We sampled 12 reservoirs located in southern Spain (Fig. 1) between July 2016 and August 2017 once during the summer stratification and once during winter mixing. In Table 1, we show the geographical coordinates, age, and the morphometric description of the studied reservoirs. The reservoirs were built between 1932 and 2003, for water supply and agriculture irrigation, and they are located in watersheds with different lithologies and land uses (more details can be found in León-Palmero et al., 2019, 2020). These reservoirs differ in morphometric, chemical, and trophic characteristics, covering a wide range of concentrations of dissolved organic carbon (DOC), total nitrogen (TN), total phosphorus (TP), and chlorophyll a (Table 2). All raw data for the water column were deposited in the PANGAEA database (https://doi.org/10.1594/PANGAEA.912535, last access: 14 May 2020.).

We obtained the reservoir surface area, perimeter, and volume using the following open databases: Infraestructura de Datos Espaciales de Andalucía (IDEAndalucia; http://www.ideandalucia.es/portal/web/ideandalucia/, last access: 4 February 2018).) and the Ministerio para la Transición Ecológica (https://www.embalses.net/, last access: 15 September 2019).

The mean depth was calculated as follows (Eq. 1): 1Meandepth(m)=Volumem3Surfaceaream2. The shoreline development ratio (DL) (Aronow, 1982) is a comparative index relating the shoreline length (i.e., the perimeter of the reservoir) to the circumference of a circle that has the same area. The closer this ratio is to 1, the more circular the lake. A large ratio (≫1) indicates that the shoreline is more scalloped than a low ratio. The equation is as follows (Eq. 2): 2DL=Lengthoftheshoreline(m)2πAream2. The shallowness index (m-1) was obtained by dividing the shoreline development index (DL) by the mean depth (m), as in Eq. (3): 3Shallownessindexm-1=DLMeandepth(m). We sampled the water column near the dam, in the open water of the reservoir. During the stratification and the mixing period, we selected the same location. First, we performed a vertical profile of the reservoir using a Sea-Bird 19plus CTD profiler, coupled to a Spherical Underwater Quantum Sensor (LI-193R), and a fluorimeter Turner^® SCUFA (model CYCLOPS-7) for continuous measurements of temperature (∘C), dissolved oxygen (µmol L-1), conductivity (µS cm-1), turbidity (FTU – formazin turbidity unit), density (kg m-3), photosynthetic active radiation, chlorophyll a fluorescence (µg L-1), specific conductance (µS cm-1), and salinity (psu – practical salinity units). Then, based on the temperature and oxygen profiles, we selected six to nine depths, representative of the oxic and anoxic layers and the transition between them in the different reservoirs. We took the water samples using a UWITEC sampling bottle of 5 L with a self-closing mechanism. We collected samples for the dissolved CH4 analysis in 125 or 250 mL airtight Winkler bottles in duplicate (250 mL) or in triplicate (125 mL). We filled up the bottles very carefully from the bottom to avoid the formation of bubbles and minimize the loss of CH4 during field sampling. We preserved the samples with a solution of HgCl2 (final concentration 1 mmol L-1) to inhibit biological activity and sealed the bottles with Apiezon^® grease to prevent gas exchanges. We also took samples from each depth from the chemical and biological analysis explained below. We also measured barometric pressure using a multi-parameter probe (Hanna HI 9828) for the gas saturation calculations. We calculated the saturation values (%) for dissolved oxygen as the ratio of the dissolved gas measured and the gas concentration expected in equilibrium. We calculated the gas concentration in equilibrium, taking into account the differences in temperature, salinity, and barometric pressure (Mortimer, 1956).

