Introduction

Lakes and rivers are important components of the global carbon cycle in terms of transport and storage of organic carbon, as well as emissions of CO2 and CH4 to the atmosphere. Rivers transport from land to the ocean ~ 500 TgC/yr of organic carbon (Stallard 1998). The organic carbon buried in sediments of lakes (70–140 TgC/yr) is coequal to the rate in the oceans (120 TgC/yr), and comparable to the accumulation of organic carbon on land (soils and vegetation) during the last interglacial period (230 TgC/yr) (Cole 2013). This is striking because the total area of lakes is < 4% of the Earth’s nonglaciated land area and ~ 1% of that of the oceans. Lakes and rivers also emit to the atmosphere CO2 and CH4, two potent greenhouse gases (GHGs). The CO2 emission from lakes globally was estimated to ~ 300 TgC/yr (Raymond et al. 2013) and would be six times lower than that from rivers of ~ 1800 TgC/yr (Raymond et al. 2013; Liu et al. 2022). On the contrary, the emission of CH4 from lakes (42 TgC/yr) could be approximately ten times higher than that from rivers (4.5 TgC/yr), accounting for 21% of total aquatic emissions, according to Rosentreter et al. (2021). The global CH4 emission from lakes has been recently revised downwards (31 TgC/yr) (Johnson et al. 2022) and that from rivers upwards (21 TgC/yr) (Rocher-Ros et al. 2023). The relative importance of CH4 emissions from lakes compared with those from rivers at global scale is variable from one study to another, while at regional scale, some studies suggest that integrated CH4 emissions from rivers might be higher than those from lakes, such as at the African continental scale (Borges et al. 2022). Yet, it is clear that the exchange of CO2 and CH4 between inland waters and the atmosphere is large enough to command further interest and research effort at the global scale. Further, a large source of uncertainty in the global CH4 budget is attributed to emissions from inland waters (Saunois et al. 2020), thus, reducing the uncertainty of CH4 emissions from lakes and rivers is clearly a priority (Johnson et al. 2022; Melack et al. 2022).

The uncertainty on the estimates of CO2 and CH4 emissions from lakes and rivers cannot be reduced by measurements alone, because it will be impossible to measure in situ GHG emissions in all rivers and lakes globally, given the global extent of river networks (~ 2 × 106 km by length and ~ 770,000 km2 by area) (Allen and Pavelsky 2018) and of lakes (~ 1.4 × 106 lakes > 0.1 km2 with a combined surface area of 2.7 × 106 km2) (Messager et al. 2016). It is then required to model GHG emissions in rivers and lakes using either mechanistic models (e.g. Guo et al. 2020) or statistical relations with relevant predictors represented by spatial data sets. The choice of predictors needs to be anchored on a sound conceptual understanding of drivers of GHG emissions to minimize the application of spurious correlations and avoid redundancy in predictors of multivariate models. Numerous relevant potentially applicable spatial data sets are already available such as RiverATLAS (Linke et al. 2019) and HydroLAKES (Messager et al. 2016) and are expanding rapidly (Toming et al 2020; Altenau et al. 2021; Bonnema et al. 2022; Pi et al. 2022; Sikder et al. 2023). The most readily available predictors for lakes are morphometric, so regional or global upscaling attempts of CO2 and CH4 emissions from lakes have relied on lake surface area (Holgerson and Raymond 2016; Rosentreter et al. 2021) or depth (Borges et al. 2022). Others have integrated lake productivity based on statistical distributions of chlorophyll-a by lake size (DelSontro et al. 2018), although not accounting for strong latitudinal variations of productivity (Lewis 1987), or by integrating ecoclimatic zones (Johnson et al. 2022). So, there is an increasing need to expand and possibly standardize upscaling methods (Lauerwald et al. 2023). Yet, the largest limitation to reduce the uncertainty of the estimates of GHG emissions from rivers and lakes is the scarcity of field data of GHG emissions to adequately capture the very large spatial and temporal variability, allowing developing robust and reliable statistical models (e.g., Deemer and Holgerson 2021).

There are fewer studies of CO2 and CH4 variations in mountainous lakes than in lowland lakes. Available data suggest that there is no systematic difference in CH4 emissions between mountainous and lowland lakes, while CO2 emissions seem to be lower in mountainous lakes that occasionally act as sinks for atmospheric CO2, based on data mainly collected in the Alps (Pighini et al. 2018), Rocky Mountains (McCrackin and Elser 2011; Crawford et al. 2015), Sierra Nevada (Cohen and Melack 2020), and the Tibetan Plateau (Yan et al. 2018, 2023; Wang et al. 2022). Available data sets of GHG emissions from lakes and rivers are also spatially skewed toward higher latitudes, mainly in North America and Scandinavia. These data sets cannot capture the features of GHG emissions from lakes and rivers in other parts of the globe with different climates and land cover. Yet, the majority of riverine GHG emissions are from the tropics (20 °N to 20 °S), ~ 60–80% for CO2 (Raymond et al. 2013; Borges et al. 2015; Lauerwald et al. 2015; Sawakuchi et al. 2017; Liu et al. 2022) and ~ 50–70% for CH4 (Sawakuchi et al. 2014; Rocher-Ros et al. 2023). Emissions from tropical lakes represent 37–50% of the total for CH4 (Bastviken et al. 2011; Johnson et al. 2022) and 34% for CO2 (Raymond et al. 2013). The estimate of emissions of CO2 from tropical lakes as given in existing literature has been recently questioned and seems to have been overestimated (Borges et al. 2022). The CO2 emissions from tropical lakes might, in fact, be lower than their counterparts at higher latitudes (Borges et al. 2022), while the CH4 emissions from tropical lakes are disproportionately intense compared with their higher-latitude counterparts (Bastviken et al. 2011; Johnson et al. 2022; Borges et al. 2022). Relatively lower lacustrine CO2 but higher lacustrine CH4 emissions in tropical climates than at higher latitudes are partly related to high phytoplankton production in nonhumic lakes (Morana et al. 2023) that leads to the uptake of dissolved CO2 but at the same time high CH4 production in sediments (Borges et al. 2022) due to the deposition of fresh phytoplankton detritus (Grasset et al. 2018). Similarly, prior estimates of emissions of N2O (Kroeze et al. 2010) from both rivers and lakes globally have been recently questioned and revised downwards (Hu et al. 2016; Maavara et al. 2019; Lauerwald et al. 2019). Even so, the N2O emissions modeled from nitrogen deposition rates and emission factors (Lauerwald et al. 2019) seem to provide higher estimates in tropical lakes compared with those derived from field measurements (Borges et al. 2022). Additionally, emissions of N2O from tropical lakes are lower compared with their boreal counterparts (Borges et al. 2022; 2023). This calls for a careful evaluation and a reduction of uncertainty of the GHG emissions from tropical lakes and rivers.

Until recently, the majority of research on GHG emissions from tropical inland waters was carried out in South America and in particular on the Amazon basin, as reviewed by Melack (2016) for CO2, and Melack et al. (2022) for CH4. Melack et al. (2022) summarized published CH4 fluxes and regional estimates in several types of aquatic systems (including rivers, lakes, reservoirs, and several types of wetlands), evaluated the adequacy of methods, and listed the limitations in current estimates. Melack et al. (2022) also identified regions or habitats of the Amazon basin with few or no data that include systems at altitudes above 500 m, mainly in the Andes cordillera. These systems are expected to be sensitive to climate change as increased air temperature accelerates the thawing of glaciers, leading to an increase in surface runoff (Bliss et al. 2014; LópezMoreno et al. 2014), that should affect GHG emissions from existing aquatic environments, as well as their coverage by the emergence of new lakes and ponds (Bosson et al. 2023).

