Greenhouse gas fluxes from reservoirs determined by watershed lithology, morphometry, and anthropogenic pressure

Human population growth has increased the demand for water and clean energy, leading to the massive construction of reservoirs. Reservoirs can emit greenhouse gases (GHG) affecting the atmospheric radiative budget. The radiative forcing due to CO2, CH4, and N2O emissions and the relative contribution of each GHG in terms of CO2 equivalents to the total forcing is practically unknown. We determined simultaneously the CO2, CH4, and N2O fluxes in reservoirs from diverse watersheds and under variable human pressure to cover the vast idiosyncrasy of temperate Mediterranean reservoirs. We obtained that GHG fluxes ranged more than three orders of magnitude. The reservoirs were sources of CO2 and N2O when the watershed lithology was mostly calcareous, and the crops and the urban areas dominated the landscape. By contrast, reservoirs were sinks of CO2 and N2O when the watershed lithology was predominantly siliceous, and the landscape had more than 40% of forestal coverage. All reservoirs were sources of CH4, and emissions were determined mostly by reservoir mean depth and water temperature. The radiative forcing was substantially higher during the stratification than during the mixing. During the stratification the radiative forcings ranged from 125 mg CO2 equivalents m−2 d−1 to 31 884 mg CO2 equivalents m−2 d−1 and were dominated by the CH4 emissions; whereas during the mixing the radiative forcings ranged from 29 mg CO2 equivalents m−2 d−1 to 722 mg CO2 equivalents m−2 d−1 and were dominated by CO2 emissions. The N2O contribution to the radiative forcing was minor except in one reservoir with a landscape dominated by crops and urban areas. Future construction of reservoirs should consider that siliceous bedrocks, forestal landscapes, and deep canyons could minimize their radiative forcings.


Introduction
Human population growth has increased the need for water and clean energy, promoting the construction of reservoirs for irrigation, consumption, and hydropower. The number of reservoirs has increased significantly over the past 60 years, reaching over 16.7 million dams globally (Lehner et al 2011). This trend is still ongoing, especially in countries with emerging economies, where over 3000 major hydropower dams are either planned or under construction (Zarfl et al 2015). Now it widely is accepted that inland waters, including reservoirs, despite their small global surface area, contribute much in proportion to the global carbon cycle (Tranvik et al 2009, Raymond et al 2013. Reservoirs have a radiative forcing dependent on their greenhouse gas (GHG) emissions (Barros et al 2011, Deemer et al 2016. The CO 2 emissions from inland waters (ca. 2.1 Pg C yr −1 ) are similar in magnitude to the estimate of the global uptake of CO 2 by the global ocean (2.4 Pg C yr −1 ) (Le Quéré et al 2018). Lakes and reservoirs are usually CO 2 supersaturated (Cole et al 1994), releasing 0.32 Pg C yr −1 (Raymond et al 2013).
In carbonate-poor lakes, an excess of respiration over primary production produces supersaturation, whereas, in calcareous watersheds, supersaturation is due to the loadings of inorganic carbon during the weathering (López et al 2011, McDonald et al 2013, Marcé et al 2015, Weyhenmeyer et al 2015. Inland waters are not only sources of CO 2 , but they can be significant sources of CH 4 and N 2 O (Tranvik et al 2009, Soued et al 2015 with warming potential of 34 and 298 times higher than CO 2 in a 100 year timescale (IPCC 2013). CH 4 emissions from reservoirs appear to be responsible for the majority of their radiative forcings (ca. 80% of the CO 2 equivalents) and are comparable to emissions from paddies or biomass burning (Deemer et al 2016, Samiotis et al 2018). Reservoirs, collectively considered, emit 13.3 Tg C yr −1 of CH 4 , although there is an astonishing lack of data, which severely limits our confidence in this global estimation (Deemer et al 2016). Methanogenesis is a microbial process more sensitive to temperature than other processes as, for instance, methanotrophy, respiration, and photosynthesis (Marotta et  On the other hand, the eutrophic reservoirs emit at least one order of magnitude more CH 4 than the oligotrophic ones. Indeed CH 4 emissions seem to be closely linked to primary productivity (Schmidt and Conrad 1993, Grossart et al 2011, Bogard et al 2014, Tang et al 2014, Deemer et al 2016. Phytoplankton-derived organic carbon appears to fuel higher rates of methane production than terrestrial-derived organic carbon (West et al 2012(West et al , 2016. Reservoir eutrophication is increasing worldwide as a result of the intensification of agriculture and the use of fertilizers (Canfield et al 2010, Heathcote andDowning 2012). Then, the expected increase in global temperatures along with reservoirs eutrophication might exacerbate CH 4 emissions.
