You are on page 1of 8

Identifying and Correcting Timing Errors at


Seismic Stations in and around Iran
by Ellen M. Syracuse, W. Scott Phillips, Monica Maceira, and Michael
L. Begnaud
ABSTRACT
A fundamental component of seismic research is the use of Suspected problems with station timing can be found in a
phase arrival times, which are central to event location, Earth variety of ways. The simplest way of identifying a potential
model development, and phase identification, as well as derived timing problem, which is utilized in this study, is by monitoring
products. Hence, the accuracy of arrival times is crucial. How- temporal changes in travel-time residuals on a station-by-sta-
ever, errors in the timing of seismic waveforms and the arrival tion basis (e.g., Röhm et al., 1999). For particular distributions
times based on them may go unidentified by the end user, par- of earthquakes recorded by a pair of stations, changes in the
ticularly when seismic data are shared between different organ- difference between the arrival times recorded at the two sta-
izations. Here, we present a method used to analyze travel-time tions for a single earthquake can reveal timing errors (Rowe
residuals for stations in and around Iran to identify time peri- et al., 2003; Gibbons, 2006). These methods have the advan-
ods that are likely to contain station timing problems. For the tage of not requiring the user to have access to full waveform
14 stations with the strongest evidence of timing errors lasting data. If waveform data are available, analysis of fundamental-
one month or longer, timing corrections are proposed to ad- mode surface-wave dispersion between pairs of stations can be
dress the problematic time periods. Two additional stations are
used to identify problems with timing, location, or polarity
identified with incorrect locations in the International Registry
(Weidle et al., 2013). These three methods rely on analyzing
of Seismograph Stations, and one is found to have erroneously
signals from earthquakes, limiting the ability to identify short
reported arrival times in 2011.
intervals (less than a day to multiple weeks) of concern. To
identify more ephemeral timing changes or to identify the on-
set of a timing change with greater precision, changes in the
Electronic Supplement: Figures showing residual distributions timing of the Green’s function resulting from the cross corre-
over time and associated P-wave ray paths. lation of microseismic noise over time between pairs of stations
can be analyzed (Koch and Stammler, 2003; Sens-Schön-
felder, 2008).
INTRODUCTION Apparent timing problems can have a variety of sources
that fall into four categories: (1) actual timing problems, (2) sta-
Many seismological tools rely on the accurate timing of data.
tion location misinformation, (3) problems related to phase
Global Positioning System (GPS) clocks, which are accurate to
identification, and (4) Earth structure changes. Legitimate tim-
100 ns (see Data and Resources), are now the standard for
ing problems can occur when the clock is unable to lock with a
keeping time at seismic stations. Ideally, this should result
in accuracy that is at least 105 times higher than typical sample GPS satellite for a prolonged period of time, perhaps following
rates, but in reality, station timing problems commonly degrade station vandalism (e.g., Owens et al., 1993). Stations with tele-
the quality of seismic datasets, often without the knowledge of metered data may incur a delay (e.g., Chang et al., 2012) that
users. Because shared data repositories are a fundamental data may not be properly corrected for at the archiving location. If
source for the seismic community (e.g., the Bulletin of the there is a problem with the clock at the telemetry repeater,
International Seismological Centre [ISC], the Data Manage- multiple stations within a network may be similarly affected.
ment Center at the Incorporated Research Institutions for Leap seconds have been inserted to UTC in some years since
Seismology [IRIS], the European Integrated Data Archive 1972, either following 30 June or 31 December, to account for
at the Observatories and Research Facilities for European Seis- changes in Earth’s rotational speed over time. Because GPS-
mology [ORFEUS]), even if data operators are aware of issues based times do not include leap seconds, they need to be cor-
with data quality, there is no clear way to convey this informa- rected before their insertion; not correcting for them may leave
tion to the end user. GPS clock times 1 s fast with respect to UTC.

1472 Seismological Research Letters Volume 88, Number 6 November/December 2017 doi: 10.1785/0220170113

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/88/6/1472/3950286/srl-2017113.1.pdf


by Indonesian Agency for Meterology, Climatology and Geophysics user
predominance of basin-trapped waves in the coda (Phillips
40°
0 10 20 30
and Aki, 1986).
Earthquake depth (km) Analysis of arrival-time data from the ISC for stations and
earthquakes in the Iran region reveals that timing inconsisten-
cies are likely to be present at several stations. We identify af-
fected stations and time periods, and present corrections that
35°
can be applied to address these problems in all but one case.

