toward understanding subtle instrumentation effects associated...

Toward Understanding Subtle Instrumentation Effects Associated

with Weak Seismic Events in the Near Field

by Jiří Zahradník and Axel Plešinger

Abstract Broadband observations of small earthquakes at short epicentral distancesreveal a mixture of near-field effects and instrumental artifacts. We investigated thesephenomena at a station equipped with an STS-2 and CMG-40T sensor situated almostabove shallow M 3.0 to 3.8 events (peak ground acceleration 2 × 10�1 m=sec2).The horizontal components were systematically accompanied by tiltlike disturbances,and the tilt obtained from the STS-2 records exceeded more than 10 times the valuespredicted by the source model. We also observed a so far uncommonly recognized typeof disturbance, whose shape is the first derivative of the tiltlike disturbance. The mostlikely explanation seems to be clipping of high-frequency signal peakswithin the sensorsystem. A computational model of a broadband feedback velocimeter as a lineardynamic system with saturation proved this interpretation on a qualitative level.Generally, any asymmetry in the transfer of high frequencies in the feedback velo-cimeterwould produce a long-period disturbance of this type.Users of near-fault broad-band velocigrams may numerically simulate the disturbances, without any knowledgeof their physical nature, and subtract them from the records. The decontaminatedrecords still may have a strange, bow-shaped form, related to the near-field rampand the static displacement (of the order of 1 × 10�5 m in this article). The effectsstudied in this article seem to have a general character, for apparently any feedback-controlled broadband velocimeter.

Introduction

Earthquake seismology relies on understanding pro-cesses taking place at rupturing faults. Of particular impor-tance are observations in epicentral regions where specificnear-field phenomena provide invaluable insight into thesource physics. They include, for example, permanentdisplacements, tilts, and rotations. With recent advancesin density and configuration of broadband networks andwith improvement of the quality of the instruments, manyvaluable near-fault records have also been obtainedfor weak earthquakes. Some of them are obscured bylong-period disturbances whose origin has not yet beenfully understood. Most likely they represent a mixture ofsubtle ground motion and instrumental effects, whosecorrect deciphering is challenging. Identification andpossible removal of the disturbances has a broader impactthan just into tiny details of the earthquake source. Aproper treatment of the disturbances should be a part ofnewer data acquisition and quality control procedures. Thisis particularly important in real-time applications, such asquick moment-tensor calculations, identification of faultplanes, and release of the shake maps, where disturbedrecords might bias even gross estimates of the earthquakeparameters.

Near-field effects (such as a permanent displacement,tilt, and rotation) have been well-known in theory and alsoobserved during strong earthquakes for quite a long time(Bouchon and Aki, 1982), but their observational analysisfor small earthquakes is still quite rare. This article is focusedjust on small events at short epicentral distances.

Long-period disturbances have already received atten-tion in relation to tilts (Wielandt and Forbriger, 1999; Kalkanand Graizer, 2007), baseline corrections (Boore et al., 2002),strong-motion data processing (Graizer 2005), coseismicdeformation and permanent displacement (Boroschek andLegrand, 2006; Wu and Wu, 2007), nonvertical installationof sensors (Pillet and Virieux, 2007), and sensor nonlinearity(Delorey et al., 2008). The present article is a follow-up ofour previous investigation of weak events at near stations inGreece, often accompanied by tiltlike disturbances on CMG-3T and LE-3D/20s records (Zahradník and Plešinger, 2005).

A strong motivation for this article came from the recentearthquake swarm in West Bohemia, on the border betweenthe Czech Republic and Germany. Long-period disturbancesoccurred quite systematically on the Streckeisen STS-2 re-cords ofM ∼ 3 events. Fortunately, a station in the epicentralregionwas equipped alsowith aGuralp CMG-40T instrument.

59

Bulletin of the Seismological Society of America, Vol. 100, No. 1, pp. 59–73, February 2010, doi: 10.1785/0120090087

The two collocated sensors opened a way to a better under-standing of tiltlike disturbances, similar to those we observedin Greece. In particular, we tried to answer the fundamentalquestion of whether we actually recorded a ground-motiontilt, or an instrumental artifact. Evenmore importantly, duringthe study a new kind of disturbance also emerged. Its shapewas the first derivative of the tiltlike disturbance. Consideringthat the tiltlike disturbancewas formally equivalent to a step ininput acceleration, the new kind was equivalent to a step ininput velocity. Because the velocity step had no plausiblephysical explanation in terms of the near-field effects, it hadto be investigated as a purely instrumental artifact.

Therefore, the general objective of this article is to betterunderstand how subtle ground-motion effects (small perma-nent displacements and tilts) are mixedwith subtle instrumen-tation effects. More specifically, we try to use two collocatedbroadband feedback velocimeters, STS-2 and CMG-40T, tostudy disturbances duringMw 3 to 3.8 earthquakes, recordedat epicentral distance of 2 km. We are mainly interested ina new kind of a long-period disturbance, called U-shapeddisturbance, whose shape is the first derivative of the tiltlikedisturbance.

As a tool, we use simple source models and syntheticseismograms, together with modeling a broadband feedbackvelocimeter as a linear dynamic system. We show in thisarticle that very large tiltlike disturbances on STS-2 recordsare definitely due to an (unknown) instrumental effect. Thenew U-shaped disturbance is interpreted as caused by instan-taneous clipping of high-frequency signals within the instru-ment, excited by high-frequency ground motions. We alsodemonstrate how to remove these disturbances from the re-cords through their numerical modeling, even without exactknowledge of their physical origin. As a result of subtractingthe modeled disturbance, we show that a permanent ground-motion displacement can be resolved from the near sourcerecordings of small earthquakes, for example, of the order of1 × 10�5 m for theMw 3.6 earthquake studied in this article.

Understanding of the disturbances gained in this articleimplies that similar effects are likely to be present at nearrecordings of weak events anywhere in the world, on differ-ent types of broadband feedback velocimeters. Therefore, atthe end of the article we give also two examples from Japanfor shallow crustal earthquakesM ∼ 4, recorded at epicentraldistances 1 to 13 km with a disturbance on STS-1 and STS-2.

