Local Variability of CO 2 Partial Pressure in a Mid-Latitude Mesotidal Estuarine System ( Tagus Estuary , Portugal )

Estuaries play a crucial role in regional carbon cycling. Until now, accurate estimations of the impact of environmental variables on estuarine air–water CO2 fluxes have been mostly characterized by a low spatial-temporal sampling resolution. This study reports on the variations of CO2 partial pressure (pCO2) and related environmental parameters, at both tidal and seasonal temporal scales, in the surface seawater of a station located in the lower section of the Tagus estuary, Portugal. The study was carried out from February to December 2007. Air–water CO2 fluxes suggest that the lower estuary acted as a relatively weak source of CO2 to the atmosphere, with an average rate of 7.2 mol·m−2·year−1, with highest fluxes occurring in winter. Over a tidal cycle, pCO2 was mainly influenced by tidal-induced mixing. Results suggest an influence of upper and central estuary inputs with higher pCO2 values. pCO2 varied seasonally, with values decreasing from ~890 μatm in winter to ~400 μatm in summer and increasing again to ~990 μatm in autumn. The generalized linear model (GLM) applied to the data set explained 69.3% of the pCO2 variability, pointing to the thermodynamic effect of temperature and biological activity as the most relevant processes in CO2 dynamics. Tidal variation of pCO2 corresponded to ~35% of its seasonal variability, denoting the importance of tide conditions on the dynamics of inorganic carbon. Results showed distinct patterns in the dynamics of CO2 at the tidal scale. This outcome suggests that disregarding tidal variability in the use of seasonal data sets may lead to significant errors in annual carbon budget estimations.


Introduction
Inner estuaries act as sources of CO 2 to the atmosphere due to their heterotrophic ecosystem metabolic status [1][2][3][4].A compilation of available air-water CO 2 fluxes in inner estuaries indicates that these near-shore ecosystems emit CO 2 to the atmosphere at an average of 20.8 mol•C•m −2 •year −1 [5][6][7].Scaling up this value yields CO 2 emissions of 0.10 Pg•C•year −1 [8], which is much lower than published values by Chen and Borges [9], for example, mainly because of the contribution of a considerable amount of unpublished or new data from Asia and the Arctic [10].Oliveira et al. [11] have shown that the Portuguese western near-shore ecosystems emit 0.3 Tg•C•year −1 to the atmosphere, corresponding to ~0.3% of the total estimated CO 2 worldwide emissions for near-shore ecosystems [8].A relevant feature of these compilations is the great variability of CO 2 partial pressure (pCO 2 ), ranging between 143 and 11,335 µatm, and yielding air-water fluxes from 2.2 to 76.0 mol•C•m −2 •year −1 .This variability in near-coast areas is significantly greater over space and time than typically observed in open ocean environments.Factors controlling pCO 2 within estuaries may vary among systems according to environmental conditions.Moreover, it has been reported that pCO 2 can present considerable spatial and temporal variations within the same estuary [12][13][14], due to the hydrodynamic and geomorphological complexity of such coastal areas.
Estuaries cover extensive areas of coastlines worldwide, playing a pivotal role in regional carbon cycling [2,12].Previous studies have not been able to determine accurately the impact of environmental variables on estuarine air-water CO 2 fluxes, mostly due to limited spatial-temporal sampling resolution.Consequently, studies with a high temporal resolution provide further clues into estuarine mechanisms, both natural and anthropogenic, that are susceptible to influence or even control significant carbon flows, as well as information on current CO 2 source/sink patterns.
This paper focuses on the study of CO 2 variability at different temporal scales of a mid-latitude estuarine system, the Tagus estuary (SW Portugal), contributing to characterize the dynamics of air-water CO 2 fluxes with higher temporal resolution.The Tagus estuary is a particularly relevant system to perform such a CO 2 study.The estuary is one of the largest on the west coast of Europe and supports important human communities and natural resources.Thus, this paper characterizes pCO 2 variability on a tidal and seasonal scale and tackles the important processes controlling those variations, using data collected throughout the year 2007 in one monitoring station.Air-water CO 2 exchanges were also estimated, and a generalized linear model was applied to evaluate the effect of different environmental variables on pCO 2 variability.

Study Area
The Tagus estuary is located on the densely populated west coast of Portugal, in the Iberian Peninsula (38 • 36 -39 • N, 08 • 54 -09 • 24 W) (Figure 1).The estuary is a relatively shallow and wide mesotidal lagoon-type estuary, connected with the Atlantic Ocean via a narrow and deep channel.It has extensive intertidal areas that comprise 20 to 40% of its total inner estuarine area.A fringe of salt marsh vegetation, covering approximately 13 km 2 , occupies the upper littoral zone.Hydrographic conditions are mainly determined by a considerable riverine input and the inflow of saline water from the Atlantic.Saline waters can be transported from 60 to 120 m depth since the coastal adjacent area is subjected to frequent upwelling events.The main freshwater source is the Tagus River with flows varying monthly from ~145 to ~815 m 3 •s -1 [15], and other less important riverine discharges as the Sorraia and Trancão rivers.There are also a significant number of effluents from urban, industrial, and agricultural sources.The residence time of the waters in the estuary varies between 26 and 8 days [15].Such features lead to a pronounced river plume extending in the coastal area [16].Besides the fresh water inlets, an important driver of this mesotidal system hydrodynamics is the astronomic tide with a dominant semidiurnal period and maximum amplitude reaching 4.8 m in spring tide.Additionally, storm surges favoring a significant elevation of the water level (relative to astronomical tide) caused by meteorological forcing are reported to be frequent in the Tagus estuary lower section [17,18].Wind predominance is from the south and southwest during winter, rotating progressively to the northwest and north during spring, and maintaining this direction throughout summer.