The anthropogenic production of nitrogen fertilizer has doubled the inputs of this element to the Earth's surface, changing the nitrogen cycle. This change likely exceeds all the other human interventions in the cycles of nature (Gruber andGalloway 2008, Schlesinger 2009), but in comparison with the carbon cycle has received less attention (Battye et al 2017). Changes in land-use from forestal to agricultural or urban can boost the production of N 2 O due to nitrogen loadings into the aquatic systems (Seitzinger et al 2000, Mulholland et al 2008, Beaulieu et al 2011. N 2 O is produced aerobically by nitrification and anaerobically by denitrification depending on oxygen availability (Canfield et al 2010). In reservoirs, the few available data suggest that they are relevant in agricultural landscapes (Beaulieu et al 2015). Unfortunately, the importance of the reservoirs in global N 2 O emissions is practically unknown. Deemer et al (2016) estimated, using a very scarce database, that the global N 2 O emission from reservoirs is 0.03 Tg N yr −1 accounting for 4% of the radiative forcing in a 100 year timescale.
Fluxes of CO 2 , CH 4 , and N 2 O have been reported mostly for tropical and boreal reservoirs, lacking the data of these fluxes in the Mediterranean biome, where the reservoirs are the preponderant aquatic ecosystems (Naselli-Flores 2003, Barros et al 2011, Lehner et al 2011, Morales-Pineda et al 2014, Deemer et al 2016. In this region, reservoirs provide drinking and irrigation water (Naselli-Flores 2003, Morales-Pineda et al 2014); consequently, they are close to agriculture and urban areas having high human pressure. Therefore, we need more simultaneous measurements of CO 2 , CH 4 , and N 2 O emissions in Mediterranean reservoirs submitted to contrasting anthropogenic pressure to get more accurate estimates of the global reservoir radiative forcing.
Here, we simultaneously measured the fluxes of CO 2 , CH 4 , and N 2 O in a group of Mediterranean reservoirs. We covered the vast idiosyncrasy of temperate reservoirs to obtain their radiative forcings in terms of CO 2 equivalents. We hypothesized that reservoirs located in anthropogenic landscapes would have higher radiative forcings than forestal reservoirs. Besides, we postulated that CH 4 emissions would be the main responsible for the positive radiative forcings.

Study reservoirs
We sampled 12 reservoirs between July 2016 and August 2017 in the South of Spain (figures 1(a), (b)). The reservoirs are located in watersheds with diverse lithology (figures 1(c), (d); supplementary figures 1-12 is available online at stacks.iop.org/ERL/15/044012/ mmedia), different land-use (figures 1(e), (f); supplementary figures 13-24), morphometries, and ages (supplementary table 1). We quantified the CO 2 , CH 4 , and N 2 O fluxes using a PICARRO Cavity Ring-Down Spectroscopy (CRDS) gas analyzer connected to a floating chamber during the stratification (summer) and mixing (fall-winter) periods at one representative location. The reservoirs were built between 1932 and 2003, and they also differ in chemical and trophic characteristics with a range of chlorophyll-a from 0.6 to 18.6 μg l −1 and a range of dissolved organic carbon (DOC) from 0.79 to 4.95 mg l −1 . More basic details on the study reservoirs in León-Palmero et al (2019) and supplementary tables 1 and 4. We collected data on reservoir area, capacity, age, watershed lithology, and land use from open databases (more details in supplementary methods).
