Drought conditions disrupt atmospheric carbon uptake in a Mediterranean saline lake

Inland saline lakes play a key role in the global carbon cycle, acting as dynamic zones for atmospheric carbon exchange and storage. Given the global decline of saline lakes and the expected increase of periods of drought in a climate change scenario, changes in their potential capacity to uptake or emit atmospheric carbon are expected. Here, we conducted continuous measurements of CO₂ and CH₄ fluxes at the ecosystem scale in an endorheic saline lake of the Mediterranean region over nearly 2 years. Our focus was on determining net CO₂ and CH₄ exchanges with the atmosphere under both dry and flooded conditions, using the eddy covariance (EC) method. We coupled greenhouse gas flux measurements with water storage and analysed meteorological variables like air temperature and radiation, known to influence carbon fluxes in lakes. This extensive data integration enabled the projection of the net carbon flux over time, accounting for both dry and wet conditions on an interannual scale. We found that the system acts as a substantial carbon sink by absorbing atmospheric CO₂ under wet conditions. In years with prolonged water storage, it is predicted that the lake's CO₂ assimilation capacity can surpass 0.7 kg C m² annually. Conversely, during extended drought years, a reduction in CO₂ uptake capacity of more than 80 % is expected. Regarding CH₄, we measured uptake rates that exceeded those of well-aerated soils such as forest soils or grasslands, reaching values of 0.2 µmol m⁻² s⁻¹. Additionally, we observed that CH₄ uptake during dry conditions was nearly double that of wet conditions. However, the absence of continuous data prevented us from correlating CH₄ uptake processes with potential environmental predictors. Our study challenges the widespread notion that wetlands are universally greenhouse gas emitters, highlighting the significant role that endorheic saline lakes can play as a natural sink of atmospheric carbon. However, our work also underscores the vulnerability of these ecosystem services in the current climate change scenario, where drought episodes are expected to become more frequent and intense in the coming years.

1 Introduction

Inland saline lakes are diverse and play a crucial role in the global carbon cycle, serving as dynamic zones for carbon dioxide exchange with the atmosphere (Li et al., 2022; Liao et al., 2024) and long-term sinks of organic and inorganic carbon (Anderson and Stedmon, 2007; Song et al., 2013; Li et al., 2017). In limnology, however, the ecological importance of these systems has only recently been recognized, despite accounting for approximately 44 % of global lake volume and 23 % of lake surface area (Messager et al., 2016). Saline lakes vary in size from ephemeral ponds to extensive deep-water bodies, such as the Caspian Sea (Eugster and Hardie, 1978). These lakes are characterized by salinity levels that exceed 3 parts per thousand and are notably isolated from a direct marine influence (Williams, 2002; Wang et al., 2018). They are found in endorheic (hydrologically landlocked) basins across a wide array of climates, spanning cold to warm/hot arid regions in all continents, including Antarctica (Williams, 2002; Wang et al., 2018). As terminal points in many hydrological networks, they collect significant amounts of not only salts but also nutrients and organic and inorganic carbon (Anderson and Stedmon, 2007; Song et al., 2013; Batanero et al., 2017; Li et al., 2017; Liao et al., 2024).

Past estimations about the role of saline lakes on global carbon fluxes suggested that these lakes might function as hotspots for the CO₂ emission (Duarte et al., 2008). However, more recent works point out that saline lakes have lower partial pressures of CO₂ than freshwater lakes (Wen et al., 2017), and some systems appear to take up CO₂ during the winter (Li et al., 2022) or even annually (Yang et al., 2021) due to physicochemical reactions and increased activity of primary producers. Therefore, more seasonal studies on CO₂ fluxes in saline lakes are needed to understand the conditions when these systems behave as sinks or sources of CO₂. Variations in CO₂ and CH₄ flux estimates across different studies of water bodies are primarily due to the highly variable data obtained from discrete sampling (Li et al., 2022) or because of differences in sampling seasons at the intra-annual scale (Liao et al., 2024). Meanwhile, gathering continuous time series data on CO₂ and CH₄ sequestration and emission fluxes over years is needed for accurate assessment of the net carbon balance in inland water systems (Martínez-García et al., 2024). Nevertheless, long-term, uninterrupted, and direct monitoring of greenhouse gas flux dynamics at the ecosystem level is relatively scarce in aquatic ecosystems, and this is particularly true for saline lakes. To the best of our knowledge, only a couple of studies have reported continuous year-round direct measurements at the ecosystem scale for CO₂ fluxes (Yang et al., 2021; Li et al., 2022). However, saline lake characteristics differ with latitude (Hammer, 1986) and could have very different behaviours regarding carbon exchanges depending on climate conditions.