DATA AND METHODS


30°
First-arrival times for P and S phases were downloaded from
the ISC database (see Data and Resources) for stations and
earthquakes occurring between 2000 and 2016 within the
25° study region (Fig. 1). Only earthquakes with 10 or more P-
arrival times recorded at these stations are retained. This results
40° 45° 50° 55° 60° in a dataset consisting of 16,500 earthquakes, and 333,000 P-
arrival times and 105,000 S-arrival times recorded at 299
▴ Figure 1. Map of stations and sources used in this study. stations.
Earthquakes are shaded by depth. Small gray triangles show sta-
These earthquakes are relocated using a modified version
tions with no identified timing or location errors. White triangles
of the algorithm tomoDD (Zhang and Thurber, 2003, 2006)
show stations with identified timing errors. Squares show stations
using a 3D P- and S-wave velocity model for the region devel-
with identified location errors (corrected locations plotted here).
oped by Maceira et al. (2016) from the joint inversion of seis-
The diamond shows Station SHGR, which appears to be affected
mic arrival times, surface-wave dispersion, and gravity data. We
by misassociated arrivals.
note that, although similar and overlapping in many cases, the
datasets used in this study and Maceira et al. (2016) differ in
Station location misinformation may occur if a station is that publicly available data of all qualities are included in this
moved without updating the coordinates of the station. Röhm study, whereas the Maceira et al. (2016) study includes only
et al. (1999) found that within the ISC Bulletin, several sta- earthquakes with expected epicentral accuracies of 50 km or
tions show a change in mean residual following a data gap, better from a longer time period. The use of the ISC dataset
which may result from station down time while it is being is intended to demonstrate that the method of error identifi-
moved. The ISC does report multiple sets of coordinates cation described in this study can be applied to readily available
for some stations through time, but it is often unclear if these arrival-time catalogs. We also note that similar results, albeit
changes in coordinates are due to the station being physically with larger travel-time residuals and slightly more residual scat-
moved or due to the initially reported locations being incor- ter, are obtained if a 1D velocity model is used for relocation, as
rect. Alternatively, a station location may be incorrectly re- opposed to the 3D model used here. Although differential
ported for the entire duration of data from a station. travel times for pairs of earthquakes recorded at the same sta-
Problems related to phase identification may arise if there tion are typically used in the tomoDD algorithm, they are
is a change in the noise level at a station. Because higher levels omitted here so that an earthquake location resulting from data
recorded by stations not containing any timing errors during
of noise can obscure the onset of a phase, a phase arrival may be
the earthquake is unaffected by possible timing errors at other
picked later than its actual onset when noise levels are elevated.
times during the study period. Because bad data are the focus of
Alternatively, the use of a noncausal filter can cause phase on-
this study, data are weighted based on only their estimated pick
sets to be identified too early. Changes in how data are dissemi- uncertainty (i.e., less reported precision for an arrival time im-
nated may also lead to apparent timing shifts (Weidle et al., plies a larger pick uncertainty), and there is no maximum al-
2013), and changes in equipment (e.g., upgrading from a lowable residual in the inversion.
short-period to a broadband sensor) will affect the waveforms After relocation, earthquake origin time versus P-wave
recorded, which may cause an apparent shift when phases are residual is plotted for each station (Fig. 2). Stations with no
identified. obvious timing problems (Fig. 2a) have residuals with a con-
Changes in the subsurface beneath a station can also lead stant moving-window average and scatter over time. Stations
to apparent timing changes. For example, seasonal changes in with a small number of arrivals are not considered further, due
groundwater at Merapi Volcano have been tied to 10% changes to too few data points for meaningful statistics. Stations that
in seismic velocity at some depths (Sens-Schönfelder and We- appear to have clock drift (Fig. 2b) cannot be corrected for in a
gler, 2006), which may mimic clock drift. Velocity changes have straightforward manner (see the AKDM, GURO, and MLAZ
also been identified via progressive phase delay between earth- section). One station (see the SHGR section) appears to have
quake doublets in central California and are also thought to residuals that generally follow an expected distribution but has
result from near-station groundwater changes because of the a period with some large residuals. Stations with large devia-

Seismological Research Letters Volume 88, Number 6 November/December 2017 1473

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/88/6/1472/3950286/srl-2017113.1.pdf


by Indonesian Agency for Meterology, Climatology and Geophysics user
(a) (c) 60
6
40
4
Residual (s)

2 20

0 0

–2 –20
–4
–40
–6 IBAF SHGR
–60
2004 2006 2008 2010 2012 2014 2016 2004 2006 2008 2010 2012 2014 2016

(b) 8 (d) 8
6 6
4 4
Residual (s)

