Identifying Deep Seismogenic Sources in Southern Piedmont (North-Western Italy) via the New Tool TESLA for Microseismicity Analysis

Abstract

The analysis of earthquake source mechanisms is key for seismotectonic studies, but it is often limited to traditional methods plagued with issues of precision and automation. This is particularly true in low-seismicity areas with deep and/or hidden seismogenic sources, where the identification of precise source mechanisms is a difficult and non-trivial task. In this study, we present a detailed application of TESLA (Tool for automatic Earthquake low-frequency Spectral Level estimAtion), a novel tool designed to overcome these limitations. We demonstrated TESLA’s effectiveness in defining source mechanism analysis by applying it to seismic sequences that occurred near Asti (AT), in the Monferrato area (Southern Piedmont, Italy). Our analysis reveals that the observed clusters consist of two distinct seismic sequences, occurring in 1991 and 2012, which were activated by the same seismogenic source. We relocated a total of 36 events with magnitudes ranging from 1.1 to 3.7, using a 3D velocity model, and computed 12 well-constrained focal mechanism solutions using the first motion polarities and the low-frequency spectral level ratios. The results highlight a relatively small seismogenic source located at approximately 5 km north of Asti (AT), at a depth of between 10 and 25 km, trending SW–NE with strike-slip kinematics. A smaller cluster of three events shows an activation of a different fault segment at around 60 km of depth, also showing strike-slip kinematics. These findings are in good agreement with the regional stress field acting in the Monferrato area and support the use of investigation tools such as TESLA for microseismicity analysis.

1. Introduction

Understanding and characterizing active fault structures is critically important for a thorough and accurate seismic hazard assessment, as they represent potential seismogenic sources of future earthquakes. Identifying such sources and assessing their potential is therefore crucial for earthquake preparedness and risk mitigation [1,2,3,4,5]. Detecting deep and active faults can be challenging when using conventional methodologies, such as seismic (reflection or refraction) profiling, magnetotelluric surveys, gravimetric measurements, and surface geological observations. In many cases, seismic data are limited to shallow depths (<10–15 km); wide-angle reflection seismic surveys, necessary for imaging the deep crust, are often unavailable due to their high cost and the time required for organization. Additionally, magnetotelluric and gravimetric methods may lack the resolution needed to clearly image deep fault structures. Geological field evidence may be lacking when faults are buried beneath thick sedimentary cover, with no clear surface expression or outcrops [6]. This is particularly evident in regions where fault activity is low or where faults are deeply buried beneath thick sedimentary units. In such seismotectonic contexts, the reliance on conventional seismic catalogues and analysis and/or geological surveys usually fails to provide the resolution needed for detailed fault mapping. This challenge becomes even more intricate when dealing with complex seismogenic structures [7,8,9,10].

Earthquakes typically occur as a result of cyclical stress release when stress accumulation along fault planes exceeds the rock strength, resulting in a sudden rupture and energy release, either as discrete events or via seismic sequences or swarms ([11,12] and references therein). Especially in complex fault systems, the temporal and spatial evolution of seismicity reflects the interplay of local stress conditions and the mechanical properties of the fractures/faults, as observed in multi-scale phenomena ranging from localized rock bursts to large tectonic earthquakes [13].

To face this problem, microseismicity analysis, meaning earthquakes typically with magnitudes below 3, emerges as a key tool for unveiling these otherwise undetectable seismogenic structures, offering an alternative or complementary perspective to conventional approaches [14,15,16]. Although individual microearthquakes release minimal energy and are rarely felt, their high occurrence rates and spatial distribution offer valuable insights into fault geometries and seismogenic structures to define the active stress field [17,18,19]. Low-energy earthquake locations and their fault plane solutions can identify geometrical attributes of seismogenic faults, reveal fault interactions, and track the time evolution of the seismic cycle over time. Moreover, in low-seismically active areas, microseismic analysis may constitute the only available evidence of the ongoing rupture processes. In particular, focal mechanisms are a valuable tool for characterizing earthquake sources, offering insights into the geometry and kinematics of faulting. Parameters such as strike, dip, and rake describe the orientation and the slip along the fault plane that generated the seismic event. Additionally, the analysis of focal mechanism solutions allows the inference of the regional stress field.