The carbon hydrochemistry in permanent saline lakes, especially in mountainous and Arctic latitudes such as the Tibetan Plateau or Svalbard, is largely influenced by surface ice formation (Anderson et al., 2024; Rysgaard et al., 2012, 2013; Wu et al., 2014; Yan and Zheng, 2015). In contrast, saline lakes in arid and semi-arid endorheic basins, including Mediterranean climates, are typically shallow, often ephemeral, and/or hypersaline due to evaporation exceeding precipitation (García et al., 1997; Batanero et al., 2017; Saccò et al., 2021). The lower depths and higher surface-to-volume ratio, driven by drought conditions, induce significant physicochemical fluctuations in these inland saline water bodies, spanning from diurnal to interannual scales (Comin et al., 1990; García and Niell, 1991; García et al., 1997; Batanero et al., 2017). Consequently, the precipitation regime and subsequent changes in groundwater levels determine the ecology of saline lakes in arid and semi-arid regions. However, research on the interannual variability of carbon fluxes in saline lakes affected by seasonal flooding and drought is lacking. This knowledge gap urgently requires focused research to elucidate the impacts of climatic variability on the carbon dynamics of these ecosystems, which have been identified as particularly vulnerable to climate fluctuations (Tweed et al., 2011). Furthermore, recent studies highlight a global decline in lake water storage in most endorheic basins and in the Sahara, Arabia, and the southern Europe basin in particular (Wang et al., 2018), a situation expected to worsen with more severe droughts in a climate change scenario, leading to lower water levels and prolonged desiccation periods (Wurtsbaugh et al., 2017; Hassani et al., 2020).

In this study, we carried out continuous and interannual measurements of CO₂ and CH₄ fluxes at the ecosystem level in Spain's saline lake Fuente de Piedra using the eddy covariance (EC) method. Serving as a model of a Mediterranean shallow saline lake, it is characterized by sporadic episodes of water retention but is predominantly dry during the summer. It is worth noting that Fuente de Piedra Lake was designated as a Ramsar site in 1983, ensuring a rich history of monitoring water storage and the various meteorological drivers of CO₂ and CH₄ fluxes. The objectives of our work are (a) to quantify carbon exchanges during the dry and the flooded conditions, determining its role as a carbon source or sink; (b) to evaluate main drivers promoting carbon exchange behaviours; and finally (c) to use available records of the time series for the main drivers of fluxes to model the annual net carbon flux of the system in the past. Our research aims to enhance our understanding of the carbon dynamics and the impacts of climate change on the net carbon balance in Mediterranean intermittent endorheic lakes.

Figure 1 Location of the Fuente de Piedra Lake (province of Málaga, southern Spain). Dot inside the polygon in the right panel shows the location of the eddy covariance tower. The areas within the footprint contributing the 90 % to measured fluxes are delimited inside polygons for daytime (blue) and night-time (red). The cross indicates the location of the piezometer used for measuring groundwater level. Source image: Sentinel-2. Wet image was taken on 17 February 2021.

2 Material and methods

2.1 Study site

Fuente de Piedra is a shallow and saline lake located in an endorheic basin in the province of Málaga, Andalusia, Spain (37.11° N, 4.77° W; elevation 410 m: Fig. 1). It spans approximately 17 km², measuring 6.8 km in length and 2.5 km in width, with a maximum depth of 1.5 m. We take advantage of Fuente de Piedra Lake's inclusion in the Ramsar Convention in 1983. This designation ensures a rich history in the monitoring of water storage and the meteorological drivers discussed in this article. Such a comprehensive dataset allows for the back projection of the net carbon flux of the system over time, incorporating both dry and wet conditions at an interannual scale. This lake is recognized as a vital habitat within a protected wetland at various levels – regional (as a natural reserve), European (designated as a special bird protection area), and international (acknowledged as a Ramsar site) – and offers an exemplary nesting ground for the pink flamingo (Phoenicopterus roseus), largely due to its shallow waters. Among primary producers, diatoms constitute the largest fraction of primary producers of phytoplankton throughout the year, being dominated by Hantzschia amphioxys, Amphora coffeaeformis, Stauronensis amphioxys, Cocconeis placentula, and Entomoneis sp. and several species of Navicula and Nitzschia sp. (García and Niell, 1993).