Sampling
Sampling was performed at a fixed station located in the lower section of the Tagus estuary (38°41.34′N, 09°13.39′W) (Figure 1).The lower estuary is connected to the coastal area by a channel, with depths over 30 m, having Lisbon (the Portuguese capital city) on its right bank.Sampling was conducted from February to December 2007 covering four seasons and included two tidal conditions (neap and spring).Surface water sampling (~1 m) was carried out at neap and spring tide every hour (approximately 13 h) with a Niskin horizontal bottle and a conductivity-temperature-depth (CTD) probe.A total of 101 observations were recorded.Sampling dates, hydrological information, and meteorological conditions are detailed in Table 1.
Table 1.Sampling dates and ranges of hydrologic and wind data recorded for each sampling period in the Tagus estuary fixed station during 2007.Q represents Tagus river flow and u10 the wind speed referenced to a height of 10 m (expressed in m•s −1 , see Section 2.4).

Sampling
Sampling was performed at a fixed station located in the lower section of the Tagus estuary (38 • 41.34 N, 09 • 13.39 W) (Figure 1).The lower estuary is connected to the coastal area by a channel, with depths over 30 m, having Lisbon (the Portuguese capital city) on its right bank.Sampling was conducted from February to December 2007 covering four seasons and included two tidal conditions (neap and spring).Surface water sampling (~1 m) was carried out at neap and spring tide every hour (approximately 13 h) with a Niskin horizontal bottle and a conductivity-temperature-depth (CTD) probe.A total of 101 observations were recorded.Sampling dates, hydrological information, and meteorological conditions are detailed in Table 1.

Meteorological Data
Wind speed and direction were measured in situ with a Vaisala ® meteorological station (Datalogger Campbell Scientific CR510) (Campbell Scientific, Logan, UT, USA) coupled with a MetOne 034A (Campbell Scientific, Logan, UT, USA) anemometer.Continuous measurements were acquired with 1-min intervals at 11 m height.Wind speed was referenced to a height of 10 m (u 10 ) [19] and a standard deviation of ±2 m•s −1 was assumed as the wind speed error.Atmospheric CO 2 data were measured at the Terceira Island's reference station (Azores, Portugal, 38.77 • N, 27.38 • W), as part of the network of the National Oceanic and Atmospheric Administration (NOAA)/Climate Monitoring and Diagnostics Laboratory (CMDL)/Carbon Cycle Greenhouse Gases Group (CCGG) [20].Dickson et al. [21] algorithms were used to convert observed atmospheric CO 2 content in mole fraction (in dry air) to wet air values.Direct measurements of atmospheric CO 2 partial pressure made on-board was only available for some sampling periods.Significant correlations were found out between Terceira data and shipboard data (r 2 = 0.910, p < 0.05, n = 45) and the discrepancies occurred only between 3 and 13 µatm.The impact of using Terceira data on this study were considered negligible.

Chemical Analysis
Chlorophyll a (Chl a) was determined by filtering triplicate aliquots of ~150 mL water through Whatman GF/F(Sigma-Aldrich, Kawasaki, Japan) filters (0.7 µm) under a 0.2 atm vacuum, and immediately frozen and later extracted in 90% acetone for analysis in a fluorometer Hitachi F-7000 (Hong Kong, China), calibrated with commercial solutions of Chl a (Sigma-Aldrich, Kawasaki, Japan).The variation coefficient was 1.8%.Dissolved oxygen (DO) was analyzed following the Winkler method [22] using a whole-bottle manual titration, and the coefficient of variation associated with the method ranged from 0.08 to 0.25%.Total alkalinity (TA) samples were filtered through Whatman GF/F (0.7 µm) filters, fixed with HgCl 2 and stored (refrigerated not frozen) until use.Samples were then titrated automatically with HCl (~0.25 M HCl in a solution of 0.45 M NaCl) past the endpoint of 4.5 [21], with an accuracy of ±2 µmol•kg −1 .The respective accuracy was controlled against certified reference material (batch 77, Marine Physical Laboratory, Scripps Institution of Oceanography, San Diego, CA, USA).Measured TA values (2260.91 µmol•kg −1 ) differ from 2.7% from the theoretical values (2199.33 µmol•kg −1 ).

