the apparent speed of the air-craft vp measured from seismic cross-correlogram and the vobtained by nonlinear fitting, we can calculate the angle be-tween the flight direction and the seismic array (θ in Figure 6)by θ = arccos(v/vp). The estimated flight parameters are:

• Emitted frequency f0: 69.2 Hz

• Aircraft speed v: 96.3 m/s

• Flight distance s: 1662.6 m

• Closest distance h: 701.3 m

• Flight direction (relative to the seismic array) θ : 85.8◦

Note that the first four parameters can be obtained with onesingle geophone. After consulting a professional pilot, we be-lieve that these obtained flight parameters are within a rea-sonable range. It is most likely the recorded seismic signaloriginated from a passenger plane maintaining a holding pat-tern before landing at Changi airport, when it just passed ourobservation array. To the best of our knowledge, this is thefirst practical example of using seismic data to estimate air-craft flight parameters.

Figure 7: Doppler effect shown in spectrogram of the first or-der harmonic observed at channel 6. The blue dashed linedenotes the nonlinear curve fitting of the picked peaks fromspectrogram.

CONCLUSIONS

A Investigation of the aircraft-induced ground motions observedby three-component geophones suggests potential applicationsfor estimating elastic properties of the near surface and simul-taneously for aircraft detection and monitoring. The verticalgeophone component is more sensitive to the aircraft signalsthan the horizontal components. High spatial resolution es-timation of the transmission coefficient variation at the air-ground interface can be obtained by a deconvolution method.By fitting the observed Doppler effect in seismic recording, weproposed a new way to estimate aircraft flight parameters withgeophone data.

ACKNOWLEDGMENTS

We would like thank Aaron Fong (Singapore Airlines) for theuseful discussions. The authors acknowledge the EDB PetroleumEngineering Professorship for financial support. We wouldlike thank Cambridge Sensing Pte Ltd for financial support.Gang Fang is additionally supported by National Natural Sci-ence Foundation of China (grant 41504109).

Seismic aircraft footprint

REFERENCES

Chang, J. P., S. A. de Ridder, and B. L. Biondi, 2016, High-frequency rayleigh-wave tomography using traffic noisefrom long beach, californiatomography using traffic noise:Geophysics, 81, B43–B53.

Dou, S., N. Lindsey, A. M. Wagner, T. M. Daley, B. Freifeld,M. Robertson, J. Peterson, C. Ulrich, E. R. Martin, and J. B.Ajo-Franklin, 2017, Distributed acoustic sensing for seis-mic monitoring of the near surface: A traffic-noise interfer-ometry case study: Scientific reports, 7, 11620.

Fang, G., Y. E. Li, Y. Zhao, and E. R. Martin, 2020, Urbannear-surface seismic monitoring using distributed acousticsensing: Geophysical Research Letters.

Gill, T. P., 1965, The doppler effect: Logos Press, AcademicPress.

McDonald, J. A., and T. T. Goforth, 1969, Seismic effects ofsonic booms: Empirical results: Journal of geophysical re-search, 74, 2637–2647.

Oliver, J., and B. Isacks, 1962, Seismic waves coupled to sonicbooms: Geophysics, 27, 528–530.

Riahi, N., and P. Gerstoft, 2015, The seismic traffic footprint:Tracking trains, aircraft, and cars seismically: GeophysicalResearch Letters, 42, 2674–2681.

Sedunov, A., A. Sutin, N. Sedunov, H. Salloum, A.Yakubovskiy, and D. Masters, 2016, Passive acoustic sys-tem for tracking low-flying aircraft: IET Radar, Sonar &Navigation, 10, 1561–1568.

Sutin, A., H. Salloum, A. Sedunov, and N. Sedunov, 2013,Acoustic detection, tracking and classification of low flyingaircraft: 2013 IEEE International Conference on Technolo-gies for Homeland Security (HST), IEEE, 141–146.

Zhang, Y., Y. E. Li, H. Zhang, and T. Ku, 2019, Near-surfacesite investigation by seismic interferometry using urbantraffic noise in singapore: Geophysics, 84, B169–B180.