Resolving the location of small intracontinental earthquakes using Open Access seismic and geodetic data: lessons from the 18

A low-magnitude earthquake was recorded on January 18, 2017, in the T´en´er´e desert in northern Niger. This intraplate region is exceptionally sparsely covered with seismic stations and the closest open seismic station, G.TAM in Algeria at a distance of approximately 600 km, was unusually and unfortunately not operational at the time of the event. Body-wave magnitude estimates range from m b 4 . 2 to m b 4 . 7 and both seismic location and magnitude constraints are dominated by stations at teleseismic distances. The seismic constraints are strengthened considerably by array stations of the International Monitoring System for verifying compliance with the Comprehensive Nuclear Test-Ban-Treaty. This event, with magnitude relevant to low-yield nuclear tests, provides a valuable validation of the detection and location procedure for small land-based seismic disturbances at signiﬁcant distances. For seismologists not in the CTBT system, the event is problematic as data from many of the key stations are not openly available. We examine the uncertainty in published routinely-determined epicenters by performing multiple Bayesloc location estimates with published arrival times considering both all published arrival times and those from open stations only. This location exercise conﬁrms lateral uncertainties in seismologically-derived location no smaller than 10 km. Coherence for InSAR in this region is exceptionally high, and allows us to conﬁdently detect a displacement of the order 6 mm in the time-frame containing the earthquake, consistent with the seismic location estimates, and with a lateral length scale consistent with an earthquake of this size, allowing location constraint to within one rupture length (  5 km) – signiﬁcantly reducing the lateral uncertainty compared with relying on seismological data only. Combining Open Access-only seismological and geodetic data, we precisely constrain the source location, and conclude that this earthquake likely had a shallow source. We then discuss potential ways to con-tinue the integration of geodetic data in the calibration of seismological earthquake location.

to the network displayed in Figure 2. The best signals are found on stations to the 78 North East; in Eastern Europe and Central Asia -a distribution that will be the result 79 of both the network coverage, and the orientation of the focal mechanism and resultant 80 radiation pattern (note that the focal mechanism for this earthquake is unknown). the GEOFON data center at GFZ Potsdam. Each of these arrays has an aperture of 89 only a few km, with the intention that short period signals (e.g. 1-4 Hz) are coherent 90 between sensors and that the SNR of signal arrivals can be improved by delay-and-91 stack beamforming (e.g. Rost and Thomas, 2009). Similarly, estimating the coherence 92 or relative power of beams in di↵erent directions allows us to estimate the backazimuth 93 and apparent velocity of incoming wavefronts. This assists in algorithms to associate 94 detections and helps to build confidence that a given signal detection is indeed associated 95 with our event hypothesis, on the basis of directional coherence of arrivals.  The remaining three panels of Figure 2 show signals for the P -arrivals at arbitrarily 114 chosen teleseismic 3-component stations (in Czechia, Saudi Arabia, and Kenya). We note 115 that the SNR for the signals at many of these stations is relatively poor, and that im-116 provement through stack-and-delay is not possible for non-array stations. The waveforms 117 shown in Figure 2 also highlight the potential subjectivity in identifying the onset of a 118 particular phase arrival, with the majority of arrivals being emergent, especially in terms 119 of identifying a confirmed signal above the level of noise. We see no unmistakable depth 120 phases, which would o↵er a high-precision constraint on the event depth. A few stations 121 show multiple bursts of energy but there is insu cient evidence at any station to label 122 with confidence the later arrivals as depth phases.

123
Summarising the available seismological data, we are left with a comparatively sparse 124 set of phase observations, of variable, but often limited, precision. The advantages in 125 signal identification and arrival precision that arise from the enhanced processing of 126 small aperture arrays is clear. But only a few of the operators of these stations make 127 their waveform data Open Access (see Figure 2). Similarly, many of the more isolated 128 three-component stations, vital for filling gaps in azimuthal and epicentral coverage, 129 remain closed to the general public. Combined, these pose the question of how reliant 130 high-precision earthquake location is on closed-access data, and how well characterised 131 events such as the Ténéré earthquake can be, using only Open Access seismic data.  Figure 3 shows the epicenters listed in Table 1 together with their published 95% con-    In the case of remote continental earthquakes, with a sparsity of near-field seismological 290 data, the recently-developed global coverage of satellite radar o↵ers an additional dataset 291 to which may help constrain earthquake locations, and complement those constraints 292 available from seismology. The limiting factor in locating an earthquake using satellite 293 geodesy is not directly the magnitude of the earthquake, but instead the amplitude of 294 the surface deformation, and whether any signal can be detected. Whilst the Ténéré 295 earthquake is lower magnitude than typically studied using InSAR (e.g., Weston et al,  To supplement the available seismic data, we process interferometric synthetic aper-309 ture radar (InSAR) images for the source region using data from the European Space 310 Agency's Sentinel-1 satellites. We use acquisitions that span the earthquake date, and 311 construct interferograms using all potential pairs where the earthquake occurs within a 312 timespan of up to four consecutive acquisitions ( Figure 5). Processing was carried out 313 using the LiCSAR system (Lazecký et al, 2020, to which readers are directed for a full 314 description of the processing approach). Each interferogram is processed using multilook-315 ing factors of 5 in range and 20 in azimuth, with interferograms therefore having a spatial 316 resolution of ⇠ 100 ⇥ 100 m per pixel. Data are then subject to spatial filtering using an 317 adaptive power spectrum filter. Due to the remote location, only ascending track data 318 were being routinely acquired at the time of our study earthquake, with a 12-day repeat 319 time. Coherence in the region at such short temporal baselines is good -the region is 320 unvegetated desert, and whilst migratory sand can cause problems for radar interferom-321 etry, this does not appear to be the case around our earthquake, although we note that 322 the dune fields to the south and southwest show markedly lower coherence. In  mentary Material, Figure S2 shows average coherence prior to spatial filtering at 12, 24 324 and 36 day temporal baselines across the whole Sentinel-1 archive. As demonstrated by 325 Figure 5 and Figure 6, after spatial filtering, coherence at the wavelengths of interest 326 for earthquake-related processes is extremely high. Given the lack of major topographic 327 features, there is minimal topographically-correlated atmospheric noise, although all in-328 terferograms are subject to long-wavelength noise presumed to result from a combination 329 of atmospheric variations and orbital e↵ects (see Figures 5 and 6). One SAR acquisition 330 (20161216) features NE-SW orientated bands which are clearly unrelated to either the 331 regional tectonics or our study earthquake. Although the exact origin of these features 332 is uncertain, they are most likely to be atmospheric rolls. Some of the interferograms 333 shown in Figure 6, which do not span the earthquake, show significantly higher levels of 334 noise, which we presume to be atmospheric in origin, showing that even here, atmospheric Coppersmith, 1994), is common to all interferograms that span the earthquake date, and 343 is not present in any interferograms that do not span the earthquake (see Figure 6 for 344 examples). We are therefore confident that this signal relates to our study earthquake, 345 despite the small amplitude of the observed signal.  Figure 7a). To remove long-wavelength atmo-349 spheric e↵ects, and to isolate signals at wavelengths likely to be related to a m b 4.3 earth-350 quake, we spatially filter the InSAR data using a 4-pole Butterworth filter, bandpassed 351 between 15000 and 500 metres (Figure 7b).