Estimated Parameters
Current velocity and tidal height data at the Tagus estuary were calculated by running the MOHID model (information online at www.mohid.com) in hindcast mode, prescribing real forcing for river discharge, tide, and wind.The MOHID Water Modelling System is a 3D marine modelling platform that has been applied to the Tagus estuary in simultaneous hydrodynamic and biogeochemical studies (e.g., [16,[23][24][25]).pH values corrected to in situ temperature were calculated from TA and in situ pH and temperature [26], using the carbon dioxide constants of Millero et al. [27].TA values were normalized to a salinity of 35, being designated as nTA.Seawater pCO 2 was calculated from in situ temperature, corrected pH and TA using the carbonic acid dissociation constants given by Millero et al. [27] and the CO 2 solubility coefficient of Weiss, s [28].Errors associated with pCO 2 calculations were estimated to be ±10 µatm (accumulated errors on TA and pH).
The air-water CO 2 fluxes (CO 2 Flux) were computed according to the following equation [21]: where k is the gas transfer velocity and ∆pCO 2 is the air-water difference of pCO 2 .Positive CO 2 flux indicates the emission of CO 2 from water to the atmosphere and negative indicates otherwise.
To determine the most likely value for k, calculations were based on a set of parameterizations: (1) Carini et al. [29] relationship (hereinafter referred to as C96), based on a SF 6 release experiment in the Parker River estuary; (2) Raymond and Cole [30] relationship (hereinafter referred to as RC01), based on a compilation of published k 600 values in various rivers and estuaries using different methodologies (floating chamber, natural tracers (CFC, 222 Rn) and purposeful tracer SF 6 ); and (3) Borges et al. [31] relationship (hereinafter referred to as B04) based on floating dome CO 2 measurements in the Scheldt estuary, taking into account the contribution of the water current from the conceptual relationship of O'Connor and Dobbins [32].
All exploratory analysis and statistical modelling were implemented using the statistical software R 2.7.1 (Insightful Corp., 1999).A logarithmic transformation of the response variable pCO 2 to linearity was necessary in order to perform subsequent analyses (Shapiro-Wilk test for normality, p < 0.05).The ANOVA (Analysis of Variance) was tested for significant differences in the following factors: tide (2 levels, neap and spring), tidal current (2 levels, ebb and flood), hours (variable was categorized into 3 levels, morning, noon, and afternoon), and season (4 levels, spring, summer, autumn, and winter).All explanatory variables were considered statistically different at a significance level of 0.05.Model building by variable selection was necessary to eliminate multicollinearity between independent variables.The backward elimination procedure was applied to obtain an ideal subset model.This method entered all four factors and their first-order effects (interactions between factors).Non-significant variables were removed according to the Akaike information criterion (AIC) [33].
A generalized linear model (GLM) with a log link function was applied to the entire data set to estimate the effects of various environmental variables (S, T, DO, and Chl a) on the response variable pCO 2 .The GLM, after a backward selection using the AIC, included the following factors: season, tide, tidal sequence, hours, and three interactions (between season and tide, season and tidal sequence, and hours and tide).When applying one-way ANOVAs for each of the four factors, the season and the tide were considered statistically significant (p < 0.05).An analysis of deviance was performed to assess the goodness-of-fit of the model.

Results
Ranges for measured physical-chemical and biological parameters are presented in Table 2.

Hydrodynamics
Generally, the Tagus river inputs reflect the common pattern of a mid-latitude system, with higher discharges in winter and lower ones in spring/summer (Table 1), despite the high degree of hydrological regime regularization imposed by a significant number of dams upstream.However, a relatively high discharge (109 m 3 •s −1 ) at neap tide was noticed during summer.Inversely, reduced river discharges of only 17-18 m 3 •s −1 were observed in autumn (Table 1).The impact of river inflows on the temperature and salinity at the estuary channel is depicted in the T-S diagram (Figure 2).Colder waters were observed in winter and autumn, whereas warmer waters were detected during the productive period (spring and summer).The occurrence of a relatively intense discharge in winter (370 m 3 •s −1 ; Table 1) resulted in lower salinities (Table 2; Figure 2), whereas the exceptional high discharge in summer (Table 1) produced a decrease in salinity (Table 2; Figure 2).Tidal amplitude (Table 1) under neap conditions varied between 1.3 m (winter) and 2.1 m (summer) and under spring conditions between 3.0 m (summer) and 3.9 m (spring).

Tidal Variability
Data were recorded in the fixed station in autumn 2007, when river discharge reached the lowest flow values under both tide conditions (Table 1), to assess the semidiurnal tidal variability of pCO 2 in the estuary.Figure 3 shows the semidiurnal tidal cycle effect on T, S, DO, Chl a, nTA, pH, and pCO 2 parameters at spring and neap tide.
Most sampling campaigns were carried out at flood during the spring tide.Results showed that salinity tracked the tidal advection with its maximum and minimum practically coincident with high and low tide (Figure 3A).The semidiurnal tidal cycle was followed by other variables as well.In general, maximum values of T, Chl a, and pH were attained simultaneously with minimum values of nTA at high tide (Figure 3A-C).This occurrence indicated that warmer, and biologically more productive marine waters impoverished in alkalinity, were transported to the estuary by the flooding tidal current.During neap tide most of the sampling occurred at ebb, but a slight lag between the salinity maximum and the high tide was observed.Minima values of S and T were reached at the initial hours of flood (Figure 3E).Such features were reflected on the trends of Chl a, DO, pH, and nTA (Figure 3F,G).The plots of temperature, pH, nTA, DO, and Chl a versus salinity (Figure 4) in autumn revealed that warmer and more productive offshore waters with higher pH dominated the lower estuary at flood (spring tide).The influence of colder, more oxygenated, less productive, and carbonate-enriched Tagus estuarine waters was observed during ebb (Figure 4A-E).
(Table 1) under neap conditions varied between 1.3 m (winter) and 2.1 m (summer) and under spring conditions between 3.0 m (summer) and 3.9 m (spring).