Here, we report the dissolved concentrations of CO2, CH4 and N2O in 15 lakes located in the northern region of Ecuadorian Andes along an elevational gradient from 2213 to 4361 m above sea level, as well as a gradient of lake surface area (0.003–6.1 km2) and depth (0.9–74 m) (Fig. 1). Most lakes were located in the páramos of Salve Facha and Antisana y Mojanda. We also report a large suite of biogeochemical variables that allow us to characterize the lakes and provide some insights into the biogeochemical and physiographic drivers of the variability of GHGs in the sampled lakes, such as water temperature, specific conductivity, oxygen saturation level (%O2), stable isotope composition of carbon (δ13C-DIC) in dissolved inorganic carbon (DIC), nitrate (NO3−), nitrite (NO2−), ammonium (NH4+), total alkalinity (TA), major cations (Na+, Mg2+, Ca2+, K+), dissolved silicate (DSi), Fe, stable isotope composition of O and H of H2O (δ18O-H2O and δ2H-H2O), and dissolved organic carbon (DOC) and particulate organic carbon (POC) concentrations. We explore modelling of CO2 and CH4 emissions to the atmosphere from the lakes using our data in highland lakes and literature data for lowland lakes (totaling about 30 lakes for each gas), statistical relations with relevant predictors from HydroLAKES (Messager et al. 2016), and recently published estimates of ebullitive CH4 emissions (Borges et al. 2022). Finally, the upscaled lacustrine CO2 and CH4 emissions are compared with the emissions from rivers based on a recent estimate at the scale of the Amazon basin that included data from the mountainous Andean streams (Chiriboga and Borges 2023).

Fig. 1
figure 1

Fifteen sampled highland lakes in Northern Ecuador

Full size image

Material and methods

Site description

A vast amount of headwaters can be found in the Andes, which contribute 82% of sediments and solutes to the Amazon basin (McClain and Naiman 2008). Páramos are Andean mountain ecosystems that are located discontinuously in the Neotropics at elevations between 3000 m and up to 4300 m, mainly from Venezuela to northern Peru. The páramos occupy more than 30,000 km2 in South America. Ecuador is the country with the highest proportion of its territory covered by páramos (with 7%) Páramo lakes constitute a primary source of drinking water for large populations in Ecuador, Colombia, and Venezuela. In Ecuador, the air temperature is relatively constant, the average annual rainfall is 1345 mm and occurs throughout most of the year; only ~ 12% of the days are dry. These conditions have favored the formation of peat and a large number of wetlands and lakes. The páramo usually presents deep and organic rich soils (30% by volume) and its vegetation is dominated by grasslands (Zapata et al. 2021).

Lake sampling

The lakes were sampled at different periods from April 2019 to March 2022, with an inflatable boat, during daytime (early morning to late afternoon), and using three strategies: (1) Lake Salve Fecha was sampled at a single location (close to the edge of the lake) during four occasions in 2019 and 2020 (1 April 2019, 1 September 2019, 1 October 2019, and 1 January 2019) as part of the sampling of the Napo river network (Chiriboga and Borges 2023); (2) Lakes Santa Lucía, Secas, and Humbolt were sampled on two occasions (24–25 September 2021 and 17–21 December 2021), along a transect across the longest length with equidistant points (n = 9, 6, 3, respectively); (3) the remaining 11 lakes (Cubilche 1 and 2, Cuicocha, Loreto, Mojanda, Negra, Papallacta, San Pablo, Sucus, Yaguarcocha, and Yanacocha) were sampled at only one station in the center of the lake on one occasion, from 22 March 2022 to 28 March 2022. Admittedly, CO2, CH4 and N2O in surface waters of lakes show variations in time (from subdaily to seasonal) and horizontally (from shore to the deepest part frequently in the center) (Fernandez et al. 2020; Ngochera and Bootsma 2020; Mandryk et al. 2021; Borges et al. 2022; Peacock et al. 2023), but a “one station in the center” snapshot survey of lakes has frequently been used to investigate broad geographical gradients as a function, for instance, of climate or catchment land cover (Juutinen et al. 2009; Finlay et al. 2010; Lapierre and del Giorgio 2012; Kortelainen et al. 2020; Casas‐Ruiz et al. 2021; Rodríguez-Cardona et al. 2023). Albeit not optimal, the “one station in the center” snapshot survey approach is a tradeoff between the detail of the description of within lake horizontal/temporal variability and a broad description of geographical gradients, with typical resources allocated to sampling effort. A statistical summary [average ± standard deviation (st. dev.), coefficient of variation (CV), median, first quartile (Q1), second quartile (Q2), minimum–maximum range] of the partial pressure of CO2 (pCO2) and the dissolved CH4 and N2O concentrations in Lakes Salve Fecha, Santa Lucía, Secas, and and Humbolt is given in Table 1. The data collected in Lake Salve Facha provided an indication of the temporal variability (during four different sampling occasions) at a single site, while data collected in Lakes Santa Lucía, Secas, and Humbolt provided an indication of the spatial variability (multiple sampling points), although also integrating temporal variability (during two different sampling occasions). Despite a relatively important variability indicated by the minimum–maximum and Q1–Q3 range, the median and average values of the three gases tended to converge, except for Lake Humbolt, possibly because of a lower number of samples (n = 6) (Lewis 1993). The CV indicated that CH4 was more variable (35–89%) than pCO2 (29–66%), while N2O was the least variable (6–16%). The variability of the three GHGs, as given by the CV, was highest in the smallest lake (Humbolt).

Table 1 Statistical summary (average ± st. dev., coefficient of variation (CV), median, first quartile (Q1), second quartile (Q2), minimum–maximum) of CH4 concentration (nmol/L; relative analytical precision ± 10.9%), partial pressure of CO2 (pCO2; ppm, relative analytical precision ± 5.2%), and N2O concentration (nmol/L, relative analytical precision ± 5.8%) in lakes were multiple samplings were carried out (Salve Fecha, Santa Lucía, Secas, and Humbolt). SA, surface area
Full size table

Sample collection and analysis

Water temperature, specific conductivity, pH, and %O2 were measured in surface water with a YSI multiparameter probe (ProPlus). Water samples for CH4 and N2O analysis were collected in duplicate borosilicate serum bottles (Weathon; volume 40 mL) with a 2L polyethylene bottle with the bottom cut and fitted with a silicone tubing at the stopper (Abril et al. 2007), poisoned with 100 µL of a saturated solution of HgCl2 and sealed with a butyl stopper and crimped with an aluminum cap. After overnight equilibration, measurements were made, on a headspace (Weiss 1981) (15 mL of high-purity N2 into the 40 mL sample bottles), with a gas chromatograph (SRI 8610C) with a flame ionization detector for CH4 and electron capture detector for N2O calibrated with CH4:N2O:N2 gas mixtures (Air Liquide Belgium) with mixing ratios of 1, 10, and 30 ppm for CH4, and 0.2, 2.0, and 6.0 ppm for N2O, with a precision of ± 10.9% for CH4 and ± 5.8% for N2O (based on duplicate samples).

The pCO2 in the water was measured in the field with a Li-Cor Li-820 infrared gas analyzer based on the headspace technique from four 60-mL polypropylene syringes filled directly with surface water. A headspace of 30 mL was created with ambient air that was equilibrated in the syringe with 30 mL of water by vigorous shaking during 5 min. The water temperature after equilibration was measured immediately after injection of the headspace with a portable thermometer (Testo 925). The pCO2 in the atmosphere was measured by injecting ambient air sampled with an additional polypropylene syringe. The Li-Cor Li-820 was calibrated with pure N2 and CO2:N2 gas mixtures (Air Liquide Belgium) of 388, 804, 3707, and 8146 ppm. The final pCO2 value was computed taking into account the partitioning of CO2 between water and the headspace, as well as equilibrium with HCO3− (Dickson et al. 2007) using water temperature measured in situ, after equilibration, and TA. The precision of pCO2 measurement was ± 5.2%.

The CO2 concentration is expressed as partial pressure in parts per million (ppm) and as dissolved concentration for CH4 (nmol/L). N2O data are presented as percent of saturation level (%N2O, where atmospheric equilibrium corresponds to 100%), computed from the global mean N2O air mixing ratios (https://www.esrl.noaa.gov/gmd/hats/combined/N2O.html), using the Henry’s constant (Weiss and Price 1980).