Salinity levels in the lake vary significantly, ranging from oligosaline (5 ppt) to hypersaline conditions (>200 ppt), influenced by the annual hydrological cycle (Batanero et al., 2017). This cycle is delineated into two distinct phases: a pooling phase during autumn and winter (December to March) and an evaporative and drought phase spanning spring and summer (April to November). The lake primarily receives water from groundwater inflow (Rodríguez-Rodríguez et al., 2006), complemented by contributions from two streams (Fig. 1) and surface runoff from surrounding farmlands. Notably, the stream entering from the northeast adds nutrients. However, sediment samples distributed across the lake and analysed through combustion (Heiri et al., 2001) showed it to be homogeneous in organic carbon (0.21±0.07 mg C), nitrogen (0.015±0.004 mg N), and the C : N ratio (14.4±2.3).

2.2 Field measurements of greenhouse gas fluxes and meteorological drivers

We employed the eddy covariance method to quantify the exchanges of CO₂, CH₄, and energy (sensible and latent heat) every 30 min from August 2021 to May 2023. Thus, the eddy covariance system was operated for more than 21 months, including two dry periods in summer. An open-path eddy covariance (EC) system was strategically positioned atop a tower, 3.1 m above ground level, on the western bank of the lake (Fig. 1). This setup included two open-path infrared gas analysers: the LI7500 for CO₂ and water vapour and the LI7700 for CH₄ (LI-COR Inc., Lincoln, NE, USA). Wind vector components (u, v, w) and sonic temperature were accurately measured using a sonic anemometer (R. M. Young 81000V, Traverse City, MI, USA). These data were recorded on a data logger (CR1000, Campbell Scientific, Logan, UT, USA) at a frequency of 10 Hz.

In addition to gas measurements, we measured a comprehensive suite of environmental and soil state variables every 10 s to capture the conditions over a representative ground surface area and collected every 30 min average by a data logger (CR1000, Campbell Scientific, Logan, UT, USA). Air temperature (T_a) and relative humidity (RH) were monitored using a thermohygrometer (HMP 45C, Campbell Scientific, Logan, UT, USA). Net radiation (R_n) was quantified by a net radiometer (NR Lite, Kipp & Zonen, Delft, Netherlands). Soil heat flux (G) calculations, including the energy stored in the soil, were facilitated by one heat flux plate (HFP01SC, Hukseflux, Delft, Netherlands) placed at 8 cm depth, complemented by three pairs of soil temperature probes (TCAV, Campbell Scientific, Logan, UT, USA) situated at depths of 4 cm and lateral distances of 3.20, 6.34, and 8.90 m from the tower.

The groundwater level (GWL) was monitored daily using a piezometer situated within a well in the salt flats, approximately 2 km south of the EC tower and on the opposite side of the lake (37.1071° N, 4.7631° W; Fig. S1 in the Supplement). The location of the piezometer coincides with the central and deepest part of the lake when there is water. Negative GWL values indicate conditions when the lake is entirely lacking surface water. Positive values measured by the piezometer were used to determine the lake depth. Furthermore, data on daily precipitation (PPT), air temperature, and incident solar radiation (spanning wavelengths from 350 to 1100 nm) were acquired from a meteorological station located adjacent to Fuente de Piedra Lake, in Sierra de Yeguas (37.1383° N, 4.8358° W; 467 m a.s.l.). The tower setup and instruments were maintained (mainly cleaning lenses of the open-path sensors) every 2 weeks.

2.3 Greenhouse gas flux data processing, quality control, and partitioning

Half-hourly means (48 measurements per day), variances, and covariances of greenhouse gas fluxes, adhering to the principles of the Reynolds decomposition, were calculated using the EddyPro^® 7.0.7 software (LI-COR), according to international standards and protocols (Sabbatini et al., 2018). The data processing protocol encompassed the following steps: (1) axis rotation for tilt correction using the double-rotation method (Wilczak et al., 2001), (2) turbulent fluctuations calculated using the block-averaging method, (3) time lag that was compensated by covariance maximization with default, (4) Webb–Pearman–Leuning (WPL) correction of air density fluctuation (Webb et al., 1980), (5) despiking and raw data statistical screening (Vickers and Mahrt, 1997), and (6) spectral corrections of high- and low-pass filtering effects. Regarding the latter, high-frequency loss due to path averaging, signal attenuation, and sensor separation was compensated according to Moncrieff et al. (1997), whereas low-frequency loss due to finite time-averaging length and detrending was corrected according to Moncrieff et al. (2004). Quality check flags were calculated for flux data according to the widely adopted methodology combining two tests: the steady-state test and the developed turbulent conditions test. Over the study period, we only selected high-quality fluxes (flag value =0) measured when the open-path sensors were totally clean according to their respective automatic gain control (AGC) values (AGC value equal to or above 56 for open-path LI-7500A CO₂/H₂O analyser and AGC value equal to or higher than 20 for LI-7700).