We stored the Winkler bottles in the dark at room temperature until analysis in the laboratory. We measured dissolved CH4 using headspace equilibration in a 50 mL airtight glass syringe (Agilent P/N 5190–1547) (Sierra et al., 2017). We obtained two replicates for each 150 mL Winkler bottle and three replicates for each 250 mL Winkler bottle. We took a quantity of 25 g of water (±0.01 g) using the airtight syringe and added a quantity of 25 mL of a standard gas mixture that had a methane concentration similar to atmospheric values (1.8 ppmv) to complete the volume of the syringe. The syringes were shaken for 5 min (Vibromatic, Selecta) to ensure mixing, and we waited 5 min to reach complete equilibrium. Then, the gas in the syringe was injected manually into the gas chromatograph (GC; Bruker^® GC-450) equipped with a hydrogen flame ionization detector (FID). We calibrated the detectors daily using three standard gas mixtures with CH4 mixing ratios of 1952, 10 064, and 103 829 ppbv, made and certified by Air Liquide (France). We calculated the gas concentration in the water samples from the concentration measured in the headspace using the Bunsen functions for CH4 (Yamamoto et al., 1976; Wiesenburg and Guinasso, 1979). The precision in the quantification of the gas mixture of CH4 used in the headspace equilibrium (1.8 ppmv) expressed as the coefficient of variation was 3.7 % (n=123). The precision of the measurement of the dissolved CH4 concentration, which included the analytical processing of the samples and the equilibration step, was 3.6 % for four to six replicates of each sample. We calculated the saturation values (%) as the ratio of the concentration of the dissolved gas measured to the gas concentration expected in equilibrium considering the temperature, salinity, and barometric pressure of each reservoir. We used the atmospheric gas concentrations provided by the Global Greenhouse Gas Reference Network website (https://www.esrl.noaa.gov/gmd/ccgg/index.html, last access: 20 September 2019), which is part of the National Oceanic and Atmospheric Administration (NOAA) Earth System Research Laboratory in Boulder, Colorado. We calculated the 2016 global mean atmospheric concentrations for CH4 (Dlugokencky, 2019) from the 2016 global monthly mean. The differences among these values and the local atmospheric concentrations are assumed to be small compared with the high dissolved concentrations obtained in the studied reservoirs.

From the discrete sampling, we selected thee or four representative depths of the epilimnion, metalimnion (oxycline), and hypolimnion and bottom layers for nutrient analysis during the stratification period. We also selected three or four equivalent depths during the mixing period. In total, we analyzed 77 samples: 41 samples from the stratification period and 36 samples from the mixing period. We determined total nutrients using unfiltered water, while we filtered the samples through pre-combusted 0.7 µm pore-size Whatman GF/F glass-fiber filters for the dissolved nutrients. We acidified the samples for dissolved organic carbon (DOC), total dissolved nitrogen (TDN), and total nitrogen (TN) with phosphoric acid (final pH <2). We measured DOC, TN, and TDN by high-temperature catalytic oxidation using a Shimadzu total organic carbon (TOC) analyzer (Model TOC-VCSH) coupled to a nitrogen analyzer (TNM-1). We calibrated the instrument using a four-point standard curve of dried potassium hydrogen phthalate for DOC and dried potassium nitrate for TN and TDN (Álvarez-Salgado and Miller, 1998). We analyzed two replicates and three to five injections per replicate for each sample. We purged the DOC samples with phosphoric acid for 20 min to eliminate all the dissolved inorganic carbon. The precision of the DOC measurements expressed as the mean coefficient of variation was 3.0 %. The mean precision for the TN and TDN was 8.2 % and 2.9 %, respectively.

We measured the NO3- concentration in duplicate with 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 (APHA, 1992). The mean coefficient of variation was 0.5 %. We measured NO2- concentrations by inductively coupled plasma optical emission spectrometry (ICP-OES). Dissolved inorganic nitrogen (DIN) was calculated as the addition of the NO3- and NO2- concentrations. The detection limits for the NO2- concentration was 1.4 µmol L-1. We measured total phosphorus (TP) concentration in triplicate using the molybdenum blue method (Murphy and Riley, 1962) after digestion with a mixture of potassium persulfate and boric acid at 120 ∘C for 30 min (APHA, 1992). The precision in the quantification of the TP concentration was 11.1 %.

