GPR signature of Quaternary faulting: A study from the Mt. Pollino 1 region, southern Apennines, Italy. 2

. With the aim of unveiling evidence of Late Quaternary faulting, a series of ground penetrating radar (GPR) 12 profiles were acquired across the southern portion of the Fosso della Valle-Campotenese normal fault (VCT) located 13 at the Campotenese continental basin (Mt. Pollino region), in the southern Apennines active extensional belt (Italy). 14 A set of forty-nine 300 MHz and 500 MHz GPR profiles, traced nearly perpendicular to this normal fault, were 15 acquired and carefully processed through a customized workflow. The data interpretation allowed us to reconstruct a 16 pseudo-3D model depicting the boundary between the Mesozoic bedrock and the sedimentary fill of the basin, which 17 were in close proximity to the fault. Once reviewing and defining the GPR signature of faulting, we interpret near- 18 surface alluvial and colluvial sediments dislocated by a set of conjugate (W- and E-dipping) discontinuities that 19 penetrate inside the underlying Triassic dolostones. Close to the contact between the continental deposits and the 20 bedrock, some buried scarps which offset wedge-shaped deposits are interpreted as coseismic ruptures, subsequently 21 sealed by later deposits. Our pseudo-3D GPR dataset represented a good trade-off between a dense 3D-GPR volume 22 and conventional 2D data, which normally requires a higher degree of subjectivity during the interpretation. We have 23 so reconstructed a reliable subsurface fault pattern, discriminating master faults and a series of secondary splays. This 24 contribution better characterizes active Quaternary faults in an area which falls within the Pollino seismic gap and is 25 considered prone to severe surface faulting. Our results encourage further research at the study site, whilst we advise 26 our workflow ideal also for similar regions characterized by high seismic hazard and scarcity of near-surface 27 geophysical data.


Introduction 32
A "seismic gap" is an area surrounded by regions struck by large earthquakes in historical or recent times. Such 33 earthquake-free areas are characterized by the presence of seismogenic faults, whose past activity or possible 34 quiescence is inferred on the basis of morpho-structural and/or paleoseismological data. The "seismic gaps" (McCann 35 et al., 1979) show an apparent lack of historical seismicity but are candidate regions for the occurrence of large 36 earthquakes in the near future (Mogi 1979;Plafker and Galloway 1989;Cinti et al., 1997;Galadini and Galli, 2003). characterized by Late Quaternary continental syn-tectonic sedimentation (Fig. 1a-c). 25 The paleoseismological trenching and radiocarbon dating document in the region the occurrence of paleo-earthquakes 26 with 6.5 < Mw < 7.0 and a recurrence time interval of ~ 1200 years (Cinti et al., 1997(Cinti et al., , 2002(Cinti et al., , 2015aMichetti et al., 27 1997Michetti et al., 27 , 2000. But this high magnitude interval contrasts with the historical seismicity records, reporting a single 28 significant Mw 5.2 event occurred in 1693 (Tertulliani and Cucci, 2014). In the last three decade's instrumental 29 seismicity recorded only two moderate seismic sequences climaxed in the Mw 5. 6 Mercure (1998 The Campotenese basin and its VCT boundary fault is an example that summarizes the aforementioned issues: 1) lack 3 of availability of paleoseismological data as the basin is entirely located within the Mt. Pollino National Park, thus 4 requiring prior authorization from authorities; 2) lack of availability of publically accessible geophysical data; 3) no 5 fresh recent surface displacements within the Holocene deposits have been observed along its trace. For all these 6 reasons, the VCT represents an ideal case study suitable to test our working method. 7 We have conducted an explorative GPR field campaign across a VCT sector, suggested by discontinuous and smooth 8 geomorphic scarps, as a screening tool for the definition of its possible Quaternary displacement history. The 9 objectives of the paper are to: i) review and describe geophysical characteristics associated with a peculiar GPR 10 signature of faulting, and propose a reference methodological workflow; ii) specifically check the efficiency of GPR 11 prospecting to locate the VCT fault and to depict its subsurface pattern and spatial continuity at shallow depth; iii) 12 provide new data to eventually relate the occurrence of Mw > 6.0 seismic events; iv) pave the way for other local 13 geophysical studies and identify interesting sites for future ground-truthing and/or paleoseismological trenching; v) to 14 have direct application and impact to the planning of future mitigation strategies for the reduction of surface faulting 15 risk in the nearby urbanized areas. 16