Tidal Variability
Data were recorded in the fixed station in autumn 2007, when river discharge reached the lowest flow values under both tide conditions (Table 1), to assess the semidiurnal tidal variability of pCO2 in the estuary.Figure 3 shows the semidiurnal tidal cycle effect on T, S, DO, Chl a, nTA, pH, and pCO2 parameters at spring and neap tide.
Most sampling campaigns were carried out at flood during the spring tide.Results showed that salinity tracked the tidal advection with its maximum and minimum practically coincident with high and low tide (Figure 3A).The semidiurnal tidal cycle was followed by other variables as well.In general, maximum values of T, Chl a, and pH were attained simultaneously with minimum values of nTA at high tide (Figure 3A-C).This occurrence indicated that warmer, and biologically more productive marine waters impoverished in alkalinity, were transported to the estuary by the flooding tidal current.During neap tide most of the sampling occurred at ebb, but a slight lag between the salinity maximum and the high tide was observed.Minima values of S and T were reached at the initial hours of flood (Figure 3E).Such features were reflected on the trends of Chl a, DO, pH, and nTA (Figure 3F,G).The plots of temperature, pH, nTA, DO, and Chl a versus salinity (Figure 4) in autumn revealed that warmer and more productive offshore waters with higher pH dominated the lower estuary at flood (spring tide).The influence of colder, more oxygenated, less productive, and carbonate-enriched Tagus estuarine waters was observed during ebb (Figure 4A-E).Estuarine levels of CO2 follow the tidal cycle in autumn: an increase in pCO2 values at ebb, both under spring and neap tide conditions (Figure 3D,H), indicates the transport of CO2-enriched estuarine waters to the coast.
Such features are also evidenced on the plots of pCO2 against salinity (Figure 4F) and, although not so obviously, with tidal elevation (Figure 4G).For this period of the year, the tidal influence on the variability of pCO2 was statistically significant (Fisher's test, p < 0.001, n = 26).Considering all samples for 2007 (n = 101), the difference of the average ranges of tidal pCO2 was 208 µatm.This value was calculated as the average of the difference between the lower and higher values at each tidal cycle.Such features are also evidenced on the plots of pCO 2 against salinity (Figure 4F) and, although not so obviously, with tidal elevation (Figure 4G).For this period of the year, the tidal influence on the variability of pCO 2 was statistically significant (Fisher's test, p < 0.001, n = 26).Considering all samples for 2007 (n = 101), the difference of the average ranges of tidal pCO 2 was 208 µatm.This value was calculated as the average of the difference between the lower and higher values at each tidal cycle.

Seasonal Variability
Seasonal variation is detected in all the analyzed parameters, a pattern detected in the box plots with the physical-chemical and biological water properties for the four periods sampled in 2007 (Figure 5).Temperature and salinity increased from winter to spring/summer.Temperature ranged from 14.1 to 19.2 °C and salinity from 21.2 to 35.3 (Figure 5A,B; Table 2).Salinity values were strongly controlled by river discharges and showed considerable variability (Figure 5B; Table 1), showing lower values in winter.Maximum Chl a concentrations were measured during the productive period, peaking at 10.2 mg•m −3 , reached in spring tide, associated with high variability.During the nonproductive period, the values of Chl a hardly attained 1 mg•m −3 (Figure 5C; Table 2).pH varied from 7.88 to 8.19, with the highest values recorded during spring and summer (Figure 5D; Table 2).The lowest values of nTA were also attained during spring and summer (Figure 5E, Table 2).
Mean values of pCO2 ranged from 402 to 990 µatm throughout the year and displayed maximum values during the non-productive period (Figure 5F; Table 2).The amplitude of pCO2 seasonal signal was 588 µatm.The relationship between surface water pCO2 and temperature (Figure 6) showed different patterns regarding the productive and non-productive periods: pCO2 was higher at the low temperature during autumn and winter, and showed the opposite pattern during the productive period, with lower pCO2 associated with higher temperatures.