However, despite advances in seismic network coverage and data processing techniques, determining reliable focal mechanisms for low-magnitude events remains a major challenge [20]. A main difficulty in microseismicity analysis is often the limited quantity and quality of data, due to the high noise content, which can hinder the application of traditional P-wave polarity inversion methods [21,22,23]. In this study, we apply a specific and innovative technique for analyzing microseismicity. Our goal is to demonstrate that, through the use of advanced analytical tools, microseismic data can provide crucial and otherwise unobtainable information about active faults, otherwise not detectable. Specifically, we employ TESLA (Tool for automatic Earthquake low-frequency Spectral Level estimAtion [19]), a recently developed tool designed to estimate low-frequency amplitude spectral levels of both P- and S-waves. By adopting a hybrid approach that combines P-wave polarity data with spectral level ratios, we are able to complete the computation of more robust focal mechanisms, particularly for low-magnitude events.

We apply this method to study a small area to the north of Asti, in the Monferrato region (Southern Piedmont, Italy), where the low-energy seismicity and the scarcity of well-constrained focal mechanisms have hindered a comprehensive identification of the deep and tectonically active seismogenic sources (Figure 1). We focused on a small seismic cluster of 36 earthquakes, with local magnitudes ($M_L$) ranging from 1.1 to 3.7. We demonstrate that TESLA is a tool that can significantly refine microseismicity analysis and focal mechanism solutions, ultimately contributing to a more accurate understanding of active crustal deformation, as undertaken in this study, where our results suggest the presence of a previously not-constrained seismogenic source.

2. Seismotectonic Context of the Southern Piedmont

The complex tectonic framework of the Southern Piedmont is a result of the tectonic process that involved the Alps and Apennines, as well as their crustal-scale juxtaposition. This regional complexity is part of the broader, highly intricate geodynamic framework of the northwestern Italian peninsula that experienced the N–S to NNE–SSW convergence of the European and African plates [2,24,25]. This convergence led to a continental collision throughout the Tertiary Period, closing the Alpine Tethys Ocean and causing the Adria microplate to rotate counterclockwise. Consequently, a dipping continental subduction zone formed, with the European lower crust emplaced beneath the Adria microplate. An ongoing debate persists concerning the European slab geometry (detached versus continuous), partly attributed to observed variations in lower crustal structures. However, recent studies [26,27,28] provide new insights, revealing strong lateral variations in the 3D crustal structure across the arc, transitioning from a preserved slab in the south to a smooth crustal root in the north. The resulting highly accurate Moho map indicates an eastward deepening of the European Moho, which is consistent with a subducted slab beneath the Adria upper plate. In the southwest Piedmont region, Moho steps reorient perpendicularly to the belt, shifting from west to east as the Alpine arc transitions into the deeper (>45 km) Adria continental crust. Further east, the European Moho plunges even further, reaching depths exceeding 75 km [26,29,30].

The current stress field in the Western Alps is highly complex, displaying local variations and reflecting the interplay of major plate interactions and diverse local geological structures [25,27]. The complex stress field results primarily in strike-slip faults, along with extensional forces in the internal Western Alps and compressional forces at the belt’s edges. This aligns with geodetic data showing a notable counterclockwise rotation in southeastern Adria and slow (<1 mm/yr) horizontal movements across the Western Alps.

Regarding historical seismicity in the Monferrato region, a very low occurrence of seismicity is found. Catalogues dating back to the year 1000 show only three seismic earthquakes with an estimated maximum magnitude compatible with 5.0 [31], as can be seen in Figure 1. Regarding instrumental seismicity, the Western Alps are the most seismically active region, predominantly characterized by frequent and low-energy events ($M_L$ < 3.0). The epicentral distribution of this seismicity correlates with the primary geological and geophysical features, forming two distinct arc-shaped structures extending northward. The first is the western Briançonnais arc [32] that follows the Apenninic Front Line, with earthquake depths largely confined to the uppermost 12 km. Conversely, the second structure, the eastern Piedmont arc [32], located at the Alps–Po Plain boundary, aligns with the western edge of the “Ivrea Body” anomaly [33,34] and it exhibits slightly deeper seismicity, typically ranging from 8 to 20 km. The Ivrea Body is a high-density crust–mantle structure located beneath the Western Alps, characterized by strong seismic, gravity, and magnetic anomalies [35]. It is bounded by the Insubric Line, a major fault marking the tectonic boundary between the Adriatic plate and the Alpine orogenic wedge. The Ivrea Body is interpreted as part of a mantle wedge that acts as an indenter in the collisional system [21,27]. Notably, the Monferrato area is situated at the center-south of the Piedmont region, between Liguria and the Po Plain. This region serves as a crucial transition zone between the Alps and the Apennines. The Monferrato region typically exhibits low seismic activity, mainly concentrated at the contact between the Apennines and the Po Plain. Despite this, it has notably experienced some of the largest earthquakes recorded in Northern Italy over the past few decades, with magnitudes ($M_d$) of between 4.0 and 4.8 (occurring in 2000, 2001, and 2003).