Tectonic setting and seismicity 17
The Campotenese continental basin is located in the northernmost Calabria region south-west of the Mt. Pollino 18 calcareous massif (southern Italy, Fig. 1 in the case of a normal fault); ii) reflection packages thickening as they approach the fault strands; iii) abrupt lateral 34 dip variation of the reflections; iv) peculiar reflection package geometries, with contorted reflection patterns 35 resembling "colluvial wedges", which McCalpin (2009) defines as deposit due to "subsidence and sedimentation of 36 the hangingwall and erosion of the morphological scarp in the footwall"; v) localized strong GPR signal attenuation 37 due to the presence of conductive media within the main fault zone. 38 Based on the research and criteria reviewed above, we carried out a near-surface interpretation of faulting based on 39 the co-existence of most of these features along several adjacents analyzed GPR profiles. These conditions strengthen 40 the interpretation of each profile and aids to highlight the spatial continuity of the interpreted structures over linear 1 distances of at least many tens, or hundreds, of meters. 2

GPR and GNSS survey 3
Three different geophysical field campaigns carried-out during the 2014-2015 years, a dataset of 49 GPR profiles was 4 acquired in the southern sector of the ROCS across the VCT fault segment (Fig. 1b-c), covering a buffer zone of ~ 5 400 m and ~ 200 m respectively along and across the fault strike (area of ~ 8 Ha), for a total linear length of GPR 6 profile about 4100 m collected using a Common Offset (CO) configuration (Fig. 2). 7

FIGURE 2 HERE 8
We used a Zond 12e GPR system equipped with 300 and 500 MHz antennae. The lower frequency antennae was 9 ultimately preferred and considered the best trade-off between maximum resolution and achievable signal penetration 10 (in our case ~ 4 m) concerning the surveyed materials and wanted subsurface structures. The GPR was equipped with 11 an odometer wheel to measure the radar profiles' length and with a Topcon GR-5 Global Navigation Satellite System 12 (GNSS) receiver to achieve accurate positioning of GPR traces and profile. Considering the scarce presence of 13 obstacles across the survey site and the good satellite coverage, we opted for a Network Real-Time Kinematic 14 positioning (NRTK, connected to the NETGEO network), measuring coordinates and elevations with centimetre 15 accuracy, and stored directly within the SEG-Y GPR files. 16 Three datasets were acquired after preliminary fieldwork and collection of geological structural data at the surface and 17 which allowed us to infer the possible location of the fault trace. The average NE-SW direction of the GPR lines was 18 initially planned with the primary purpose of intersecting the VCT fault perpendicularly to its SW-NE strike, as 19 reported by literature and visible by surface evidence. This solution theoretically allows a more reliable interpretation 20 of the investigated structure by reducing the effect of the apparent dip-direction and dip-angle of both stratifications 21 and faults. 22 The acquisitions carried out in 2014, first resulted in twelve SW-NE GPR profiles collected in the southern sector of 23 the basin (CMT light-blue lines in Fig. 2a), which was a flat land characterized by Quaternary alluvium. The second 24 acquisition encompassed four additional radar profiles collected in the same area, and another nine radar profiles 25 progressively moving to north, which were collected with slightly different and converging orientations in the central 26 sector (CMT green lines Fig. 2a). This solution was pursued for two main reasons: 1) to avoid directly surveying the 27 outcropping dolostones (only partially crossed with two northernmost profiles) characterizing two hills h1 and h2 28 (dashed white polygons in Fig. 2), and thus focussing only on the sedimentary cover which is our target for possible 29 Quaternary faulting; 2) to optimize, through a preliminary GPR data interpretation, the future acquisition schemes by 30 figuring out the dip direction of the buried geologic structures of interest. In fact, similarly to the interpretation of 31 reflection seismic profiles, the "apparent dip" of reflections in bidimentional radar profiles should be considered to 32 achieve a reliable 3D conceptual model.  Fig. 2a, north "n" and south "s") were extended in NNE-SSW and NE-SW directions, respectively, 1 crossing h1 for several tens of meters (max profile length ~220 m) throughout the basin. The GPR profiles were 2 recorded using a trace step of 0.05 m and a profile inter-distance of 25 m for dataset "n" and 10 m for dataset "s", 3 respectively. A detailed summary of the acquisition parameters used for the GPR surveys is reported in Table I. For  4 these two datasets, the profile spacing and positioning are more regular and accurate, thanks to a preparatory transects 5 planning using a GIS project. Thus, we later staked out their initial and final positions during the fieldwork through 6 the differential Global Navigation Satellite System (GNSS). The results of the accurate GPR traces positioning 7 achieved during the GNSS campaigns were also later used for GPR data processing, visualization, and interpretation. 8