2 2
0 0
–2 –2
–4 –4
–6 –6
AKDM IRAZ
–8 –8
2012 2014 2016 2004 2006 2008 2010 2012 2014 2016
Year Year
▴ Figure 2. Examples of the distribution of P-wave residuals over time for (a) a station with no apparent timing errors (Station IBAF,
located at 31.5902° N, 55.56730° E), (b) a station with apparent clock drift over multiple years (AKDM, located at 38.3285° N, 42.98000° E),
(c) a station with erroneously reported arrival times (Station SHGR, located at 32.1083° N, 48.80130° E), and (d) a station with suspected
timing errors from June 2005 through September 2006 (Station IRAZ, located at 35.4046° N, 49.92900° E). The gray line shows the 20-point
moving average of residuals through time.

tions in moving-window average over time (≳1 s) are investi- the time period are likely due to these earthquakes being poorly
gated further (Fig. 2d and Ⓔ Figs. S1–S14, available in the located or the velocity model used for relocation poorly rep-
electronic supplement to this article). For these stations requir- resenting earth structure along these paths; these stations are
ing further investigation, their operation is separated into a removed from further consideration.
background time period and one or more time periods of in- Following this analysis, 14 stations remain, with time peri-
terest. The background time period is considered to contain no ods of interest ranging from two months to three years. If clock
timing error and is either the longest time period with a con- errors do exist at these stations, P- and S-wave residuals during
stant moving-average residual or the time period with the a time period of interest should be offset by the same amount
smallest average residual; in most cases, these two definitions from the background. However, the large scatter in S-wave re-
give the same background time period. The time period of in- siduals relative to P-wave residuals, combined with a lack of S-
terest contains residuals for which moving average deviates wave arrival data for many time periods of interest, prevents the
from the background and is suspected of containing a persis- comparison of P-wave residuals over time to S-wave residuals
tent timing error. Two additional stations with uniformly large over time from yielding meaningful results.
scatter in residuals over time are also investigated (Ⓔ Figs. S15 For the remaining 14 stations, timing corrections are cal-
and S16). culated for the time periods of interest, based on the concept of
For each station requiring further investigation, the travel-time station corrections. Typically, a travel-time station
source–station distribution is plotted, with ray paths during correction is calculated to accommodate for the effect of near-
the time period of interest highlighted. In some cases, the time station velocity structure that is unaccounted for in a broader
period of interest corresponds to an isolated swarm of earth- velocity model or to approximate the effect of the elevation of
quakes (Fig. 3a). For such cases, the residual deviations during the station, if elevation is not directly accounted for in the ana-

1474 Seismological Research Letters Volume 88, Number 6 November/December 2017

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/88/6/1472/3950286/srl-2017113.1.pdf


by Indonesian Agency for Meterology, Climatology and Geophysics user
(a)40° ation, which is found to yield the best results. For stations with
identified location errors (see description in the Station Loca-
tion Corrections section), the corrected locations are used in
this inversion. After station corrections are determined, they
are applied and held fixed in a separate final inversion for
35°
source location only; residuals reported for stations after cor-
3 rection are from this final inversion.
Residual (s)

2
30° RESULTS AND DISCUSSION
1

0 Station Timing Corrections


SZD1 The results of the correction calculations are given in Table 1
25° and are described below for each station. Positive corrections
indicate that the station clock was modeled as being slow dur-
(b)40° ing the time period of interest, and negative corrections indi-
cate that it was modeled as being fast. Corrections should be
added to the arrivals from an affected station during time peri-
ods of interest. Background information for each station is ob-
tained from the International Registry of Seismograph
35°
Stations, maintained by the ISC (see Data and Resources).
10 For all stations discussed here, the average residual for a station
8 during its background time period is 0 s. Figures showing P-
Residual (s)