Earthquake hypocenters in this area are generally confined to the uppermost 20 km of depth. The seismogenic sources are hypothesized to be associated with the main pre-Burdigalian tectonic structures of the Tertiary Piedmont Basin and linked to the larger deformation of the Apenninic chain [36,37]. This includes the pervasive basement involvement characterized by thrust-related structures and even potential NW-verging structures beneath the Western Alps. Additionally, a NW–SE-oriented fold, featuring a short limb directed towards the NNE, is observed [38] buried under the Po Plain, corresponding to a Northern Apennines structure. The Monferrato area in Southern Piedmont is characterized by a thick Tertiary Piedmont Basin sedimentary succession that overlies the Ligurian and the Sub-Ligurian basement [38,39]. This complex geological framework showcases significant basement involvement, featuring reactivated extensional Triassic–Jurassic faults and structures like the Monferrato Arch. Mapped across the provinces of Turin and Alessandria [37,38,39], this arch itself displays thrusting and detachments through thrust-related structures, affecting both Mesozoic carbonates and the Cenozoic succession. Since the complexity of this low-defined structural setting, the seismogenic activity of these specific structures is not yet fully understood or unequivocally linked to the specific seismic activity.

Different seismic sources from the active thrust faults appear associated with the August 2000 ($M_d$ 4.8) and July 2001 ($M_d$ 4.0) seismic sequences that occurred in the Monferrato region. These sources exhibit distinct depths and kinematic characteristics. Improved earthquake location efforts reveal that this seismic activity is primarily organized within a SW–NE-oriented volume at depths ranging from 8 to 20 km, involving the underlying basement units. This alignment is further supported by analyses of focal mechanisms, where most solutions show a strike-slip faulting component [36].

Our understanding of seismogenic sources in the Monferrato areas is notably limited. While larger, unconstrained structures are reported to the north [37,38,39], specific sources for the Monferrato have not been found, to our knowledge. This evidences an urgent necessity for high-resolution studies to accurately define fault structures and assess their seismic potential in the region. This endeavor is not only complicated by the region’s inherently complex geology but also by the intrinsic challenge in identifying the seismogenic structures, particularly because they can be trapped or buried at depth by the thick sedimentary units that form and conceal the Tertiary Piedmont Basin.

3. Materials and Methods

To maximize data availability and ensure the best comprehensive analysis, we merged and cross-referenced seismic events data from catalogues produced by three distinct seismic networks, which operated at different times in the region and featured varying station densities: the Istituto Nazionale di Geofisica e Vulcanologia (INGV) [40] catalogue, the Regional Seismic Network of Northern Italy (RSNI) [41] catalogue, and the Catalogo delle Localizzazioni ASSolute (CLASS) [42]. This was conducted to enhance and complete our database in terms of seismic events and stations, as well as to extend our analysis further back in time. Each of the three catalogues provides not only hypocentral and magnitude parameters, but also the arrival times of P- and S-waves for the seismic stations that recorded the events. These arrival times formed the core dataset for this study, as they are essential for accurate hypocenter relocation.

By merging and cross-referencing the events from these three catalogues, we focused on the Monferrato area, where a small seismic cluster of 36 earthquakes, with local magnitudes ($M_L$) ranging from 1.1 to 3.7 occurred. All events took place to the north of Asti, an area where no known seismogenic structures are indicated and no detailed seismological analyses are available. Further analysis of the temporal distribution within this cluster revealed that it is composed of two distinct seismic sequences, occurring in 1991 and 2012–2013, respectively.

For the microseismicity analysis, we used the novel tool TESLA, developed by Adinolfi et al. [19], which automatically calculates P- and S-wave low-frequency spectral levels. These can be inverted alongside P-phase polarities to more effectively constrain focal mechanisms or estimate seismic moment. TESLA achieves this by systematically exploring various signal windows for computing displacement spectra and then evaluating the optimal displacement spectra through quantitative criteria to estimate low-frequency spectral levels, as explained in [19]. This adaptive approach is key to optimizing spectral analysis, particularly for microseismicity.