If these effects represent an important general phenom-enon, then why have they not been broadly recognized yet?The reason obviously is that for interpreters they usuallyrepresent merely noise, so such records are excluded fromanalyses. Much worse are situations when the disturbancesare simply overlooked in routine data processing, becausethen they may potentially bias any long-period sourcestudies. That is why newer automated data quality proce-dures should include identification of the disturbed recordsand, possibly, also their correction. As a by-product, thearticle might be a challenge also for the producers ofnew instruments: intrinsic clipping due to high-frequency

motions near the source is to be better treated even if theinstrument is primarily developed for precise recording ofweak low-frequency ground motions.

Stations and Events

The West Bohemia region has been well-known for itsearthquake swarms (Neunhöfer and Meier, 2004). About20,000 events were instrumentally recorded there duringthe last crisis in October 2008. The activity is tightlycompacted in a fault segment of a 3 × 3 km size, thus provid-ing similar, systematically repeating recordings. A 13-stationshort-period permanent network, the WEBNET, monitors theregion and provides data for highly accurate locations andfocal mechanism studies (Horálek et al., 2000, 2008; Fischerand Horálek, 2003; Fischer and Michálek, 2008). The citedarticles provide all necessary details about regional tectonics,local site conditions, instrument installations, etc. One of thestations, NKC (50.2331° N, 12.4479° E), is situated almostabove the October 2008 swarm. This station, belonging tothe Czech Regional Seismic Network, is also equipped withtwo broadband instruments, a Streckeisen STS-2 and a GuralpCMG-40T. Some other broadband stations have been inoperation near the source region at epicentral distances upto 25 km, but none of them provided disturbances as largeand clear as those observed at NKC.

The 2008 data set can be briefly characterized as follows(T. Fischer, personal comm., 2009): nine of the 9–28 Octoberevents had local magnitudes between 3.0 and 3.8. Within anaccuracy of 1 km, sufficient for this article, the epicenterposition of all events was the same (50.21° N, 12.45° E). Theirdepths varied between 8 and 10 km; the epicentral distance ofthe NKC station was 2 km. Focal mechanisms were retrievedfrom amplitudes and polarities at the local stations by severalapproaches, providing strike 166° to 177°, dip 55° to 72°, andrake �19° to �47° (A. Boušková, personal comm., 2009).These ranges included both the intraevent and the intra-approach variations. Independently, our solution for twoevents obtained by waveform inversion of near-regional re-cords provided similar strike/dip/rake values; for example,the values of 169°=59°= � 41° were obtained for theMw 3.6 event of 28 October. From the practical point of view,we have a set of events with almost identical hypocenters andmechanisms, resembling a repeatable laboratory experiment.

Records

We visually inspected the records of all nine ML > 3

events and of many smaller ones (see the Data and Resourcessection). We found great similarity not only among theirwaveforms, but also among the disturbances. We demon-strate these effects by a representative example with a verygood signal-to-noise ratio: records of theMw 3.6 event of 28October 2008 at 08:30:11 UTC.

60 J. Zahradník and A. Plešinger

Data Processing

We used continuous and triggered STS-2 records sampledat 20 Hz and 100 Hz, respectively (corresponding antialiascorner frequencies fc � 8 and 40 Hz), and continuousCMG-40T records sampled at 250 Hz (fc � 100 Hz). Thethree-component raw files were transformed from counts tometers per second, but neither filtering nor instrumentalcorrection was performed in order to keep the whole broad-band spectra untouched. On the contrary, the instrumentalresponse was convolved with the calculated ground motionsto be compared with the records (see later discussion in theSynthetic Seismograms section). Such a forward approachhas been strongly preferred against deconvolving the instru-ment response from the records. As a next step, the velocityrecordswere integrated in the time domain. This so-called rawdisplacement is proportional to ground acceleration forfrequencies f < f0 and to ground displacement for frequen-cies f > f0, where f0 is the instrumental low-frequencycorner (1=120 Hz for STS-2, 1=30 Hz for CMG-40T).

Long-Period Disturbances

Even a brief look at a typical integrated STS-2 veloci-gram (raw displacement record) shows an impressive pattern(Fig. 1): Both horizontal components are dominated by avery large long-period disturbance. It can be characterizedas a smooth baseline change, or step, lasting about 120 sec.The amplitude of the step is much larger than the P- andS-wave signals on both the EW- and NS-components. Thevertical component (Z) has a smaller disturbance whoseamplitude is comparable to the seismic signal. Its shape(hereafter called U-shape) strongly differs from the horizon-tal components: it looks like the first derivative of the EWand NS disturbances. The collocated CMG-40T records havesmall disturbances, too (e.g., on the Z-component in Fig. 1b);however, none of them is as large as the huge steps on thehorizontal components of the STS-2.

Ramps

The other important feature of the raw displacementrecords is the steady amplitude increase between the P- andS-wave onsets (see Fig. 2). This is a typical ramp due to thenear-field source effect (e.g., equation 4.29 of Aki and Ri-chards, 1980). Lasting roughly 1 sec, it has nothing to dowiththe considerably shorter source duration (of about 0.2 sec). AtNKC station the ramp can be easily identified on all threecomponents (on both STS-2 and CMG-40T records). Whybe concerned about the ramp? First, note that the ramp ampli-tude is by far not negligible with respect to the P- andS-amplitudes. It represents a first-order effect of a broad spec-tral content. As such, the ramp brings an important piece ofinformation about the focal mechanism, independent of theP- and S-wave amplitudes, because the correspondingnear-field term has its own directional dependence. It canbe used to check an assumed focal mechanism or facilitate

its retrieval from waveforms if only few near-source stationsare available for weak events. Second, a clear ramp in theintegrated record indicates that the ground motion has a static(permanent) displacement of comparable size, theoretically ofinfinite duration after the S-wave transient signal. Usually thestatic displacement cannot be readily seen. This is because it isa low-frequency effect; at f < f0 the raw displacement recordis proportional to ground acceleration, not to ground dis-placement. Later we explain where the static displacement isencoded in the original records.

Peak Values

With 20 Hz sampling we get underestimated peakvalues. On the other hand, differentiating the triggeredSTS-2 (100 Hz sampling) and the continuous CMG-40Tvelocigrams (250 Hz sampling), we arrive at almost identicalpeak accelerations of about 2 × 10�1 m=sec2 in the S-wavegroup of the largest observed events, without any visible clip.The peak velocities reached 3 × 10�3 to 5 × 10�3 m=sec.Such velocities are often recorded during stronger distantevents, without any long-period disturbance; their accelera-tion is, however, much smaller. As an interesting example letus mention the 26 December 2004, Mw 9 Sumatra earth-quake, recorded at NKC with the same peak velocity asthe studied M 3 to 3.8 events, while the acceleration wastwo orders of magnitude smaller than in the local events.This is a typical feature of the disturbances, already dis-cussed in relation with similar events in Greece (Zahradníkand Plešinger, 2005): the high-frequency content of theground motion is a prerequisite for this phenomenon.