6 wave residuals over time before and after correction and ray
30° paths for the corresponding time period of interest are given
4
2 in Ⓔ Figures S1–S14, ordered alphabetically by station name.
0
IRAZ ASAO, NASN, and THKV
25° Station ASAO is located near Ashtian, Markazi Province, Iran;
40° 45° 50° 55° 60°
Station NASN is located near Nain, Isfahan Province, Iran;
▴ Figure 3. Examples of ray paths for potential periods of inter- and Station THKV is located near Karaj, Alborz Province,
est for stations where (a) there is a strong correlation between Iran. Data for these three stations have been reported to
source location and residual, indicating that further investigation the ISC since 2001 (Ⓔ Figs. S1, S11, and S13). These stations
is not required (Station SZD1, 29.4863° N, 60.86550° E) and have average residuals of −0:7 to −4:3 s prior to 2004. Addi-
(b) residual is independent of source location, prompting further tionally, arrivals are less frequently reported for the earlier time
investigation (Station IRAZ). Small circles show the locations of periods of interest than for 2004 and later. This reduction in
all earthquakes recorded at the given station, and line shade in- residuals, increase in pick frequency, and apparent shift in sta-
dicates the magnitude of the residual. The color version of this tion timing correspond to the replacement of digitizers at these
figure is available only in the electronic edition. three stations that were known to have had timing issues (E.
Bergman, personal comm., 2017). The corrections for ASAO
and THKV are similar, 4.5 and 4.2 s, respectively. The calcu-
lytical method being used. The correction can differ for P and S lated correction for NASN is 7.1 s, although the relatively small
waves, and it typically applies to the entire duration of the number of arrivals for the time period of interest leads to a
dataset. relatively large uncertainty in the corrections. Following the
For each time period of interest, a timing correction is correction, the mean residual for the period of interest at each
calculated. This is done by creating a dummy station for each of these three stations is 0.6 s or less.
time period of interest and attributing the affected arrivals to
the appropriate new station. Then, station corrections are cal- BASK
culated for all dummy stations in a joint inversion for earth- Station BASK is located near Baskale, Van Province, Turkey,
quake location and station correction. This inversion uses a and data have been reported to the ISC between mid-2011
modified version of the algorithm used for the initial residual and mid-2015 (Ⓔ Fig. S2). For earthquakes from January
calculation. Here, it is modified to calculate station corrections through April 2012, the average uncorrected residual at BASK
for stations selected by the user, that is, only the new dummy is −1:3 s, with almost no earthquakes having positive P-wave
stations. For each selected station, a single correction is calcu- residuals. Although most earthquakes in the time period of in-
lated to apply to both P and S arrivals by including both phases terest did occur in a similar area to the northwest of the station,
in the inversion. This inversion is only continued for one iter- P-wave residuals for stations at all azimuths are negative during

Seismological Research Letters Volume 88, Number 6 November/December 2017 1475

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/88/6/1472/3950286/srl-2017113.1.pdf


by Indonesian Agency for Meterology, Climatology and Geophysics user
Table 1
Results of the Correction Calculations
Latitude Longitude Elevation T1 T2 Correction Meanpt rmspt rmspa Meanft rmsft rmsfa
Station (°) (°) (m) (yyyy/mm) (yyyy/mm) (s) (s) (s) (s) (s) (s) (s)
AKDM 38.3285 42.98000 1662 — — — — — 1.079 — — —
ASAO 34.5480 50.02531 2217 2000/01 2003/12 2.5 −1.829 3.635 2.430 −0.233 2.960 2.413
BASK 38.0523 44.00400 1931 2012/01 2012/04 2.0 −1.331 1.639 4.048 −0.067 0.926 4.023
BNDS 27.3993 56.17130 1500 2012/04 2014/06 3.5 −2.102 3.035 2.190 −0.100 2.176 2.018
GURO 38.5509 42.03220 1388 — — — — — 1.273 — — —
HAKT 37.5579 43.70710 2153 2007/11 2008/03 −5.8 3.699 5.473 2.050 0.150 4.137 1.944
HAKT 37.5579 43.70710 2153 2011/01 2011/02 −7.1 4.480 5.259 2.050 0.105 1.940 1.944
HKZM 35.3775 48.90450 2334 2012/01 2012/02 −3.5 2.735 2.966 1.290 0.237 0.908 0.945
IGAR 32.4063 52.04740 1910 2005/01 2005/12 1.6 −1.306 2.640 1.965 0.185 2.256 1.921
IHSH 37.3050 47.26360 2147 — — — — — 8.493 — — 2.280
IKLH 33.3190 51.57870 2185 2005/01 2005/12 1.5 −1.516 2.718 2.590 0.175 2.217 2.449
IKRD 36.7759 59.51460 2245 — — — — — 8.358 — — 3.214
IPIR 32.6841 50.89170 2587 2005/01 2005/12 1.6 −1.544 2.747 2.256 0.023 2.230 2.170
IRAZ 35.4046 49.92900 1920 2005/06 2006/09 −2.0 3.943 6.124 2.991 0.115 4.541 2.622
IZEF 32.8956 52.32910 2321 2005/01 2005/12 1.5 −1.366 2.813 2.683 0.207 2.447 2.564
MLAZ 39.1410 42.54950 1581 — — — — — 1.075 — — —
NASN 32.7990 52.80800 2800 2000/01 2003/12 2.5 −4.302 6.732 3.320 0.573 4.340 3.266
QAM 33.7606 51.43970 1864 2000/01 2006/12 −0.9 0.606 1.779 2.191 0.087 1.642 1.495
SHGR 32.1083 48.80130 150 2011/01 2011/12 — −13.25 27.67 9.648 0.530 2.007 1.890
THKV 35.9157 50.87881 1795 2000/01 2003/12 2.8 −0.602 4.013 3.128 0.613 4.232 3.140
TUTA 39.4019 42.81370 2154 2010/03 2012/04 3.6 −3.196 3.749 1.664 −0.172 1.670 1.638
Corrections, mean residuals, and root mean square (rms) residuals are given in seconds. Station elevations are given in meters
above sea level. Subscript “p” indicates values prior to correction; “f” indicates final values after correction; “t” indicates values
just for the time period of interest; and “a” indicates values for all times.