Samples for δ13C-DIC were collected in 12-mL Exetainer vials (Labco) and poisoned with 50 µL of a saturated solution of HgCl2. Prior to the analysis of δ13C-DIC, a 2-mL helium headspace was created and 100 µL of phosphoric acid (H3PO4, 99%) was added in the vial to convert carbonate (CO32−) and bicarbonate (HCO3−) ions to CO2. After overnight equilibration, up to 1 mL of the headspace was injected with a gastight syringe into a coupled elemental analyser—IRMS (EA-IRMS, Thermo FlashHT or Carlo Erba EA1110 with DeltaV Advantage). Calibration was performed with certified standards (NBS-19 or IAEA-CO-1 and LSVEC). The obtained data were corrected for isotopic equilibration between dissolved and gaseous CO2 (Gillikin and Bouillon (2007). Reproducibility of measurement based on duplicate injections was typically better than ± 0.2‰.

Water was collected in surface water with a 2-L polyethylene bottle, and a known volume was filtered through 47 mm diameter GF/F Whatman glass fibber filters. The filtrated water was collected and further filtered through polyethersulfone syringe encapsulated filters (0.2 µm porosity) for NO3−, NO2−, NH4+, TA, major elements (Na+, Mg2+, Ca2+, and K+), as well as DSi and Fe, δ18O-H2O, δ2H-H2O and DOC. An additional water filtration was made on 25 mm diameter GF/F Whatman glass fibber filters for POC analysis.

Water samples for nutrient analysis (NO3−, NO2−, and NH4+) were stored frozen (−20 °C) in 50-mL polypropylene vials. NO3− and NO2− were determined with the sulfanilamide colorimetric with the vanadium reduction method (American Public Health Association 1998), and NH4+ with the dichloroisocyanurate–salicylate–nitroprussiate colorimetric method (Standing committee of Analysts 1981). Detection limits were 0.3, 0.01, and 0.15 µmol/L for NH4+, NO2− and NO3−, respectively. Precisions were ± 0.02 µmol/L, ± 0.02 µmol/L, and ± 0.1 µmol/L for NH4+, NO2− and NO3−, respectively.

Samples for TA analysis were stored at ambient temperature in polyethylene 55-mL vials. TA was measured by open-cell titration with HCl 0.1 mol/L according to Gran (1952), and data quality was checked with certified reference material obtained from Andrew Dickson (Scripps Institution of Oceanography, University of California, San Diego, CA), with a typical reproducibility better than ± 3 µmol/kg. DIC was computed from TA and pCO2 measurements using the carbonic acid dissociation constants for freshwater of Millero (1979) using the CO2sys package.

Samples for δ18O-H2O and were δ2H-H2O stored at ambient temperature in polypropylene 8-mL vials. δ2H-H2O was measured on H2 gas derived from a high-temperature (1030 °C) Cr-based reactor by automated injections of water using a TriPlus autosampler on an elemental analyzer (Thermo Flash HT/EA; Thermo Finnigan) coupled to a continuous-flow isotope-ratio mass spectrometer (Delta V Advantage; Thermo Finnigan). δ18O-H2O values were measured on a Thermo GasBench II coupled to a Thermo Delta XP IRMS after equilibration with CO2. The long-term uncertainty for standards was ± 0.1‰ for δ18O and ± 2.0‰ for δ2H-H2O.

Samples for major elements were stored at ambient temperature in 20-mL scintillation vials and preserved with 50 μL of HNO3 (65%). Major elements were measured with inductively coupled plasma MS (ICP-MS; Agilent 7700x) calibrated with SRM1640a from National Institute of Standards and Technology, TM-27.3 (lot 0412) and TMRain-04 (lot 0913) from Environment Canada, and SPS-SW2 Batch 130 from Spectrapure Standard. The limit of quantification that in this case also corresponds to the background equivalent concentration was 0.038 mg/L for Na+, 0.005 mg/L for Mg2+, 0.077 mg/L for K+, 0.029 mg/L for Ca2+, 0.185 mg/L for DSi, and 0.37 µg/L for Fe. The precision was ± 4.1% for Na+, ± 3.8% for Mg2+, ± 4.9% for K+, ± 2.4% for Ca2+, ± 5.8% for DSi, and ± 4.6% for Fe.

Samples for DOC analysis were stored at ambient temperature and in the dark in 40-mL brown borosilicate vials with polytetrafluoroethylene (PTFE) coated septa and poisoned with 50 µL of H3PO4 (85%). DOC concentration was measured with a wet oxidation total organic carbon analyzer (IO Analytical Aurora 1030W), with a typical reproducibility better than ± 5%.

Filters for POC analysis were decarbonated with HCl fumes for 4 h and dried before encapsulation into silver cups; POC was measured on an EA-IRMS (Thermo FlashHT with DeltaV Advantage), with a reproducibility better than ± 5%. Data were calibrated with certified (IAEA-600: caffeine) and in-house standards (leucine, tuna muscle tissue) that were previously calibrated versus certified standards.

Flux computations and upscaling

The dissolved concentrations of CO2 and CH4 were estimated for each of 12,549 lakes in HydroLAKES (Messager et al. 2016) from statistical relations as function of lake surface area (LSA), derived from our own data in Ecuadorian highland lakes and data available from literature in the floodplain lakes of the Central Amazon (Abril et al. 2013; Amaral et al. 2020; Barbosa et al. 2016; Mitchell et al. 2018). We excluded from the analysis reservoirs that represent 0.06% of lake counts (n = 8) and 11% of total lacustrine surface area (2773 km2) in the HydroLAKES data set.

The pCO2 (in ppm) was computed from two linear regressions as a function of LSA (in km2) according to:

$$\begin{gathered} {\text{log}}\left( {{\text{pCO}}_{{2}} } \right)\, = \, - \,0.{1293}\left( { \pm \,0.0{346}} \right) \, \times {\text{ LSA}}\, + \,{3}.0{63}( \pm \,0.0{73}) \hfill \\ \quad (r^{{2}} \, = \,0.{58};\,p\, = \,0.00{39};\,n = \,{12}) \hfill \\ \end{gathered}$$

for an elevation > 500 m (highlands), excluding three lakes (Cuicocha, Loreto, and Papallacta) and

$$\begin{gathered} {\text{log}}\left( {{\text{pCO}}_{{2}} } \right)\, = \, - \,0.000{3745}\left( { \pm \,0.000{1642}} \right) \times {\text{LSA}}\, + \,{3}.{5}0{8}( \pm \,0.0{61}) \hfill \\ \quad (r^{{2}} \, = \,0.{26};\,p = \,0.0{376};\,n = \,{17}) \hfill \\ \end{gathered}$$

for an elevation < 500 m (lowlands).

The CH4 (in nmol/L) was computed from a linear regression as function of LSA (in km2) according to:

$$\begin{gathered} {\text{log}}\left( {{\text{CH}}_{{4}} } \right)\, = \, - \,0.{3753}\left( { \pm \,0.0{426}} \right) \, \times {\text{ log}}\left( {{\text{LSA}}} \right)\, + \,{2}.{545}( \pm \,0.0{66}) \hfill \\ \quad (r^{{2}} \, = \,0.{71};\,p\, < \,0.000{1};\,n\, = \,{34}) \hfill \\ \end{gathered}$$

for both highland and lowland lakes.

The water temperature (WT in °C) was computed from the air temperature (AT in × 10 °C) from HydroLAKES, according to:

$$\begin{gathered} {\text{log}}\left( {{\text{WT}}} \right)\, = \,0.00{2192}\left( { \pm \,0.0000{69}} \right) \, \times {\text{ AT}}\, + \,0.{89}0{5}( \pm \,0.0{151}) \hfill \\ \quad (r^{{2}} \, = \,0.{98};\,p\, < \,0.000{1};\,n\, = \,{23}) \hfill \\ \end{gathered}$$

The flux of CO2 (FCO2) and CH4 (FCH4) between surface waters and the atmosphere was computed according to:

$$F = \,k \times \, \Delta G{,}$$

where k is the gas transfer velocity and ΔG is the air–water gradient of a given gas.