We determined the chlorophyll a concentration and the abundances of cyanobacteria and photosynthetic picoeukaryotes in all the depths sampled during the discrete samplings (n=178). We determined the chlorophyll a concentration by filtering the particulate material of 500 to 2000 mL of water through pre-combusted Whatman GF/F glass-fiber filters. Then, we extracted the pigments from the filters with 95 % methanol in the dark at 4 ∘C for 24 h (APHA, 1992). We measured chlorophyll a (Chl a) absorption using a Perkin Elmer UV Lambda 40 spectrophotometer at the wavelength of 665 nm and for scattering correction at 750 nm. The detection limit was 0.1 µg L-1.

To obtain the cumulative chlorophyll a in the whole water column (mg Chl a m-2), from the discrete depths, we summed the concentration of Chl a from each stratum using the trapezoidal rule (León-Palmero et al., 2019), as indicated in Eq. (4): 4CumulativeChla=∑k=1nXik⋅Zk+1-Zk-12, where Z stands for the depth considered, and n is the number of depths sampled. Zk stands for the n sampled depth; Xij is the Chla concentration (µg L-1) at the depth Zk.

We determined in triplicate the abundances of cyanobacteria and photosynthetic picoeukaryotes using flow cytometry using unfiltered water. We collected and fixed the samples with a mixture of 1 % paraformaldehyde and 0.05 % glutaraldehyde for 30 min in the dark at 4 ∘C. Then, we froze the samples in liquid nitrogen and stored them at -80 ∘C until analysis. We analyzed the samples in the FACSCalibur flow cytometer equipped with the BD CellQuest Pro software for data analysis. We used yellow–green 0.92 µm latex beads (Polysciences) as an internal standard to control the cytometer performance every day. We used different signals for groups determination: the side scatter (SSC), chlorophyll a (red fluorescence – FL3), phycoerythrin (orange fluorescence – FL2), and phycocyanin (blue fluorescence – FL4), following the protocols and indications for data analysis of previous works (Cellamare et al., 2010; Collier, 2000; Corzo et al., 1999; Gasol and Giorgio, 2000; Liu et al., 2014). In Fig. S13 in the Supplement, we show a cytogram of the populations of cyanobacteria and photosynthetic picoeukaryotes. The mean coefficient of variation for the abundances of cyanobacteria and photosynthetic picoeukaryotes was 8.8 % and 11.4 %, respectively.

We estimated gross primary production (GPP), net ecosystem production (NEP), and ecosystem respiration (R) by measuring temporal changes in dissolved oxygen concentration and temperature using a miniDOT (PME) submersible waterlogger during the stratification period. We recorded measurements every 10 min for 24–48 h during the same sampling days. Briefly, the equation for estimating free-water metabolism from measurements of dissolved oxygen was established by Odum (1956) (Eq. 5): 5ΔO2/Δt=GPP-R-F-A, where ΔO2/Δt is the change in dissolved oxygen concentration through time, F is the exchange of O2 with the atmosphere, and A is a term that combines all other processes that may cause changes in the dissolved oxygen concentration as horizontal or vertical advection, and it is often assumed to be negligible. The calculations were performed as in Staehr et al. (2010). The physical gas flux was modeled as follows (Eq. 6): 6FgO2m-2h-1=kO2meas-O2sat, where F is the physical gas flux, and k (m h-1) is the piston velocity estimated following the equation of Jähne et al. (1987) and the indications of Staehr et al. (2010). O2meas is the actual oxygen concentration (mg mL-1), and O2sat is the oxygen concentration in water in equilibrium with the atmosphere at ambient temperature and salinity.

