1. Introduction
Non-Aqueous Phase Liquids, or NAPLs, are organic liquid contaminants that are poorly soluble in water such as oil, gasoline, and petroleum products. NAPLs tend to contaminate soil and groundwaters for very long periods of time and need to be removed with active (e.g., pump-and-treat, soil vapour extraction) or passive (natural attenuation) remedial strategies [
1]. The efficiency of both approaches relies on environmental monitoring [
2] and on a suitable geochemical tracer of ongoing transport, dilution, degradation, and volatilization phenomena [
3,
4]. Radon gas can be used to identify NAPLs in soil and groundwater because of its preferential partitioning in organic phases [
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23]. The resulting radon deficit (regarding the site-specific radon background concentration in soil or groundwater) can be utilized as an NAPL indicator [
17].
Moreover, it is possible to estimate the saturation of the residual NAPLs either in the vadose or saturated portion of an aquifer. NAPL saturation (
SNAPL) in the saturated aquifer, on top of the “NAPL source zone”, or downgradient in the plume, can be calculated using the following equation [
17]:
where:
ΔC∞: radon deficit = equilibrium radon concentration in NAPL polluted aquifer/equilibrium radon concentration in clean aquifer.
KNAPL/W: radon partition coefficient between NAPL and water at Roma average temperature (70, [
17,
22])
It is worth noting that this equation, Equation (7) in ref [
17], is affected by a typo in the numerator where “Δ
S∞” is reported in place of “Δ
C∞”.
Although many works applied this approach to estimate the amount of the residual NAPLs in soil and groundwater (see for example, [
8,
17,
24,
25]), to our knowledge no references are available for a statistical treatment of radon and NAPL concentration, in terms of principal component analysis. This research is a first example of the potential of such a kind of approach to infer processes accounting for relationships among relevant parameters (radon, NAPLs, water table fluctuations) to successfully manage the reclamation of a contaminated area. Since groundwater monitoring and sampling is routinely carried out once every two–three months by companies responsible for remediation actions, it is particularly useful to have a tool to establish relations among variables and understand the evolution of contamination with standard sampling times.
Two study sites, with different hydrogeological features, age of the spillage, composition of residual NAPLs, and clean-up procedures are employed here to show the advantages and drawbacks of this method. The factor analysis is here applied only to groundwater and not soil variables’ data, which were not available with the same frequency; however, the same approach can be extended to soil contamination in case of a larger dataset.
5. Discussion
The groundwater radon concentrations in sites 1 and 2 were used to estimate average values for all monitoring wells over the study period. Radon data from clean wells of site 1 (those with higher radon levels and nearly no NAPLs, PZ03, PZ07, PZ14, PZ25, and PZ26) were averaged and used as a reference for calculating the radon deficit of dirty wells, as the ratio of the radon concentration in NAPL polluted aquifer and the equilibrium radon concentration in clean groundwater. The latter (119 ± 11 Bq L
−1) was compared and verified with a corresponding value (
222Rn
C, 129 ± 10 Bq L
−1) calculated from the average
226Ra specific activity of the Alban Hills Hydrogeological Unit (A
Ra, 85 Bq kg
−1, see
Table 3), according to Equation (2), using a mineral density (r,) of 2.65 kg/L, a porosity (n) of 0.35, and an emanation radon efficiency (E) of 0.2 [
23]:
Since the two values agree within the error range, it was possible to determine with good approximation the NAPL saturation of contaminated wells in site 1 according to Equation (1) and show areas with higher saturation in a schematic map (
Figure 5a). The well PZ13 and the adjacent area where the residual source zone is presumably located were characterized by the highest NAPL saturation (2.39%), while the wells located downgradient (PZ02, PZ04, PZ09, and PZ19) showed a saturation around or above 1%, apart from PZ19. PZ19 well was included in the plume zone, even if its saturation was much lower (0.42), because its apparent low saturation is calculated from a radon concentration affected by the proximity of PZ19 to a recharge well. Actually, the introduction of water treated with activated carbon filters absorb, not only the NAPLs, but also radon, which may not have time to reach the full equilibrium with its parent
226Ra in the rock aquifer (low residence time).
Other wells surrounding the main plume or working as pumping wells were less polluted (PZ06, PZ8, PZ17, PZ18, and PZ19). The plume was elongated in the direction of the groundwater flow (NNW-ESE) and was perfectly retained in the fueling station area. Since the monitoring of the water table elevation and the groundwater quality was not continuous but periodical, it was difficult to demonstrate direct relationships between the rainfall distribution and the depth of the groundwater table on one side and the dissolved NAPL on the other. Using factorial analysis can help clarify these interactions.
Moreover, a recent study [
22] carried out in a nearby area with a similar environmental and hydrogeological setting invoked a possible mechanism of contaminant (MTBE) remobilization from soil grains (mainly zeolites) linked to rainfall washing the terrain. Using the approach presented in [
33], a dissipation half-life of 23 days was estimated for MTBE in the groundwater [
22]. A similar process could be applied to site 1 where residual NAPLs, located at a depth of 6–12 m, could be removed, dissolved, transported, and affected by degradation and volatilization phenomena (natural attenuation).