Analyzing P- or S-wave amplitudes in the time domain becomes significantly challenging when dealing with hundreds of earthquakes from dense seismic networks. It is thus of primary importance to develop data analysis tools that are robust, automatic, and suitable for measuring seismological observables. TESLA is a fast, configurable, and automatic tool that is capable of processing large datasets without extensive manual operations from the analysts. In this work, we computed 67 P/S-wave low-frequency spectral level ratios using TESLA. These levels serve as additional observables that, alongside P-wave polarities, help to better constrain low-magnitude focal mechanisms. Furthermore, the low-frequency spectral levels calculated by TESLA can be used to estimate the seismic moment and moment magnitude.

For the seismic analysis, we defined a specific workflow that is shown in Figure 2, and which we explain as follows.

• Database Integration and Development. The P- and S-wave arrival times from the three catalogues (RSNI, CLASS, and INGV) were merged into a unified database. To ensure consistency and maximize the quality of the dataset, a priority scheme was applied; phase picks from the RSNI catalogue were given precedence, as this catalogue generally catches more events and provides more constrained locations due to the higher station density close to the Monferrato area, followed by those from CLASS, and finally INGV. We preferred to use this order due to the number of available events and the uncertainty associated with the arrival time reading of the seismic phases.
  To improve the quality of the S-wave picks, which are generally more difficult to identify accurately as they occur on the P-wave coda, we applied a quality check procedure based on the modified Wadati relation [43]. We calculated differential arrival times between station pairs and applied Huber regression to estimate a robust slope, which corresponds to the V_p/V_s ratio for the study region. This ratio was determined to be 1.70. Outliers that did not align with the fitted regression line were automatically traced back to the corresponding S-phase arrivals, which were then removed from the dataset. This allowed us to retain the most reliable arrival information for each seismic station, resulting in an average number of 19 P-wave arrivals and 12 S-wave arrivals for each event.
• Earthquake Location. Following the quality process, we computed absolute event locations using the program NonLinLoc [44], using a 3D local velocity model [33]. The use of [33] 3D velocity model enhances earthquake location accuracy by accounting for complex subsurface velocity variations, as occur in the Monferrato area, which standard 1D models cannot capture [28,45]. The relocation procedure employed a total of 53 seismic stations, the closest ones shown in Figure 1, from various local and national networks, providing a robust geometric constraint for the hypocentral solutions that minimize the azimuthal gap.
• Seismological Observables for Focal Mechanism. An automated polarity picking algorithm was applied to the same waveforms dataset to extract clear P-wave polarities, based on the identification of the first motion. TESLA was employed to automatically compute the low-frequency spectral levels of both P- and S-waves. Seismic waveforms for the events were then downloaded via the INGV FDSN web service, using data from available seismic stations. The Fourier analysis was constrained in the range of frequencies between 0.5 and 40 Hz by exploring waveform time windows with lengths from a minimum of 0.3 s to a maximum of 1.6 s. This range was selected to fully contain the expected rupture duration of the events, while allowing TESLA to explore multiple window lengths and positions around the phase arrival. As described in [19], unlike approaches that require a predefined time window, TESLA automatically selects the most appropriate window based on waveform complexity and the quality of the spectral fit. The selection process is driven by quantitative criteria—such as signal-to-noise ratio and spectral misfit—and ensures statistically valid and consistent results. TESLA was able to successfully complete 213 source spectra, out of which 95 corresponded to P-waves and 118 corresponded to S-waves, for events with magnitudes ranging from $M_L$ 1.6 to 3.3, showing good results throughout the magnitude range (Figure 3). The amplitude spectrum is modeled by fitting the seismic source model proposed by Brune (1970) [46] and revised by Boatwright (1980) [47]. This fitting procedure infers the parameters Ω₀, $f_c$, and Q using the Levenberg–Marquardt algorithm [48], a robust iterative method for nonlinear least-squares problems. TESLA provides estimates of the fit parameters, their uncertainties, and the goodness of fit. Prediction accuracy is measured by the Mean Absolute Percentage Error (MAPE), which is scale-independent and allows comparison across differently scaled spectral data. Only spectra with an MAPE below 60% are selected to ensure reliable results. Additional information can be found in Adinolfi et al. (2023) [19].
• Focal Mechanism Computation. By combining these two observables, P-wave polarities and spectral level ratios, we performed a joint inversion using the BISTROP algorithm [23]. Representative examples of the focal mechanism fits obtained with BISTROP are presented in Figure 4.
  The inversion was performed using a Bayesian framework, where the search begins with a uniform prior distribution over the strike, dip, and rake parameter space. The algorithm performs a full grid search, evaluating the probability for all combinations with a 2° step in strike, dip, and rake. The focal mechanism solution corresponds to the Maximum A Posteriori (MAP) probability solution. To assess the quality of our focal mechanism solutions, we evaluated the distribution of Kagan Angles (KA) [49], defined as the smallest rotation angle needed to align the principal axes of one solution with those of another, for all solutions computed by BISTROP with an a posteriori probability exceeding 90%, relative to the MAP solution. As demonstrated in Adinolfi et al., 2022 [20], this threshold offers a practical balance between confidence and resolution, allowing the comparison to focus on the most probable and well-constrained solutions. The median of the KA distribution provided a measure of the consistency of the solution set, reflecting how tightly clustered the acceptable solutions are around the selected mechanism. We considered focal mechanisms with a KA median below 30° to be well-constrained [20] because of the high level of confidence in their reliability.
• Seismotectonic Interpretations. In the final step, we interpreted the focal mechanism solutions within the broader tectonic framework of the study area. We analyzed the number, depth distribution, and spatial arrangement of the earthquake locations to identify and characterize distinct seismogenic sources. Kinematic styles (e.g., strike-slip, normal, reverse), as well as fault geometries and orientations, were inferred from the focal mechanism parameters and compared across the obtained solutions. These observations were then integrated with regional stress field data and existing geological knowledge to refine the tectonic interpretation and evaluate the potential role of newly identified fault segments within the broader seismotectonic context.