The k normalized to a Schmidt number (Sc) of 600 (k600 in cm/h) was derived from the parameterization as a function wind speed (Cole and Caraco 2018):

$$k_{{{6}00}} = {2}.0{7} + 0.{215} \times u^{{{1}.{7}}} ,$$

where u is wind speed (m /s) that was extracted from WorldClim (Fick and Hijmans 2017) for each lake in HydroLAKES.

k600 was converted to k for the Sc at in situ temperature (Wanninkhof 1992) according to:

$$k = k_{{{6}00}} \times \, \left( {{\text{Sc}}/{6}00} \right)^{{ - 0.{5}}} .$$

A Monte Carlo error analysis (10,000 iterations) was conducted by propagating errors in the slope and constant of the equations of the gas concentrations and WT (refer to above equations), an uncertainty of ± 65% and ± 40% on dissolved CH4 concentration and pCO2, respectively, to account for within-lake variability (based on Table 1), and an error of ± 30% for k600 based on the reported variability across different lakes of k at low wind speeds (Cole and Caraco 2018).

Results and discussion

We sampled 15 lakes in Ecuador along an elevational gradient (2213–4361 m) (Fig. 1). The surface area of the sampled lakes ranged between 0.003 and 6.1 km2, and lakes at higher elevation tended to be smaller (Fig. 2). Lakes were generally shallow, with depths ranging between 0.9 and 72 m, with a tendency toward shallower lakes at higher elevation (Fig. 2). Lakes Cuicocha, Loreto, and Papallacta are crater lakes (hereafter referred to as “volcanic”), and the others are of glacial origin. Among the latter, San Pablo and Yaguarcocha are surrounded by the cities of Otavalo and Ibarra, respectively (hereafter referred to as “urban”).

Fig. 2
figure 2

Lake area (km2), lake depth (m), water temperature (Temp. °C, analytical precision ± 0.1 °C), stable isotope composition of O of H2O (δ18O-H2O in ‰, analytical precision ± 0.1‰), specific conductivity (cond. µS/cm, analytical precision ± 0.1 µS/cm), and total alkalinity (TA in mmol/L, analytical precision ± 0.003 mmol/L) in surface waters of 15 highland lakes in Northern Ecuador (Fig. 1) versus elevation (m). Solid grey lines show fits to the data (Table S2). Dotted blue line indicates the δ18O-H2O elevation gradient for Ecuador based on the global rain water isoscape given by Inguaggiato et al. (2010)

Full size image

Water temperature ranged between 10.2 and 15.7 °C and followed a clear elevational gradient (Fig. 2). The specific conductivity in the sampled lakes was generally low (20–183 µS/cm), except for the two “urban” lakes (San Pablo and Yaguarcocha) (294–298 µS/cm) and the three “volcanic” lakes (Cuicocha, Loreto and Papallacta) (708–722 µS/cm) (Fig. 2). The conductivity of the three “volcanic” lakes was lower than in crater lakes in active volcanic systems in Ecuador characterized by conductivity values > 10,000 µS/cm (Inguaggiato et al. 2010; Melián et al. 2021). Conductivity was negatively related to elevation for the non-“volcanic” lakes, possibly reflecting the combination of decreasing lake size and temperature, and increasing soil organic matter content and precipitation with increasing elevation. In the Andean páramo, the soil organic carbon increases with elevation because of the decrease in temperature that leads to a decrease of organic matter mineralization in soils enhancing organic matter accumulation in soils (Poulenard et al. 2003). The high organic matter content of soils and lower temperature are not favorable to rock weathering leading to soil water with low mineral content (low conductivity). Additionally, high elevation páramo areas are characterized by high precipitation leading to dilution of solutes in surface waters, which should additionally contribute to decrease conductivity. TA showed very similar patterns to those of conductivity, with the three “volcanic” lakes characterized by higher TA values (6.05–6.16 mmol/L), followed by the two “urban” lakes (3.04–3.07 mmol/L), and the remaining lakes with the lowest TA values (0.19–1.99 mmol/L).

The cation sequence was dominated by Ca2+ or Na+ followed by Mg2+ and K+ (Table S1). The lakes with a high total cation content (21–147 mg/L and 1035–7916 µeq/L) were dominated by Na+ (Na+ > Ca2+ > Mg2+ > K+), characteristic of sodium bicarbonate lakes. Lakes with a lower total cation content (~ 5 mg/L and ~ 235 µeq/L) were dominated by Ca2+ (Ca2+ > Na+ > Mg2+ > K+) (Table S1). This might indicate that the cation content of the sample lakes was related to the weathering of primarily volcanic rocks that dominate in the area (rhyolite, andesite, dacite, and diorite) (Mosquera et al. 2022). The three “volcanic” lakes (Cuicocha, Loreto, and Papallacta) were characterized by higher total cation content (144–147 mg/L 7778–7916 µeq/L) than the other 12 lakes (Table S1). The composition of the three “volcanic” lakes in divalent ions (Ca2+ and Mg2+) relative to Na+ was consistent with the other 12 lakes (Fig. S1). However, the three “volcanic” lakes had lower K+, DSi and TA relative to the other 12 lakes (Fig. S1), which might possibly indicate different weathering mechanism in the three “volcanic” lakes compared with the other 12 lakes: either a noncongruent dissolution of aluminosilicate minerals or alternatively the weathering of a higher portion of sodium rich volcanic rocks in the three “volcanic” lakes compared with the other 12 lakes (Mosquera et al. 2022). The Fe content was negatively related to Na+, and the three “volcanic” lakes had lower Fe relative to Na+ compared with the other 12 lakes. This pattern might be possibly related to the chelation of Fe to dissolved organic matter, as Fe was correlated to DOC (Fig. S2).

The δ18O-H2O values were negatively related to elevation and paralleled elevational gradient based on the global rain water isoscape (Inguaggiato et al. 2010) (Fig. 2). This might possibly indicate that the sampled lakes were mainly recharged by local precipitation (meteoric recharge). The offset between observed δ18O-H2O values in the sampled lakes and the precipitation elevational gradient most probably resulted from fractionation during evaporation. The observed δ2H and δ18O-H2O values in the sampled lakes deviated from the local meteoric water line (LMWL) (Inguaggiato et al. 2010), confirming substantial fractionation during evaporation (Fig. S3).