Methods of Analysis of the Records

Two tools have been used to analyze the records: (1) for-ward simulation of a step response, and (2) synthetic seis-mograms.

Step Response

By step response we understand the integrated output ofthe broadband velocigraph with a simple steplike input. Wemay consider a step in input acceleration, velocity, or dis-placement (Fig. 3a). Note equivalent cases: for example, thevelocity step is equivalent to the pulselike acceleration andalso equivalent to an infinitely linearly growing displace-ment. The method is straightforward: knowing the poles andzeros of the transfer function of an instrument, we calculateits response to the unit-amplitude input step (Fig. 3b). Then,by trial and error, or by the least-squares method, the long-period trend of the record is fitted by manipulating just twoparameters: the onset time and the amplitude of the step.

Synthetic Seismograms

This is a tool enabling forward simulation of the records,including ground motion and instrumental response. We use

Toward Understanding Subtle Instrumentation Effects Associated with Weak Seismic Events 61

Figure 1. Long-period disturbances at NKC station for a Mw 3.6 event (28 October 2008 at 08:30:11 UTC). Shown is the integratedvelocity output (raw displacement). (a) STS-2 record. Note the U-shaped disturbance on the vertical component, contrasting with thepredominating steplike disturbance on the horizontal components. Seismic signals need a zoomed view (Fig. 2). (b) CMG-40T record fromthe same pier; the only clear disturbance is on the vertical component. This is a typical pattern of all inspected events.


the discrete wavenumber method (code AXITRA of Bouchon,1981 and Coutant, 1989; and code ISOLA of Sokos and Zah-radnik, 2008), providing complete wavefield, including allnear-field terms of the Green function. A local 1D layered

crustal model is adopted (Novotný, 1996). A point-sourcemodel is used because with a single near station a finite-extent model is not available and the directivity effect is notinvestigated. The source is described by its hypocenter posi-tion (50.21° N, 12.45° E , depth 8 km) and focal mechanism(strike=dip=rake � 169°=59°= � 41°). The slip rate is mod-eled by a triangle of a duration of 0.2 sec. The ground motionis calculated up to 5 Hz. This frequency range is narrowerthan in real data, but it is sufficiently broad for qualitativemodeling of the near-field ramp between the P and S phasesand for the static displacement (Fig. 4). The instruments aredescribed by the poles and zeros of their transfer function,and the synthetic ground motion is convolved with it. Ifthe interest is only in the static ground-motion displacementand its spatial variability, it is not necessary to calculateseismograms. Then we use the static solution for a crack inhalfspace (Okada, 1992), code Coulomb 3.1 (Toda et al.,2005). Horizontal gradient of the static displacement on ver-tical components provides a model estimate of the ground-motion tilt.

Results of Analysis of the Records

Tiltlike Disturbances on Horizontal Components

It was demonstrated in Figure 1 that the main distur-bances on the integrated horizontal components of STS-2have the form of a smooth steplike baseline change. Thus,the first approximation can be made in terms of the instru-ment response to a step in input acceleration. As seenfrom Figure 5a, the EW and NS components can be well

Figure 2. Detail of the same records as shown in Figure 1,STS-2 and CMG-40T superimposed. The near-field ramp betweenthe P- and S-wave onsets is well developed on all three components(time 2 to 3 sec). Note the almost identical records from both in-struments until the onset of the disturbances in the S-wave group.

0 40 80 120 160 200

time (seconds)

-100

0

100

200

300

400

0 40 80 120 160 200time (seconds)

-2

0

2

4

6

8

10

raw

dis

pla

cem

ent.

(mm

)

0 40 80 120 160 200

Time (s)

-0.4

0

0.4

0.8

1.2

(1)

(2)

(3)

acceleration velocity displacement

(1)

(2)

(3)

(a) (b)

Figure 3. Input step and instrument response. (a) Unit step in input acceleration (1), velocity (2), or displacement (3). (b) Response of twoinstruments, STS-2 (solid line) and CMG-40T (dotted line) for the steps (1)–(3); shown is the integrated velocity output, also called rawdisplacement, or mass channel.


matched by the acceleration steps of 2:5 × 10�6 and �1:6×10�6 m=sec2, respectively.

In principle, an acceleration step of amplitude A on ahorizontal component may be related with a step in tilt,Θ � A=g, where g is the acceleration of gravity (Wielandtand Forbriger, 1999). Thus, the observed steps of the orderof A ∼ 1 × 10�6 m=sec2 would correspond to a tilt of theorder of Θ ∼ 1 × 10�7 radians. The ground-motion tilt canbe estimated from the calculated synthetic motions near theNKC station. They provided the vertical static displacementof about �2:0 × 10�5 m, with a horizontal gradient around0:6 × 10�8 radians. This is an order of magnitude smallerthan the Θ value derived from the recorded disturbances. Itindicates that, partially or completely, the tiltlike disturbanceson STS-2 records are an instrumental artifact.

A confirmation comes from the collocated CMG-40Tinstrument. The comparison of STS with CMG is legitimate,thanks to an experiment in which a technician stood on the

pier and generated a tiltlike response on both STS andCMG sensors, roughly proportional to their accelerationsensitivities. Nevertheless, as proved in Figure 5b, the tiltmeasured by CMG-40T was three to five times smaller thanon the STS-2, if any. It corresponds to the acceleration stepsof �0:5 × 10�6, �0:5 × 10�6 m=sec2 for the EW and NScomponents, respectively, still larger than the theoreticalestimate, but these disturbances are already close to thelong-period noise level.

U-Shaped Disturbances on Vertical Components

On the Z component of the STS-2 we observe a distur-bance whose shape is the first derivative of the tiltlike distur-bance (Fig. 1). It can be matched with the STS-2 response to astep in velocity of �0:45 × 10�5 m=sec, if combined with adisplacement step of �2:0 × 10�5 m (Fig. 5a). Such a staticdisplacement is in good agreement with the computationalestimate previously discussed. On the other hand, the velocitystep must be an instrumental artifact. The vertical componentof the CMG-40T record yields similar estimates: �0:6×10�5 m=sec and �2:0 × 10�5 m.