GPR data processing and results: 10
The processing sequence was customized after testing several workflows and parameters. We aimed to remove random 11 and coherent (e.g. ringing) noise and enhance the data quality to better visualize the geometry of the buried reflections 12 and their discontinuity in signal amplitude and phase. The first step was an accurate Quality Control (QC) of the 13 profile coordinates and topographic profiles. Although the generally favorable environmental conditions (e.g. good 14 satellite coverage, no forested areas etc..) of the site for a GNSS survey, some measurements were occasionally 15 suffered a degradation of positional accuracy (e.g. temporary scarce satellite coverage or poor communication via 16 Network Transport of RTCM via Internet Protocol -Ntrip). For some traces therefore the coordinates and elevation 17 field records that were outliers (Fig. 3a) were corrected using various strategies (e.g. replacement, interpolation, or 18 smoothing, Fig. 3b). 19 profiles were comparable. Although the metre resolution of the DTM is unable to represent centimetre topographic 25 variations, the comparison confirmed an excellent match of the topographic profiles at a meter scale, so that the DTM 26 data were integrated to correct the GNSS measured topography when the accuracy of GNSS recordings were 27 excessively degraded. With the topographic profiles corrected, the raw GPR data (Fig. 3c, illustrating the profile 28 cmt5s) were initially processed with the Prism software (Radar System, Inc., http://www.radsys.lv/en/index/) using a 29 basic processing sequence, to analyze the main characteristics of data and optimize a customized processing flow. The 30 processing sequence was later improved through ReflexW software (https://www.sandmeier-geo.de/reflexw.html, see 31 Table II for details on the processing algorithms and parameters). The workflow included a time-zero correction, 32 dewow, amplitude recovery, velocity analysis, background removal, bandpass filtering, F-K filtering, 2D time 33 migration, topographic correction, and time-to-depth conversion. The amplitude recovery was operated through a 34 "gain function" including a linear and an exponential coefficient (g(t) = (1+a*t)*e (b*t) ) to enhance the amplitude 35 (reflectivity) contrasts as well as preserving the horizontal and vertical amplitude variations already visible in the raw 36 data (Fig. 3a). This amplitude recovery function was used across all the profiles with slight customization depending 37 on the datasets (details in table II). The entire processing flow was applied to all the available radar profiles, again 1 with occasional filtering adaptations aiming to remove local pervasive signal ringing (e.g. due to low antennae-ground 2 coupling). Particular care was dedicated to the migration process, whose algorithm was decided after extensive tests 3 on several radar profiles to select the best migration strategy. 4   TABLE 2  The workflow therefore, suggests a challenging imaging task due to velocity variation happening not only in depth as 25 well as laterally across the different media. This sharp change of reflectivity and velocity at a distance of about 13-14 26 m (Fig. 4d) represents a complex problem for the efficiency of 1D migration algorithms standardly used for GPR 27 imaging. Such considerations has lead testing a 2D migration algorithm, by creating and using a 2D velocity model 28 obtained for each radar profile through a hyperbolic diffraction fitting tool (Fig. 5a). Single velocity points have been 29 fitted for each area displaying hyperbolic diffractions, while in the remaining parts of the radar profiles we have 30 arbitrarily included presumed velocity adaptation only to obtain a regular grid of points to spatially interpolate the 2D 31 models. The 2D migrated radar profiles, in comparison to the 1D approach, resulted in improved imaging of GPR 32 profiles, displaying a more accurate collapse of the hyperbolic diffractions into point sources and an improved 33 relocation of dipping reflections, with a refinement of their geometry and an increase of their continuity. A good-34 quality imaging result is visible on the central sectors of radar profiles displaying strong reflectivity and reflections 35 with improved continuity, but also many phase breaks and displacements. Despite steep topographic gradients, sharp 36 lateral velocity variation and the reflection heterogeneity might cause imaging issues that need to be treated using . We believe we have reached a good compromise for our purposes. In our case, a considerable improvement, 1 can be seen along the hill-slope and flatter areas (profile cmt1n_a, Fig. 5b) which are of greatest interest for the study 2 aimed at detecting possible earthquake ruptures within the Quaternary deposits. The improved imaging of reflection 3 geometries is therefore fundamental for the interpretation and detection of geophysical signatures of faults. 4