Quantification of CO 2 , CH 4 and N 2 O fluxes We measured CO 2 , CH 4 , and N 2 O fluxes using a highresolution laser-based CRDS (PICARRO G2508) coupled to a floating chamber. For each reservoir in each sampling period, we took 3-5 measurements for 40 min. We calculated the daily (from 10 am to 4 pm) average and the standard error from these measurements. We obtained the flux calculation using the equation (1) (Zhao et al 2015): where Flux water-air (μmol m −2 s −1 ) is the flux from the water surface to the atmosphere; the b (ppmv s −1 ) value is the slope of the linear regression between the time and the concentration of each gas inside the chamber; the V (m 3 ) is the floating chamber volume; the A (m 2 ) is the floating chamber area; the P 0 (Pa) is the barometric pressure; the R is the gas constant (8.314 m 3 Pa K −1 mol −1 ); and T 0 (K) is the ambient temperature. We checked that the slope was significantly different from zero for each measurement using a two-tailed t-Student test. We also calculated the coefficient of determination (R 2 ) for each measurement, accepting those whose R 2 >0.85 (Moseman-Valtierra et al 2016). We measured ambient temperature, barometric pressure (HANNA HI 9828), and wind speed (MASTECH MS6252A) at the beginning of each flux measurement. Determination coefficients (R 2 ) for CO 2 fluxes were always >0.85. For CH 4 fluxes, most cases R 2 were >0.85, but it decreased until 0.65 when ebullition events were relevant. In these cases, we computed the b value using the end-point concentrations and the time interval between them (equation (2)) (Zhao et al 2015): where [ppmCH4] f and [ppmCH4] i are the CH 4 concentration in the floating chamber at the end and the beginning of the time considered; t f and t i are the time at the end and the beginning of the measurement. For N 2 O flux measurements, most of R 2 values were low (even when the regression was significantly different from zero). For those cases, we first checked the analyzer precision (<25 ppb). If the changes were larger than the analyzer precision, we assumed these fluxes were different from zero. We also compared the N 2 O fluxes with the percentage of saturation of dissolved N 2 O in the water column. Details for the To obtain the reservoir radiative forcings we summed the corresponding forcing due to CO 2 emissions, the warming potential (GWP) of CH 4 in terms of CO 2 equivalents, and the warming potential of N 2 O in terms of CO 2 equivalents. We used 34 to convert CH 4 in CO 2 equivalent and 298 to convert N 2 O in CO 2 equivalent in a 100 year time horizon, including the climate-carbon feedbacks (IPCC 2013).

C, N and P analysis in the water column
We sampled the epilimnion of each reservoir for C, N and, P analysis. We measured total nutrient concentrations using unfiltered water, while we filtered through 0.7 μm pore-size Whatman GF/F glass-fiber filters samples for dissolved nutrients. We acidified with phosphoric acid (final pH<2) the samples for DOC, total dissolved nitrogen (TDN), and total nitrogen (TN). We measured DOC, dissolved inorganic carbon (DIC), TN, and TDN by high-temperature catalytic oxidation using a Shimadzu total organic carbon (TOC) analyzer (Model TOC-V CSH) coupled to nitrogen analyzer (TNM-1) (Álvarez-Salgado and Miller 1998). The instrument was calibrated using a four-point standard curve of dried potassium hydrogen phthalate for DOC, dried sodium bicarbonate and sodium carbonate for DIC, and dried potassium nitrate for TN and TDN. We analyzed two replicates and three to five injections per replicate for each sample. Samples for DOC analysis were purged with phosphoric acid for 20 min to eliminate DIC.
We measured the NO 3 − concentration using the ultraviolet spectrophotometric method, using a Perkin Elmer UV-Lambda 40 spectrophotometer at wavelengths of 220 nm and correcting for DOC absorbance at 275 nm (Baird et al 2012). We measured NH 4 + and NO 2 − concentrations by inductively coupled plasma optical emission spectrometry. Total phosphorus concentration was measured by triplicate using the molybdenum blue method (Murphy and Riley 1962) after digestion with a mixture of potassium persulphate and boric acid at 120°C for 30 min (Baird et al 2012). We also measured dissolved CH 4 and N 2 O by headspace equilibration in a 50 ml air-tight glass syringe by duplicate in the water column (Sierra et al 2017a(Sierra et al , 2017b. We analyzed simultaneously the concentration of dissolved CH 4 and N 2 O using gas chromatography (more details in supplementary methods).