Seasonal Variability
Seasonal variation is detected in all the analyzed parameters, a pattern detected in the box plots with the physical-chemical and biological water properties for the four periods sampled in 2007 (Figure 5).Temperature and salinity increased from winter to spring/summer.Temperature ranged from 14.1 to 19.2 • C and salinity from 21.2 to 35.3 (Figure 5A,B; Table 2).Salinity values were strongly controlled by river discharges and showed considerable variability (Figure 5B; Table 1), showing lower values in winter.Maximum Chl a concentrations were measured during the productive period, peaking at 10.2 mg•m −3 , reached in spring tide, associated with high variability.During the non-productive period, the values of Chl a hardly attained 1 mg•m −3 (Figure 5C; Table 2).pH varied from 7.88 to 8.19, with the highest values recorded during spring and summer (Figure 5D; Table 2).The lowest values of nTA were also attained during spring and summer (Figure 5E, Table 2).
Mean values of pCO 2 ranged from 402 to 990 µatm throughout the year and displayed maximum values during the non-productive period (Figure 5F; Table 2).The amplitude of pCO 2 seasonal signal was 588 µatm.The relationship between surface water pCO 2 and temperature (Figure 6) showed different patterns regarding the productive and non-productive periods: pCO 2 was higher at the low temperature during autumn and winter, and showed the opposite pattern during the productive period, with lower pCO 2 associated with higher temperatures.
The effect of different environmental variables (S, T, DO, and Chl a) on pCO 2 response was assessed by applying the GLM to the complete data set, and the following equation was obtained: where E is the expected (mean) value of pCO 2 in µatm, T in • C, DO in mg•L −1 , and Chl a in mg•m −3 .The respective model explains 69.3% of pCO 2 variability, where temperature and chlorophyll a accounted for 58.4% and 7.6% deviance, respectively (Table 3).The effect of different environmental variables (S, T, DO, and Chl a) on pCO2 response was assessed by applying the GLM to the complete data set, and the following equation was obtained: where E is the expected (mean) value of pCO2 in µatm, T in °C, DO in mg•L −1 , and Chl a in mg•m −3 .The respective model explains 69.3% of pCO2 variability, where temperature and chlorophyll a accounted for 58.4% and 7.6% deviance, respectively (Table 3).

Air-Water CO2 Fluxes
Daily average CO2 air-water fluxes for each tidal cycle sampling, and respective fluxes against the wind speed, are presented in Figure 7.These values were estimated according to the three mentioned parameterizations (Section 2.6).In general, the highest fluxes were observed during more intense wind speed, except for three samplings during the productive period (Figure 7B).The averaged CO2 fluxes estimated by the three parameterizations ranged from 2 mmol•m −2 •d −1 in spring to 56 mmol•m −2 •d −1 in winter, following a seasonal pattern (Fisher's test, p < 0.001, n = 101).In general, the lowest fluxes were obtained when applying the C96 relationship, followed by RC01 and B04 with the highest estimates (Figure 7A).
On average, the fluxes estimated by the RC01 and B04 algorithms differed from those obtained from C96, respectively by 22% and 36%.An exception was observed when the wind speed was low, as in spring (24 April) (Figure 7A).

Air-Water CO 2 Fluxes
Daily average CO 2 air-water fluxes for each tidal cycle sampling, and respective fluxes against the wind speed, are presented in Figure 7.These values were estimated according to the three mentioned parameterizations (Section 2.6).In general, the highest fluxes were observed during more intense wind speed, except for three samplings during the productive period (Figure 7B).The averaged CO 2 fluxes estimated by the three parameterizations ranged from 2 mmol•m −2 •d −1 in spring to 56 mmol•m −2 •d −1 in winter, following a seasonal pattern (Fisher's test, p < 0.001, n = 101).In general, the lowest fluxes were obtained when applying the C96 relationship, followed by RC01 and B04 with the highest estimates (Figure 7A).
On average, the fluxes estimated by the RC01 and B04 algorithms differed from those obtained from C96, respectively by 22% and 36%.An exception was observed when the wind speed was low, as in spring (24 April) (Figure 7A).

Tidal Variability
The evolution of monitored environmental variables was distinct along tidal cycles and under different tide conditions (neap/spring) in the Tagus estuary, clearly revealing significant tidal variability.These patterns were shaped by the effects of tide amplitude alone, or in combination with a tidal asymmetric behavior, since flood is typically one hour longer than ebb in the estuary, leading to higher current velocities during ebbs [34].An examination of the temporal distribution of environmental variables together with tidal data from autumn 2007 revealed some typical features of dynamics of the estuary, namely, a dominant east-west current transporting CO2-rich water from the inner to the lower estuary, while the reverse tidal current imported saltier and more productive (phytoplankton-rich) seawater into the estuary (Figures 3 and 4).When river discharge was minimum (autumn 2007), salinity amplitude in surface water varied between 1.6 and 8.4 along a 13-h cycle.Considering that vertical stratification was absent (corroborated by CTD profiles, data not shown), variations in salinity were mostly controlled by horizontal mixing and advection.The effect of such physical processes along the tidal cycle was noticed in the concentration trends of the water properties.Similar water advection control was also reported in the Guadalquivir estuary (SW Spain), for properties such as TA, dissolved inorganic carbon and apparent oxygen utilization [35].
As previously noted, the range of pCO2 variation also suggested that this property changed as a function of tide amplitude.Higher variability was, in general, observed in association with spring tide data (Figure 5F).For example, when higher tide amplitude occurred (Δ = 2.5 m) on 17 April