FIGURE 5 HERE 5
A successive import of the processed SEG-Y data was done into the seismic interpretation software OpendTect Pro 6 v.6.4 (Academic license courtesy of dGB Earth Science, https://www.dgbes.com), which was used first for global 7 quality control of processing operations (correctness of topographic correction and datum plane, coordinates accuracy 8 and matching, profiles orientation and intersection) and for three-dimensional (3D) visualization of all the profiles 9 ( Fig. 6a). The three-dimensional GPR project was subsequently integrated with geological and structural maps, DTM, sticks picked on displaced reflections and correlated across adjacent radar profiles. In particular, we used the "surface 14 geometry" tool to extract the properties of each single mesh building up the surfaces, and obtaining the "dip" and "dip 15 azimuth" data. Subsequently, such values have been automatically saved in an attribute table, which can then be 16 queried to reconstruct the "synthetic" stereonets. 17 18

GPR data description and interpretation 19
The 3D MOVE project allowed us to extract 2D and 3D data visualizations to better figure out the relationships 20 between the main reflections identified on the different GPR profiles (Fig. 6a). The workflow aimed to reconstruct 21 and model the three-dimensional surfaces including both horizons and high-angle discontinuities. 22 higher frequency content zone in the 2D spectrum of Fig. 6c). However, its reflection pattern is not spatially 29 homogenous, being characterized by oblique and sub-parallel reflections. The latter are interpretable as dolostone beds 30 of moderate (25-30°) W and E "apparent" dip on the respective sides of the surveyed dolostone hills. In addition, these 31 reflections are frequently cut and slightly displaced by apparent high-angle (60-65°) phase discontinuities, highlighted 32 by a dense hyperbolic diffractions pattern (radar profile cmt2n, Fig. 7a), suggesting intense fracturing and little faults 33 displacing the dolostone (Fig. 7b). This radar signature was recorded not only in correspondence of the outcropping 34 carbonate but also in the transition slope areas covered just by a thin soil layer (Figs. 7b,c). In the southern side of h1, 35 an outcrop with thin microbialitic laminae allows one to measure the attitude of the bedding (NNW dip, ~ 30-35° dip 36 angle) as well as two sets of major and minor joints (SW and SE dip and dip angle of ~ 40-45°, respectively) fitting 1 with GPR reflections. 2 Apart from its internal heterogeneities, the GPR signature of the Triassic dolostones can be considered as a well-3 defined depositional facies (fc1) (Sangree and Widmier, 1979;Huggenberger, 1993;Beres et al., 1999;Jol and 4 Bristow, 2003). A different radar signature fc2 is defined for the profile sectors on the sides of fc1. This second facies 5 is characterized by prominent laterally-continuous and sub-parallel reflections in the very shallow depth range (< 1 m, 6 just beneath the direct arrivals), stratigraphically sealing underlying reflections 1-3 m deep: the latter are more 7 discontinuous, wavy, and contorted, with moderate to low reflectivity and encompassing sparse diffraction hyperbolas 8 (in unmigrated data, Fig. 7a). This reflection pattern onlaps onto a generally prominent wavy reflection (Fig. 7a,b), 9 which typically marks the transition to strong signal attenuation deeper in the section. 10 The reflection package belonging to fc2 corresponds to the alluvial/colluvial deposits ( Fig. 7b-d), outcropping on the 12 sectors with flat topography, which represent the GPR profile sectors we've carefully inspected to find for geophysical 13 evidence of Quaternary faulting. A key-layer for this research is the described prominent, wavy reflection, as it can 14 be recognized in many radar profiles. The related interpretation is not straightforward in the absence of direct data 15 (e.g. boreholes and/or paleoseismological trenches) or at least without additional geophysical data. A strong GPR 16 reflection suggests significant variation of the dielectric constant between the two media so that most of the incident 17 energy is reflected back to the receiver at the surface. This wave behaviour is potentially explained by several resting on a fractured substratum. Its top surface is progressively deepening towards the W, thus providing increased 33 space for settling sediments and thus a gradual thickening of deposits is observed from E to W. 34 In light of the above considerations, we interpret the prominent, wavy GPR reflection as a buried top layer of carbonate 35 fault splays to the West associated to the VTC segment. Thus, analyzing the geophysical characteristics of the 2 prominent, wavy reflection in terms of a structural interpretation, the main peculiar characteristic is the clear "stepped" 3 geometry of some sectors (Figs. 5b, 6b, 7b, 8b), namely breaks of its continuity associated to lateral sharp variations 4 of depth (linked to sediment growth and onlaps). We also notice other geophysical features, which can be observed in 5 the stratigraphy of overlying deposits (fc2): some reflections are semi-continuous to discontinous (sharp variation in 6 signal amplitude and phase) and display evident lateral variation of the dip angle. 7

