This study conducted some simple data processing to identify potential anomalies in Rn concentrations related to seismic activity. In particular, some simple elaborations were adopted, which were similar to those that allowed other researchers to identify significant evidence of correlations between gas anomalies and seismic activity in tectonically active areas [2, 21, 45, 46]. The frequency of Rn activity for both springs is described by the typical Gaussian distribution (Fig. 3). Hence, the recorded fluctuations (whether temporal, diurnal or seasonal) are within the range included in M ± 2σ, where M and σ are the mean and standard deviations of the time series, respectively, [47]. Instead, anomalies in Rn concentrations are defined as significant deviations from the mean value; specifically, it is commonly assumed that signals related to earthquakes fall outside the so-called 2-σ confidence interval [48].
Based on these studies, the mean values and the standard deviations of Rn activity, which are useful for the detection of outliers, have been calculated for the SG (SG0, Fig. 4) and for the SRF (Fig. 5). The acquired time series are shown in Figs. 4 and 5 with three-hour moving averages, which were adopted to eliminate data noise. The mean values and the ± σ and ± 2σ thresholds are also displayed in graphs by solid and dashed lines, respectively. These graphs also show seismic events of Mw ≥ 3.5 (taken from the previously described database).
In Fig. 4, SG0 represents the entire SG time series recorded between April 2017 and December 2019. Since this series is uneven, the sequence has been separated into four sections (SG1–SG4 in Fig. 4) to improve the accuracy of the processing. No intervals exceeding the ± 2σ thresholds in relation to the seismicity of the study area are observed. However, decomposing the SRF time series is not necessary, as the data show a relatively limited variation from the mean value of about 5000 Bq/m3 (Fig. 5). Additionally, in this case, as for the SG, there are no intervals exceeding the ± 2σ thresholds in relation to seismic activity. The values exceeding ± 2σ are due to different conditions: (1) at the SG, they are related to sensor emersions above the water surface (Fig. 4), while (2) at the SRF, they are represented by the initial values of the time series that are presumably attributable to another condition (before the start of monitoring), whose course is not known (Fig. 5).
Additionally, a different data processing procedure was considered. According to Dobrovolsky et al. [49], two parameters must be treated together to calculate the strain radius (R in km) of the effective precursory manifestation zone: the magnitude of the earthquake (Mw) and the distance between the epicentre and the measuring site (D in km). In this way, these authors defined a relationship to identify the interactions between gas-geochemical and seismic signals using the following empirical equation:
$$R = 10^{0.43 \times M}$$
(1)
where R is the strain radius of the precursory manifestation zone and M is the moment magnitude (Mw).
The conceptual basis of Eq. (1) is that an approximately circular region around the epicentre of the earthquake should undergo elastic crustal deformation prior to earthquakes [50]. Therefore, precursory signals were expected for events where R ≥ D. Considering two earthquakes of the same magnitude, a closer seismic event affects the Rn activity in the monitoring site more significantly than a distant one.
Equation (1) was applied to Mw ≥ 3.5 earthquakes that occurred at the two study sites (between April 2017 and December 2019). Epicentral distances from the two sites were also determined (Fig. 6). This processing allowed the identification of five seismic events for the SG and six seismic events for the SRF, in which a possible Rn response was expected in terms of geochemical precursor signals.
The response could not be verified for three of the five seismic events identified at the SG due to the lack of Rn data, while no interaction was observed for the four Montecilfone seismic events at the SRF. The non-interaction between the gas content and the Montecilfone earthquakes has a geodynamic explanation; these seismic events occurred in a different plate (i.e. in the Adria plate subducting towards the SW beneath the Apennines) from the Apennine belt where the monitoring sites are located [51]. Therefore, it is not surprising that the Montecilfone seismic sequence did not cause anomalies in the inner sector of the Apennines.
Overall, three seismic events produced expected Rn activity responses as presented in Fig. 7. Specifically, the Mw 3.8 L’Aquila earthquake, which occurred on March 31st, 2018, was detected at the SG; the Mw 4.4 Balsorano earthquake, which occurred on 7 November 2019, produced responses at both springs; and the Mw 3.9 San Leucio del Sannio earthquake, which occurred on 16 December 2019, produced a response at the SRF. These two latter events present comparable R and D values at the SRF. For this reason, they fall on the straight line (Fig. 6, SRF) that separates the expected interaction area (R ≥ D) from the unexpected interaction area (R < D). Specifically, an increase in Rn concentration (which lasted for about 12 h) of ≈ 1200 Bq/m3 (7% with respect to the average value) and another one of ≈ 1500 Bq/m3 (8% with respect to the average value) were observed at the SG seven and 11 days before the L’Aquila and the Balsorano earthquakes, respectively (the black arrow in the SG, Fig. 7A, B). For the latter earthquake, a decrease of approximately 2000 Bq/m3 (after the peak) was also observed. Another increase in Rn concentration of ≈ 500 Bq/m3 (10% with respect to the average value, which lasted about six hours) was recorded at the SRF about 17 days before the Balsorano earthquake (the black arrow on the left in the SRF, Fig. 7C). A different Rn behaviour was identified at the SRF before the San Leucio del Sannio earthquake. The preparation phase of this seismic event is illustrated by a clear increasing trend during the month of November of ≈ 1000 Bq/m3 (20% with respect to the average value: the black arrow at the SRF on the right, Fig. 7). In all the observed cases, it can be inferred that changes in the strain field before seismic activity may have affected gas and fluid migration, causing an increase in Rn content ranging from 7 to 20% with respect to (background) pre-anomaly values. Therefore, the recorded pre-seismic signals could be explained as the result of dilation and/or contraction of fracture systems that led to changes in the gas flow behaviour; however, it is believed that a more accurate analysis of time lags between the Rn signals and the occurrence of earthquakes is necessary, as the identified interactions are too limited in number to clarify this aspect. Additionally, processes of dilation and/or contraction are expected to drive some changes in the chemical content of groundwater, whose intensity would be inversely correlated with the groundwater resource volume in the corresponding aquifers due to different dilution rates. Furthermore, some transient and evident perturbations in Rn concentrations before the San Leucio del Sannio earthquake were detected, but these were attributed to the users’ field interventions. The sampling frequency of 10 min allowed the acquisition of accurate data through which it was possible to measure the duration of the anomalies until the signals returned to stable background values.
Considering the anomalies recorded on the Rn time series at the two sites, the correlation between the selected events (for which a response was expected) and the hydrogeochemical time series acquired in the SG and SRF areas was investigated. Indeed, the monthly sampling of some selected springs in the two areas made it possible to detect some anomalies in springs with deeper input [12, 14]. Box-and-whiskers statistical analyses performed on the hydrogeochemical data of these springs revealed anomalous values (outliers) in Li, V, Cr and Cs concentrations in the Raiano spring (Fig. 2A) before the Balsorano earthquake, for which anomalies in Rn content were recorded at both stations (i.e. the SG and SRF in Fig. 7).
The recorded variations and trends identified in the hydrogeochemical time series of the Rn-monitored SG and SRF did not show clear relationships with seismic events. Nevertheless, a potential response could exist that is represented by the behaviour of some elements and ions, such as Cl, SO4 and Fe, and parameters such as electrical conductivity and temperature at the SG, where limited changes were recorded. Moreover, the sampling frequency of the water chemistry (every month) did not allow comparisons with the Rn time series (one measurement every 10 min). Therefore, the availability of longer time series data is essential for the correct evaluation of the variations in the acquired series.