To quantify the sampling area of flux measurements, a footprint model was estimated using the method by Kljun et al. (2004; Fig. 1). Data periods when the wind came from the adjacent terrestrial environment (251 to 59°) were rejected, representing between 45 % and 70 % of the available daytime and night-time data respectively during the dry season (GWL ≤0) and 30 % of the available daytime and night-time data respectively during the wet seasons (GWL >0). Overall, for the nearly 2 years of measurements, 18 % and 8 % of the potential daytime data were of good quality for CO₂ and CH₄ fluxes respectively. For night-time, the available data were reduced to 10 % and 5 % respectively. In order to provide additional information regarding the turbulent flux quality (Moncrieff et al., 1997), we calculated the energy balance closure (ratio of the sum of sensible and latent turbulent fluxes, H+ LE, to the net radiation minus soil heat flux, R_n−G) using all the available data during the drought period. The obtained value was 76 % (R²=0.64; n=3117) and is within the range of most of the reported FLUXNET sites (Wilson et al., 2002; Stoy et al., 2013).

2.4 Predicting greenhouse gas fluxes as responses to meteorological drivers

We examine the relationship of CO₂ and CH₄ fluxes in response to groundwater level, serving as a proxy for long-term water storage, with air temperature as the main factor regulating respiration in the system and incident solar radiation modulating the photosynthetic rate. Since these variables were measured every 24 h, we processed half-hourly CO₂ and CH₄ fluxes collected to construct 24 h integrated values. It is important to note that gaps in long-term data records are inevitable when using eddy covariance, primarily due to instrument failure and insufficient turbulence (Falge et al., 2001; Baldocchi, 2003; Aubinet et al., 2012). Additionally, data coverage is often reduced when wind originates from undesirable sectors, though this does not necessarily compromise measurement quality (Goloub et al., 2023). Despite these limitations, most studies derive their primary results from annual or seasonal CO₂ and CH₄ balances using various gap-filling procedures. In contrast, we exercise greater caution with our limited data coverage, being very restrictive in quantifying daily fluxes. We selected dates with over 50 % of the anticipated data points, particularly those with more than 25 valid measurements well-distributed throughout the day, to calculate integrated daily flux. Before integrating daily flux values, we filled gaps in the half-hourly CO₂ and CH₄ flux data using linear interpolation for missing values. This selection criterion aimed to accurately represent the daily pattern in flux measurement distribution. Then, a trapezoidal integration of the values measured every 30 minutes was performed to calculate the daily flux.

To analyse the relationship between integrated daily fluxes and their potential environmental predictors, we employed a linear regression combined with a forward model selection technique (Aho et al., 2014). This method involved sequentially fitting a series of regression models, each incorporating different combinations of predictors and their interactions. The process began with simple models, each containing only one primary predictor, and gradually increased in complexity to include all possible interactions among predictors. We then used the Akaike information criterion (AIC) to evaluate and identify the most effective model from the set. The model with the lowest AIC was selected as the best fit, indicating it provides the most useful balance between model complexity and explanatory power. After selecting the model, we examined the influence of the predictors by analysing the slope β coefficients at a significance level of alpha =0.05, using a 95 % confidence interval to determine whether these coefficients were significantly different from zero. Additionally, while groundwater level (GWL) was measured daily, our model selection process also aimed to identify which daily measurements of air temperature and incident solar radiation (mean, minimum, or maximum) were the best predictors. This method ensured that the chosen model is robust and relevant to the ecological scales being studied.

Finally, the chosen candidate model for predicting the fluxes in the system was fitted using available records for the studied drivers, covering the period from 2001 to the present at Fuente de Piedra Lake. This approach aimed to retrospectively model the fluxes in the system. Detailed instructions on how to run these analyses are in the R script available in DRYAD (see details in the “Data availability” section).