U-Shaped Disturbances on Horizontal Components

A visual inspection of the EW component of the STS-2record in a 20-sec window after the seismic signal (Fig. 5a)suggests a mixture of the effects dominating on NS and Z.Indeed, a good approximation is obtained with superpositionof the acceleration step of 2:5 × 10�6 m=sec2 and the veloc-ity step of �3:0 × 10�5 m=sec. The latter velocity step is sixtimes larger than on the Z component; it is perhaps surpris-ing, not obvious from visual inspection, because the accel-eration step is dominant. Other events show velocity stepresponses on all three components. No disturbances of thiskind were identified above the noise level on horizontal com-ponents of the CMG-40T records.

Because the STS-2 is a symmetrical triaxial system con-sisting of three identical inclined pendulums yielding outputsin the U, V, W coordinate system (Melton and Kirkpatrick,1970), we also applied the inverse rotation to get from therecorded (recombined) conventional E, N, and Z compo-nents back to the actual U, V, and W channels. All of themrevealed a mixture of tilt-type (acceleration step) as well asU-type (velocity step) responses; no singular behavior of anyof the three pendulums has been found.

A Partial Conclusion

A partial conclusion is that the disturbances on the hor-izontal components of the CMG-40Tmay relate to true tilt, buta poor signal-to-noise ratio makes this conclusion uncertain.Unfortunately, no events were found in the analyzed data setto make a more specific conclusion of this kind. On the otherhand, quite certainly the STS-2 experienced a much largertilting, definitely not due to the earthquake ground motion.We can only speculate about possible reasons; for example,

50 51 52 53 54 55-1.0E-004

-8.0E-005

-6.0E-005

-4.0E-005

-2.0E-005

0.0E+000

2.0E-005D

isp

lace

men

t (m

)

50 51 52 53 54 55Time (s)

Time (s)

Time (s)

-1.0E-004

0.0E+000

1.0E-004

2.0E-004

3.0E-004

Dis

pla

cem

ent

(m)

50 51 52 53 54 55-1.2E-004

-8.0E-005

-4.0E-005

0.0E+000

4.0E-005

Dis

pla

cem

ent

(m)

Z

N

E

static displacement

P S

Figure 4. Synthetic ground-motion displacement for the NKCstation, epicentral distance 2 km, and a typical Mw 3.6 event at thedepth of 8 km. The instrument response is not considered. Note thenear-field ramp between the P- and S-wave group. The strike, dip,and rake angles were varied within their uncertainty ranges to showthat the static displacement (highlighted on Z) remains very stable.


an additional local tilt might have been provoked by the high-frequency ground vibrations in the instrument feet, or perhapseven inside the case; all this remains hitherto unclear. The bigadvantage is the collocation of the two instruments; the absentevidence of a large tilting in the CMG-40T records excludesmany possible speculations about the large disturbances in thehorizontal components of the STS-2. For example, there is noreason for the speculation about a tilt due to local effects pro-voked by the high-frequency vibrations in a small underlyingblock of rock, such as we made in our study in Greece (Zah-radník and Plešinger, 2005). There is also no need to speculatein western Bohemia about large tilting due to undergroundfluid motion (such as magma motion studied by Gambinoet al., 2007), or due to liquefaction of the sand around theseismometer base (Kinoshita, 2008). The U-shaped distur-

bances are new. They are most clear on the vertical compo-nents of the STS-2 and CMG-40T, where they are notmasked by tilt, but generally they may exist on all threecomponents. Being equivalent to the instrument responseto a step in velocity (pulse in acceleration), for which we haveno theoretical explanation in terms of the source effect, theU-shaped disturbances are understood as a pure artifact,calling for an explanation.

Physical Model of the U-Shaped Disturbance

Seismograph: Dynamic System with Nonlinearity

A physical model potentially producing long-period dis-turbances is the seismograph understood as a time-invariantdynamic system with internal nonlinearities. We used

Figure 5. (a) Forward modeling, STS-2, and (b) forward modeling, CMG-40T, are the same records as in Figure 1 (solid line), withdisturbances formally matched by various step responses detailed in the text (gray line). (c) Deconvolution, CMG-40T, Z component. Leftpanel is deconvolution of the raw-displacement to the input velocity, filtered by a 4-pole 1 Hz low-pass filter; a spurious permanent step isrevealed. Middle panel is the raw-displacement deconvolved to input displacement; static displacement is masked by the ramp due to thespurious velocity step. Right panel is the same as in the middle, but after removal of the ramp; now the static displacement is revealed.


MATLAB Simulink (2002) for modeling such a system.Seeking for a plausible nonlinear behavior we assume thatsaturation in the forward path is a possible mechanism.To validate this assumption we analyze possible saturationeffects in a simplified model of a 30-sec force-feedback velo-cimeter with a 1-sec inertial sensor, proportional-integral-derivative (PID) feedback, and an upper cut-off frequencyof 100 Hz. The instrument setup is schematically shownin Figure 6. The setup could be easily rescaled to any otheranalogical system, for example, a 120-sec to 80 Hz PID-feedback velocimeter with a 6-sec inertial sensor, the resultswould qualitatively remain the same.

To demonstrate how a PID-feedback velocimeter reactsto impulsive near-field ground motions comprising static dis-placement and a tilt in situations without and with saturation,we show in Figure 7 the respective responses to the syntheticseismogram for the N-component of the 28 October 2008,Mw 3.6 event. The synthetic ground displacement signal(Fig. 7a) produces at the output of the amplifier of the mass-to-ground transducer output a high-frequency signal (Fig. 7b)with peak amplitudes �2:1 and 1.7 (dimensionless). No sat-uration occurs, and the long-period component in the inte-grated velocigram (raw displacement record, output of themass position channel) (Fig. 7d) represents the natural, linearresponse of the system to the static displacement in the syn-thetic seismogram. Figure 7c shows the output of the ampli-fier in the case of symmetric saturation on a level of �1:5.The asymmetric signal peaks exceeding this level are cut off,and the system reacts to this subtle internal asymmetry verysensitively by producing a spurious long-period, U-shapeddisturbance (Fig. 7e). The natural system response to thestatic ground displacement is masked by this disturbance.

In most situations typical for observations near the seismicsource it is impossible to reveal the clipped peaks responsiblefor this disturbance in the original broadband velocigram. If atilt is added to the input, regardless whether of natural orinstrumental origin, its effect is also invisible in the velocityoutput but becomes apparent in the raw displacement record(Fig. 7f), as a long-period disturbance with permanent offsetgradually predominating the saturation disturbance.