We calculated the hourly net ecosystem production (NEPhr) and the daytime net ecosystem production (NEPdaytime) following Eqs. (7) (Cole et al., 2000) and (8): 7NEPhrgO2m-3h-1=ΔO2gm-3h-1-F/Zmix,8NEPdaytimegO2m-3daylightperiod-1=meanNEPhrduringdaylightgO2m-3h-1×Lighthours(h). NEPhr is directly derived from the changes in dissolved oxygen (ΔO2), after accounting for physical gas flux with the atmosphere (F). Zmix is the depth of the mixed layer (m), which was inferred from the temperature profile as the upper mixed zone where the temperature remains constant. NEPdaytime is the portion of NEP between sunrise and sunset, when the photosynthesis takes place. We obtained the exact light hours from an online solar calculator (https://es.calcuworld.com/calendarios/calcular-salida-y-puesta-del-sol/, last access: 24 May 2018). We established the start and the end time for photosynthesis as 30 min before sunrise and 30 min after dawn (Schlesinger and Bernhardt, 2013). We obtained hourly R (Rhr), R during the daytime (Rdaytime), and R throughout the whole day (Rday), following Eqs. (9), (10), and (11), respectively: 9RhrgO2m-3h-1=meanNEPhrduringdarknessgO2m-3h-1,10RdaytimegO2m-3daylightperiod-1=RhrgO2m-3h-1×Lighthours(h),11RdaygO2m-3d-1=RhrgO2m-3h-1×24(h). 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 throughout the day. Finally, we obtained the GPP and NEP for the day (Eqs. 12 and 13): 12GPPgO2m-3d-1=NEPdaytime+Rdaytime,13NEPgO2m-3d-1=GPP-Rday.

We selected three or four representative depths for determining the abundance of the functional genes of the epilimnion, metalimnion (oxycline), and hypolimnion and bottom layers during the stratification period. We also selected three or four equivalent depths during the mixing period. In total, we analyzed 41 samples from the stratification period and 36 samples for the mixing period. We pre-filtered the water through 3.0 µm pore-size filters and extracted DNA following the procedure developed by Boström et al. (2004) for environmental samples. During the DNA extraction protocol, we combined a cell recovery step by centrifugation of 12–20 mL of the pre-filtered water, a cell lysis step with enzyme treatment (lysozyme and proteinase K), and, finally, the DNA recovery step with a co-precipitant (yeast tRNA) to improve the precipitation of low-concentration DNA. DNA was quantified using a DNA quantitation kit (Sigma-Aldrich) based on the fluorescent dye bisbenzimide (Hoechst 33258). Extracted DNA served as the template for PCR and quantitative PCR (qPCR) analysis to test the presence and abundance of the mcrA gene and the phnJ gene. For PCR analysis, we used the recombinant Taq DNA Polymerase (Thermo Fisher Scientific) using the Mastercycler X50 thermal cycler (Eppendorf). We ran the qPCR plates using SYBR Green as the reporter dye (PowerUp™ SYBR™ Green Master Mix, Thermo Fisher Scientific) in the Applied Biosystems 7500 Real-Time PCR System and the 7500 Software. In both cases, PCR and qPCR, we designed the standard reaction mix recipes and the thermocycling conditions using the provider specifications and primer requirements. We chose specific primers from studies performed in natural samples of freshwater. We used pure cultures as positive controls (more details below).

We targeted the alpha subunit of methyl-coenzyme reductase (mcrA) as a genetic marker to determine the existence and abundance of methanogenic Archaea in our samples. This gene appears to be an excellent marker, since all known methanogens have the methyl-coenzyme M reductase, which is the enzyme responsible for the conversion of a methyl group to CH4 (Grabarse et al., 2001). We used specific primers from West et al. (2012), adapting their procedure. The forward primer was mcrAqF (5′-AYGGTATGGARCAGTACGA-3′), the reverse primer was mcrAqF (5′-TGVAGRTCGTABCCGWAGAA-3′), and the annealing temperature was 54 ∘C. The expected size of the PCR product was ∼200 bp (bp – base pair). We used a culture of Methanosarcina acetivorans (ATCC 35395) as a positive control. We tested all the samples (n=77). We also tested the presence of the phnJ gene, which encodes a subunit of the C–P lyase complex (Seweryn et al., 2015; White and Metcalf, 2007). This enzyme cleaves C–P bonds in phosphonate compounds, releasing methane, and changes in response to the phosphate availability (Yao et al., 2016a). We ran the amplification with a pair of primers previously used by Fox et al. (2014) and Yao et al. (2016a). The forward primer was PhnJoc1 (5′-AARGTRATMGAYCARGG-3'), and the reverse was PhnJoc2 (5′-CATYTTYGGATTRTCRAA-3′), adapting the PCR procedure from Yao et al. (2016a). The annealing temperature was 52.5 ∘C, and the positive controls were run using a pure culture of Rhodopseudomonas palustris (ATCC 33872). The expected size of the PCR product was ∼400 bp. We checked the result of the amplification by running 1.5 % (w/v) agarose gel electrophoresis. If we did not detect amplification in the PCR or qPCR samples, we changed the standard procedure by increasing the DNA amount and the primers' concentration to corroborate the negative results. We tested all the samples (n=77).