this period. The calculated correction for this time period both periods, residuals tend to increase over time and more
is 2.0 s. rapidly return to background values, suggesting that these
may both be cases of clock drift. Given the relatively short du-
BNDS ration of these periods, this is difficult to confirm with cer-
Station BNDS is located near Bandar Abbas, Hormozgan tainty; they are treated as periods of constant timing offset
Province, Iran, and data have been reported to the ISC since for correction calculation. The timing corrections for the
2004, with a six-month gap from late 2012 through early 2013 two periods are among the largest of any station discussed here:
(Ⓔ Fig. S3). The period of interest spans April 2012 through −5:2 and −6:6 s.
June 2014, and it contains average residuals of 2.1 s in the un-
corrected data. After applying the timing correction of 2.9 s, HKZM
the average residual during the time period of interest is Station HKZM is located west of Razan, Hamadan Province,
brought down to 0.1 s. Iran, and data have been reported to the ISC between 2011
and 2013 (Ⓔ Fig. S5). During the period of interest from
HAKT January through February 2012, all P-wave residuals are pos-
Station HAKT is located near Hakkari, Hakkari Province, itive and greater than 1 s, with an average P-wave residual of
Turkey, and data have been reported to the ISC since 2005 2.9 s. Although many of the residuals during this period are for
(Ⓔ Fig. S4). This station has two short periods of interest last- earthquakes in a cluster to the southeast of the station, P-wave
ing five months (November 2007–March 2008) and two residuals are also positive and large for earthquakes to the north
months (January–February 2011), both of which contain pre- and west of the station.
dominantly positive P-wave residuals. Although a very short
deviation in residual distribution is often due to an aftershock IGAR, IKLH, IPIR, and IZEF
series or a swarm, the large residuals occur for earthquakes at a Stations IGAR, IKLH, IPIR, and IZEF are located near Ghar-
range of azimuths and distances for both of these periods. In neh, Kalahroud, Kuh-e Pir Pir, and Zefreh, Isfahan Province,

1476 Seismological Research Letters Volume 88, Number 6 November/December 2017

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/88/6/1472/3950286/srl-2017113.1.pdf


by Indonesian Agency for Meterology, Climatology and Geophysics user
Iran, respectively. Data have been reported to the ISC since (a) 40°
2005 (Ⓔ Figs. S6, S7, S8, and S10). With the exception of
IGAR, these stations are part of the Esfahan (Isfahan) subnet-
work of the University of Tehran network. In uncorrected
data, the three stations have average residuals of −1:3 to −1:6 s
35°
for earthquakes in 2005. The corrections calculated for these
stations are also similar to each other, 2.0–2.1 s. The consis- 15
tency between the corrections and affected time periods, com- 10

Residual (s)
bined with their being part of the same subnetwork, suggests a 5
common cause to their errors, perhaps a telemetry delay that 30° 0
–5
was improperly accounted for. Station IZEF also displays a sec-
–10
ond deviation in residuals in November and December 2011, –15
but this is not investigated further because it is caused by a IHSH
typical distribution of residuals that is contaminated by several 25°
large residuals. (b) 40°
IRAZ
Station IRAZ is located near Razequan, Markazi Province,
Iran, and data have been reported to the ISC since 2004
(Ⓔ Fig. S9). The period of interest spans from June 2005 35°
through September 2006, and it contains generally positive 15
P-wave residuals, averaging 4.8 s, with more scatter than the 10