FIGURE 8 HERE 8
These broken reflection packages present truncantions (e.g. Smith and Jol, 1995), displacements, and hyperbolic 9 diffraction events (insets of Fig. 8b1,b2). Such peculiar GPR signature is therefore compatible with coseismic 10 displacement due to Late Quaternary surface faulting events (Fig. 8b). Contorted reflections across the main 11 discontinuities frequently show localized strong attenuation of GPR signal (Fig. 8b). The attenuation might be linked 12 to their high dip-angle, causing a minor amount of energy being reflected back to the antenna, but, more likely, due to 13 the presence of conductive fine soils nearby faulted zones (e.g. circle 1 in Fig. 8b) Fig. 8b). Using all such stratigraphic evidence and geophysical markers of faulting, we have therefore 17 interpreted and classified synthetic (W-dipping, blue) and antithetic (E-dipping, red) normal faulting events (Fig. 8b). that show a good degree of continuity from north to south (Fig. 9). For the studied sector of the VCT, we have 31 reconstructed the tridimensional fault-network and the geometry of the associated synsedimentary deposits at a metric 32 scale of observation (Fig. 9). 33 34 5. Discussion 1

Inferences from subsurface 3D model 2
The perspective view of Fig. 9a shows a 3D structural scheme of the main tectonic lineaments at the basin scale 3 displaying a NW-SE faults strike (modified after Brozzetti et al., 2017a) in relation to the GPR investigated area (white 4 rectangle). Our GPR interpretation enriches many of the details such a former structural scheme across the southern 5 VCT segment. We highlight an en-echelon system of two main SW and NE-dipping faults as well an articulated set 6 of extensional meso-faults within the Quaternary sediments. The high-angle GPR discontinuities identified in the 7 study (e.g., Fig. 9b) show a considerable continuity in the NW-SE direction (accurate 3D structural recontruction in 8 Fig. 9c), dissecting not only Quaternary alluvial-colluvial deposits (except for the very shallow fc2 layers), but also 9 deeper stratigraphic layers. 10 The reconstructed faults mark a horst-graben structure, mostly buried within the Campotenese basin, which locally 11 emerges from the Quaternary deposits. In the investigated area it corresponds to a NNW-SSE elongated topographic 12 high (h1 and h2 in Fig. 2a) Fig. 8b), separated by a right step-over of about 0.5 km from the segment that borders the eastern basin (Figs. 2c, 8a). 19 Thus, also the fault-set d3 and d4 located on the eastern part of h1 and h2, can be hierarchically classified as a network 20 of minor splays embedded in the southern junction zone between the two VCT segments (Fig. 9c). 21 The three-dimensional model (Fig. 9a,c) highlights that these faults, despite having a typical Apenninic NW-SE trend length, area and depth), which are often difficult to assess. These scale relations can also be applied also to Quaternary 33 scarps originated by cumulative coseismic faulting produced by medium-strong earthquakes (generally M > 6). 34 Nevertheless, only through paleoseismological analysis, by sampling and dating the stratigraphy at different levels, is 35 it possible to date and distinguishing the amount of slip of each seismic event. But in cases like the VCT, the GPR 36 data assume a key-value since provided key fault parameters where no direct information on the nature of the surveyed 37 deposits and no accurate dating is available. Our GPR interpretation by itself doesn't allow one to extract any date for 1 a single earthquake, nor identify a succession of past seismic events and neither establish recurrence times (Galli, 2 2020  GPR survey did not require special authorizations and was relatively fast and low cost. The pseudo-3D configuration 41 was an efficient compromise between spatial coverage and duration of the data acquisition (four days of fieldwork). 1 On the other hand, the data processing was non-trivial, requiring about six months overall to set up an optimized 2 workflow, due to challenging data characteristics, such as the steep and rugged topography and the sharp lateral 3