In very special situations, saturation does not necessarilyyield the effect demonstrated in Figure 7e. This is the case ofperfectly symmetric saturation of perfectly symmetric inputsignals, when spurious pulses of equal amplitude in bothpolarities appear and their effects mutually cancel each other.Therefore, when speaking about disturbances due to satura-tion, we implicitly mean asymmetric saturation. Asymmetrycaused by the signal, of course, not asymmetrical saturationof some active element in the system, because this is veryunlikely if not entirely impossible.

Simulation of saturation in the feedback loop has shownthat its consequence is qualitatively quite the same as that ofsaturation in the forward path, but its effect is much stronger.The explanation is simple. Denote the frequency response ofthe feedback and forward path by β�f� and K�f�, respec-tively. The overall frequency response of the closed-loop sys-tem is S�f� � K�f�=�1� β�f�K�f��. In the frequency bandin which jβ�f�K�f�j ≫ 1, we have jS�f�j ∼ 1=jβ�f�j;hence, the transfer properties are governed by the feedback.Up to seismic frequencies for which jβ�f�K�f�j ≫ 1, anynonlinearities in the forward path (for instance, those of themechanical inertial sensor) are efficiently suppressed, butany nonlinearities in the feedback manifest themselves undi-minished. In reality, nonlinearity in the feedback is unlikely

Figure 6. Schematic diagram of a 0.033–100 Hz force-feedback velocimeter with PID feedback. The feedback path is considered to belinear. In the forward path, the largest high-frequency signals appear on the output of d.c. amplifier A; saturation is therefore modeled in thispoint. The setup allows recording of the individual outputs for any sort of theoretical, synthetic, and actual input ground motions. Tilt issimulated by a parabolic ramp of input displacement.


because PID feedback is implemented, except the integratingoperational amplifier, exclusively by passive, linear elements(capacitors, resistors). We perhaps could consider possiblesubtle nonlinearities in the magnet-coil assembly of the

force-feedback transducer at high frequencies, but that wouldbe pure speculations. Even if there were any, we have noknowledge about their nature nor do we know how to quan-tify them.

Figure 7. Responses (integrated velocigrams) of the system simulated in Figure 6 to a near-field ground motion. (a) Input ground dis-placement (N-component of the synthetics shown in Fig. 4); (b) amplifier output, no clipping; (c) same as (b), amplifier output symmetricallyclipped; (d) natural (linear) system response to ground motion comprising a static displacement, no clipping; (e) same as (d), slight instan-taneous symmetric clipping according to (c), producing a U-shaped disturbance, such as that in Figure 1a; (f) combined effect of symmetricsaturation and a tilt.


Although producing huge long-period disturbances, theinstantaneous saturation itself is a tiny effect that cannot berecognized in the output as a series of cut peaks, a patterncommonly known in the observatory practice as clipping.Note also that the saturation we propose for explanationof the observed records is not simply related to the standardclip level (such as routinely estimated from the frequencydependence of the instrumental sensitivity to displacement,velocity and acceleration), because the peak motions of thedisturbed records are well below such a level. In fact thevelocity clip level, VCL, of a feedback velocimeter is not afrequency independent parameter but decreases with the firstpower of frequency beginning at some threshold frequency,fCL. For example, for the 40 Vpp version of a KS-2000,the standard VCL � 10 mm=sec, fCL ∼ 4 Hz, and for f �40 Hz the clip level is 10 times lower, that is, 1 mm=sec.According to the technical specifications of a standardSTS-2 and CMG-40T, fCL ∼ 20 Hz and VCL ∼ 13 mm=secfor both instruments, and the clip level drops to about5 mm=sec at 50 Hz. These values are not much higher thanour observed peak velocities. It is also quite possible thatparasitic resonances in the inertial sensor at very high fre-quencies, at which jβ�f�K�f�j approaches 1 and the transferproperties of the system are no more governed by the feed-back, are the cause of saturation at even lower levels than thepresumed high-frequency (f > fCL) clip level. The complexnature of intrinsic saturation cannot be better investigated onthis phenomenological level.

To summarize the section on simulation, the physicalmodel has one main advantage: compared with the formalmodeling in the preceding sections, where an additional spu-rious pulse (of unknown nature) had to be added to the groundmotion, here the input motion remains completely realistic.We just allow for an internal nonlinearity (instantaneoussaturation) at high frequencies (f > ∼10 Hz); the resultingsignal asymmetry introduces a small spurious d.c. componentinto the system. The response of the system to this d.c. com-ponent appears in the integrated velocigram (raw displace-ment record) as a U-shaped disturbance which heavilydestroys the very low (f < 0:01 Hz) seismic frequencies.

Any Other Effect?

To explain the U-shaped disturbance in the integratedoutput, we need an acceleration pulse (velocity step) inthe input. We provided a possible explanation of such a pulseas due to intrinsic instrumental saturation at high frequencies.Is there any other physical effect that could act on the seis-mograph in the same way as an acceleration pulse? Consid-ering the horizontal components, equation (3) of Pillet andVirieux (2007) seems to suggest two possible cases: (1) atransient tilt, and (2) transient rotation (the second timederivative of the rotation angle with respect to the verticalaxis). The transient tilt may have similar time variation asground velocity, and there are two possible pulselike velocitytransients: those due to the near- and intermediate-field terms

of the Green function. By the near-field pulselike velocitytransient, we mean the trapezoidal velocity ramp. Simplepoint-source calculations show that (similarly to the perma-nent tilt) this transient tilt is too small, unable to explain theobserved U-shape disturbances. By the intermediate-fieldpulselike velocity transient, we mean pulses at the P- andS-wave arrival, with the duration comparable to the sourceduration or smaller. We cannot rule out relatively largeamplitudes of such tilt transients, but modeling the high-frequency tilt (and rotation) goes already beyond the scopeof this article. Moreover, the fundamental problem with suchan interpretation would be the systematic occurrence of theU-shaped disturbance on the vertical component. Its expla-nation through the transient tilt would imply a nonverticalinstallation (equation 6 of Pillet and Virieux, 2007), whichhad not been clearly indicated in our observations.

Removing the U-Shaped Disturbance, and Retrievingthe Permanent Displacement

Put aside for a while any possible cause of the U-shapeddisturbances, try to remove them from the record, and askwhether the decontaminated record is compatible with thesynthetic ground motion of Figure 4 as regards the near-fieldramp and the static displacement. For brevity, we concentrateon the (simpler) vertical component.

