Near-field fault slip of the 2016 Vettore Mw 6.6 earthquake (Central Italy) measured using low-cost GNSS

The temporal evolution of slip on surface ruptures during an earthquake is important for assessing fault displacement, defining seismic hazard and for predicting ground motion. However, measurements of near-field surface displacement at high temporal resolution are elusive. We present a novel record of near-field co-seismic displacement, measured with 1-second temporal resolution during the 30th October 2016 Mw 6.6 Vettore earthquake (Central Italy), using low-cost Global Navigation Satellite System (GNSS) receivers located in the footwall and hangingwall of the Mt. Vettore - Mt. Bove fault system, close to new surface ruptures. We observe a clear temporal and spatial link between our near-field record and InSAR, far-field GPS data, regional measurements from the Italian Strong Motion and National Seismic networks, and field measurements of surface ruptures. Comparison of these datasets illustrates that the observed surface ruptures are the propagation of slip from depth on a surface rupturing (i.e. capable) fault array, as a direct and immediate response to the 30th October earthquake. Large near-field displacement ceased within 6–8 seconds of the origin time, implying that shaking induced gravitational processes were not the primary driving mechanism. We demonstrate that low-cost GNSS is an accurate monitoring tool when installed as custom-made, short-baseline networks.


GNSS results from the 30 th October earthquake
We recorded positional data using custom-made GNSS units that are designed around a low-cost U-blox NEO-M8T GNSS receiver chip, and built to be compact, robust, weatherproof, and power-efficient. The GNSS chips allow GNSS constellations to be tracked using a single frequency over the L1 band and provide observation data of pseudo-range and carrier phase at rates of up to 10 Hz. These single-frequency observations allow the c 60 cm d b a 13       position of one receiver relative to another to be continually resolved with centimetre spatial accuracy for baselines up to approximately 10 km 12,13 . The temporal accuracy is sub-microsecond 14 , as GNSS is fundamentally dependant on very precise timing. The GNSS units were assembled with the purpose to monitor surface deformation resulting from earthquakes, but the underlying technology could be applied to many potential monitoring applications within the geosciences, such as volcanic deformation, mass wasting processes, glacial motion or anthropogenic ground movements.
Following the 24 th August Amatrice earthquake we deployed four GNSS units along the Mt. Vettore fault as two footwall-hangingwall pairs, to create two baselines with lengths of 1,286 m and 1,870 m ( Fig. 1, Table 1). The along-strike separation of these two baselines is 6.2 km. To eliminate the potential for gravitationally-induced effects in the displacement signal, the locations of the GNSS units were chosen on stable bedrock. The two hangingwall units, MV02 and MV04, are located on a hill within the basin below MV01, and on the Castelluccio plain, respectively. All units recorded positional data at 1 Hz continually to a data card from the time they were installed; there is no triggering mechanism to capture particular seismic events. We deployed the units with the intent to measure the near-field post-seismic deformation of the Mt. Vettore fault following the Amatrice earthquake; by chance, both our baselines were located on a section of the Mt. Vettore fault which subsequently ruptured during the 30 th October earthquake. The duration of the data presented here is 1 minute, starting from 06:40:00 UTC, that covers 18 seconds preceding the earthquake and 42 seconds following it (Fig. 2a,b, Supplementary Table S1). The data describe the displacement of the hangingwall GNSS unit relative to the corresponding footwall unit as three components: horizontal east-west, horizontal north-south, and vertical up-down. The resolved horizontal displacement ("horizontal throw") is derived from the two horizontal components, and corresponds to the incremental relative displacement maps shown as insets in Fig. 2a Data from both GNSS pairs show an absence of pre-seismic deformation, as the hangingwall is stationary relative to the footwall for the time preceding the sudden onset of co-seismic displacement. The initiation of relative displacement occurs between 06:40:20 and 06:40:21 for baseline MV01-MV02 and 06:40:21 and 06:40:22 for baseline MV03-MV04. The data then indicate rapid south-westward and downward displacement of the hangingwall relative to the footwall over the following 2-4 seconds. We infer that the data recorded within this period reflect a composite signal of increasing co-seismic displacement together with superimposed high frequency ground motions associated with differential shaking. We estimate rise time (the duration between onset and cessation of co-seismic fault displacement) to be ca. 2-4 s for both baselines assuming that co-seismic displacement on the fault surface ended when all three components of displacement reached their finite offset values (Fig. 2). The 1 Hz temporal frequency of the displacement record effectively limits the resoultion of this approximation to integer seconds. Estimated slip velocities are 0.13-0.26 ms −1 and 0.26-0.52 ms −1 for baselines MV01-MV02 and MV03-MV04 respectively. Following this short period of large, rapid displacement, the data then indicate only very slight relative movement for the remaining 35-37 seconds. We interpret small fluctuations in displacement during this latter period as the remaining effects of differential ground shaking, as they do not produce further finite displacement. Table 1 summarises our interpreted values of finite displacement during this event from the two hangingwall-footwall pairs. The finite displacement observed at baseline MV03-MV04 close to the along-strike centre of the rupture trace exceeds that observed along baseline MV01-MV02 located towards the north-eastern tip of the rupture trace. Furthermore, a footwall fault splay behind site MV01 ruptured during the October 30th earthquake. This rupture was not visited on foot, but could be seen from across the valley, and it likely accommodated a portion of the total slip, which is not represented by the GNSS data on baseline MV01-MV02.

