Monthly salinity profiles in wells show a shallower and thinner saline interface towards the coast, until there is an abrupt change (close to 1 m thickness). The freshwater (salinity <1 ppt) thickness is 25 m in the continental aquifer (in well P4), and reduces towards the coast where there is brackish groundwater with a 2 ppt concentration (Fig. 3a). Salinity profiles show a stable fresh/brackish water thickness without discontinuities or stratification. Figure 3b shows that in rainy months (May–November), the aquitard stores freshwater derived from rainfall, attenuating the recharge in the regional aquifer. The salinity in the aquitard decreases by the freshwater storage.
Time series of the aquifer head, salinity and hydraulic gradient in the study area are shown in Fig. 4. Hurricanes were not registered during the study period, although the signal of northern cold fronts can be seen from October 2017 to March 2018. Figure 4a shows the water surface astronomical variation in the sea level and the aquifer, confirming that the astronomical tide propagates significantly into the latter. Figure 4b is a plot of the meteorological head (MH), composed by (1) the effect due to the aquifer recharge from precipitation and (2) the nonastronomical tide propagation into the aquifer. Precipitation from rainy months (June–October, Fig. 4d) recharges the aquifer, as can be seen from the MH signal (1) during specific short-events in early September and late October, and (2) for the rainy period during which MH slowly increases from May to October (Fig. 4b). This aquifer response is clear at the wells farther from the coast (P4, P7b, P8 and P9), but not at P5 and P7a, the wells closer to the coast, where the ocean influence predominates and the aquifer is confined. Zavala-Hidalgo et al. (2010) report that the monthly mean sea level was the lowest in July and highest in October, which is consistent with the monthly mean MH from well P5 and P7a, as shown in Fig. 4b.
The rainy season commonly is related to increments in the hydraulic gradient and an increment in the groundwater flux rate towards the coast (Fig. 4c). These increments in the discharge flux rate related to precipitation can be appreciated in Fig. 4c,d. In fact, given the monthly average sea-level variation mentioned in the preceding, the hydraulic gradient in wells is lower during the higher sea-level period (September–November, Fig. 4c), but also coincides with the largest freshwater recharge period. From mid-December to the end of April the hydraulic gradient in the wells is consistently high (0.5 m/km in P7a), which may be due to a low monthly-averaged sea level, even if the precipitation is low in that period. Finally, the effects of the cold fronts from December to April, associated with strong northerly winds which produce 20–30-cm storm surges lasting 1–4 days (Torres-Freyermuth et al. 2017), can be seen in the MH signal (Fig. 4b), in particular at the stations closer to the coast and a lesser extent in station P4 and P7b.
The salinity records (Fig. 4e,f) at the upper and bottom saline interface show a variable diurnal response related to the astronomical tide, except at BSI in P8 well, where the ocean influence is minimal. At P7a-BSI the signal decreases during some periods, coincident with the behaviour of the monthly salinity profiles in P7a shown in Fig. 3c.
Throughout the study period, salinity at P7a BSI shows a seasonal dynamic behaviour, ranging from a maximum of 27.7 ppt on July 10th to 13.5 ppt on April 12th (Fig. 4f); this suggests that the BSI moves seasonally up and down, as registered in the electrical-conductivity logger located in a fixed position where the diurnal variability is small. The high values in June are associated with (1) the end of the dry season, (2) a low monthly-averaged sea level and (3) a relative increase in the hydraulic gradient. In contrast, the low salinity values occur in three periods: (1) a decrease in October (lowest value of 13.5 ppt on October 20, 2017), which coincides with both the period of the year with highest meteorological head and the largest freshwater input to the aquifer, (2) gradual quasi-monotonic decreases of salinity from July to September 2017 and (3) from February to April 2018, possibly associated with the decrease of the sea level and apparently an increase in freshwater storage, i.e., thickening and deepening of the freshwater layer near the coast.
Sea level forcing
The tidal forcing, arising from planetary motions influenced by the rotation of the earth and the orbits of the moon around the earth and the earth around the sun, can be defined by a set of spectral lines, i.e., the sum of a finite set of monochromatic sinusoids with specific frequencies and phases (tidal constituents, Pawlowicz et al. 2002). Power spectra (PS) of the head data in the monitoring wells correspond to the PS of the astronomical tide (Fig. 5a). The peaks’ frequencies, in cycles per hour (cph), coincide with the main lunar diurnal constituent O1 (0.0387 cph), the diurnal lunisolar K1 (0.0417 cph), and main lunar semidiurnal M2 (0.0805 cph). These values are coincident with values reported in local studies (Vera et al. 2012; Torres-Mota et al. 2014) and express the most important forcing wave components of the astronomical tide in the study zone.