Such an experiment is demonstrated in Figure 8. Basedon the preceding analysis, we know that the vertical compo-nent of the STS-2 integrated record is disturbed by the veloc-ity step �0:45 × 10�5 m=sec. We calculate the instrumentresponse to such a step and subtract it from the record; thisis what we call the decontaminated record. At the same time,we use the synthetic ground motion of Figure 4, convolve itwith the STS-2 response, and compare it with the decon-taminated record. As a result (Fig. 8) the near-field rampis matched fairly well, thus validating the assumed focalmechanism. The following part of the synthetics (after Swave) is much simpler than the true record, due to the simplestructural model, but the long-lasting trend of the record iswell approximated, too.

This figure brings an important message to practicalusers of the broadband records: Although the disturbancehas been removed, and the decontaminated record is wellfit by the synthetics, it has a strange, bow-shaped form. Sucha form is nothing abnormal. It is simply the instrumentresponse to the near-field ramp and the static displacement,similar to the response to a displacement step (Fig. 3).A careful look at the record also shows that its bow-shapingstarts at the S-wave arrival (after the ramp), at the level of thetheoretically predicted static displacement, �2:0 × 10�5 m.This is easy to explain: Onset of the static displacementis relatively abrupt, representing a high-frequency dis-placement feature (f ≫ fo), thus it is reproduced in theoutput. On the contrary, the following constancy of thestatic displacement, representing a low-frequency displace-ment feature, is naturally distorted by the instrument quite


considerably. This example illustrates that some apparentlystrange records, such as the bow-shaped record previouslydescribed, might be completely free of any disturbance, anddo provide information about the static displacement.

The reason why we were so much interested in the staticdisplacement, and appreciated its agreement with the theo-retical source model, is simple. It validated usability of thesource model to estimate realistic values of the tilt in the pre-ceding sections.

Finally, we have to answer the legitimate question whythe forward simulation has been preferred against deconvo-

lution of the instrument effect. The answer should be givenseparately for the studied steps in acceleration, velocity, anddisplacement. The simplest task is to recognize presence andsize of acceleration steps; the integrated output of the broad-band instrument (raw displacement), if properly rescaled toacceleration, shows them directly. This is what we can see ata first glance in Figure 1a; the scale shown to the right of thatplot enables direct reading of the size of the step, because atvery low frequencies the raw displacement is proportional toinput acceleration. In other words, just the integration of thebroadband output and multiplication by a constant factor issufficient to get the equivalent permanent input acceleration.Regarding the permanent velocity and permanent displace-ment input, the broadband output can be deconvolved usingthe frequency response of the instrument, but success of thisinverse problem strongly depends on the signal-to-noise ratioand presence of disturbances. For instance, the vertical com-ponent of the CMG-40T record shown in Figure 1b couldsuccessfully be deconvolved to the input velocity (see theleft panel in Fig. 5c): the deconvolution revealed the samesize of the velocity step as inferred by forward modeling(�0:6 × 10�5 m=sec). However, in the record deconvolvedto input displacement (middle panel in Fig. 5c), the coseis-mic static displacement was hidden in the superimposedramp corresponding to the spurious velocity step. After cau-tious retrieval of the onset time of the ramp and its removal(1.22 sec after the P-onset; this time exactly corresponded tothe maximum amplitude in the original broadband veloci-gram), the deconvolution provided a permanent displace-ment of the same order as in our forward modeling (seethe right panel in Fig. 5c). In this sense forward modelingand deconvolution have thus been equivalent approaches.However, in any case we have to distinguish carefully be-tween superimposed effects (spurious steps and the naturalstatic displacement) and to interpret them correctly. In situa-tions with lower signal-to-noise ratios, deconvolution tendsto instabilities so that forward modeling uses to be the pref-erable choice.

Examples from Other Earthquakes

Disturbances such as those discussed in this article canbe found in many earthquake records from various broad-band feedback velocimeters. As a rule, to make them visibleduring routine inspection of the records, it suffices to inte-grate the original velocigrams, and/or to low-pass filter them.Here we present two examples from the Japan F-net network(see the Data and Resources section).

Japan, 11 March 2002

At 06:54 UTC on 11 March 2002 (Fig. 9a), a shallowcrustal event of Mw 3.9 (34.1° N, 134.5° E) was recorded atISI station (34.06° N, 134.45° E) by an STS-1 sensor at anepicentral distance of about 1 km. This was a case where thetiltlike disturbance dominated the integrated record, making

40 50 60 70 80Time (s)

-8E-005

-6E-005

-4E-005

-2E-005

0E+000

2E-005

Raw

dis

pla

cem

ent

(m)

Z

STS-2

40 50 60 70 80Time (s)

-8E-005

-6E-005

-4E-005

-2E-005

0E+000

2E-005

Raw

dis

pla

cem

ent

(m)

ZCMG-40T

(a)

(b)

Figure 8. (a) STS-2, (b) CMG-40T. Demonstration of a bow-shaped record (solid line) obtained by removing the U-shapeddisturbance from the vertical component of the studied earthquake(cf. Fig. 1). It is well fitted by the synthetic record (gray). Theunusual, bow-shaped form of the record comes from static dis-placement, seen in the synthetic ground motion (dotted), and fromthe instrument response.


the seismic signal almost invisible. The duration of the dis-turbance was consistent with the corner period of the STS-1instrument (360 sec). A puzzling feature was the presenceof a similar disturbance on the vertical component. Similarcases elsewhere (Pillet and Virieux, 2007) were interpretedas due to a nonvertical installation.

Japan, 14 May 2006

At 16:42 UTC on 14 May 2006 (Fig. 9b), a shallowcrustal event of Mw 4.3 (34.2° N, 135.2° E), was recordedat NOK station (34.16° N, 135.34° E) by an STS-2 sensorat an epicentral distance of about 13 km. The integrated re-cord looked undisturbed; however, application of the 0.01 Hzlow-pass filter revealed a clear U-shaped disturbance on thevertical component and a superposition of an U-shaped andtiltlike disturbance on both horizontal components.

We have no examples of these effects from strongerand/or more distant events (e.g., M6 events at near-regionaldistances of the order of ∼100 km); these events rather pro-duce obvious clipping at small distances, or the records atmore distant stations are free of clip, but the high-frequencywaves are attenuated, so the records are free of the distur-bances. In this sense, the range of distances (∼ < 10 km) andmagnitudes (<4) typical for these phenomena is quitenarrow. There might be exceptions from this simple rule, sothe records should always be analyzed with caution.