Tidal Variability
The evolution of monitored environmental variables was distinct along tidal cycles and under different tide conditions (neap/spring) in the Tagus estuary, clearly revealing significant tidal variability.These patterns were shaped by the effects of tide amplitude alone, or in combination with a tidal asymmetric behavior, since flood is typically one hour longer than ebb in the estuary, leading to higher current velocities during ebbs [34].An examination of the temporal distribution of environmental variables together with tidal data from autumn 2007 revealed some typical features of dynamics of the estuary, namely, a dominant east-west current transporting CO 2 -rich water from the inner to the lower estuary, while the reverse tidal current imported saltier and more productive (phytoplankton-rich) seawater into the estuary (Figures 3 and 4).When river discharge was minimum (autumn 2007), salinity amplitude in surface water varied between 1.6 and 8.4 along a 13-h cycle.Considering that vertical stratification was absent (corroborated by CTD profiles, data not shown), variations in salinity were mostly controlled by horizontal mixing and advection.The effect of such physical processes along the tidal cycle was noticed in the concentration trends of the water properties.Similar water advection control was also reported in the Guadalquivir estuary (SW Spain), for properties such as TA, dissolved inorganic carbon and apparent oxygen utilization [35].
As previously noted, the range of pCO 2 variation also suggested that this property changed as a function of tide amplitude.Higher variability was, in general, observed in association with spring tide data (Figure 5F).For example, when higher tide amplitude occurred (∆ = 2.5 m) on 17 April (Table 1, spring data), the amplitude of pCO 2 over the tidal cycle was 330 µatm.By contrast, on 24 April when the tidal range was only 1.4 m (Table 1), pCO 2 amplitude was only 100 µatm.Differences in variability were not, however, so significant in summer and autumn, a pCO 2 amplitude difference of ~21% being observed between tides, contrasting with 70% obtained in spring.
The tendency of elevated pCO 2 at low tides in the Tagus channel can also be attributed to biogeochemical processes, namely, the inorganic carbon respiration in the upper/central estuary.High loadings of organic matter and nutrients (data not shown) originating in the shallow inner estuarine areas could reach the study site at low tide, supporting in situ production of CO 2 .Several authors have concluded that benthic mineralization processes can be a major source of CO 2 to the water column in different coastal systems (e.g., Hammond et al. [36], Cai et al. [37], and Forja et al. [38]).The Tagus estuary is no exception [39][40][41][42].Moreover, the relatively low correlation between pCO 2 and Chl a (r = −0.574,p < 0.05) agrees with the GLM results (6.5% deviance for Chl a).
The thermodynamic impact of temperature on pCO 2 variation using an average temperature for each sampling period [43] yields a correction of 2% of the measured value on average, suggesting that the difference in the magnitude of pCO 2 between tides is not affected by thermodynamic effects.Hence, the variability of CO 2 on a tidal scale in the Tagus estuary is essentially influenced by variables (e.g., salinity) and processes such as tidal advection and horizontal mixing of the two water masses-estuarine water and oceanic water.Biological activity seems to account to a lesser extent to such variability.More accurate studies on the dynamics of the lower Tagus estuary would be necessary to expand on these assumptions.However, such studies require an increase in the sampling frequency and/or the presence of mooring buoys with physicochemical sensors, both limiting by their operational and financial costs.

Seasonal Variability
All studied variables reflect typical seasonal variations of a mid-latitude system (Figure 5).The highest observed pCO 2 value in winter was 890 µatm, corresponding to an oversaturation of ~230% of pCO 2 (calculated using the annual mean of atmospheric pCO 2, air of 389.2 µatm).During spring and summer pCO 2 values dropped to 402 µatm and increased again to a maximum of 990 µatm in autumn, showing an oversaturation of ~250%.These values fall within the ranges reported for other coastal systems [9,11,44].
The GLM applied to the entire data set (n = 101; Table 3) points to the thermodynamic effect of temperature and, to a lesser extent, to the biological activity, as the main factors controlling pCO 2 variability.The slope of the plot for pCO 2 vs. T is 19.4 µatm• • C −1 (r = 0.487, p < 0.05) in summer (Figure 6), close to the thermodynamic value of 13 µatm• • C −1 determined empirically by Takahashi et al. [43].This suggests that pCO 2 is independent of temperature in that period of the year.The negligible phytoplankton production might have contributed to the elevated pCO 2 values during autumn and winter.Actually, when a statistical GLM was exclusively applied to the productive period data set (n = 52), pCO 2 variability was found to be satisfactorily explained by chlorophyll a and dissolved oxygen (33% of the variability), therefore associated with biological activity.Although the interpretation of GLM results should be done with caution due to the relatively low number of observations under analysis, the decrease of pCO 2 values during the productive period is, most likely, the result of phytoplankton blooms and the related photosynthetic assimilation of CO 2 .
Studies on the inorganic carbon systems in coastal areas have reported high pCO 2 associated with upwelled waters, suggesting that upwelling is a source of both nutrient-rich and high CO 2 levels [45,46].This occurrence, however, is not evident in the results, probably due to the strong influence of inner estuarine waters, frequently characterized by the same combination of nutrients and high pCO 2 .