Residual (s)
background. After applying the timing correction of 1.5 s, the 5
residuals during the period of interest continue to have more 30° 0
scatter than the background, and P-wave residuals for shorter –5
–10
windows within the period of interest have averages of up to
–15
2 s, suggesting a secondary source of shorter-term timing errors. IKRD
25°
QAM 40° 45° 50° 55° 60°
Station QAM is located in Ghamsar, Isfahan Province, Iran,
and data have been reported to the ISC from 2005 through
▴ Figure 4. Plots of ray paths shaded by residual for stations
(a) IHSH and (b) IKRD, which indicate that the reported locations
2006 and 2010 through 2015 (with two arrivals reported be-
for these stations are incorrect. The correct locations should
tween 2006 and 2010), an ∼14-month gap from mid-2012
be in the direction of the most negative residuals (south for IHSH
through mid-2013 and a 4-month gap in 2014 (Ⓔ Fig. S12).
and east for IKRD). Outliers from this pattern (particularly large
Prior to 2007, residuals have an average value of 0.6 s. Follow-
positive residuals for some earthquakes south of the reported
ing a timing correction of −0:9 s for this earlier period, resid-
location for IKRD) indicate earthquakes that are significantly
uals are consistently distributed for all time periods.
mislocated using the reported station location or individual bad
picks. Refer to the Station Location Corrections section and
TUTA
Table 1 for correct coordinates of these stations. The color
Station TUTA is located south of Tutak, Ağrı Province, Tur-
version of this figure is available only in the electronic edition.
key, and data have been reported to the ISC between 2007 and
2014 (Ⓔ Fig. S15). During March–April 2010, all but one
is no identifiable period of larger-than-average residuals, the
residual for this station are negative, with an average value
root mean square (rms) residual for this station is 2.5 times
of −3:2 s. Although the most negative residuals are for a cluster
of earthquakes to the southwest of the station, residuals are that of all stations in this dataset. Residuals for earthquakes
negative for earthquakes at all azimuths to the station during surrounding the station are clearly azimuthally dependent
this period. After the application of the 3.6 s timing correction, (Fig. 4a and Ⓔ Fig. S15), suggesting that the reported station
the residuals are brought in line with those of the background location is likely incorrect, and the station may actually be lo-
periods, with similar scatter throughout. cated south of the reported location. Google Earth imagery also
shows no evidence of a seismic station at the coordinates given
by the ISC (38.30670° N, 47.26300° E), whereas seismometer
Station Location Corrections housings are evident at many other stations in the region (see
IHSH Data and Resources). Approximately 1° south of the reported
Station IHSH is located near Hashtrood, East Azerbaijan location, at 37.3050° N, 47.2636° E, what appears to be a seis-
Province, Iran, and data have been reported to the ISC by mometer housing is visible in Google Earth imagery.
the University of Tehran since 2005, with no data reported Using this new location with an elevation of 2147 m
between 2007 and mid-2013 or since 2015. Although there (based on the elevation for these coordinates in Google Earth)

Seismological Research Letters Volume 88, Number 6 November/December 2017 1477

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/88/6/1472/3950286/srl-2017113.1.pdf