We also have no data to systematically investigate pos-sible effects of the local geological conditions. It seems that,ironically, rock sites might rather support the disturbancesthan to suppress them, because the high-frequency groundmotions propagate on rock sites with less attenuation thanon soft sites.

Discussion and Conclusion

Near-fault observations have to be interpreted with greatcaution. High-frequency ground motion in the proximity ofthe source may create instrumental artifacts. For example,they include permanent (static) tilts exceeding by an orderof magnitude the values estimated by the source models. Thisarticle brings new examples of this phenomenon fromM 3 to3.8 events recorded at an epicentral distance of 2 km by a120-sec STS-2 sensor. The collocated 30-sec CMG-40Tsensor measured a considerably smaller tilt, if any. A newtype of disturbance, so far not well recognized, has beenstudied. It is a long-period disturbance corresponding to aspurious step in input velocity, equivalent to a pulse in accel-eration. This disturbance has been found in the integratedbroadband velocigrams of more than nine ML > 3 events,with almost identical U-shape. In general it has been presenton all three components of both the STS-2 and CMG-40Trecords. Theoretically, there is no room for such an effect inthe translational earthquake ground motions. A likely expla-nation of this phenomenon in terms of an instrumental arti-fact is instantaneous saturation of high-frequency signals in

the feedback instruments. Such a saturation produces narrow,one-sided (nonzero mean value) pulses to which the feedbacksystem responds like a spurious step in ground velocity(pulse of ground acceleration). In the integrated velocigram(raw displacement record), this response manifests itself in ashape called U-disturbance for brevity in this article. In gen-eral, any asymmetry in the transfer of high-frequency signalsin the force-feedback system, possibly conjoined with intrin-sic parasitic resonances, may produce a U-type disturbance.

Once we correctly recognize the type of disturbance,we can formally simulate it independently of the unknowndetails of internal processes in the seismograph. This taskis relatively easy because all disturbances we faced were for-mally equivalent to a very simple perturbation of the input,just steps in acceleration or velocity. Therefore, the standardfrequency response represented by poles and zeros enablesstraightforward modeling of the onset time and step size ofthe disturbance, by trial and error, or the least-squaresmethod. In general, the step size varies among events, as wellas from one component to the other, which is why theprocedure should be applied independently to all three com-ponents of each event. The simulated disturbance can be sub-tracted from the record. Quality of the decontaminated recordobviously depends on its signal-to-noise ratio.

An important practical message for the analysts is thateven instrumentally uncorrected records free of any distur-bance may have an apparently strange, bow-shaped form.Just these records bring important information about near-field effects, including the static displacement (of the orderof 1 × 10�5 m in this article). The bow-shaped form of therecord comes from the instrument response to the ramp and(infinitely long-lasting) static displacement; as an essentiallyzero-frequency effect, it cannot be directly seen on the inte-grated broadband output, because the output is proportionalto displacement only above the instrumental low-frequencycorner.

The long-period disturbances seem to be a generalphenomenon accompanying almost any type of broadbandrecordings close to the source of weak events. During thisand previous work we encountered it on CMG-3T, STS-1,STS-2, and KS-2000M. Each record might have a differentappearance due to the instrument pass-band and due to acombination of various types of disturbances. It is a matterof training to properly recognize and simulate records suchas this, in particular to properly distinguish between near-field effects and instrumental artifacts. Collocated instru-ments of different types are very helpful for this task.

Removal of the disturbances is absolutely necessarybefore any use of the records in source studies using frequen-cies of the order of the instrumental corner frequency orlower. That is why future work should concentrate on iden-tification and removal of the disturbances in the frame ofautomated procedures of the data acquisition and qualitycontrol, with special importance for real-time determinationof the source parameters. It is clear that routine removal ofthe disturbances for applications not particularly interested in


Figure 9. Examples of long-period disturbances at other stations and instruments. (a) Integrated STS-1 velocigram of an Mw 3.9 event(Japan, ISI station). (b) Low-pass filtered integrated STS-2 velocigram of an Mw 4.3 event (Japan, NOK station).


source details may have an even simpler form than wediscussed in this article: just denoising the record by subtrac-tion of the whole time-varying long-period trend (withoutclassifying the individual disturbance types); then, however,subtle source effects, such as permanent displacement, willbe lost.

It is very well-known that the STS-2 is a highly sensitiveinstrument useful for recording small near or distant events atstation locations with very quiet background noise condi-tions (McNamara et al., 2005). Does the presence of the dis-turbances mean that this instrument is not a proper choice fornear events with magnitudes up to Mw 3.8? First of all, weemphasize that none of the analyzed records had been satu-rated in the standard sense, that is, in the form of a series ofequal-sized cut peaks directly visible in the broadbandrecord. Second, if we want to recognize and study the near-field effects of small earthquakes, we need a broadbandsensor with very low intrinsic noise at low frequencies. Allthis means that an instrument such as the STS-2 (or CMG-3T,Trillium 120, etc.) is not a bad choice; or, better speaking, itis the only possibility. The problem is, in fact, that contem-porary instruments highly sensitive at low frequenciesapparently have not been constructed in a way to avoidinstantaneous internal electronic clipping due to groundmotions including very high frequencies. And, ironically,just these instantaneous high-frequency saturations may re-sult in distortion of the low-frequency content of the records,as shown in the present article (U-shaped disturbances).Maybe a new generation of broadband sensors will be freeof these problems.

Data and Resources

The STS-2 seismograms from NKC station of the CzechRegional Seismic Network are available from the web pageof the Institute of Geophysics, Academy of Sciences of theCzech Republic, http://www.ig.cas.cz/en/seismic‑service/waveform‑request/. The CMG-40T seismograms from thesame station were obtained from the WEBNET group belong-ing to the same institute. Data from Japan, stations ISI andNOK, were downloaded from the F-net (NIED) web pagewww.hinet.bosai.go.jp/fnet/. Focal mechanism of the mainstudied event was calculated by waveform inversion withsoftware ISOLA (http://seismo.geology.upatras.gr/isola/).Details of the solution, agency code UPSL, can be foundin the Moment Tensors section of the European-Mediterra-nean Seismological Center (http://www.emsc-csem.org/).All web pages referenced here were last accessed Octo-ber 2009.