We conducted 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). We performed the Shapiro–Wilk test of normality analysis and Levene's test for homogeneity of variance across groups. We performed a one-way analysis-of-variance test (ANOVA) when the data were normally distributed. In case the data did not meet the assumptions of normality, we used the paired Kruskal–Wallis rank-sum (K–W) or Wilcoxon (V) tests. We analyzed the potential sources of dissolved CH4 using simple regression analysis and generalized additive models (GAMs) (Wood, 2006). A GAM is a generalized model with a linear predictor involving a sum of smooth functions of covariates (Hastie and Tibshirani, 1986, 1990). The model structure is shown in Eq. (4): 4yi=f1x1i+f2x2i+…+fnxni+∈i, where fj is the smooth functions, and ∈i is independent identically distributed N (0,σ2) random variables. We fit smoothing functions by penalized cubic regression splines. The cross-validation method (generalized cross-validation – GCV – criterion) estimates the smoothness of the functions. We fitted the models to minimize the Akaike information criterion (AIC) and the GCV values. We calculated the percentage of variance explained by the model (adjusted R2) and the quality of the fit (deviance explained). We also fixed the effect of each predictor to assess the contribution of the other predictor on the total deviance explained. Then, the sum of the deviance explained by two predictors can be different from the deviance explained by the model due to interactive effects.

We found pronounced differences in the concentration of dissolved CH4 of the studied reservoirs among depths and seasonal periods (Figs. 2–4 and S1–9). The concentration of dissolved CH4 ranged up to 4 orders of magnitude, from 0.06 to 213.64 µmol L-1, during the summer stratification (n=96), and it was less variable during the winter mixing (n=84), ranging only from 0.02 to 0.69 µmol L-1. All depths were consistently supersaturated in CH4, during both the stratification and mixing period (Table S1 in the Supplement). The dissolved CH4 concentration and the percentage of saturation values were significantly higher during the stratification period than during the mixing period (V=78, p value <0.001, and V=78, p value <0.001, respectively). These differences in the concentration of dissolved CH4 are coherent with the differences found in the CH4 emissions from these reservoirs in the stratification and mixing periods (León-Palmero et al., 2020). The wide range in CH4 concentrations found in this study covers values reported in temperate lakes (Donis et al., 2017; Grossart et al., 2011; Tang et al., 2014; West et al., 2016), to those found in tropical lakes and reservoirs (Murase et al., 2003; Naqvi et al., 2018; Okuku et al., 2019; Roland et al., 2017). In the surface mixing layer during the stratification period (i.e., epilimnion), we found values from 0.06 to 8.18 µmol L-1 (Table S1), which is about 80 times the maximum values found in the surface waters of Lake Kivu (Africa) by Roland et al. (2017) and similar to the concentrations reported in subtropical and tropical reservoirs (Musenze et al., 2014, and references therein).