Figure 2 Time series of (a) air temperature and incident solar radiation, (b) groundwater level and precipitation (PPT), and (c) CO₂ and (d) CH₄ flux, collected at Fuente de Piedra Lake during 2021, 2022, and 2023.

3 Results

3.1 Time series of greenhouse gas emissions and meteorological drivers

At Fuente de Piedra Lake, we observed significant seasonal variations in meteorological conditions, as illustrated in Fig. 2a. Air temperature (T_a) and incident solar radiation exhibited consistent trends throughout the study period, with mean daily values of 17±7 °C for T_a and 18±8 MJ m⁻² d⁻¹ for incident solar radiation. In contrast, these two environmental variables generally followed asynchronous patterns with groundwater level (GWL) and precipitation events, except for a brief peak in precipitation and an increase in GWL between August and September 2021 (Fig. 2a and b). Particularly during the summer months (July, August, September), the highest daily air temperature values coincided with GWLs beneath the surface and a lack of precipitation. Also, during dry conditions, a salt crust several centimetres thick developed on the sediment (see Fig. 3b). For instance, the minimum T_a recorded was 2 °C in January 2023, corresponding with a period of frequent precipitation and GWL above the surface. Conversely, the maximum T_a of 34 °C occurred in August 2021, during a period when the GWL was 24 cm below the surface.

Figure 3 Aerial photos of Fuente de Piedra Lake during a period of maximum water availability (wet, a) and during a typical dry episode (b). Note that for some periods during the dry episodes, a salt crust forms, covering practically the entire extent of the lake. The figure shows the daily pattern of CO₂ and CH₄ fluxes during wet conditions (c and e respectively) and dry conditions (d and f respectively). Water availability is measured in terms of groundwater level. Source image: Sentinel-2. Wet image was taken on 17 February 2021, and dry image was taken on 13 May 2021.

After the study period, we were able to collect 4128 measurements for CO₂ fluxes and 2425 measurements for CH₄ fluxes, which were relatively well-distributed over the study period. We observed a strong correspondence between water storage in Fuente de Piedra Lake and its capacity to assimilate atmospheric CO₂. The CO₂ flux patterns can be categorized into two distinct lake states: a flooded lake (groundwater level >0 cm, indicated from purple to blue colour in Fig. 2c and d) and a dry lake (groundwater level <0 cm, shown from orange to yellow colour). During flooding conditions, the lake acted as a CO₂ sink, with fluxes ranging from 0 to −30 µmol m⁻² s⁻¹. The CO₂ assimilation capacity increases with groundwater level and incident solar radiation, particularly from January to June.

Notably, the flooded conditions in the 2 years of the study showed marked differences in groundwater level. In 2022, we recorded the highest CO₂ uptake of 30 µmol m⁻² s⁻¹ and an averaged value of −3.11 µmol m⁻² s⁻¹ during May, when the groundwater level was at its peak, 40 cm above the surface. This period was also characterized by high variability in CO₂ flux, occasionally showing peaks of CO₂ outgassing approaching 20 µmol m⁻² s⁻¹. The peak CO₂ uptake in 2023 was approximately half of what was observed in 2022. Although this peak occurred in March rather than May, it followed a similar trend to the groundwater level (GWL), which was also about half of the peak level observed in 2022 (∼20 cm above the surface) resulting in averaged CO₂ uptake values of −3.66 µmol m⁻² s⁻¹. In contrast, under dry conditions, Fuente de Piedra Lake ceases CO₂ uptake, occasionally transitioning to minor CO₂ emissions. Notice that during extreme rainfall pulses within dry conditions, when Fuente de Piedra Lake remained relatively dry with the groundwater level below the surface, we observed notable net CO₂ emissions. For example, a heavy rainfall event in September 2021 (36 mm d⁻¹) resulted in CO₂ emissions reaching up to 10 µmol m⁻² s⁻¹, with a high elevation of the groundwater level above the surface (from −20 to near 20 cm). Also, in October of both 2021 and 2022, subsequent rainfall events (12 and 22 mm d⁻¹, respectively) corresponded with CO₂ emissions of 7 and 5.5 µmol m⁻² s⁻¹, respectively. In these cases, emissions occurred under negative GWL conditions.

In the case of CH₄, the flux was relatively stable throughout the whole study period, generally acting as a sink, with values fluctuating between −0.2 and 0.1 µmol m⁻² s⁻¹ and an average of −0.05 µmol m⁻² s⁻¹ for the study period (Fig. 2d). Furthermore, the measured CH₄ fluxes did not seem to align clearly with the meteorological variables examined (Fig. S1).