Biological analyses and reservoir metabolism
We determined chlorophyll-a concentration by collecting the particulate material of 500 to 2000 ml of water by filtering through 0.7 μm pore-size Whatman GF/F glass-fiber filters, then extracting the filters with 95% methanol in the dark at 4°C for 24 h (Baird et al 2012). We measured pigment absorption using a Perkin Elmer UV-Lambda 40 spectrophotometer at wavelengths of 665 and 750 nm for scattering correction.
We recorded dissolved oxygen concentration and temperature using a miniDOT (PME) submersible water logger during the stratification period. We got measurements every 10 min for 24-48 h. We established the start and ended time for photosynthesis as 30 min before sunrise and 30 min after dawn (Schlesinger and Bernhardt 2013). We calculated the respiration rate during the night (the period between 60 min after dawn and 60 min before sunrise) (Staehr et al 2010), and we assumed that the respiration rate overnight was similar to the respiration rate over the day. The equations used to calculate lake metabolism were taken from Staehr et al (2010).

Statistical tests
We performed all the statistical analysis in R (R Core Team 2014) using the packages car (Fox and Weisberg 2011), nortest (Gross and Ligges 2015), and mgcv (Wood 2011). More details on T-test and generalized additive models (GAMs) (Wood 2006) in supplementary methods.
Results and discussion CO 2 , CH 4 and N 2 O fluxes We found that some reservoirs were sinks (fluxes<0) and other sources (fluxes>0) for CO 2 and N 2 O fluxes, but all reservoirs were CH 4 sources (figure 2, supplementary table 2). The daily average of CO 2 fluxes ranged from −131.97 to 393.11 mg C m −2 d −1 during the stratification period (figure 2(a), orange dots) and from −52.51 to 149.62 mg C m −2 d −1 during the mixing period (figure 2(a), blue dots). We measured the lower value in the Jándula reservoir (#11) consistently in both periods. We did not find significant differences between the stratification and mixing periods ( figure 2(b); supplementary table 3). The median of these fluxes in both periods was 114.00 mg C m −2 d −1 , similar to previous data for northern temperate reservoirs flux variability in these Mediterranean reservoirs was more comprehensive than the variability found in boreal lakes and reservoirs (Soued et al 2015).

CO 2 flux drivers
To determine the main drivers (predictors) of GHG fluxes in the study reservoirs, we used GAMs (see supplementary methods, supplementary tables 4 and 5). The inputs of dissolved inorganic and organic carbon and net ecosystem metabolism (i.e. the budget between photosynthesis and respiration) are considered the main drivers of CO 2 fluxes in lakes and reservoir (Tranvik et al 2009, McDonald et al 2013, Marcé et al 2015, Weyhenmeyer et al 2015. In fact, the non-calcareous area in the watershed and the reservoir respiration were the main drivers of CO 2 fluxes during the stratification period with a fit deviance of 93.4% (Log 10 (CO 2 +150)=−6.6 10 -4 non-calcareous area+1.70 log 10 (Respiration rate) 0.35 ) and an explained variance of 91% (figure 3(a); supplementary table 5). The non-calcareous area in the watershed was a linear function inversely related to CO 2 fluxes ( figure 3(b)), and most of the deviance was 90.7%, whereas respiration only explained the 9.4%. Unlike this linear function, reservoir respiration showed a power function with the CO 2 flux ( figure 3(c)). The smaller the calcareous watershed, the lower the export of DIC is. Indeed, we found a significant and negative relationship between non-calcareous area and the DIC concentration in the reservoirs irrespectively of the sampling period (linear regression results n=24, R 2 =0.50, p<0.001) ( figure 3(d)). This result agrees with previous studies showing that a significant fraction of CO 2 emissions in boreal lakes is related to inorganic carbon loading from watershed (Weyhenmeyer et al 2015). In other Mediterranean reservoirs, carbonate weathering was also related to CO 2 supersaturation and, consequently, to CO 2 evasion (López et al 2011, Marcé et al 2015. We obtained a significant and positive function between reservoir respiration and the concentration of chlorophyll-a during the stratification period (linear regression results n=12, R 2 =0.43, p<0.05) (figure 3(e)), but not with the concentration of DOC (linear regression results n=12, p= 0.64). Overall, we show, for the first time, a remarkable and direct link between watershed lithology and the CO 2 fluxes from reservoirs. CH 4 flux drivers CH 4 emissions from a reservoir depend on its net production (i.e. the budget between methanogenesis and methanotrophy) and its storage capacity into the water column. Dissolved CH 4 storage is related to water mean depth (i.e. the higher the hydrostatic pressure, the higher storage capacity is) and temperature (i.e. the lower temperature, the higher solubility is) (Keller andStallard 1994, West et al 2016).