Comparison of GNSS and other datasets
Co-seismic offsets derived from the regional GPS network 9 (Fig. 1a)      We visited the GNSS sites and inspected nearby scarp surfaces of the Mt. Vettore fault on 28-29th October (i.e. prior to the M w 6.6 event), and again in the two days following the 30th October earthquake. Our field observations four hours after the earthquake confirmed finite surface rupture displacements had intersected both GNSS baselines (Fig. 1b,c) and that significant slip had occurred on an antithetic fault system to the west of MV02 (Fig. 1d). The measured total displacement of ruptures intersecting baseline MV03-MV04 are similar to the GNSS displacements (−1.21 m downward and 0.44 m horizontal), while those rupture displacements intersecting baseline MV01-MV02 were far less than the corresponding GNSS displacements (−0.06 downward and 0.05 horizontal). These two comparisons are highly informative as they provide a unique insight into two typical situations encountered when comparing near-field displacements to finite surface slip. Firstly, the case of baseline MV03-MV04 where the tectonic system is relatively simple and a good agreement exists between near-field displacement and finite surface displacement. Secondly, the case of MV01-MV02, where the tectonic system is complex and disparities exist between near-field displacement and finite surface displacement that in this case is due to the presence of multiple closely spaced faults with differing orientations and slip vector azimuths.

Timing of ground rupture
Data from the Italian National Seismic Network 5 record the hypocentre origin time for the 30 th October earthquake as 06:40:17.36 ±0.02 UTC. The accelerometer IT.CLO at Castelluccio village ( Fig. 1), installed as part of the Italian Strong Motion Network 15 , recorded an arrival time of 06:40:20.24 UTC, with a duration of 8.86-9.71 seconds 16 (Fig. 2c). The initiation of acceleration at IT.CLO preceded the rupture-induced co-seismic displacement at our GNSS baselines by no more than 1-2 s. At IT.CLO the peak ground acceleration in the north-south and up-down components occurred at approximately 06:40:23.50, and the east-west component is also close to its maximum value at that time which is near-synchronous with the interpreted cessation of co-seismic displacement at each of our baselines (Fig. 2c).
By comparing the timing of key events across different datasets, we can make the following observations on the evolution of the co-seismic sequence. Surface and near-surface rupture initiated almost immediately (1-2 s) after the onset of shaking at the Castelluccio accelerometer IT.CLO, and significant rupture ceased within five seconds or less after the onset of shaking at IT.CLO. Most, if not all, of the GNSS measured co-seismic displacement occurred prior to the time of peak ground acceleration. Finite displacement across both GNSS baselines was achieved rapidly, within ca. six to eight seconds of the hypocentre origin time (Fig. 2c), demonstrating that the observed surface ruptures are the result of dynamic earthquake slip.
The timing and magnitude of the co-seismic displacements measured using GNSS in the near-field show that the observed surface ruptures are part of a capable fault array on which the surface rupture developed as a direct and immediate response to the propagation of slip from depth during the 30th October earthquake. The implication of this unique observation is that measurements of surface slip obtained through earthquake geology and palaeoseismology can reliably inform physical models of earthquakes where rupture mechanics are required and to aid interpretation of the palaeoseismic record. Alternative mechanisms for the partial or complete generation of some of the surface displacements of this earthquake sequence have been proposed, involving shaking induced shallow landslides and deep seated gravitational slope deformation on a large scale [17][18][19][20][21] . Such mechanisms are very unlikely to be the cause of the displacements we have observed, as landslides and large-scale slope deformation are known to occur progressively over typical time periods of 45-220 seconds 22,23 , whereas we have shown that finite co-seismic displacement in the near-field occurred rapidly, within six to eight seconds of the hypocentre origin time.
The use of low-cost GNSS receivers located in close proximity to capable faults can significantly enhance our understanding of co-seismic processes and are complementary to existing far-field networks. There is additional scope for further technical development and improved deployment of GNSS receivers, and we are currently testing a number of future enhancements. To provide greater insight into earthquake propagation and surface rupture processes we need to increase the temporal resolution of co-seismic measurements, and are already now using a higher GNSS recording frequency of 10 Hz in other active tectonics projects. Similarly, we are improving spatial resolution by deploying a larger number of receivers in closer proximity to recent surface ruptures, to help investigate spatial variability in near-field displacements during future earthquakes. Remotely operated units that require minimal maintenance will significantly reduce deployment costs.

Methods
The low-cost GNSS units consist of a single frequency Ublox NEO-M8T GNSS receiver chip with a patch antenna and a 4 GB data logger using universal asynchronous receiver/transmitter (UART). The GNSS chip is configured to output observation data packets of approximately 600 bytes, consisting of code pseudorange, carrier phase, Doppler and signal-to-noise ratios to the data logger at 1 Hz. The packet size is dependent on the number of satellites currently in view at each receiver and the GNSS constellations enabled, however typical data rates at 1 Hz are of the order of 50 MB/day. Each unit is rated at approximately 0.7 W and is powered by a 12 volt 55 Ah lead acid battery which provides power for approximately 40 days. There is no 'earthquake detection' as part of the system design; the units log data continuously until power is depleted or the memory of the logger is full. Hence this configuration requires no maintenance or interaction duration operation other than to change the battery and to retrieve data from the data logger. Additional batteries can be installed in parallel to lengthen this maintenance interval. We are developing future versions of these units which will record dual frequency observation data at 10 Hz and be entirely self-sufficient in terms of power and data transfer.
The units were already installed prior to the 30 th October earthquake as part of a longer-term project to monitor seismicity following the 24th August Amatrice earthquake. Following retrieval of data from the GNSS units the recorded observation data from each hangingwall receiver and its footwall counterpart was post-processed using static relative positioning to calculate the position of the hangingwall receiver relative to its footwall counterpart with time. The initial position of the footwall receiver of each footwall-hangingwall pair was calculated using Precise Point Positioning (PPP). The recorded observation data were supplemented with navigation, ionospheric, precise orbit and clock data from the International GNSS service (www.igs.org).

Data Availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.