3.2 Fluxes of CO₂ and CH₄ at a daily scale

When examining the daily scale during wet conditions, it becomes evident that CO₂ assimilation predominantly takes place during daylight hours, specifically between 09:00 and 14:00 (local time) (Fig. 3). It should be noted that the few emission values observed within this time frame correspond to the emission occurrences described earlier for the dry season, which promptly followed rainfall events. In the case of dry conditions, it is noteworthy that a salt crust forms over the lake, leading to a near cessation of CO₂ fluxes. In the case of CH₄, there is also a discernible increase in assimilation fluxes between 09:00 and 14:00 LT. This observed pattern is consistent in both wet and dry conditions.

Figure 4 Prediction of CO₂ flux as a response to groundwater level with a slope of −0.115 (a), the interaction between groundwater level and daily maximum temperature (b), and the interaction between groundwater level and incident solar radiation (c). Model has an R² of 0.73. Using existing time series for the model predictors, it has been possible to reconstruct the estimated CO₂ fluxes, as well as the annual cumulative value of CO₂ removal since 2001 for Fuente de Piedra Lake (d).

3.3 Model predictions of 24 h integrated flux values in the study system

No evident relationship between CH₄ flux and the environmental predictors studied was found during the study period. Additionally, measurements obtained from the EC tower resulted in a substantial number of gaps in the CH₄ time series, making it impossible to establish a predictive model for these fluxes. On the contrary, we were able to adjust a robust regression model for CO₂ integrated daily flux. A total of 26 CO₂ integrated daily flux values were obtained for the sampling period, analysing those dates when more than 25 valid measurements were available. After AIC model selection routine, the candidate predictive model for daily CO₂ integrated flux was determined to include groundwater level, maximum daily air temperature, and mean incident solar radiation. Additionally, the model incorporated interactions between groundwater level and maximum daily air temperature, as well as between groundwater level and incident solar radiation (a summary table of the model is available in the Supplement). Indeed, the fitted statistical model has a relatively high explanatory capability (adjusted R²=0.73).

The model confirmed a positive effect of groundwater level on enhancing CO₂ assimilation in the system (Fig. 4a; ${\mathit{\beta }}_{\mathrm{GWL}}=-\mathrm{0.115}$, 95 % CI $=-\mathrm{0.214}$ to −0.016; note that assimilation corresponds with negative values of the CO₂ flux). While the isolated impact of air temperature increase on CO₂ assimilation could not be determined (${\mathit{\beta }}_{T_{\mathrm{a}}}=+\mathrm{0.07}$, 95 %CI $=-\mathrm{0.053}$ to +0.192), the model identified an antagonistic interaction between air temperature and groundwater level (${\mathit{\beta }}_{\mathrm{GWL}\times Ta}=+\mathrm{0.010}$, 95 %CI $=+\mathrm{0.005}$ to +0.015). As air temperature increases, the positive effect of GWL on CO₂ assimilation diminishes (Fig. 4b). Conversely, mean daily incident solar radiation was found to promote CO₂ assimilation (${\mathit{\beta }}_{\mathrm{Rad}}=-\mathrm{0.1}$2, 95 %CI $=-\mathrm{0.21}$ to −0.03), with a pronounced synergy between mean daily incident solar radiation and the presence of water (${\mathit{\beta }}_{\mathrm{GWL}\times \mathrm{Rad}}=\mathrm{95}\mathit{\%}$CI $=-\mathrm{0.0116}$ to −0.003), notably enhancing the capacity for CO₂ assimilation (Fig. 4c).

Using time series data for groundwater level, daily maximum air temperature, and mean daily incident solar radiation at Fuente de Piedra Lake, we generated retrospective predictions of CO₂ assimilation capacity in the system dating back to 2001 (Fig. 4d). The model predictions closely aligned with the observed values for the study period when using the time series data for the predictors, supporting the robust predictive ability of our model (Fig. S2). Our estimates indicate a pronounced fluctuation in CO₂ assimilation capacity according to hydrological variations. In years with higher groundwater level and prolonged water storage, the model predicted an exceptionally high capacity for atmospheric CO₂ assimilation of the lake, with annual values surpassing 0.7 kg C m⁻² yr⁻¹ (e.g. in 2011, 2012, 2014, 2020). In contrast, during years marked by extended droughts, a substantial reduction in CO₂ assimilation capacity was modelled, with predicted reductions exceeding 80 % compared to wetter years (Fig. 4d). These drought periods resulted in a reduction of the assimilation capacity to less than a third of the levels recorded in wet years (e.g. from 2006 to 2010).