Air-Water CO 2 Exchange
The results showed significant variation in the magnitude of estimated CO 2 fluxes emitted from the Tagus, depending on the algorithm used to estimate the gas transfer velocity (Figure 7).Lower CO 2 fluxes are obtained using the C96 algorithm, due to the lower k values, probably because of strong fetch limitation and low tidal currents in the Parker River estuary [29].Fluxes estimated by the three algorithms differed significantly, similarly to what has been observed in numerous CO 2 studies [31,47,48].It has been pointed out that the greatest source of uncertainty in the calculation of gas fluxes arises from the rate term k in both open and coastal environment processes [31].In shallow coastal sites, however, where important sources of turbulence exist (e.g., boundary friction, current velocity), one or more of these processes may control the air-water exchanges.In macrotidal estuaries specifically, tidal currents were found to significantly affect k, especially under low wind conditions [31,[48][49][50].
Inherent differences in various measurement techniques is another source of uncertainty, leading to the emergence of new alternatives for standard formulations for the solubility and gas transfer velocities across air-water surfaces (e.g., [51][52][53][54]).However, such formulations require more comprehensive simulations of atmosphere-ocean interactions, and further calibration and validation.For this reason, some authors [31,[48][49][50]55] concluded that a simple parameterization of k as a function of wind speed should be site-specific in estuaries, and that the use of generic relationships to estimate air-water fluxes could lead to substantial errors.Nevertheless, the variability of the calculated air-water CO 2 fluxes throughout the tidal cycle is larger (50-80%) than the variability of fluxes using different parameterizations, as a result of fluctuations in both ∆pCO 2 and wind speed.Hence, for estimates of CO 2 sinks and sources in the Tagus estuary, or for areas and seasons characterized by strong diurnal cycles of temperature and productivity, it is preferable to perform measurements that account for short time variability.
Despite the different values coming from the various parameterizations, CO 2 fluxes show a clear seasonal pattern.The highest fluxes were attained in winter (24.5-65.5 mmol•m −2 •d −1 ) and the lowest in spring (0.4-21.3 mmol•m −2 •d −1 ).Another common feature was the highest emissions, coincident with the highest wind speed (Figure 7B), indicating that CO 2 exchanges were, to a large extent, driven by wind.Some exceptions were noticed during the productive period (Figure 7B), when high wind speed was associated with a decrease in pCO 2 .This phenomenon is explained by low ∆pCO 2 (Figure 5F), most likely due to carbon uptake by primary producers, as suggested by the higher values of chlorophyll a observed for that period (spring and summer, Figure 5C).
Throughout the year 2007, the part of the Tagus estuary directly influenced by marine waters acted as a weak source of CO 2 to the atmosphere, at an average rate of 7.2 mol•C•m −2 •year −1 (average value for the three parameterizations).This value is within the range reported for near-shore ecosystems [9,11].

Conclusions
Several interrelated factors and processes, occurring at tidal and seasonal scales, control the CO 2 dynamics in the lower estuarine areas under direct influence of marine waters.The generalized linear model applied to the lower Tagus estuary 2007 data set (n = 101) explained 69.3% of the pCO 2 variability, pointing to the thermodynamic effect of temperature and biological activity as the most important processes in CO 2 dynamics.The tidal variation of pCO 2 was found to correspond to ~35% of the seasonal range of variability, emphasizing the importance of tidal fluctuations on carbon variability in the estuary.As a consequence, the use of seasonal data sets without daily quantifications and characterizations of tidal trends may lead to significant errors in annual carbon fluxes and budget calculations.Lower pCO 2 values consistently occurred in high tides.Thus, the use of "single-point" monthly or seasonal data collected during low tide to calculate a CO 2 gas-flux budget might overestimate the flux of CO 2 to the atmosphere at this specific location.Similarly, the use of monthly or seasonal data collected during high tides at high solar radiation could underestimate the flux of CO 2 to the atmosphere.
Overall, at the time of sampling, the lower Tagus estuary acted as a relatively weak source of CO 2 to the atmosphere, at an average rate of 7.2 mol•m −2 •year −1 (average value for the three parameterizations), with highest fluxes in winter.On average, the fluxes estimated by the RC01 and B04 algorithms differed from those obtained from C96, respectively by 22% and 36%.Despite the different values coming from the various parameterizations, CO 2 fluxes showed a clear seasonal pattern: highest fluxes in winter (24.5-65.5 mmol•m −2 •d −1 ) and lowest in spring (0.4-21.3 mmol•m −2 •d −1 ).Another common feature was the highest emissions to be, in general, coincident with the highest wind speed, indicating that CO 2 exchanges are, to a large extent, driven by wind speed.
The results strongly illustrate the need for higher temporal resolution data for the assessment of coastal carbon fluxes for this estuary, supporting previous similar claims by other researchers referring to other ecosystems [14,46,[56][57][58].Therefore, the combination of seasonal point and transect evaluations with periodic diurnal measurements of system parameters will improve annual carbon budget calculations and may provide better numeric modelling of water quality.Furthermore, it may provide additional insights on why the continental shelves might have switched from a source to a sink of CO 2 over the last century [59].
Finally, the methodological approach presented here can be improved in future works, to reflect recent findings in the study of CO 2 air-water fluxes.Such enhancements can include, for example, the calculation of the upwelling index to obtain a better insight on the influence of this source of CO 2 at the monitored areas, as done by Ribas-Ribas et al. [45], as well as the estimate of the uncertainty propagation associated with pCO 2 calculations, as proposed by Orr et al. [60].