To assess and quantify these processes, principal component analysis was applied to the groundwater data of site 1, but not to the soil variables that were not available with the same frequency. The testing provided three main factors accounting for 85% of the total variance (
Table 4). It is worth noting that the interpretation of factors is not always based on very high scores, but it is still supported by other descriptors.
Factor 1 explaining the 50% of variance could be interpreted as the radon deficit process, which is the preferential solubility of radon in the NAPLs. A negative score of the groundwater depth (corresponding to a rainy period) was calculated with a negative value of radon and positive scores for MTBE, ETBE, and the total hydrocarbon concentrations. This implies that rainwater (and local leaks from municipal water pipelines) removed the residual NAPLs trapped in the aquifer pores, causing an increase in MTBE, ETBE, and the total hydrocarbon concentration in the groundwater where radon favorably dissolved with a reduction in the water radon contents. This process was already proposed in [
23] for this site and [
22] for a close fueling station in Roma.
Factor 2 accounted for a further 20% of variance (
Table 4). Since a direct correlation was found between the depth of the groundwater and the concentrations of ETBE and total hydrocarbons, with approximately zero scores for radon and MTBE levels, we can attribute this component to a degradation process of the first two substances in aerobic conditions. Precipitations (consistent with a reduction in the groundwater depth) introduced oxygen and nutrients into the aquifer promoting rapid biodegradation of ETBE and total hydrocarbons in the vadose zone, with reduced efficiency for MTBE [
34,
35]. The additions of biosurfactants (natural extracts of marine algae and plants) improved the separation and dispersion of hydrocarbon blobs, allowing the development and growth of natural microorganism that further degraded the NAPLs.
Factor 3 explained a supplementary 15% of variance (
Table 4); it could be interpreted as volatilization of organic products (mainly MTBE) and radon degassing in the unsaturated part of the aquifer, also linked to the action of soil vapor extraction practices. MTBE is characterized by a higher vapor pressure and thus volatilizes more easily from the residual NAPL [
36]. The low score of the groundwater table depth demonstrated no role of the small aquifer fluctuations.
NAPL saturation of polluted wells in site 2 was estimated according to Equation (1), based on the radon deficit values. The average radon level (17.1 Bq L
−1) of the groundwater extracted from PZ01, PZ09, SN01, SN03, and SN09 was used as a reference. The NAPL saturation in groundwater, represented on a schematic map (
Figure 6a), ranged from about 8 to 54%, with the highest values in correspondence with the lower radon activity concentrations and stronger contamination. Such high saturations, compared with those calculated for site 1, are well explained by the recent age of this diesel spill.
The presence of dissolved NAPLs was presumably associated with blobs of residual droplets of diesel and, to a lesser extent, gasoline in the soil pores that were mobilized by rainfall and groundwater table fluctuations. Their location was probably at a depth of 1–2 and 2–3 m below ground level, as suggested by traces of VOCs detected using a photoionization detector (PID) during the drilling of monitoring wells PZ02, PZ04, PZ06, and PZ07 in 2019, as notified by MARES. It is worth noting a good correspondence between higher NAPL saturation and the presence of soil VOCs.
The statistical analysis was also applied to the groundwater data of site 2, providing a different scenario (
Table 5). The three main components accounted for 80% of the total variance. Factors 1, 2, and 3 explained about 37%, 23%, and 20% of variance, respectively. The first two components could be ascribed to the radon deficit process driven by rainfall (factor 1) and groundwater table fluctuations (factor 2) dynamics. Factor 3 could explain the biodegradation in anaerobic conditions.
The interpretation of both factors 1 and 2 as a radon deficit process induced by the removal of the residual NAPLS was due to the shallow depth of the groundwater table (about 2 m) and its large annual fluctuations (up to 70 cm). In this case, it was possible to distinguish the contribution of rainfall that washed out the total hydrocarbons from the unsaturated zone and the effect of the groundwater table and the related capillary fringe on MTBE. Rainfall (factor 1) was effective with the total hydrocarbons related to the recent diesel spill and located at shallower depth where they were absorbed by backfill materials containing expanded clay, as demonstrated by the inverse correlation of the water table depth and radon concentration with the total hydrocarbons. The direct correlation of the groundwater depth with MTBE and ETBE contents suggests that these substances (linked to a previous gasoline spill) are located below the groundwater table and consequently are not removed by rainfall. In case of factor 2, the groundwater depth is inversely correlated with MTBE, consistent with the assumed location of residual MTBE at or just below the ground water surface. The direct correlation between the water table and total hydrocarbons strengthens the hypothesis that the total hydrocarbons are placed above and mobilized by rainfall. The absence of significant correlation with ETBE could be explained by its low degradation in anaerobic condition [
37], No correlation with radon is evident, but this could be explained by the smoothing effect of the groundwater flow reversal on the transport and location of NAPLs and in turn on radon levels.
