CO2, CH4 and N2O fluxes along an altitudinal gradient in the northern Ecuadorean Andes: N2O consumption at higher altitudes

Tropical forest soils are an important contributor to the global greenhouse (GHG) budget and understanding this ecosystem function is of vital importance for future global change and climate research. In this study, we quantified soil fluxes of carbon dioxide (CO2), methane (CH4) and nitrous oxide (N2O) of four tropical forest sites located along an altitudinal gradient from 400 to 3010 m a.s.l. on the western flanks of the Andes in northern Ecuador. We assessed the physicochemical soil properties 20 influencing these fluxes during the dry season, as well as the bulk isotopic signature of N2O. The CO2 fluxes ranged between 55.3±12.1 and 137.6±32.8 mg C m h, with the highest and lowest emissions at the highest strata, at 3010 and 2200 m a.s.l., respectively. CH4 fluxes at all sites exhibited a net consumption of atmospheric CH4 and ranged between -74.4±25.0 μg C m 2 h at 2200 m a.s.l. to -46.7±14.7 μg C m h at 3010 m a.s.l. Net fluxes of N2O ranged between -5.1±1.9 and 13.2±31.3 μg N m h, with a marked net sink at 2200 and 3010 m a.s.l., whereas a net source at 400 m. pHwater and nitrate (NO3) content 25 at 5 cm depth were able to explain 83% of the observed temporal (daily measurements) and spatial (four forest sites) variability of the CO2 fluxes; indicating that an increase in pHwater and NO3 contents lead to an increase in CO2 emissions. For CH4 fluxes, it was not possible to obtain a statistically significant model to identify the physicochemical soil drivers responsible for the CH4 consumption. For N2O, bulk density and pHwater at 5 cm depth were negatively correlated to the N2O fluxes, but able to explain only 36% of the temporal and spatial variability. In addition, the bulk isotope N2O data confirmed that N2O reduction 30 was at the basis of the observed net soil sink at higher altitudes. Finally, the soil GHG budget showed that all studied soils were net sources of GHG’s. CO2 emissions represented the largest component of the total soil GHG budget, CH4 consumption was quite consistent along the elevation gradient, whereas N2O was highly variable, and the transition from sources to net https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c © Author(s) 2020. CC BY 4.0 License.

sinks at higher altitudes represented the biggest change in the net GHG balance. Overall, for non-CO2 GHGs, we noticed a transition from a net source to a net GHG sink along altitude. 35

Introduction
In 2014, the Intergovernmental Panel on Climate Change (IPCC) released its Fifth Assessment Report (AR5) indicating that the atmospheric concentrations of the three major biogeochemical greenhouse gases (GHGs) (carbon dioxide -CO2, methane -CH4 and nitrous oxide -N2O) have reached unprecedented levels. As such, current atmospheric concentrations indicate that in 2018, CO2 (407.8±0.1 ppm), CH4 (1869±2 ppb) and N2O (331.1±0.1 ppb) concentrations were 147%, 256% and 123%, 40 higher compared to pre-industrial levels (before 1750) (WMO, 2019). A large proportion of CO2 emissions are sourced by human activities, particularly fossil fuel burning and land use change. Similarly, the increase in CH4 emissions is mainly due to fossil fuel and agriculture (60%), as well as for N2O, of which food production is the largest contributor to the N2O increase (Syakila and Kroeze, 2011).