352
The resulting stack shows a clear, coherent line-of-sight displacement of up to 6 mm.

353
Only one lobe of the deformation field is clearly visible, and although there are indica-354 tions on the filtered stack of opposite-polarity displacement lobes to the northeast and 355 southeast of the main deformation lobe, these are insu ciently clear to permit the deter-356 mination of a focal mechanism. We visually assess that the causative fault plane most 357 likely lies to the southeast or northeast of the peak in displacement. The lack of a clear 358 four-lobe pattern of deformation argues against a pure strike-slip mechanism, and we 359 infer that the earthquake therefore involved either dip-or oblique-slip faulting.

360
The deformation pattern shows no clear discontinuities in phase, either on the stack 361 or on individual interferograms, suggesting that the rupture did not break the surface, 362 and that the top of the fault rupture patch is buried. That there is an observable signal 363 at all, however, from such a small-magnitude event, indicates that the earthquake must 364 have been shallow ( 5 km; see Mellors et al (2004); Dawson and Tregoning (2007)), 365 consistent with the lack of any clearly separated depth phases in the seismic data (see 366 Figure 2). In the case of this earthquake, located in the sandy Ténéré desert, we consider 367 it likely that the earthquake ruptured to the top of the consolidated bedrock, but that the 368 deformation signal is subsequently blanketed by overlying less consolidated sandstones, 369 less able to sustain coseismic rupture. this earthquake give a hypocentre -the point of rupture initiation. In contrast, geodetic 386 data like that used here has no capacity to constrain the earthquake initiation, or its rup-387 ture process, in time, as the displacement seen in the interferograms is the result of the 388 complete earthquake rupture. In this case, we are unable to solve robustly for a causative 389 fault plane from the InSAR data, but even if we could, the earthquake hypocentre could 390 still lie anywhere on that rupture plane. For larger earthquakes, with rupture lengths 391 of > 5 km, this can pose additional location problems. However, for a small-magnitude 392 event like the 2017 Ténéré earthquake, where the rupture length is likely to be < 5 km, 393 this discrepancy between the seismological hypocentre and the geodetic fault rupture will 394 be small, compared to the uncertainties in seismological location.

395
Seismological locations are subject to uncertainty in the solid-Earth velocity structure 396 along the full ray path from source to receiver. In the case of the locations shown in Figure   397 3 and 4, the relative travel-time di↵erence between all the locations shown is < 0.5 s 398 for regional arrivals and < 0.2 s for teleseismic arrivals. As demonstrated in Figure 2, 399 the majority of arrivals are emergent, and picking a precise onset is usually subject to 400 uncertainties on at least this magnitude. This is then compounded by the variation in we are unable to use the InSAR data to quantitatively estimate a comparable geodetic 414 magnitude, we note that that surface displacement wavelength of the deformation imaged 415 using InSAR is perhaps longer than would be expected, particularly at the lower end of the 416 range of m b estimates. As the InSAR deformation field captures all deformation between 417 the two acquisition dates, we cannot rule out the possibility that the displacement seen is 418 enhanced by some level of aseismic process. However, this would be rare for an earthquake 419 of this magnitude.

420
The consideration of both InSAR and seismological data for small magnitude earth-tory. Geophysics 39 (5) Figure 1). Left panels show 3 vertical component velocity waveforms, filtered between 1 -3 Hz, from 3-component instruments RAYN, MORC, and LODK. Vertical red line shows the predicted P -wave arrival, based on the NEIC location. Lower panels show data from three small-aperture seismic arrays (Bucovina, Mount Meron, and Makanchi), again filtered between 1 -3 Hz. Top panel shows the beamformed waveform, based on the NEIC location. Lower panels show sweeps through slowness and azimuth space (c.f. Davies et al, 1971), with colour indicating array coherence using the F -statistic (e.g. Blandford, 1974). White lines show the predicted slowness and azimuth for P -wave arrivals from the Ténéré earthquake.      Figure 3. Contours show 95% interval ellipses determined using di↵erent seismic arrival subsets, as described in Figure 4.