The DOC concentrations ranged between 1.2 and 12.8 mg/L, and DOC data points appeared to be grouped in clusters as a function of several variables such as elevation, lake area, lake depth, conductivity, and water temperature (Fig. 3). The three “volcanic” lakes (Cuicocha, Loreto, and Papallacta) were characterized by the lowest DOC concentrations (1.2–2.3 mg/L), the two “urban” lakes (San Pablo and Yaguarcocha) were characterized by intermediate DOC concentrations (4.7–5.4 mg/L), and the remaining 10 lakes were characterized by the highest DOC concentrations (8.3–12.8 mg/L). DOC was positively correlated to elevation and negatively related to temperature (excluding the three “volcanic” lakes). This might reflect the higher inputs of soil DOC in lakes more strongly influenced by inputs from the páramos. The negative relation between DOC and conductivity is consistent with the idea that soils richer in organic matter tend to lead to less mineralized water. Low inputs of soil DOC in the three “volcanic” lakes are possibly related to a low ratio of catchment area to lake surface area, reducing the inputs from soils to the lake. Additionally, the DOC content could be partly modulated by lake size, as indicated by the negative relation between DOC concentration and lake area (Fig. 3). Small lakes might have a greater connectivity and potential exchange of organic carbon with the riparian wetlands and surrounding terrestrial landscape along its periphery relative to lake surface area than larger lakes (Staehr et al. 2012). The two above explanations of DOC patterns are compatible as lake size decreases with elevation and the smaller lakes were, thus, located in the páramos with organic carbon rich soils. The POC concentrations ranged between 0.4 and 5.7 mg/L, and the POC data points were less clearly clustered than DOC data points and did not correlate with elevation, lake area, lake depth, or water temperature but did show a relationship with conductivity (Fig. 3). One of the two “urban” lakes (San Pablo) was characterized by the highest POC value (5.7 mg/L), while the three “volcanic” lakes were characterized by some the lowest POC values (0.4–0.6 mg/L). The correlation of POC with conductivity (excluding the three “volcanic” lakes) might possibly reflect a general dilution pattern by meteoric water of solutes and POC. Variations of DIN were modest (3.6–11.0 µmol/L) and DIN did not significantly correlate with elevation, lake area, lake depth, conductivity, and water temperature. DIN was dominated by NH4+ (70–98%) that ranged between 3.4 and 9.2 µmol/L. NH4+ did not significantly correlate with elevation, lake area, lake depth, conductivity, and water temperature. NO3− ranged between 0.1 and 1.9 µmol/L and did not significantly correlate with elevation and temperature but correlated positively to lake area, lake depth, and conductivity. These correlations might possibly reflect a stronger impact in smaller and shallower water bodies of soil–water inputs with a low NO3− content related to denitrification in highland soils (Pineda et al. 2021). The NH4+ and NO3− values we report in Ecuadorian lakes were consistent with those in 51 páramo lakes in the Colombian Andes investigated by Zapata et al. (2021) who reported an average NH4+ concentration of 36 µmol/L (range 20–60 µmol/L), and an average NO3− concentration of 1.1 µmol/L (range 0.1–12.8 µmol/L).

Fig. 3
figure 3

Particulate (POC in mg/L, relative analytical precision ± 5%) and dissolved (DOC in mg/L, relative analytical precision ± 5%) organic carbon, dissolved inorganic nitrogen (DIN in µmol/L), NO3− (µmol/L, analytical precision 0.1 µmol/L) and NH4+ (µmol/L, analytical precision 0.02 µmol/L) in surface waters of 15 highland lakes in Northern Ecuador (Fig. 1) versus elevation (m), lake surface area (km2), lake depth (m), specific conductivity (cond. µS/cm, analytical precision ± 0.1 µS/cm), and water temperature (temp. °C, analytical precision ± 0.1 °C). Solid grey lines show fits to the data (Table S2)

Full size image

The pCO2 values ranged between 208 and 10,069 ppm (Fig. 4). The pCO2 data points appeared to be grouped into the same three clusters observed for DOC (Fig. 4). The three “volcanic” lakes (Cuicocha, Loreto, and Papallacta) were characterized by higher pCO2 values (3269–10,069 ppm), the two “urban” lakes (San Pablo and Yaguarcocha) by the lowest pCO2 values (208–254 ppm), and the other páramo lakes by intermediary pCO2 values (644–2152 ppm). All lakes had pCO2 values above atmospheric equilibrium (400 ppm) and, thus, acted as sources of CO2 to the atmosphere, except for the two “urban” lakes that acted as sinks of atmospheric CO2. The highest pCO2 values in the three “volcanic” lakes seemed to be in part related to geogenic CO2 inputs, reflecting the influence of postvolcanic activities, such as emission of volcanic gases and input of hydrothermal water. The δ13C-DIC values in surface waters of the “volcanic” lakes were extremely high (~ 3.5‰) (Fig. S4), and magmatic inputs likely provide the dominant imprint on δ13C-DIC signatures as also reported in Lakes Kivu (~ 1.0‰) and Sonachi (~ 9.0‰) (also a crater lake) (Borges et al. 2012). The δ13C values of volcanic CO2 in gas bubbles in Ecuadorian crater lakes range between −2% and −6‰ (Inguaggiato et al. 2010; Melián et al. 2021). Consequently, the positive δ13C-DIC values observed in the three “volcanic” lakes also probably reflect the uptake of 12C by primary production and enrichment in 13C in the residual DIC. This effect might be possibly enhanced by long water residence time typical of small crater lakes devoid of outlets. The two “urban” lakes were also characterized by enriched δ13C-DIC signatures, also possibly indicative of uptake of 12C by primary production consistent with the low pCO2 values (and high POC values). A general increase of δ13C-DIC was observed from (HCO3− + CO32−):DIC = 0.5 (half of the DIC is in the form of CO2) toward (HCO3− + CO32−):DIC = 1 (all of the DIC is in the form of HCO3− + CO32−). This pattern possibly reflected a gradient of lakes from systems with low rock weathering due to dominance of deep organic soils (low HCO3−) and high CO2 from respiration leading to (HCO3− + CO32−):DIC close to 0.5 with low δ13C-DIC and lakes with higher rock weathering leading to (HCO3− + CO32−):DIC closer to 1 with high δ13C-DIC.

Fig. 4
figure 4

Partial pressure of CO2 (pCO2 in ppm, relative analytical precision ± 5.2%), dissolved CH4 concentration (nmol/L; relative analytical precision ± 10.9%), N2O saturation level (%N2O in %, relative analytical precision ± 5.8%), and O2 saturation level (%O2 in %, analytical precision ± 0.1%) in surface waters of 15 highland lakes in Northern Ecuador (Fig. 1) versus elevation (m), lake surface area (km2), lake depth (m), specific conductivity (cond. µS/cm, analytical precision ± 0.1 µS/cm), and water temperature (Temp. °C analytical precision ± 0.1 °C). Dotted horizontal lines indicate equilibrium of respective gases with the atmosphere and solid grey lines show fits to the data (Table S2)

Full size image

The pCO2 values in the páramo lakes were positively correlated to elevation and negatively correlated to lake area, lake depth, temperature, and conductivity (excluding the three “volcanic” lakes—Cuicocha, Loreto and Papallacta). This was indicative of higher lacustrine pCO2 values in páramo lakes bordered by soils with a higher organic carbon content, as shown above for DOC (Fig. 3). Indeed, pCO2 was positively correlated to DOC, negatively to %O2 and POC (Fig. 5). This possibly indicated a gradient of inputs of soil water with a low O2 and conductivity and high DOC and pCO2 among the páramo lakes that were more important at high elevations owing to soils with a higher organic matter content. Alternatively, the negative relation between pCO2 and %O2 could also reflect an internal production of CO2 within the lakes (and concurrent consumption of O2) by respiration of allochthonous organic matter, mostly in dissolved form, indirectly leading to a positive relation between pCO2 and DOC. The implications of the relation between pCO2 and DOC are discussed in more detail hereafter. Conversely, in lakes with lower inputs of soil DOC (the “urban” lakes at lower elevation), the lower DOC concentrations might have allowed the development of phytoplankton (lesser light limitation, e.g. Karlsson et al. 2009; Seekell et al. 2015) contributing to the decreasing pattern of pCO2 with decreasing DOC and increasing POC. Additionally, the lakes at higher elevation tended to be shallower and smaller (Fig. 2). As mentioned above to explain the pattern of DOC and lake size, smaller lakes should have a greater connectivity and potential exchange of CO2 from soils along their periphery relative to lake surface area (Staehr et al. 2012). Also, there is a stronger fetch limitation of k in smaller lakes (Wanninkhof 1992) that can possibly contribute to the accumulation of CO2 in lake surface water of smaller lakes, while higher k in larger lakes can promote degassing and bring CO2 levels closer to atmospheric equilibrium. These processes combined possibly contributed to the explanation of the negative relation between pCO2 with lake surface area as well as with lake depth (Fig. 4). Since the CO2 dynamics were related to organic matter degradation (in soils or in lake), %O2 correlated to pCO2 (Fig. 5) and %O2 patterns mirrored those of pCO2: %O2 was positively related to lake area, lake depth, and water temperature but was not significantly correlated to conductivity and elevation (Fig. 4).