by Indonesian Agency for Meterology, Climatology and Geophysics user
reduces the rms residuals from 8.493 to 2.280 s, supporting this each station. Although sinusoidal (or similar) curves could
correction to the reported location for station IHSH. A 1° er- be fit to the residuals for each station and used as station cor-
ror in station location, such as that seen here, corresponds to an rections, one could not determine the correct absolute times of
approximately 15 s error in predicted P-wave travel time for the arrivals. Therefore, no correction is solved for here.
earthquakes along the axis of the mislocation direction, which
corresponds to the residual distribution observed when using SHGR
the registered ISC location. The error in reported station co- Station SHGR has data reported to the ISC from mid-2004
ordinates appears to have been introduced during the registra- through mid-2015. In general, residuals at SHGR are small
tion of the station location with the ISC, likely as a simple (< 10 s) and tightly centered on 0 s. In 2011, most residuals
mistyping of coordinates. Spot checking of the pick files sent continue to be small and tightly centered on 0 s, but 30% of the
to the ISC (see Data and Resources) indicate that the Univer- P-wave residuals have a magnitude greater than 10 s (Fig. 2c
sity of Tehran uses a location similar to the one determined and Ⓔ Fig. S17). Upon further investigation, it appears that
here in their routine locations, not the location registered with almost all of these large-magnitude residuals are associated with
the ISC, based on the reported event-station distances and an arrival time that is the same as that reported for the same
azimuths. earthquakes, but from either Station ISHM (584 km away),
ISHV (1011 km away), SHRO (792 km away), or SHRT
IKRD
(1086 km away). Because these affected earthquakes are not
Station IKRD is located near Kardeh, Razavi Korasan Prov-
equidistant to SHGR and any of these stations, it is clear that
ince, Iran, and data have been reported to the ISC since
somehow their times were being erroneously reported for
2005. Similar to IHSH, there is no clear change in the tem-
poral pattern of residuals over time, but the rms residual is 2.5 SHGR. Upon spot checking data files from the Iranian Seis-
times that of all stations in the dataset, and there is a clear mological Center at the University of Tehran, which reports
azimuthal dependence of residuals (Fig. 4b and Ⓔ Fig. S16). many arrival times for Iranian stations to the ISC (see Data and
Google Earth imagery shows no evidence of a seismic station at Resources), including the five stations discussed here, there are
the coordinates given by the ISC (36.7759° N, 58.5146° E). no such erroneous arrivals in the original data files. Data for
Additionally, the elevation of these coordinates as reported many Iranian stations are also reported by the University of
in Google Earth is 668 m lower than the elevation given Tehran to the Centre Sismologique Euro-Méditerranéen
for the station by the ISC. The pattern of residuals suggests (CSEM) and then on to the ISC (see Data and Resources).
that the actual station location is east of the reported location. It appears that the erroneous attribution of arrivals to SHGR
As for IHSH, the ∼30 s difference between the most positive occurred in the incorporation of the University of Tehran
and the most negative P-wave residual calculated for this arrivals into the CSEM catalog, and these arrivals are then in-
station indicates that the correct location should be ∼1° from corporated into the ISC catalog. To address this problem, any
the reported location. Indeed, exactly 1° further east, at arrival for SHGR during 2011 that matched the time of any of
36.7759° N, 59.5146° E, Google Earth shows similar infrastruc- these four stations was removed from the dataset (90 P and S
ture to what is observed at other stations in the region, and the arrivals in total) before the final relocation inversion was per-
elevation differs by only 40 m from that reported by the ISC. formed.
Using this location reduces the rms residuals from 8.355 to
3.214 s for Station IKRD. As with IHSH, the event-station
distances and azimuths for IKRD given in the original Uni- Effect on Earthquake Locations
versity of Tehran pick files submitted to the ISC agree better Accurate earthquake locations rely on the accurate timing of
with our corrected location than with the registered ISC loca- data from stations. For the dataset presented here, applying
tion, indicating that the source of the incorrect location is the station timing corrections and corrected station locations
likely a mistyping of the correct station coordinates upon regis- has a relatively small effect on earthquake locations and seis-
try at the ISC. micity patterns overall. However, only a small proportion of
arrival times is affected by these corrections. Although 17
out of 299 stations have corrections applied (5.4%), this
Other Station Issues
AKDM, GURO, and MLAZ amounts to 0.48% of the arrival times being affected by a tim-
These three stations are in southeastern Turkey and have re- ing correction, 0.29% of them being affected by a station lo-
ported data from late 2012 through 2016, with the exception cation correction, and 0.02% removed as erroneous picks. Out
of GURO, which also has a small amount of data for 2011. of the 16,500 earthquakes analyzed as part of this study, only
The P-wave residuals for these stations suggest that they expe- 2183 are associated with arrivals that have had corrections ap-
rienced the same clock drift over the duration, with a roughly plied. Of these 2183 earthquakes affected by the corrections,
sinusoidal pattern (Fig. 2b). On average, residuals are most neg- the average earthquake shift is 7 km, with a maximum horizon-
ative in early 2013 and early mid-2016, and they are most pos- tal shift of 48 km and a maximum depth shift of 59 km. 39% of
itive in late 2014/early 2015, with an ∼0:7 s difference these earthquakes move 5 km or more from their locations
between the lowest and highest monthly median values for using uncorrected data.

1478 Seismological Research Letters Volume 88, Number 6 November/December 2017

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/88/6/1472/3950286/srl-2017113.1.pdf