4. Results and Discussion

Our catalogue, selected within the Asti area, consists of 36 relocated events, characterized by an average RMS residual of 0.28 s, and with mean horizontal and vertical uncertainties close to 1.5 km. Over the 36 seismic events, waveform data were available for 27, corresponding to the events belonging to the 2012–2013 sequence. We successfully computed 12 high-quality fault plane solutions by combining P-wave polarity readings with P/S low-frequency spectral level ratio analysis. These events correspond to local magnitudes ranging from 1.8 to 3.3 $M_L$. No solutions were obtained for events with magnitudes below 1.8 $M_L$ due to the low signal-to-noise ratio and limited availability of high-quality waveforms, which is typical for lower-magnitude events. Additionally, no solutions were derived for the 1991 seismic sequence due to network limitations at the time and the unavailability of waveform data. Starting from the early 2000s, both the permanent national network and the local seismic networks in NW Italy experienced an increase in the number of stations and benefited from technological advancements, greatly improving data coverage and quality.

Using both the spatial distribution of earthquakes and the similarity of their solutions, we separate the events into Cluster 1 and Cluster 2 (Figure 5 and Figure 6). For most events the polarity observations across stations were coherent, with an average of 8 polarity readings and five spectral level ratios contributing to each inversion. Spectral level fits are well adjusted for all events across the magnitude range. The solutions are well constrained, since the KA value median result is below 25°, and in some cases as low as a few degrees, indicating a high consistency among the different solutions obtained for each event, as we can see in Figure 5. All the computed focal mechanisms suggest predominantly strike-slip faulting.

We can observe specific features in the focal mechanism solutions of the two clusters (Figure 5). Cluster 1 is given by strike-slip solutions with a small reverse component and north-facing compressional quadrants. The solutions of Custer 2 correspond to strike-slip solutions with a small extensional dip-slip component with north-facing extensional quadrants. Locations show (Figure 6) that the two clusters are spatially separated by about 12 km in the map, with Cluster 2 being located to the north of Asti, and at depths between 10 and 25 km. Cluster 1 is located to the north-west of Asti, at depths ranging between 50 and 60 km, thus suggesting the activation of a deeper fault segment. In addition to their spatial separation, the two clusters also differ temporally. Cluster 2 includes the 1991 sequence and events that occurred between September 2012 and January 2013. In contrast, the three events of Cluster 1 are spread over a two-year period, with occurrences in April 2012, May 2013, and December 2013.