4 Discussion

In agreement with previous research on permanent saline lakes of the Tibetan Plateau (Li et al., 2022), we demonstrate that a model Mediterranean shallow saline lake serves as a substantial carbon sink, absorbing atmospheric CO₂ when flooded. In particularly wet years, expected uptake values can exceed 1.2 kg C m⁻² yr⁻¹. Sporadic instances were also observed where the system acted as a CO₂ emitter, resulting from rewetting events that followed rainfall during dry conditions. Conversely, CO₂ uptake ceases during dry conditions, reducing the system's capacity in dry years to below 0.2 kg C m⁻² yr⁻¹. Longitudinal time series analysis reveals that prolonged droughts indeed hinder the ability of the system to assimilate atmospheric CO₂ due to the lack of water, but we also observed that an increase in air temperature during wet conditions moderates the CO₂ net assimilation capacity, a process likely related to the reduction of gas–water solubility with temperature. This underscores the pronounced impact of seasonal and interannual variability, ultimately dictated by drought and rainfall patterns, on the ability of the studied system to sequester atmospheric carbon. Moreover, this pattern also displayed considerable variability at the daily scale, closely correlating with fluctuations in incident solar radiation over daily cycles. In this regard, the CO₂ assimilation capacity of the system peaked during those hours of maximum incident solar radiation. While measurement of CO₂ (and CH₄) fluxes at multiple scales is challenging and requires specialized equipment (i.e. eddy covariance sensors), our research proposes an alternative proxy. By integrating data from environmental predictors at various scales, we have been able to reconstruct the behaviour of CO₂ exchanges between Fuente de Piedra Lake and the atmosphere. In essence, we estimate CO₂ flux through the continuous measuring of accessible environmental variables (i.e. the amount of water, air temperature, and incident solar radiation).

In shallow, well-mixed, and oxygenated systems like Fuente de Piedra Lake, the photosynthetic capacity of the phytoplankton community is closely linked to the water column height (i.e. groundwater level; Batanero et al., 2017), promoting CO₂ assimilation as the extent of the habitat for these communities expands (Wetzel, 2001). Related to the latter, a significant synergy exists between water storage in the ecosystem and incident radiation, serving as a proxy for the photosynthetically active radiation upon which photosynthesis depends. This interaction occurs on both a daily scale, associated with variations in light intensity following day–night cycles, and an annual seasonal scale, largely determined by changes in daylight hours throughout the year. Notably, during the night, the net exchange of CO₂ between the water and the atmosphere in Fuente de Piedra Lake is negligible. This could be attributed to the absence of photosynthesis during night-time. Additionally, the high salinity inherent to these environments constrains methanogenesis, which is the least energy-efficient carbon mineralization process in the redox sequence (reviewed in Soued et al., 2024). Considering the above, it appears to offer a plausible explanation for why microbial respiration does not surpass inorganic carbon assimilation through photosynthesis in systems like Fuente de Piedra Lake during wet conditions, despite the high content of dissolved organic carbon ranging between 1.00 and 13.59 mmol C L⁻¹ (Batanero et al., 2017).

What is more, despite the limited CH₄ flux data, our results position Fuente de Piedra Lake as a CH₄ sink. Rough estimates determined that Fuente de Piedra could take up on average 1.83 and 3.70 mg C m⁻² d⁻¹ during wet and the dry conditions respectively (Fig. S3). Such values are even higher than those measured in typical well-aerated soils as in forests or grasslands, with average rates of 0.4–1.26 mg C m⁻² d⁻¹ (Murguia-Flores et al., 2021; Perez-Quezada et al., 2021). Twice the value of uptake during dry conditions compared to wet ones appears to be consistent with some proposed mechanisms promoting CH₄ reduction according to the existing literature, since the increase of temperature together with gas diffusivity due to loss of water may increase methane oxidation in a similar way to terrestrial ecosystems (Chen et al., 2011; Rafalska et al., 2023). However, caution is needed when interpreting our results, as the dynamics of methane fluxes could become very complex in an intermittent system like Fuente de Piedra. On the one hand, just as methanogenic activity is inhibited by salinity (Herbert et al., 2015), methanotrophic activity has also been observed to be significantly reduced by salinity in terrestrial systems (Ho et al., 2018). However, methane oxidation processes associated with aquatic prokaryotes may be more resistant to salinity (Khmelenina et al., 2010; Deng et al., 2017), especially if the variation is gradual (Osudar et al., 2017). Thus, further measurements and analysis are needed to estimate the role of methane oxidation and the relevance of intermittent saline lakes as CH₄ sinks in a climate change scenario.