Shallow systems are prone to have warmer waters, higher sediment exposure enhancing significantly CH 4 ebullition rates, and, consequently, less capacity to store CH 4 (Keller and Stallard 1994, Marotta et al 2014, Aben et al 2017. In the study reservoirs, we obtained that water temperature and reservoir mean depth were the main drivers of the CH 4 emissions with a fit deviance of 65.0% and an explained variance of 59% (Log 10 (CH 4 flux+1)=6.6 10 -2 Temperature −0.82+ 2.5 10 -4 e (8.44/log 10 (mean depth) ) ( figure 4(a); supplementary table 3). CH 4 emission rate was a linear and positive function of water temperature ( figure 4(b)) and accounted for 38.1% of the fit deviance. CH 4 emission rate resulted in a negative exponential function of the reservoir mean depth (figure 4(c)) with fit deviance of 27.6%. At mean depths shallower than 16 meters, the CH 4 emissions increased exponentially (i.e. 1.2 in figure 4(c)). CH 4 emissions depended on concentration of CH 4 in the surface waters following a power function (figure 4(d)) (n=24, R 2 =0.87, p<0.001). Previous studies have shown that CH 4 concentration in the water column is related to chlorophyll-a concentration (Schmidt and Conrad 1993, Grossart et al 2011, Bogard et al 2014, Tang et al 2014. We also found a positive and linear relationship between the concentration of chlorophyll-a and the concentration of CH 4 in the surface waters (n=24, R 2 =0.19, p<0.05) (figure 4(e)), but not directly with the emissions. Recent studies point out the eutrophication as the primary driver of CH 4 emissions (Deemer et al 2016, Beaulieu et al 2019. N 2 O flux drivers Nitrogen loading derived from human activities affects N 2 O emissions from inland waters (Seitzinger et al 2000, Mulholland et al 2008, Baulch et al 2011, Beaulieu et al 2011. It is widely acknowledged that the N 2 O production increases in streams and reservoirs located in agricultural and urban landscapes as a consequence of nitrate loading (Mulholland et al 2008, Baulch et al 2011, Beaulieu et al 2011, 2015. In the study reservoirs, consistently, the GAMs result showed that the TN concentration was the main driver of N 2 O fluxes along with the wind speed (Log 10 (N 2 O flux+240)= 0.72 e 0.21 TN +1.30 Wind speed 0.21 ) with a fit deviance of 82.7% and an explained variance of 79.8% (figure 5(a); supplementary table 5). N 2 O fluxes were an exponential function of TN concentration and explained most of the deviance 52.2% ( figure 5(b)). Wind speed showed a positive power function with the fluxes and only explained 13.9% ( figure 5(c)).
We determined the anthropogenic pressure in the reservoir watershed as the ratio of the area with crops plus the urban area divided by the forest area (i.e. the anthropogenic land-use ratio). We found a significant and positive relationship between this land-use ratio and the concentration of TN in the reservoir waters (n=24, R 2 =0.60, p<0.001) (figure 5(d); supplementary table 6). Both crops and urban areas increased the nitrogen concentration in their different compounds (TN, TDN, NO 3 − and NO 2 − ) (supplementary table 6). The urban area, in square kilometer or in its percentage relative in the watershed, showed a higher slope than the slope in the crop areas ( figure 5(e)). Therefore, the impact of urban development on nitrogen inputs is even higher than the influence of crop areas. N 2 O fluxes were a nonlinear function of the anthropogenic land-use ratio ( figure 5(f)). For anthropogenic land-use ratios higher than 1 (i.e. crops and urban areas predominance over the forest area), the N 2 O fluxes increased exponentially ( figure 5(f)). In contrast, we observed that for watersheds with forestal coverage more extensive than ∼40% of watershed, the N 2 O emissions decreased drastically, even becoming an N 2 O sink ( figure 5(g)). Other authors also found than boreal forest reservoirs acted as N 2 O sinks (Hendzel et al 2005). Therefore, the relevance of the nitrogen inputs from watershed on N 2 O fluxes is mostly dependent on the anthropogenic land-use. Our results suggest an exponential influence of the agricultural and, mainly, urban areas in the watershed on the N 2 O emissions. However, most of the previous studies have been mainly focused on the agricultural effects (Baulch et al 2011, Musenze et al 2014, Beaulieu et al 2015, relegating the urban influence on the background.