The obtained focal mechanism solutions exhibit a high degree of coherence among the fault plane solutions, reinforcing the reliability of the applied methodology. Moreover, the results are in good agreement with the regional stress field inferred from previous studies [25,27,50], resulting mainly in strike-slip faulting in the Southern Piedmont area.

High-resolution insights into the spatial dimension and geometry of Cluster 2 reveal a small seismogenic source located at a depth of approximately 10–25 km, oriented SW–NE with strike-slip kinematics. The hypocentral alignment matches the NE–SW nodal plane observed in the focal mechanism solutions. The length of the fault corresponds to a few kilometers, with a maximum analyzed magnitude of 3.7. This particular structure is consistent with a blind fault, previously neither evidenced nor described, and it is located approximately 5 km to the north of Asti.

Well-constrained focal mechanism solutions consistently corroborate the prevailing strike-slip motion. This type of source does not appear to be associated with the thrust faults characteristic of the Po Plain subsurface, such as the Monferrato Arch. The hypocentral depths, extending down to 20 km, suggest active structures affecting the basement within the upper crust. These characteristics, including orientation, depth, and kinematics, strongly resemble the seismic sequences studied by Massa et al. [36]. In particular, these seismic sequences, occurring in August 2000 and July 2001, were located 10–15 km southeast of Asti at depths ranging from 8 to 20 km, showing strike-slip faulting along mainly a SW–NE alignment.

The recognition of two distinct seismic sequences (1991 and 2012) provides evidence supporting the reactivation of the same fault system. Specifically, concerning Cluster 2, this source appears to undergo repeated reactivations over time, as evidenced by the recurrence of microseismicity over time, with nine events with a maximum magnitude of 3.7 in 1991 and 26 events in 2012–2013 with a maximum magnitude of 3.3.

Cluster 1, comprising only three events, is located at a considerably greater depth of approximately 60 km. This cluster also exhibits strike-slip kinematics, though its dip-slip component appears compressive, indicating a slight divergence from Cluster 2 solutions. Due to the limited number of events (three), full interpretations are not possible. Crucially, however, it is evident that these events do not belong to the source activated by Cluster 2. Their much deeper occurrence suggests that such strike-slip earthquakes can affect the deep crust, which extends down to 60 km in the area investigated. The kinematics inferred from the focal mechanisms of both Cluster 1 and Cluster 2 align well with the faulting type and orientation inferred from the regional stress field in the Monferrato area, a consistency supported by the existing scientific literature [25,30,50]. In fact, the area deformation is dominated by a strike-slip regime. These sources are in good agreement with strike-slip deformation typical of areas linked to the convergence between two plates, such as is the case for the European and Adriatic plates’ collision, with consequent crustal thickening. It is noteworthy that in this area, the Moho can reach depths between 40–60 km [33], justifying the deep recorded seismicity.

Regarding the TESLA performance, we underline that its approach offers several key advantages that enhance the efficiency and reliability of focal mechanism determination. First, TESLA is designed as a fast, configurable, and fully automated system for processing extensive seismic data, minimizing the need for manual work. For example, it computed 14 P- and 11 S-wave spectra for 17 seismic stations in just 40 s (AMD Ryzen 9 5980HS CPU @ 3.30 GHz × 8 cores, 32 GB RAM, Ubuntu 22.04.2 LTS). This automation not only saves time but also minimizes human-induced errors, making it highly convenient for rapid processing of large datasets. Moreover, TESLA’s flexible configuration allows easy adjustment of processing parameters, tailoring it to the specific characteristics of the data.

A common challenge when working with microseismicity is the high noise content within the signals. TESLA addressed this by comparing signal and noise amplitudes in the frequency domain, allowing the automatic exclusion of time windows heavily affected by noise. It compares the displacement spectrum of the signal window to that of an automatically selected pre-P noise window of equal length. Users set a minimum amplitude threshold and specify the frequency range that the signal amplitude must be above. Signal windows with SNR below this threshold are discarded.