45
Soils in terrestrial ecosystems play a vital role in the global GHG budget. Tropical forest soils, in particular, represent a net sink of carbon (C) (Pan et al. 2011), but they coincidently are the largest natural source of N2O, with an estimated contribution of 14-23% to the annual, global N2O budget (Werner et al., 2007). In general, soil CO2, CH4 and N2O production or consumption depends on microbiological processes driven by a wide range of abiotic and biotic characteristics. The combination of these processes ultimately determines if a soil is a net source or sink of GHGs. Under aerobic conditions, CO2 50 is emitted to the atmosphere by autotrophic and heterotrophic respiration of vegetation and fauna ), while CH4 is consumed by methanotrophic bacteria (Jang et al., 2006); although forest soils prone to inundation (anaerobic) emit CH4 by methanogenic microorganisms (Archaea domain). On the other hand, N2O is emitted through denitrification or a number of alternative pathways (e.g. nitrification, nitrifier-denitrification, chemodenitrification, etc.; (Butterbach-Bahl et al., 2013;van Cleemput, 1998;Clough et al., 2017)). Overall, tropical forests emit on average 12.1 t CO2-C ha -1 y -1 (heterotrophic 55 and autotrophic respiration), slightly smaller than the Net Primary Productivity (NPP) (12.5 t CO2-C ha -1 y -1 ) i.e. the net C sink of tropical forests is ~ 0.4 t CO2-C ha -1 y -1 . In aerobic conditions, CH4 fluxes vary from -1 to -40 kg CH4 ha -1 y -1 , with an average consumption of -4 kg CH4 ha -1 y -1 , while the mean rate of N2O emissions from tropical forest soils is 4.76±0.81 kg N2O ha -1 y -The understanding of the mechanisms and processes underlying GHG flux variability has greatly improved during the last decades (Butterbach-Bahl et al., 2013;Heil et al., 2016;Müller et al., 2015;Sousa Neto et al., 2011;Su et al., 2019;Teh et al., 2014). However, there is still (1) considerable uncertainty about the overall balances of many ecosystems (Castaldi et al., 2013;Kim et al., 2016;Pan et al., 2011;Purbopuspito et al., 2006), (2) a strong imbalance in field observations, skewed to the Northern Hemisphere (Jones et al., 2016;Montzka et al., 2011), and (3) a bias towards quantification of 70 emissions in lowland forests within the tropics (Müller et al., 2015;Purbopuspito et al., 2006). For instance, based on a compilation made of CO2, CH4 and N2O fluxes in South America (Table S.3), there are only two studies on emissions in upper montane forests, while they represent 11% of the world's tropical forests (Müller et al., 2015;Teh et al., 2014). To further improve our understanding of tropical forest ecosystems on the global GHG balance, environmental gradients (elevational, latitudinal, etc.) can offer great opportunities to study the influence of abiotic factors on biogeochemical processes under field 75 conditions (Bauters et al., 2017a;Jobbágy and Jackson, 2000;Kahmen et al., 2011;Laughlin and Abella, 2007), which complements the knowledge on short term responses from experimental approaches. In the case of elevational gradients, these responses are driven by abiotic variables that co-vary with elevation, which, amongst others, creates a distinctly strong climate gradient over a short spatial distance (Bubb et al., 2004;Killeen et al., 2007;Körner, 2007;Myers et al., 2000).

80
Here, we present a study of the soil-atmosphere exchange of CO2, CH4 and N2O along an altitudinal gradient in a Neotropical montane forest located on the western flanks of the Andes in northern Ecuador. We aimed (1) to determine the magnitude of the soil-atmosphere exchange of CO2, CH4 and N2O during the dry season, and (2) to assess the main climatic and soil parameters that control these fluxes. By working along this altitudinal gradient, we wanted to explore the potential effect of temperature -and other factors that co-vary with altitude -on the GHG budget of the forest soils. We expected the CO2 fluxes 85 to decrease with altitude, with higher emissions at lower altitudes in view of more ideal conditions (i.e. temperature and soil moisture); CH4 fluxes to represent net sinks that increase (i.e. negative fluxes) with altitude, and mainly explained by soil moisture; and a decrease in N2O fluxes along the altitudinal gradient, with soil moisture and nitrogen (N) content as the main explanatory variables.