The dissolved CH4 profiles showed considerable differences among depths during the summer stratification (Figs. 2a–4a and S1a–9a) but were very homogeneous during the winter mixing in all the reservoirs (Figs. 2b–4b and S1b–9b) (Table S1). Based on the differences found during the stratification period in the dissolved CH4 profiles, we sorted the reservoirs into three types. The first type of CH4 profile included six reservoirs that were characterized by an increase in the dissolved CH4 from the oxycline to the anoxic bottom, just above the sediments, where CH4 concentration reached its maximum. In these reservoirs, the oxycline may be spatially coupled to the thermocline or not. When the oxycline and the thermocline were spatially coupled, the dissolved CH4 concentration increased exponentially from the thermocline along the anoxic hypolimnion to the sediments. The reservoirs Béznar, San Clemente, and Iznájar showed this type of profile (Figs. 2a, S1a, and S2a). The existence of a sizeable almost-anoxic hypolimnion led to a massive accumulation of CH4 in this layer. The differences in the CH4 concentration between the surface and bottom waters were up to 3 orders of magnitude, as we found in Béznar (from the 0.25 to 56.17 µmol L-1; Fig. 2a), San Clemente (from the 0.23 to 45.15 µmol L-1; Fig. 1a), and Iznájar (from the 0.82 to 213.64 µmol L-1; Fig. S2a). When the oxycline and the thermocline were not spatially coupled, the dissolved CH4 concentration increased just above the sediments, where the anoxic–oxic interface was near to the bottom. The reservoirs Cubillas, La Bolera, and Francisco Abellán showed this profile type (Figs. S3a, S4a, and S5a). This accumulation of CH4 in the hypolimnion and above sediments might be related to the high rates of methanogenesis in the sediments and its subsequent diffusion to the water column. Dissolved CH4 concentration declines at the oxycline level, where the highest rates of CH4 oxidation usually occur (Oswald et al., 2015, 2016). The CH4 profiles in this group were similar to the ones found in tropical eutrophic and temperate reservoirs (Naqvi et al., 2018; West et al., 2016). The second profile type presents a small peak of metalimnetic CH4, concomitant with peaks of dissolved oxygen, chlorophyll a, photosynthetic picoeukaryotes, and cyanobacteria (Fig. 3a). In the Negratín reservoir, we found the maximum concentration of CH4 in the oxic hypolimnion. Unlike several previous works in lakes (Blees et al., 2015; Grossart et al., 2011; Khatun et al., 2019; Murase et al., 2003), we did not find a metalimnetic CH4 maximum. Khatun et al. (2019) described the existence of a metalimnetic CH4 maximum in 10 out of 14 lakes. The metalimnetic CH4 maximum may represent a physically driven CH4 accumulation due to solubility differences with the temperature at the thermocline, the epilimnetic CH4 losses by emission, and the lateral inputs from the littoral zone (Donis et al., 2017; Encinas Fernández et al., 2016; Hofmann et al., 2010). The metalimnetic CH4 maximum can also be determined by biological factors, including the light inhibition of the methane oxidation (Murase and Sugimoto, 2005; Tang et al., 2014) or the distinctive methane production by phytoplankton due to availability of nutrients, light, or precursors at this layer (Khatun et al., 2019). The third profile type included five reservoirs, in which the dissolved CH4 profile presented a CH4 accumulation more significant in the epilimnion than in the hypolimnion. The reservoirs Jándula, Bermejales, Rules, El Portillo, and Colomera showed this profile type (Figs. 4a and S6a–9a). These reservoirs had a mean CH4 concentration in the water column significantly lower than the reservoirs from the first type. Similar profiles have been reported in temperate (Tang et al., 2014) and tropical lakes (Murase et al., 2003).

We found two well-differentiated groups of CH4 data sorted by the dissolved oxygen (DO) concentration (Fig. S10), as in previous studies (Tang et al., 2014). The first dataset included the samples with a DO lower than 7.5 µmol L-1 (n=18, hereafter anoxic samples). These samples belong to the hypolimnion of the studied reservoirs during the stratification period. The second dataset included the samples with DO higher than 7.5 µmol L-1 (n=160, hereafter oxic samples). All the samples from the mixing period (n=82) and most of the samples from the stratification period (n=78) belong to this second dataset. We found significant differences (W=2632, p value <0.001) between the concentration of CH4 in the anoxic samples (median = 15.79, min = 0.35, max = 213.64 µmol L-1) and in the oxic samples (median = 0.15, min = 0.02, max = 8.17 µmol L-1). Since these two groups of samples are different, we determined their sources and drivers separately (Table S2).