Freshwater aquifer head
The tidal influence is observed in wells with distances up to ~12 km from the coast. The most distant well from the coast (P4 well) did not register the evidence of tidal influence; in the case of P7b well, harmonic analysis suggests a highly attenuated O1 and K1 signal, but the range of aquifer head variation (8 mm) is equivalent to the pressure logger accuracy; thus, this result must be taken with caution. The power spectra in these wells are four orders of magnitude smaller than sea level, suggesting that tidal effects are negligible (Fig. 5a).
The analytical model by Ferris (1952) reveals that the hydraulic diffusivity ranges from 20.5 m2/s (P7a) to 2,450 m2/s (P9; see Fig. 6), adjusting with the astronomical tidal constituents O1, K1, M2; the percent of AT propagated into the aquifer (linear correlation slope between AT and AH) ranges from 1.4 to 79%—Table 1, Fig. S1 of the electronic supplementary material (ESM)—with an expected Pearson correlation coefficient (r) from 0.77 to 0.99. This model indicates that the astronomical tide has a direct relationship with the elevation of the aquifer inland, associated with the hydraulic diffusivity. The highest correlation values are located in wells near to the coast (P7a and P5 wells), while correlation decreases inland (P8, P9 and P7b wells). The time lag between AT and AH ranged between 0 (P7a well) and 7.0 h (P7b well, see Table 1).
Cross-correlation for meteorological effects (MH and MT) shows a direct positive correlation between the meteorological tides and the aquifer head in P7a, P5, P8 and P9 wells; cross-correlation in P4 and P7b wells does not show significant correlation. The time lag between MT and MH was estimated between 0 and 3.0 h (Table 1), similar to the astronomical tide propagation time. MT and MH data are plotted in Fig. 7. The coastal zone (P7a, P5, and P9 wells) shows higher Pearson correlation (r between 0.82 and 0.71) and a moderate correlation for P8; correlation is nonsignificant in wells farther from the coast (P7b and P4, r < 0.40), suggesting that other variables, such as precipitation or other aquifer oscillations, are related to the aquifer head in this zone.
Meteorological effects propagate into the aquifer with MT percent (linear regression slope) between 40% (P8) and 61% (P7a; see Table 1), increasing near the coastline. Linear regression results indicate that the sea level rise by meteorological effects increases the aquifer head in the continent, being more noticeable near the coast (Fig. 7).
Regarding the tide propagation in coastal aquifers, previous studies have reported a direct relationship between the sea level and the aquifer head (Vera et al. 2012; Levanon et al. 2017; Dessu et al. 2018), but these studies have not considered the individual influences of both the astronomical and meteorological signals. In terms of the decay of sea level propagation into the aquifer, Fig. 8 shows that the astronomical signal (AT) experiences a faster decay than the meteorological component (MT), suggesting that the nonastronomical signal of the ocean tide, which typically has a longer period than the astronomical components analysed here (Torres-Freyermuth et al. 2017) can, theoretically, travel farther inland, as reported by Trglavcnik et al. (2018). The differences between the propagation of the forcings in the aquifer can only be observed when the astronomical and meteorological signals are analysed separately.
Power spectra (PS) in salinity present similar features as the PS in the astronomical tide: the spectra reveal the same frequencies at the main tidal components: O1 (0.0387 cph), K1 (0.0417 cph) and M2 (0.0805 cph). PS in salinity shows a similar order of magnitude to the PS in sea-level and aquifer response. This behaviour suggests that a harmonic analysis can also be an option to separate astronomical and meteorological signals in terms of aquifer salinity. P8-BSI spectra do not show similitude with the main astronomical components in the sea, suggesting that the salinity variations in the aquifer associated with the astronomical tide are strongly attenuated in the fixed logger position in this specific well (Fig. 5b).
The correlation for astronomical effects (AS and AT) shows a strong relationship between the astronomical tide and the astronomical signal in the aquifer salinity. Time lags are between −2.5 and 7 h; the time-lag in P7a BSI is abnormal because the salinity peak occurs before the high tide (−2.5 h), while in stations P7a USI and P8 USI, the time lag is estimated between 3 (P7a) and 7 h (P8). These distortions have been reported in boreholes within the coastal zones (Levanon et al. 2017), as an effect of the absence of a capillary zone. Levanon et al. 2017 suggest that burying the sensors in the aquifer is an option for filtering these distortions. In this study, sensors were installed in boreholes only, and these effects may be present in the results as observed in the PS of P7a-BSI (Fig. 5b). The astronomical signals’ responses in the study zone are similar to those in the conceptual model of Levanon et al. (2017) which was defined for unconfined aquifers, suggesting that this conceptualization can be applied in confined systems.