Figure 1 .
Figure 1.Map of Tagus estuary (SW Portugal) showing the location of the fixed sampling station.

Figure 1 .
Figure 1.Map of Tagus estuary (SW Portugal) showing the location of the fixed sampling station.

Figure 2 .
Figure 2. Temperature-salinity diagram for the surface waters sampled at the fixed station in 2007.

Figure 2 .
Figure 2. Temperature-salinity diagram for the surface waters sampled at the fixed station in 2007.Estuarine levels of CO 2 follow the tidal cycle in autumn: an increase in pCO 2 values at ebb, both under spring and neap tide conditions (Figure 3D,H), indicates the transport of CO 2 -enriched estuarine waters to the coast.Such features are also evidenced on the plots of pCO 2 against salinity (Figure4F) and, although not so obviously, with tidal elevation (Figure4G).For this period of the year, the tidal influence on the variability of pCO 2 was statistically significant (Fisher's test, p < 0.001, n = 26).Considering all samples for 2007 (n = 101), the difference of the average ranges of tidal pCO 2 was 208 µatm.This value was calculated as the average of the difference between the lower and higher values at each tidal cycle.

Figure 3 .
Figure 3. Tidal variation of: (A, E) salinity and temperature, (B, F) chlorophyll a and dissolved oxygen, (C, G) normalized alkalinity (nTA) and pH, and (D, H) CO2 partial pressure (pCO2) and tidal height, in spring and neap tide in the Tagus fixed station, for autumn 2007.

Figure 3 .
Figure 3. Tidal variation of: (A,E) salinity and temperature, (B,F) chlorophyll a and dissolved oxygen, (C,G) normalized alkalinity (nTA) and pH, and (D,H) CO 2 partial pressure (pCO 2 ) and tidal height, in spring and neap tide in the Tagus fixed station, for autumn 2007.

Figure 5 .
Figure 5. Box plots of (A) temperature, (B) salinity, (C) chlorophyll a, (D) pH, (E) normalized alkalinity (nTA), and (F) CO2 partial pressure (pCO2) for the sampling periods at the Tagus fixed station.Values are median with quartiles in boxes, 90% confidence limits are shown as bars, and observations outside 90% confidence limits are shown as diamonds (outliers).

Figure 5 .
Figure 5. Box plots of (A) temperature, (B) salinity, (C) chlorophyll a, (D) pH, (E) normalized alkalinity (nTA), and (F) CO 2 partial pressure (pCO 2 ) for the sampling periods at the Tagus fixed station.Values are median with quartiles in boxes, 90% confidence limits are shown as bars, and observations outside 90% confidence limits are shown as diamonds (outliers).

Figure 6 .
Figure 6.Distribution of CO2 partial pressure (pCO2) as a function of surface temperature for all sampling periods in the Tagus fixed station.The dashed horizontal line represents the mean atmospheric CO2 pressure (pCO2, air = 389.2µatm) for the year 2007.

Figure 6 .
Figure 6.Distribution of CO 2 partial pressure (pCO 2 ) as a function of surface temperature for all sampling periods in the Tagus fixed station.The dashed horizontal line represents the mean atmospheric CO 2 pressure (pCO 2, air = 389.2µatm) for the year 2007.

Figure 7 .
Figure 7. Average CO2 fluxes for (A) all seasons sampled during 2007, and (B) as a function of wind speed (u10).Positive values denote a flux from the water to the atmosphere.The values were calculated with three parameterizations: Carini et al. (1996) algorithm (C96), Raymond and Cole (2001) algorithm (RC01), and Borges et al. (2004a) algorithm (B04).Diamonds on panel A show average air-water fluxes for each parameterization.

Figure 7 .
Figure 7. Average CO 2 fluxes for (A) all seasons sampled during 2007, and (B) as a function of wind speed (u 10 ).Positive values denote a flux from the water to the atmosphere.The values were calculated with three parameterizations: Carini et al. (1996) algorithm (C96), Raymond and Cole (2001) algorithm (RC01), and Borges et al. (2004a) algorithm (B04).Diamonds on panel A show average air-water fluxes for each parameterization.

Table 1 .
Sampling dates and ranges of hydrologic and wind data recorded for each sampling period in the Tagus estuary fixed station during 2007.Q represents Tagus river flow and u 10 the wind speed referenced to a height of 10 m (expressed in m•s −1 , see Section 2.4).

Table 2 .
Range of physical-chemical and biological parameters in each sampling period during year 2007 in the Tagus estuary fixed station.

Table 3 .
Analysis of deviance for the generalized linear model (GLM) fitted (with Gaussian family and link log) of pCO 2 (n = 101).Abbreviations of variables are referred to in Section 2.3, Section 2.5, and Section 2.6.Terms (variables) are added sequentially (first to last).Akaike information criterion (AIC) = 1200.Coefficient estimates and corresponding standard error for the final model.

Table 3 .
Analysis of deviance for the generalized linear model (GLM) fitted (with Gaussian family and link log) of pCO2 (n = 101).Abbreviations of variables are referred to in Sections 2.3, 2.5, and 2.6.Terms (variables) are added sequentially (first to last).Akaike information criterion (AIC) = 1200.Coefficient estimates and corresponding standard error for the final model.