Factor 3 accounted for the biodegradation of MTBE and total hydrocarbons in anaerobic condition (
Table 5). The negative score of the groundwater depth indicated that rainfall supplied oxygen and nutrients promoting the activity of microorganism on these compounds [
36]. The reverse correlation with ETBE could be justified by a low degradation of ETBE in anaerobic conditions [
37].
Based on these results, it was possible to develop the conceptual models of sites 1 (
Figure 5b) and 2 (
Figure 6b).
The residual NAPL source zone is located at well PZ13 (at a depth of 6 to 11 m) where the lowest radon and highest NAPL and VOC concentrations were detected (
Figure 5), probably absorbed onto or trapped among mineral grains (for example zeolites [
22,
23]. Rainfall and, to a lesser extent, groundwater table fluctuations mobilize NAPLs that dissolve in the aquifer and are transported downstream, creating temporary plumes with a dissipation half-life of about 3 weeks [
22,
33]. The plume is perfectly confined in the fueling area by the mitigation system consisting of a series of pump-and-treat and vapor extraction wells. Radon deficit traces the NAPL location in the vadose and saturated parts of the aquifer due to its preferential solubility in organic phases; it accounts for about 50% of the variance in the groundwater data, as demonstrated by the principal component analysis. Average NAPL saturations, ranging from about 0.1 to 2.4%, are low, as expected by the age of the spill (about 20 years).
Phenomena of natural attenuation are actively working in the site; biodegradation of ETBE and total hydrocarbons in aerobic conditions justifies about 20% of variance, while volatilization of organic products, coupled with radon degassing, explains a further 15%. The use of biosurfactants certainly promoted the activity of natural microorganism and biodegradation.
Two areas of the residual NAPLs were recognized in site 2 around wells with lower radon and higher NAPL concentration (
Figure 6). The presence of two single areas probably depends on the location and dynamics of the spills (gasoline first and diesel later) and on the change in the direction of the groundwater flow over the course of the year, as demonstrated in other areas [
38]. In this site, the shallow depth of the groundwater table and its relevant fluctuations influence the mobilization of NAPLs more than in site 1, where rainfall mostly removed the residual contaminants. This hypothesis is supported by factorial analysis whose interpretation assigned 37% of the variance to rainfall (factor 1) and 23% (factor 2) to groundwater (and related capillary fringe). The radon deficit principle supported that and made it possible to calculate an average NAPL saturation of 8% to 54%, consistent with the younger age of the spills and the frequent presence of oil films in several wells. The residual NAPLs are probably located at a depth of 1–2 and 2–3 m, as confirmed by VOCs and it is plausible to expect MTBE blobs, and to a lesser extent residual ETBE, to be placed deeper than the total hydrocarbons due to the higher solubility of MTBE and the older age of the gasoline spill compared with the diesel release.
A comparison table is provided to summarize the main findings from the factorial analysis applied to sites 1 and 2. The interpretation of factors is not always based on very high scores, but it is still supported by other descriptors (
Table 6).
In site 1, rainfall more than water table fluctuation removes NAPLs from the tick vadose zone, producing the radon deficit. This is supported by the greater depth (about 18 m) of the groundwater table in site 1 and the location of the residual NAPLs in the vadose zone (at −6/−10 m and −11/−14 m), quite distant from the area of fluctuation of the piezometric level (
Table 6). At site 2, relevant fluctuations (up to 0.7 m) of a shallow water table (about 2 m below ground level) contribute to mobilize the residual NAPLs (mainly MTBE) located next to the water table surface (at −1/−2 m and −2/−3 m), promoting the radon deficit. This is demonstrated by the variance attributed to these driving factors.
Degradation in aerobic conditions, rather than in anaerobic environment, occurs in site 1, coupled with stronger volatilization and degassing phenomena. These processes are coherent with a thick vadose zone. At site 2, the shallow depth and the dynamics of the water table makes the soil where the residual NAPLS are located more frequently anaerobic. Under these contrasting environmental conditions, the biodegradation in sites 1 and 2 proceed with different rates and mechanisms. The different loadings and correlations of NAPL concentrations, strengthened by appropriate references, help to outline these scenarios.
Finally, multivariate regression was applied to develop a model for radon estimation. Following, the equations were obtained for sites 1 and 2:
with R
2 = 22.48%, R
2 (adjusted) = 20.25%,
with R
2 = 68.41%, R
2 (adjusted) = 66.23%,
Even if the regression coefficients, R2, of both equations (22.48% for site 1 and 68.41% for site 2) are not very high, we want to stress the importance of this approach to estimate the value of an unknown variable starting from others or to establish relationships among factors. With a stronger dataset, the regression coefficients would be more robust and could help in predicting unavailable data and managing mitigation actions. To our knowledge, this is the first application of such a statistical approach to this type of dataset.