by Indonesian Agency for Meterology, Climatology and Geophysics user
CONCLUSIONS Gibbons, S. J. (2006). On the identification and documentation of timing
errors: An example at the KBS Station, Spitsbergen, Seismol. Res.
Lett. 77, 559–571, doi: 10.1785/gssrl.77.5.559.
We identified 14 stations in Iran and Turkey with likely clock Koch, K., and K. Stammler (2003). Detection and elimination of time
errors between 2000 and 2016. Using the methods described synchronization problems for the GERESS array by correlating mi-
here, timing errors can be similarly identified in other arrival- croseismic noise, Seismol. Res. Lett. 74, 803–816, doi: 10.1785/
time datasets and corrections to address these errors can be gssrl.74.6.803.
solved for. We have also identified two stations with locations Maceira, M., E. M. Syracuse, E. Bergman, W. S. Phillips, M. Begnaud, S.
Nippress, and H. Zhang (2016). Improvements in earthquake loca-
that are almost certainly incorrect in the International Registry tion from joint inversion of seismic and gravity observations—
of Seismograph Stations. Applications to the Iran region, Seismol. Res. Lett. 87, 461.
Although these corrections have a relatively small effect on Owens, T. J., G. E. Randall, F. T. Wu, and R. Zeng (1993). PASSCAL
earthquake locations, one can easily image scenarios where sim- instrument performance during the Tibetan Plateau passive seismic
ilar corrections for other datasets would have a more important experiment, Bull. Seismol. Soc. Am. 83, 1959–1970.
Phillips, W. S., and K. Aki (1986). Site amplification of coda waves from
effect. For example, at a volcano monitored by fewer stations, local earthquakes in central California, Bull. Seismol. Soc. Am. 76,
the onset of a timing error at a single station could result in an 627–648.
apparent shift in earthquake locations over time and be erro- Röhm, A. H. E., J. Trampert, H. Paulssen, and R. K. Snieder (1999). Bias
neously interpreted as indicating a change in magma storage. in reported seismic arrival times deduced from the ISC Bulletin,
Geophys. J. Int. 137, 163–174.
Rowe, C. A., W. S. Phillips, and L. K. Steck (2003). Relative constraints
DATA AND RESOURCES on correction surfaces for more effective use of low-order ground
truth, Eos Trans. AGU 84 (Fall Meet. Suppl.), Abstract S21D–
The data used in this research were downloaded from the 0332.
International Seismic Centre (http://www.isc.ac.uk/iscbulletin/ Sens-Schönfelder, C. (2008). Synchronizing seismic networks with am-
search/arrivals/interactive/, last accessed January 2017). All fig- bient noise, Geophys. J. Int. 174, 966–970, doi: 10.1111/j.1365-
246X.2008.03842.x.
ures were made using the Generic Mapping Tools software Sens-Schönfelder, C., and U. Wegler (2006). Passive image interferometry
package v.4.5.9 (www.soest.hawaii.edu/gmt; Wessel and Smith, and seasonal variations of seismic velocities at Merapi Volcano, In-
1998). The other data are from the following websites: http:// donesia, Geophys. Res. Lett. L21302, doi: 10.1029/2006GL027797.
www.gps.gov/applications/timing/ (last accessed April 2017), Weidle, C., R. A. Soomro, L. Cristiano, and T. Meier (2013). Identifi-
http://www.isc.ac.uk/registries/search/ (last accessed April 2017), cation of response and timing issues at permanent European broad-
band stations from automated data analysis, Adv. Geosci. 36, 21–25,
http://www.isc.ac.uk/cgi-bin/agency-get?agency=THE (last accessed doi: 10.5194/adgeo-36-21-2013.
January 2017), and http://www.isc.ac.uk/cgi-bin/agency-get? Wessel, P., and W. H. F. Smith (1998). New, improved version of the
agency=CSEM (last accessed May 2017). For Google Earth imagery, Generic Mapping Tools released, Eos Trans. AGU 79, 579.
see www.google.com/earth/desktop/ (last accessed April 2017). Zhang, H., and C. Thurber (2006). Development and applications of
double-difference tomography, Pure Appl. Geophys. 163, 373–
403, doi: 10.1007/s00024-005-0021-y.
ACKNOWLEDGMENTS Zhang, H., and C. H. Thurber (2003). Double-difference tomography:
The method and its application to the Hayward fault, California,
The authors thank Eric Bergman and two anonymous re- Bull. Seismol. Soc. Am. 93, 1875–1889, doi: 10.1785/0120020190.
viewers for their helpful comments and input. This work
was conducted under the auspices of the U.S. Department Ellen M. Syracuse
of Energy National Nuclear Security Administration by Los W. Scott Phillips
Alamos National Laboratory under Contracts LA17-Signal- Michael L. Begnaud
Propagation-NDD2Ab and DE-AC52-06NA25396. The Los Alamos National Laboratory
views and conclusions contained in this article are those of MS F665, P.O. Box 1663
the authors and should not be interpreted as necessarily rep- Los Alamos, New Mexico 87544 U.S.A.
resenting the official policies, either expressed or implied, of the syracuse@lanl.gov
U.S. Government.
Monica Maceira
REFERENCES Oak Ridge National Laboratory
MS-6050, P.O. Box 2008
Chang, C.-H., Y.-M. Wu, D.-Y. Chen, T.-C. Shin, T.-L. Chin, and W.-Y. Oak Ridge, Tennessee 37831 U.S.A.
Chang (2012). An examination of telemetry delay in the Central
Weather Bureau Seismic Network, Terr. Atmos. Ocean. Sci. 23,
261–268, doi: 10.3319/TAO.2011.11.29.01(T). Published Online 6 September 2017

Seismological Research Letters Volume 88, Number 6 November/December 2017 1479

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/88/6/1472/3950286/srl-2017113.1.pdf


by Indonesian Agency for Meterology, Climatology and Geophysics user

You might also like