The integration of spectral level ratios alongside P-wave polarity readings offers a significant methodological advantage for focal mechanism computation. Spectral level ratios, as additional seismological observables, provide further constraints on fault plane orientation, which is especially critical when analyzing low-magnitude events with limited waveform data. The advantage of this data integration is a higher quantity of overall data, allowing us to compute focal mechanisms that would not have had a minimum amount of P-polarities available in order to attempt an inversion. An additional test was performed by computing focal mechanisms using only P-wave polarities. This approach yielded 11 solutions of different types (ranging from normal to strike-slip faulting), with one fault plane solution lost (

Table 1. Summary of focal mechanism input data and median Kagan angles (KA) for the 12 earthquakes analyzed. The table includes event origin time, magnitude, number of polarities used, and number of low-frequency spectral level ratios computed. KA values are reported both for solutions obtained using only polarities and for those obtained by combining polarities with spectral ratios. A dash (“–”) indicates that no solution was obtained using that input, while “– *” indicates that only one focal mechanism exceeded the 90% posterior probability threshold, making the computation of a median Kagan angle not applicable.

Origin Time	Magnitude	Number of Polarities	KA (Polarities)	Number of Spectral Level Ratios	KA (Polarities and Spectral Ratios)
2012-04-18 13:05:20	2.8	11	31°	8	4°
2012-10-07 06:40:10	2.7	9	25°	5	6°
2012-10-07 07:58:49	2.7	10	38°	3	17°
2012-10-10 00:35:36	2.0	3	–	4	5°
2012-10-14 20:49:39	2.4	6	57°	4	7°
2012-10-29 02:32:36	2.4	7	48°	5	10°
2012-11-20 01:33:57	2.0	6	26°	1	13°
2012-11-20 10:32:13	3.3	13	8°	5	6°
2012-11-20 16:17:23	2.5	10	83°	5	– *
2012-11-22 10:27:09	2.2	6	57°	4	6°
2013-05-07 16:53:29	1.8	5	62°	4	9°
2013-12-20 20:02:44	2.3	9	77°	10	– *

). Only three of them showed KA below 30°, with some exceeding 75°, indicating poorly constrained results. These findings highlight that the incorporation of spectral level ratios significantly enhances the quality and stability of the solutions, even when polarity data are limited. To date, no focal mechanism solutions existed for the analyzed seismic sequence, highlighting the originality and significance of our results in advancing the seismotectonic understanding of the region, as well as from a methodological perspective, thanks to the application of TESLA.

As shown in Figure 2, our method integrates several earthquake catalogues to increase P- and S-wave picks and uses a 3D velocity model with P/S low-frequency spectral level ratios and P-wave polarities to improve fault plane solution accuracy. This approach helps identify and characterize deeply buried seismogenic sources. Moreover, it is important to note that the main limitation of the amount of spectral level ratios obtained in this study was related to the number of time signals available for the analysis. As expected, events with higher magnitudes were generally recorded by more stations, resulting in a much higher number of detected signals. Nevertheless, TESLA demonstrated a consistent performance across the entire magnitude range analyzed. For the available data, the tool successfully extracted reliable spectral ratios even for lower-magnitude events, showing its potential for microseismic analysis in regions with an appropriate dense seismic network coverage.

5. Conclusions

This study demonstrates the value of analyzing microseismicity with advanced spectral techniques to improve our understanding of seismotectonic processes in low-seismicity regions. By applying the TESLA tool to a dataset of low-magnitude earthquakes in Southern Piedmont, we successfully identified two seismicity clusters of strike-slip focal mechanisms consistent with the faulting style inferred from the regional stress field. Our results confirm the presence of a previously unrecognized, seismically active fault system beneath the Asti area, with evidence of distinct fault segments at different depths. This finding can have significant implications for the tectonic interpretation of the region and a more detailed seismic hazard assessment.

Moreover, the implementation of TESLA, together with BISTROP, revealed clear advantages in computing focal mechanism solutions. The hybrid approach—combining P-wave polarity data with spectral level ratios—enabled the retrieval of reliable solutions for a low-magnitude earthquake dataset. These results validate TESLA as an automatic, effective, and robust method for microseismicity analysis, particularly in settings such as this, where traditional techniques often fail due to the limited availability of good-quality signals.

The application of advanced methodologies, such as spectral amplitude analysis and Bayesian inversion, to the 1991, 2012 Asti seismic sequences, demonstrates the necessity of moving beyond classical polarity-based techniques. These methodological approaches by using advanced tools enhance our capacity to analyze microseismicity and to detect and characterize active faults, especially in regions like the Southern Piedmont, where surface fault expressions are absent or ambiguous. As demonstrated in this study, such methods can play a valuable role in supporting earthquake hazard assessment and risk mitigation strategies, as they provide novel information on the seismogenic sources of the investigated area.