Fig. 5
figure 5

Partial pressure of CO2 (pCO2 in ppm, relative analytical precision ± 5.2%) and dissolved CH4 concentration (nmol/L, relative analytical precision ± 10.9%) versus particulate (POC in mg/L relative analytical precision ± 5%) and dissolved (DOC in mg/L relative analytical precision ± 5%) organic carbon and O2 saturation level (%O2 in %, analytical precision ± 0.1%) in surface waters of 15 highland lakes in Northern Ecuador (Fig. 1). Dotted horizontal lines indicate equilibrium with the atmosphere and solid grey lines show fits to the data (Table S2)

Full size image

Dissolved CH4 concentrations ranged between 170 and 24,908 nmol/L in the sampled lakes (Fig. 4). Dissolved CH4 concentrations were not significantly correlated to elevation, water temperature, and POC but were negatively correlated to lake area, lake depth, conductivity, DOC and %O2 (Figs. 4, 5). The patterns of CH4 paralleled those of pCO2. The smaller lakes bordered by soils with a higher organic carbon content were likely characterized by a higher CH4 production in lake sediments or in surrounding soils. Additionally, smaller lakes were generally shallower, also potentially affecting the relative removal of CH4 by bacterial oxidation compared with the sedimentary input of CH4, probably resulting in higher CH4 dissolved concentrations in shallower lakes (Borges et al. 2011; 2022).

%N2O ranged between 63.8% and 100.8%. All the sampled lakes were sinks of atmospheric N2O, except Salve Fecha, which was close to saturation (100.8%). This possibly implies a removal of N2O by denitrification due to the complete reduction of N2O to N2. %N2O did not significantly correlate to elevation, lake depth, conductivity, and temperature but correlated to lake area (Fig. 4). NO3− also correlated with lake area (Fig. 3). Similar patterns of %N2O and NO3− as a function of lake area might possibly reflect a stronger impact in smaller water bodies of soil–water inputs with low content in N2O and NO3− related to denitrification in highland páramo soils (Pineda et al. 2021). Yet, paradoxically, %N2O did not significantly correlate with NO3− but correlated with NH4+ (and thus also with DIN) (Fig. 6), although NH4+ did not correlate with lake area (Fig. 3). %N2O did not also significantly correlate to NO2− and %O2. The general lack of correlation of %N2O with other variables might possibly reflect that there was a low contribution of internal cycling of N2O within the lakes and that the N2O levels in lakes were primarily controlled by the inputs of soil–water in a low N2O level related to soil denitrification. Additionally, N2O was driven closer to equilibrium with the atmosphere in larger lakes where k increases with fetch (Wanninkhof 1992). Combined, these two processes might possibly explain the relation between %N2O and lake surface area.

Fig. 6
figure 6

N2O saturation level (%N2O in %, relative analytical precision ± 5.8%) versus NH4+ (µmol/L, analytical precision ± 0.02 µmol/L), dissolved inorganic nitrogen (DIN in µmol/L), NO3− (µmol/L, analytical precision ± 0.1 µmol/L), NO2− (µmol/L, analytical precision ± 0.02 µmol/L) and O2 saturation level (%O2 in %, analytical precision ± 0.1%) in surface waters of 15 highland lakes in Northern Ecuador (Fig. 1). Dotted horizontal lines indicate equilibrium with the atmosphere and solid grey lines show fits to the data (Table S2)

Full size image

Comparison with floodplain lakes in central Amazon

The sampled páramo lakes were generally smaller (1.1 ± 1.9 km2, average ± standard deviation) with higher DOC concentration (9.0 ± 2.2 mg/L) than the lowland floodplain lakes compiled from literature (183 ± 311 km2; 5.3 ± 1.1 mg/L) (Fig. 7). Yet, the pCO2 values in lowland floodplain lakes were higher (3115 ± 1605 ppm) than in the páramo lakes (1037 ± 615 ppm), even for overlapping values of depth, lake surface area, and DOC concentrations (Fig. 7). This pattern probably resulted from inputs of fluvial CO2 in the floodplain lakes that led to higher pCO2 values than in the páramo lakes. Bidirectional flows between rivers and floodplains determine the ecological functioning and biogeochemical fluxes of Amazonian floodplain lakes (Junk et al. 1989), including CO2 dynamics (Abril et al. 2014; Amaral et al. 2018). The pattern of lower pCO2 values in mountainous lakes compared with lowland lakes generally agrees with other studies of CO2 dynamics in mountainous lakes at higher latitude, such as in the Alps (Pighini et al. 2018), Rocky Mountains (McCrackin and Elser 2011; Crawford et al. 2015), Sierra Nevada (Cohen and Melack 2020), and the Tibetan Plateau (Yan et al. 2018, 2023; Wang et al. 2022).

Fig. 7
figure 7

Partial pressure of CO2 (pCO2 in ppm, relative analytical precision ± 5.2%), dissolved CH4 concentration (nmol/L, relative analytical precision ± 10.9%), and CH4:CO2 ratio (µmol:µmol) versus lake surface area (km2), lake depth (m), and dissolved organic carbon (DOC in mg/L, relative analytical precision ± 5%) in the sampled highland lakes in Northern Ecuador (excluding lakes Cuicocha, Loreto, and Papallacta influenced by volcanic inputs) and in lowland floodplain lakes of the Amazon basin gathered from literature (Abril et al. 2013; Albéric et al. 2018; Amaral et al. 2020; Barbosa et al. 2016; Mitchell et al. 2018). For pCO2, only data directly measured by equilibration were compiled, data computed from pH and total alkalinity were excluded as they can be prone to large errors (Abril et al. 2015). Dotted horizontal lines indicate equilibrium with the atmosphere and solid lines show fits to the data (Table S2) (blue solid lines are highland lakes; black solid lines are lowland lakes; grey dotted lines are high- and lowland lakes)

Full size image

The pCO2 values were positively correlated to water temperature across the data set merging highland and lowland lakes (Fig. S5). This correlation could be spurious, reflecting the altitudinal gradient in pCO2 and does not necessarily reflect a temperature dependence of inputs of CO2 to the lake, such as aquatic respiration within the lake or inputs of CO2 from the soils of the watershed related to soil respiration. Regarding soil respiration, it is difficult to disentangle the effects of variations of soil properties and of temperature across altitudinal gradients, since both vary along the altitudinal gradient (Badraghi et al. 2021). Pineda et al. (2021) reported a data set of soil–air emissions of CO2 in Ecuadorian Andes along an altitudinal gradient (3000–400 m). Soil–air emissions of CO2 were higher at 3000 m due to higher organic carbon content than at 400 m despite lower temperatures (~ 12 °C versus ~ 22 °C, respectively). Since soil–air CO2 emissions are a function of soil respiration, the potential CO2 inputs from soils to lakes could in fact be stronger at higher altitudes than lower altitudes; however, lower CO2 concentrations were observed in the higher altitude lakes compared with the lakes in the Central Amazon. This would suggest an additional external CO2 input in lowland lakes compared with highland lakes.

Respiration in lake sediments scales with temperature (Cardoso et al. 2014); however, respiration in the water column of lakes seems more strongly dependent on organic carbon availability than on temperature (Pace and Prairie 2005). This is in line with a lack of relation between aquatic respiration and temperature in rivers in tropical settings (for a temperature gradient between ~ 12 °C and ~ 27 °C) (Marzolf and Ardón 2021), as well as across climate zones (Bernhardt et al. 2022). So, other factors than temperature, such as organic matter availability, seem to govern the variability of aquatic respiration in lakes and rivers.

We hypothesize that the positive relation between pCO2 and temperature did not result primarily from the temperature dependence of aquatic respiration within the lakes. If the CO2 was mainly coming from soils, higher CO2 would have been expected in higher altitude lakes due to the higher soil respiration (Pineda et al. 2021) and, generally, lower lake size of sampled lakes that should have led to a stronger influence from soil inputs (Staehr et al. 2012). Hence, the hypothesis of riverine inputs of CO2 from the Amazon River in driving higher pCO2 values in lowland lakes might possibly be more appropriate and in line with the importance of bidirectional flows between rivers and floodplains for the carbon fluxes of Amazonian floodplain lakes (Junk et al. 1989; Abril et al. 2014; Amaral et al. 2018).