AT and AS are plotted in Fig. 9, showing a direct strong correlation of the astronomical signal for salinity with the astronomical tide (r between 0.76 and 0.95, Table 1). These correlations were expected because the PS of the sea level is coincident with the PS of the salinity in the monitoring wells (Fig. 5b). Results suggest that during a rising tide, the salinity in the aquifer groundwater increases, and inversely decreases during ebb. Furthermore, Fig. 9 shows that (1) the diurnal salinity variation at P7a is only 0.1 ppt at USI, and 0.8 ppt at the BSI; (2) the correlation at P7a is larger at BSI than at USI, and (3) the tidal effect in salinity occurs 5.5 h before at BSI compared to USI. These observations suggest that the astronomical tide effect in the salinity propagates in the vertical direction from the bottom to the upper saline interface (BSI to USI) and then towards the aquifer inland, as previously described by Levanon et al. (2017).
Previous studies report a direct response of the aquifer salinity to the meteorological tides; but the monitoring was implemented only in sinkholes and submarine springs (Vera et al. 2012; Parra et al. 2016), where the salinity mixing mechanism is turbulent in a free water column in the vicinities of these structures, as described by Coutino et al. (2017). In this research, the results do not show a significant relationship between MT and MS, with r < 0.50 (Table 1 and Fig. S2 of the ESM); these differences can be associated with the effect of assessing the hydrological forcings separately (especially the tide components), and the effect of different mixing mechanisms on groundwater salinity compared to highly karstified systems. In this respect, there are no previous studies reporting a similar methodology and/or similar results. Besides, the studies previously mentioned (Vera et al. 2012; Parra et al. 2016; Coutino et al. 2017; Kovacs et al. 2017) were carried out with extreme climatological events (e.g. hurricanes), which were not present in this investigation; therefore, it is not known whether extreme climatological events could generate these differences.
Figure 10 shows three selected rainfall events to compare precipitation, aquifer head and salinity (events in June, July and September 2017). Meteorological stations at MID, P9 and SSL yielded data to show the rainfall distribution throughout the study period.
Cross-correlation analysis was evaluated only for MID and P9 (transition and continent) because the conceptual model suggests a coastal confined aquifer (SSL) with poor response from precipitation; therefore, P7a and P5 were not considered in the analysis. In the case of salinity correlations, only P7a USI and P8 USI were correlated with precipitation in P9, because the recharge effects could only be identified in the upper limit of the saline interface.
Results show that short-term precipitation events at MID and the aquifer head in the study zone present a time lag between 26 and 26.5 h (moderate correlation in September 2017 with 0.50 < r < 0.7, see Table S1 of the ESM), while in station P9 the time lag ranged from 0.5 to 3.0 h, and are poorly correlated (r ≤ 0.5; see Table S1 of the ESM). This suggests that the local precipitation is not a significant forcing in terms of the vertical recharge response of the aquifer, probably due to the low permeability of the vadose zone and the aquitard layer (Fig. 3b).
The seasonal precipitation response in the aquifer was evaluated using monthly accumulated precipitation, which reflects the seasonal trend in the aquifer head due to precipitation and represents seasonal changes over the analysed period. Figure 11 shows that during rainy months (June–November 2017), the aquifer head increases by recharge, and decreases in dry months. Besides, precipitation data at P9 and MID suggest a 3-months delay in the aquifer response. Using this time lag, monthly accumulated precipitation and aquifer level (previously corrected by MT influence, M-MH) were correlated by means of a linear Pearson correlation model as shown in Table 2.
Table 2 reveals that the seasonal precipitation presents a moderate direct relationship with the mean aquifer head, similar to previous studies (Coutino et al. 2017; Kovacs et al. 2017). The values of r for P7b, P8 and P9 range between 0.50 and 0.56 with respect to precipitation at MID, and between 0.68 and 0.70 with respect to precipitation at P9. The aquifer head in P4 does not show a correlation with the precipitation data (See Fig. S3 of the ESM). These outcomes suggest that the vertical recharge is not as strongly correlated with the aquifer head as the sea level.
Cross-correlation and Pearson correlation coefficients for precipitation and salinity are nonsignificant (r < 0.50), suggesting that the salinity variation is not significantly associated with the accumulated monthly precipitation (Table 2; see Fig. S4 of the ESM). Previous studies carried out in sinkholes show an inverse relationship between precipitation and salinity due to the direct connection of the aquifer and the surface terrain (Coutino et al. 2017; Kovacs et al. 2017). In this research, the monthly salinity profiles in wells reveal that the vadose zone stores the water of precipitation during rainy months (Fig. 3), thus the saline interface does not show response to this stress. Rocha et al. (2015) report similar results, suggesting that recharge in the aquifer is not related to the variation in the saline interface.