Acknowledgments

Our thanks go first of all to our colleagues A. Bouskova, T. Fischer,J. Horalek, P. Jedlicka, J. Michalek, and J. Zednik, Institute of Geophysics,Academy of Sciences of the Czech Republic, Prague, for providing us theunique observational material for this article. We are obliged to E. Wielandt,Stuttgart, and F. Gallovic, J. Jansky, and I. Oprsal, Prague, for incentive

discussions. The examples from other stations were provided by J. Burjanekand E. Sokos. The work has been performed as part of projectsIAA300120911 and GACR 205/07/0502 (both supported by the GrantAgency of the Czech Republic) and MSM 0021620860 (supported bythe Ministry of Education, Youth and Sports, Czech Republic).

References

Aki, K., and P. G. Richards (1980). Quantitative Seismology, 1,W. H. Freeman, San Francisco.

Boore, D. M., C. D. Stephens, and W. B. Joyner (2002). Comments on base-line correction of digital strong-motion data: Examples from the 1999Hector Mine, California, earthquake, Bull. Seismol. Soc. Am. 92,1543–1560.

Boroschek, R. L., and D. Legrand (2006). Tilt motion effects on the double-time integration of linear accelerometers: An experimental approach,Bull. Seismol. Soc. Am. 96, 2072–2089.

Bouchon, M. (1981). A simple method to calculate Green’s functions forelastic layered media, Bull. Seismol. Soc. Am. 71, 959–971.

Bouchon, M., and K. Aki (1982). Strain, tilt, and rotation associated withstrong ground motion in the vicinity of earthquake faults, Bull.Seismol. Soc. Am. 72, 1717–1738.

Coutant, O. (1989). Program of numerical simulation AXITRA, ResearchReport LGIT, Grenoble.

Delorey, A. A., J. Vidale, J. Steim, and P. Bodin (2008). Broadband sensornonlinearity during moderate shaking, Bull. Seismol. Soc. Am. 98,1595–1601.

Fischer, T., and J. Horálek (2003). Space-time distribution of earthquakeswarms in the principal focal zone of the NW Bohemia/Vogtlandseismoactive region: Period 1985–2001, J. Geodynamics 35,125–144.

Fischer, T., and J. Michálek (2008). Post 2000-swarm microearthquakeactivity in the principal focal zone of West Bohemia/Vogtland:Space-time distribution and waveform similarity analysis, Stud.Geophys. Geod. 52, 493–511.

Gambino, S., O. Campisi, G. Falzone, A. Ferro, F. Guglielmino, G. Laudani,and B. Saraceno (2007). Tilt measurements at Vulcano Island, Ann.Geophys. 50, 233–247.

Graizer, V. M. (2005). Effect of tilt on strong motion data processing, SoilDynam. Earthquake Eng. 25, 197–204.

Horálek, J., T. Fischer, A. Boušková, and P. Jedlička (2000). The WesternBohemia/Vogtland region in the light of the WEBNET network, Stud.Geophys. Geod. 44, 107–125.

Horálek, J., T. Fischer, and M. Korn (Editors) (2008). Geodynamics ofearthquake swarm areas, Special Issue of Stud. Geophys. Geod. 52,455–672.

Kalkan, E., and V. Graizer (2007). Coupled tilt and translational groundmotion response spectra, J. Struct. Eng. 133, 609–619.

Kinoshita, S. (2008). Tilt measurement using broadband velocity seismo-grams, Bull. Seismol. Soc. Am. 98, 1887–1897.

MATLAB Simulink (2002). Tool for modeling, analyzing, and simulatingphysical and mathematical systems, The MathWorks, Inc., Natick,Massuchusetts.

McNamara, D. E., R. P. Buland, and H. M. Benz (2005). An assessment ofthe high-gain STS2 seismometer for earthquake monitoring in theUnited States, USGS Open-File Report, No. 2005-1437, 25 pp.

Melton, B. S., and B. M. Kirkpatrick (1970). The symmetrical triaxialseismometer—Its design for application to long-period seismometry,Bull. Seismol. Soc. Am. 60, 717–739.

Neunhöfer, H., and T. Meier (2004). Seismiciy in the Vogtland/WesternBohemia earthquake region between 1962 and 1998, Stud. Geophys.Geod. 48, 539–562.

Novotný, O. (1996). A preliminary seismic model for the region of the west-Bohemian earthquake swarms, Stud. Geophys. Geod. 40, 353–366.

Okada, Y. (1992). Internal deformation due to shear and tensile faults in ahalf-space, Bull. Seismol. Soc. Amer. 82, 1018–1040.


http://www.ig.cas.cz/en/seismic-service/waveform-request/





http://www.hinet.bosai.go.jp/fnet/

http://seismo.geology.upatras.gr/isola/

http://www.emsc-csem.org/

Pillet, R., and J. Virieux (2007). The effects of seismic rotations on inertialsensors, Geophys. J. Int. 171, 1314–1323.

Sokos, E., and J. Zahradník (2008). ISOLA a Fortran code and a Matlab GUIto perform multiple-point source inversion of seismic data, Computersand Geosciences 34, 967–977.

Toda, S., R. S. Stein, K. Richards-Dinger, and S. Bozkurt (2005). Forecast-ing the evolution of seismicity in southern California: Animations builton earthquake stress transfer, J. Geophys. Res. B05S16, doi 10.1029/2004JB003415.

Wielandt, E., and T. Forbriger (1999). Near-field seismic displacement andtilt associated with the explosive activity of Stromboli, Ann. di Geofis.42, 407–416.

Wu, Y.-M., and C.-F. Wu (2007). Approximate recovery of coseismicdeformation from Taiwan strong-motion records, J. Seismol. 11,159–170.

Zahradník, J., and A. Plešinger (2005). Long-period pulses inbroadband records of near earthquakes, Bull. Seismol. Soc. Am. 95,1928–1939.

Charles University in PragueFaculty of Mathematics and PhysicsDepartment of GeophysicsV Holesovickach 2, 180 00Prague, Czech [email protected]

(J.Z.)

Institute of GeophysicsAcademy of Sciences of the Czech RepublicBocni II/1401141 31 Prague, Czech [email protected]

(A.P.)

Manuscript received 3 April 2009


http://dx.doi.org/10.1029/2004JB003415

http://dx.doi.org/10.1029/2004JB003415

toward understanding subtle instrumentation effects associated...

Documents