variations of dielectrical properties of media (Triassic Dolostones vs Quaternary deposits). 4
Our structural reconstruction derived by GPR data interpretation shows several sets of sub-vertical discontinuities 5 within the near-surface (~ 1-4 m depth), which we interpreted as a pattern of extensional surface faulting. Such faults 6 are bounding small local "graben or semi-graben-like" structures, which cut an hypothesized Holocene age clastic 7 cover and underlying Triassic dolostones. We have also identified some chaotic and laterally discontinuous GPR-8 stratigraphic facies, interpreted as near-fault post-earthquake deposits (i.e. colluvial wedges ?). These shallow 9 structures suggest the possibility that surface faulting due to past strong earthquakes (6 < Mw < 7) occurred in relatively 10 recent times in the study area. Its traces at surface were possibly later levelled by the concurrent natural processes of 11 erosion, aggradation and, anthropogenic activities. As our results confirms the presence of seismic potential and thus 12 the possible occurrence of a large earthquake in the future, we wish the primary effect of our study to be one of raising 13 the level of attention regarding the seismic hazard in the Campotenese area, as well as prompting further research. 14 Upon ground truthing, our work may represent a preparatory study for further geophysical surveys (3D GPR and other 15 methods), as well as direct analysis including trenching, drilling, sampling campaigns and dating (e.g., luminescence, 16 radiocarbon, etc). Although a further multidisciplinary approach would be necessary to achieve a quantitative (i.e. slip 17 rates and recurrence times) assessment of the seismogenic potential of the study area, we firmly promote, particularly 18 where near-surface data is lacking, a widespread use of the presented GPR workflow on other seismic gaps worldwide. 19 20 Author contributions. ME and DC contributed equally to this work as first authors. ME, DC, CP, FB led the fieldworks. 21 ME analyzed, processed the GPR and GNSS data. ME, DC, CP, HMJ, FB contributed to the paper conceptualization 22 and writing. ME and DC managed all data in the GIS environment and within 3D interpretation programs (OpendTect,  23 Move), as well as they have created all the figures. DC realized the final 3D structural-geological model through Move 24 software. All authors reviewed and edited all the drafts. 25

26
Competing interests. The authors declare that they have no conflict of interest. 27 28 Disclaimer. Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in 29 published maps and institutional affiliations. 30 31 Special issue statement. This article is part of the special issue "Tools, data and models for 3-D seismotectonics: 32 Italy as a key natural laboratory". 33

Acknowledgments 34
We sincerely thank Leonardo Speziali, Prof. Costanzo Federico, and Roberto Volpe for their support during the field 35 operations, as well as Khayal Gavarof for his massive and valuable collaboration in data organization and processing.         which collapsed the hyperbolic diffractions (white arrows) and restored reliable reflection geometry. 30   Azimuth ranging between N 235-245° and N 062-072° for the W-dipping and E-dipping normal faults, 14 respectively. Vertical exaggeration is 2. 15 Table 1: Main information and GPR parameters used during the data collection (* the time window was 16 adapted depending on the surveyed area). 17 Table 2: Customized flow and details of the parameters used during the processing of the GPR dataset. 18