The CH4 values in the sampled highland lakes were higher (3564 ± 7186 nmol/L) than in the floodplain lakes in the Central Amazon basin (170 ± 215 nmol/L) compiled from literature (Fig. 7). Yet, CH4 concentration was correlated with depth, lake surface area, and DOC across the full data set including both highland and lowland lakes. A tighter relation between CH4 and lake morphology (depth and surface area) would suggest a dominant effect of lake size on the internal production and external inputs from soils of CH4 despite different settings across the elevational gradient in the whole Amazon basin. Such an effect seems to override the difference in water temperature between highland (12.4 ± 1.9 °C) and lowland lakes (30.7 ± 1.4 °C), despite the fact that methanogenesis shows a strong temperature dependence (Yvon-Durocher et al. 2014). The CH4 values were negatively correlated to water temperature across the data set merging highland and lowland lakes (Fig. S5). Such a correlation seemed spurious since CH4 fluxes would be expected to be positively related to temperature (Yvon-Durocher et al. 2014). In a recent meta-analysis at global scale, no relation was found between temperature and CH4 emissions from lakes (Deemer and Holgerson 2021) and rivers (Rocher-Ros et al. 2023).

The consistency across the data set of the relation between CH4 and surface area differed from pCO2, for which there were distinct relationships with surface area for highlands and lowlands, possibly reflecting local drivers specific to lowlands compared with highlands, in particular the connectivity with the Amazon river channel in the floodplain lakes. Consequently, unlike CH4, a simple “unifying” correlation between pCO2 and other variables (DOC, lake size) did not emerge for both highland and lowland lakes from our data set. Since CH4 was higher and CO2 lower in highland lakes compared with lowland lakes, the CH4:CO2 ratio was markedly higher and also correlated with lake surface area, depth and DOC (Fig. 7).

The pCO2 was not correlated with DOC across the whole basin suggesting that pCO2-DOC relationship was regionally specific, as also suggested in boreal landscapes (Roehm et al. 2009; Lapierre and del Giorgio 2012). This would suggest that the relation of pCO2 and DOC results from multiple processes and do not provide consistent and reproducible patterns across large regions as noted in boreal landscapes. This is possibly related to regional patterns in DOC composition and/or in landscape morphology and lake position in the landscape that dictate the amount and relative proportions of DOC and CO2 exports from soils to lakes (Roehm et al. 2009; Lapierre and del Giorgio 2012). This is also in line with a recent analysis of pCO2 in African lakes showing diverging pCO2-DOC relations in different type of lakes, in particular in humic and nonhumic lakes (Borges et al. 2022). In our particular case, the difference between highland and lowland pCO2-DOC patterns is most likely related to the strong influence of riverine inputs in lowland floodplain lakes. It is unclear if the correlation between CH4 and DOC is spurious or reflects a causal relation. This relation could be spurious because shallower lakes tend to have higher CH4 concentrations due to the lower effect of the removal of CH4 by methane oxidation and stronger effect of CH4 inputs from sediments or soils (Borges et al. 2011; Holgerson and Raymond 2016). Shallower lakes tended to be smaller in our data set, and smaller water bodies tend to have higher DOC concentrations because small lakes have a greater connectivity and potential exchange of organic carbon with the riparian wetlands and surrounding terrestrial landscape along its periphery relative to lake surface area than larger lakes (Staehr et al. 2012).

Emissions of CO2 and CH4 from lakes and streams of the Amazon basin

We use relationships as a function of lake surface area (Fig. 7) to scale pCO2 and dissolved CH4 concentration to all the lakes in the Amazon basin (n = 12,549) given by the HydroLAKES data set (Messager et al. 2016). The relationships combined our data in Ecuadorian highland lakes and data compiled from the literature for lowland lakes in the Central Amazon, totaling 34 lakes for CH4 and 29 lakes for CO2. For CH4, there was a robust relation with lake surface area across the whole Amazon basin (Fig. 7). For consistency, we also used a relation as a function of lake surface area to scale pCO2, but two separate relations were needed for highland and lowland lakes. Such an approach is applicable because the distribution of lake numbers and surface areas given by HydroLAKES follows a bimodal distribution, with the majority located in lowlands (elevation < 500 m) followed by highlands (elevation > 2500 m) with very few lakes located at elevations between 500 and 2500 m (Fig. S6). We upscaled CH4 diffusive fluxes using surface area as a predictor (Fig. 7), because the range of depth variations was restricted in the highland (average 8 ± 9 m range 1–28 m) and in the lowland (average 10 ± 3 m range 6–15 m) lakes, in line with previous studies in small and shallow water bodies (Holgerson and Raymond 2016). The ebullitive FCH4 was not measured in the sampled Ecuadorian lakes, and we extracted for lakes of the Amazon basin the values modeled by Borges et al. (2022) in tropical lakes globally. This model relies on a compilation of ebullitive FCH4 data in tropical lakes extrapolated to the littoral zone (depth < 10 m), for which the surface area was derived from morphometric relations based on maximum depth (Håkan et al. 2007) applied to HydroLAKES (Messager et al. 2016). The CO2 and CH4 diffusive fluxes in the streams and rivers of the Amazon basin were reported by Chiriboga and Borges (2023) and relied on statistical modelling of CO2 and CH4 concentrations applied to RiverAtlas (Linke et al. 2019). Data of ebullitive FCH4 have only been reported in nine large rivers of the Amazon basin (Sawakuchi et al. 2014). We estimated ebullitive FCH4 for the rivers and streams assuming they equaled the values of diffusive FCH4, as reported by a global compilation (n = 93) by Rocher-Ros et al. (2023).

The diffusive CH4 emission (0.0007 ± 0.0003 TgC/yr) from highland lakes (elevation > 500 m) was 18 times lower than from lowland lakes (elevation < 500 m) (0.0130 ± 0.006 TgC/yr) (Table 2). This was due to lakes in the lowlands being 11 times more numerous (12,549 versus 1116) and covering a surface area 23 times larger (22,282 versus 952 km2) than in the highlands (Fig. S6, Table 2). The average areal diffusive FCH4 (per m2) in highland lakes (0.75 ± 0.36 gC/m2/yr) was close to the FCH4 in lowland lakes (0.58 ± 0.28 gC/m2/yr) owing to similar dissolved CH4 concentrations (265 nmol/L versus 179 nmol/L) but marginally higher k values (3.5 ± 0.5 cm/h versus 2.5 ± 0.3 cm/h). The average dissolved CH4 concentration was similar in highland lakes because of equivalent average surface areas (0.85 ± 5.54 km2) in highland and lowland (1.78 ± 15.54 km2) lakes. The k values were on average marginally higher in highland lakes because of higher average wind speed (3.0 ± 0.6 m /s) than lowland lakes (1.4 ± 0.6 m /s) (Table 2). The ebullitive FCH4 in lakes (0.41 ± 0.21 TgC/yr) modeled by Borges et al. (2022) was 31 times higher than the diffusive FCH4 (0.014 ± 0.006 TgC/yr), as generally reported in lakes (Deemer and Holgerson 2021; Borges et al. 2022; Johnson et al. 2022). The total FCH4 (ebullitive and diffusive, 0.42 ± 0.21 TgC/yr) from lakes in the Amazon basin we computed was within the range (0.10–0.53 TgC/yr) modeled with one-dimensional process-based lake biogeochemical model (Guo et al. 2020) calibrated with data from lakes Calado and Janauacá (respectively) (Melack et al. 2022), and similar to the estimate of ~ 0.5 TgC/yr reported by Malhi et al. (2021) based on the extrapolation of the average fluxes from a single lake (Lake Janauacá, Barbosa et al. 2020).