Drought conditions are accompanied by an increase in air temperature, with the high air temperatures recorded immediately before the system completely dries out. We have found that this rise in air temperature leads to a reduction of the system's capacity to assimilate CO₂, even during wet conditions. A direct consequence of climatic warming is the reduction of gas solubility accentuated in saline wetlands (Batanero et al., 2022). In addition, an increase in temperature can enhance microbial metabolic rates and, therefore, biomass-specific CO₂ production (Smith et al., 2019). Given that endorheic saline lakes are fuelled by significant amounts of organic matter (Li et al., 2017; Batanero et al., 2017; Song et al., 2013), it is unsurprising that warming leads to a decrease in net primary production in the system as a result of enhanced microbial respiration and, consequently, a reduction in CO₂ assimilation capacity. In addition, carbon emissions in inland waters could increase with warming, independently of organic carbon inputs, simply because the apparent activation energy is predicted to be higher for respiration than for photosynthesis (Yvon-Durocher et al., 2010, 2012). Finally, it has been recognized that photosynthesis is often the first process to be affected by environmental stressors, with photosynthetic capacity diminishing prior to other cellular functions (Feller, 2016; Cardona et al., 2018). Specifically, carbon assimilation through the Calvin–Benson cycle exhibits particular vulnerability to both drought and elevated temperatures, occurring even when photosynthetic electron transport continues to operate effectively (Sharkey, 2005). On the whole, we show the profound synergy between global warming and intensifying drought severity and frequency, disrupting the CO₂ assimilation capacity of Mediterranean saline lakes and leading to negative feedback loops.

While the desiccation of saline lakes is not novel, with researchers highlighting the concerning increase in dry periods within many of these ecosystems over recent decades (Williams, 1993; Wurtsbaugh et al., 2017; Wang et al., 2018), our study underscores the significant implications this trend has for the ecosystem services they support. Our retrospective predictions show that in wet years, the system could exhibit a high CO₂ assimilation rate. For instance, between 2010 and 2015, we estimated that Fuente de Piedra Lake had an average assimilation rate of 0.83 (SD $=\pm \mathrm{0.2}$7) kg C m⁻² yr⁻¹, within the same range as the net assimilation observed in evergreen or deciduous forest systems worldwide (Pastorello et al., 2020) and marine salt marshes (Mayen et al., 2024). Considering the 13.6 km² area of Fuente de Piedra and assuming a constant flux across the lake surface when flooded, we could estimate around 11 t of carbon sequestered during a wet and fully flooded year. Although this upscaling exercise is crude, it highlights that Mediterranean shallow saline lakes could be a substantial carbon sink and underline the potential keeping them inundated with water during their conservation policies. Therefore, further studies across different Mediterranean shallow saline lakes are necessary to refine these estimates and enhance our understanding of their role in the global-scale carbon budget.

5 Conclusions

Our result challenges the generalized belief that inland waters primarily act as sources of greenhouse gases (Raymond et al., 2013). Conversely, the system undergoes significant reductions in its annual atmospheric CO₂ sequestration capacity during dry conditions. For instance, under severe drought conditions as observed in Fuente de Piedra from 2005 to 2009, the annual CO₂ sequestration is estimated to have fallen to less than a quarter of what was observed in more humid conditions. Climate change projections, including even the most optimistic scenarios, forecast an increase in both the frequency and duration of heatwaves and droughts in the coming years (Trenberth, 2011; Perkins-Kirkpatrick, 2020). This implies that saline lake ecosystems in arid and semi-arid endorheic basins will remain dry for longer periods or may even vanish, resulting in the loss of a significant carbon sequestration pathway. Importantly, the disappearance of saline lakes due to water scarcity has been largely attributed to anthropogenic water overuse (i.e. agriculture) rather than to macroclimatic phenomena (Wurtsbaugh et al., 2017). This seems to be the case of Fuente de Piedra Lake, as the catchment area is dominated by agricultural land. Thus, a proper water system management during drought conditions seems to be the most plausible solution to preserve the ecosystem services provided by Mediterranean saline lakes.