Reservoir radiative forcings in CO 2 equivalents
We obtained a variability range in the GHG fluxes larger than the latitudinal variability reported in previous works (Barros et al 2011). The radiative forcings due to the GHG emissions from the reservoirs differed substantially between the stratification (summer) and the mixing (fall-winter) (figure 6). Radiative forcings were substantially higher during the stratification than during the mixing. This difference could be related to the significantly higher emissions of CH 4 during the stratification than mixing (figure 2). Methanogenesis is a microbial process particularly sensitive to temperature  2). These last values were even higher than those found for tropical plantations (Laine et al 2016). In stratification, CH 4 emissions contributed significantly to the total radiative forcing (in terms of CO 2 equivalent), ranging from 3.9% to 98.32% (figure 6(a) purple sector). In contrast, the CO 2 emissions contributed to the total radiative forcing mostly during the mixing (fall and winter), accounting for up to 97% (figure 6(b) blue sector). During the mixing, the radiative forcing ranged from 28.68 mg CO 2 equivalents m −2 d −1 in Jándula reservoir (#11) to 721.65 mg CO 2 equivalents m −2 d −1 in Cubillas reservoir (#1) (supplementary table 2). The contribution of N 2 O emissions to the radiative forcing in the study reservoirs was secondary (supplementary table 2) with the exception of the Iznájar reservoir (#6) in both periods (figure 6) and the Cubillas and Jándula reservoirs (#1 and 11) during the mixing ( figure 6(b)). In the Iznájar reservoir, the N 2 O emissions accounted for up to 53.1% of the radiative forcing during the stratification, whereas during the mixing period was 22.32% (supplementary table 2). CO 2 and N 2 O emissions were driven by external factors as lithology and land-use without significant differences between mixing and stratification. In contrast, CH 4 emissions were driven by internal , Lineal relationship between the land-use ratio and the TN concentration (n=24, adj R 2 =0.62, p-value<0.001). (e), Linear relationships between the coverage of the watershed in percentage of urban area (triangles) and crops (circles). (f), Exponential relationship between the land-use ratio and the N 2 O fluxes (n=24, adj R 2 =0.60). (g), Exponential decay relationship between the coverage of the watershed in percentage of forest and the N 2 O fluxes (n=23, adj R 2 = 0.60). Orange dots stand for the fluxes during the stratification period and blue dots stand for fluxes during the mixing period. factors as water temperature and mean depth with higher emissions during stratification, which affected the total radiative forcings of the reservoirs.
Future climatic scenarios for the Mediterranean biome suggest substantial warming, a decrease in total precipitation, and extreme heat-waves and heavy precipitations (Giorgi and Lionello 2008) that likely will enhance the CH 4 emissions due to a reduction in reservoir depth (i.e. lower precipitation and higher evaporation) and an increase in the water temperatures. Climatic change may also affect nutrient loading by runoff to the reservoirs. Hydrological models for Mediterranean watersheds suggest that nutrient concentrations in reservoirs may increase despite a runoff reduction . Policies to reduce the fertilizers used in agricultural areas and, in particular, to promote the tertiary wastewater treatment in urban areas may decrease N and P loading to prevent water resources degradation and reduce GHG emissions and their subsequent radiative forcings from the already constructed reservoirs. For the construction of the projected reservoirs, the selection of optimal locations should consider that siliceous bedrock, in forestal locations, and deep canyons can minimize or even offset the GHG emissions and, consequently, reduce their radiative forcings. Figure 6. Climatic forcing due to CO 2 , CH 4 and N 2 O emissions in CO 2 equivalents in the study reservoirs in stratification (a) and mixing (b) periods. The diameter of the pie charts stands for the total climatic forcing of each reservoir. Note that scales are different for the stratification and mixing periods. Cubillas reservoir (#1) during stratification is drawn with low opacity for a better visualization of the other pie charts.