Table 2 Emissions to the atmosphere of CO2 (FCO2) and CH4 (FCH4) from the lakes and rivers of the Amazon basin
Full size table

The areal diffusive FCH4 was higher in rivers than lakes in both lowlands (6.8 ± 2.1 versus 0.6 ± 0.3 gC/m2/yr) and highlands (12.9 ± 2.9 versus 0.8 ± 0.4 gC/m2/yr) due to much higher k values (16–149 versus 3–4 cm/h), while average CH4 was higher in rivers (427 nmol/L) than lakes (179 nmol/L) in lowlands, but the opposite was observed in highlands (157 versus 265 nmol/L). Integrated diffusive FCH4 was higher in rivers than lakes in both lowlands (0.27 ± 0.08 versus 0.01 ± 0.01 TgC/yr) and highlands (0.03 ± 0.01 versus 0.001 ± 0.001 TgC/yr) because the surface area covered by rivers was higher than lakes in lowlands (40,375 versus 22,282 km2) and highlands (2338 versus 952 km2), in addition to higher areal diffusive FCH4 values. Ebullitive FCH4 was equivalent in rivers and lakes in lowlands (0.27 ± 0.08 versus 0.41 ± 0.21 TgC/yr) and highlands (0.03 ± 0.01 versus 0.01 ± 0.01 TgC/yr). The total (diffusive and ebullitive) FCH4 from rivers (0.61 ± 0.12 TgC/yr) was close to the total (diffusive and ebullitive) FCH4 from lakes (0.43 ± 0.21 TgC/yr), at the scale of the whole Amazon basin. The total (diffusive and ebullitive) FCH4 from lakes alone (0.43 ± 0.21 TgC/yr) and from lakes plus rivers (1.04 ± 0.24 TgC/yr) only represented ~ 1.4% and ~ 3.3%, respectively, of total CH4 emissions from all compartments of the Amazon basin ranging between 30 and 34 TgC/yr derived from top-down estimates based on aircraft or satellite data (Wilson et al. 2016, 2021; Pangala et al. 2017), which are mainly attributable to wetlands (Basso et al. 2021).

The CO2 emission (0.022 ± 0.015 TgC/yr) from highland lakes was 257 times lower than from lowland lakes (5.7 ± 1.9 TgC/yr) (Table 2). The average areal FCO2 from lakes was also 11 times lower in highlands (23 ± 16 gC/m2/yr) than in lowlands (255 ± 86 gC/m2/yr) owing to the lower pCO2 values (530 ppm versus 2909 ppm) despite higher k values (3.5 ± 0.5 cm/h versus 2.5 ± 0.3 cm/h). The CO2 emission from lakes from the Amazon basin we computed (5.7 ± 1.9 TgC/yr) was 4.1 times lower than the one reported by Raymond et al. (2013) (23.3 TgC/yr). This difference is related to a lower lake surface area (23,234 versus 51,629 km2) and a lower areal FCO2 (245 versus 451 gC/m2/yr) that is due to lower k values (2.5 versus 5.7 cm/h). Our averaged pCO2 value was in fact marginally higher than the one used by Raymond et al. (2013) (2811 versus 1906 ppm). The lake surface area used by Raymond et al. (2013) was derived from Downing et al. (2006), which includes lakes with a surface area down to 0.001 km2, while HydroLAKES only reports lakes with a surface area ≥ 0.1 km2. The number of lakes with a surface area < 0.1 km2 reported by Downing et al. (2006) was computed assuming a Pareto distribution which has been questioned on empirical (McDonald et al. 2012) and theoretical (Seekell and Pace 2011) grounds. The HydroLAKES data set has recently been recommended and applied in global scaling of lacustrine GHG emissions (Johnson et al. 2022; Lauerwald et al. 2023). Raymond et al. (2013) used a constant k value for several lake size classes calculated from the compilation of k values derived from tracer experiments in temperate or boreal systems. We computed the k values for each lake from local wind speed derived from WorldClim (Fick and Hijmans 2017) using the k-wind parameterization of Cole and Caraco (2018). Wind speed is zonally variable as a function of latitude, and tropical landmasses show distinctly lower values than higher latitude areas, which explained the difference in average k for lakes in the Amazon basin we computed and those used by Raymond et al. (2013). Raymond et al. (2013) used an average pCO2 of 1906 ppm for all tropical lakes based on data set of Sobek et al. (2005) that compiles mostly pCO2 data computed from pH and TA that are usually overestimated (Abril et al. 2015). The average pCO2 value proposed by Raymond et al. (2013) for tropical lakes (1906 ppm) was higher than the value we scaled for highland lakes (on average 530 ppm) but lower for lowland lakes (on average 2909 ppm).

The relative contribution of ebullitive and diffusive FCH4 to total carbon emissions (CO2 + CH4) was 31% in highland lakes and 7% in lowland lakes (Table 2). The relative contribution of FCH4 to total carbon emissions (CO2 + CH4) for the whole Amazon basin (both high- and lowlands) was low in lakes (7%) and rivers (1%).

Conclusions

Global estimates of CO2, CH4, and N2O emissions from lakes and rivers are uncertain because available data sets do not capture the inherent large variability and are insufficient to derive robust statistical functions to scale fluxes with confidence, despite the recent emergence of detailed spatial data sets of location and surface area for rivers (Allen and Pavelsky 2018; Linke et al. 2019) and lakes (Messager et al. 2016) that can be used to scale fluxes. Tropical lakes are undersampled with regards to GHG emissions compared with higher latitude systems, and this is particularly true for mountainous lakes. The CH4 and pCO2 values we report in mountainous lakes of the Ecuadorian Andes were very different from those reported in literature for the floodplain lakes of Central Amazon. This seemed to have resulted from smaller size of highland lakes that were surrounded by the páramo organic matter rich soils, which might have possibly led to higher CH4 levels in highland lakes. The higher pCO2 values in lowland floodplain lakes probably resulted from inputs of CO2 from the Amazon River, although we do not exclude other processes such as a higher respiration, in particular in lake sediments, related to higher temperatures.

We modeled the CH4 and CO2 emissions from lakes at the scale of the whole basin, using simple statistical relations based on lake surface area. Such a relation with surface area is not fortuitous but is based on well-established links between lake size and external subsidies of carbon from the surrounding landscape (Staehr et al. 2012). Given the restricted range of depth variations in the data set, surface area seemed to be a better predictor than depth to scale the CO2 and CH4 data. The resulting estimated CH4 and CO2 emissions from lakes had an inherent uncertainty, which was quantified, but should represent an improvement compared with the previous estimate of CH4 and CO2 emissions from lakes in the Amazon basin that relied on the extrapolation of the value from a single lake for CH4 (Malhi et al. 2021) or on the extrapolation of an average value of pCO2 from tropical lakes in general (Raymond et al. 2012). The uncertainty on the estimated CH4 and CO2 emissions from lakes resulted in part from the “one station in the center” snapshot survey approach, although frequently used to investigate broad geographical gradients as a function, for instance, of climate or catchment land cover (Juutinen et al. 2009; Finlay et al. 2010; Lapierre and del Giorgio 2012; Kortelainen et al. 2020; Casas‐Ruiz et al. 2021; Rodríguez-Cardona et al. 2023). Additionally, the ebullitive CH4 emissions were not measured in the Ecuadorian sampled lakes but derived from a statistical model for tropical lakes (Borges et al. 2022). The uncertainty on the estimation of CH4 and CO2 emissions from lakes of the Amazon basin could be improved in future by increasing the number of sampled lakes and better accounting for within lake spatial and temporal variability by increasing the number of samples horizontally and in time. The uncertainty on the estimation of CH4 emissions from lakes of the Amazon basin could be further improved by additional direct measurements of ebullitive CH4 emissions.

The integrated basin-wide diffusive CO2 emissions from lakes were much lower than those from the river network. The integrated basin-wide diffusive and ebullitive lake CH4 emissions were roughly equivalent to those from the river network. The total diffusive and ebullitive CH4 emissions from lakes and rivers corresponded to a very small fraction (< 2%) of the total CH4 emissions from all compartments of the Amazon basin, derived from top-down estimates, which are mainly attributable to wetlands.