Study areas
The field work was carried out along an altitudinal gradient from lowland (400 m a.s.l.) to upper montane evergreen forests (3010 m a.s.l.; Table 1) (FAO, 2017;Ministerio del Ambiente, 2015). We selected areas of well-preserved natural forests located on the western flanks of the Andes in northern Ecuador; specifically, in the Sierra region of the provinces of Imbabura and Pichincha. Four study sites ( Observations were made within one plot of about 20x20 m, established at each study site. https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c Author(s) 2020. CC BY 4.0 License.
All sites experience two rainy seasons (March -April and October -November), with a mean annual precipitation (MAP) that varies on average between 900 and 3600 mm, and an adiabatic lapse rate of approx. 5 °C per 1000 m of altitude (Table 1) 100 (Varela and Ron, 2018).

Sampling strategy
At all study areas, the sampling campaign took place from August 6 th to September 28 th , 2018, corresponding to the end of the dry season. One plot was selected for each site, and within each plot, five polyvinyl chloride (PVC) collars were installed to allow in-situ measurements using a static flux chamber method. The collars were inserted at random locations within the plots 105 but guaranteeing at least 7 m distance between each one. The insertion of the collars was performed at least 12 h before the first measurements by applying even pressure across all points to minimize effects caused by soil disturbance. The chambers consisted of a PVC pipe hermetically sealed on top with a rubber-sealed lid. The chamber area was 0.0191 m 2 and the internal volume ranged between 3.63 and 3.98 L. Each chamber was equipped with sampling ports mounted with three-way valves, and a vent tube was installed to reduce pressure interferences. 110 Gas samples were collected mid-morning and measurement cycles on each site consisted of four consecutive gas measurements once per day for one hour and during five contiguous days. The samples were taken mid-morning to avoid extreme temperatures and we consider them as representative of a whole day (Collier et al., 2014;Luo and Zhou, 2006d). For these measurements, the collars were left in place for the duration of each measurement cycle; thus, the analysis per stratum lasted 115 1 week (i.e. 1 month for all measurements in the 4 strata). However, in order to assess a both short-term and long-term variation mainly related to weather conditions, the gas measurements were done first in August and consequently repeated in the next month (September).
Adjacent to each chamber (~ 1 m), one pit was dug for soil sampling, and intact soil cores were collected using stainless steel 120 cylinders (diameter: 5.08 cm, height: 5.11 cm). The samples were taken at 5 and 20 cm depth once during the first month (August) of measurements. Each soil core was immediately packed into airtight zip-lock bags and once the sampling campaign was over, they were sent to Belgium for physicochemical soil analysis. Bulk density (ρb) was measured by oven drying (75°C for 48 h) and weighing the soil samples. Soil porosity was derived from Eq. (1), assuming a particle density of 2.65 g cm -3 . pH was measured by a potentiometric method using a pH-sensitive glass electrode, a standard reference electrode (HI 4222;Hanna 125 Instrument, Bedfordshire, UK), and a volumetric ratio soil:liquid of 1:5 for pHwater (distilled water) and pHKCl (1M KCl). NO3and NH4 + content was determined colorimetrically (Auto Analyzer 3; Bran and Luebbe, Norderstedt, Germany) after extractions performed with 1M KCl at room temperature and neutral pH. C and N concentrations (%C, %N), along with the stable N isotope signatures (δ 15 N) of the soil samples, were determined at natural abundance by a Continuous Flow Element Analyzer (Automated Nitrogen Carbon Analyzer), interfaced with an Isotope-Ratio Mass Spectrometer (Sercon 20-20; Sercon, Cheshire, UK). Moreover, the soil samples taken at 5 and 20 cm depth were combined to produce one composite sample associated to each site, and by means of the method described by the International Organization for Standardization (ISO 11277:2009), soil texture was determined. The classification was made according to the classification system of the United States Department of Agriculture (USDA, 2017); and the soil class was determined based on the classification of FAO and UNESCO: World Reference Base for Soil Resources (WRB) (FAO, 2007). 135 Daily measurements of soil moisture, expressed as water-filled pore space (WFPS), were taken per site at 5 and 20 cm depth using soil moisture sensors (EC-5, Meter Environment, Pullman, Washington, USA) and data loggers (ProCheck, Meter 140 Environment, Pullman, Washington, USA). Finally, soil temperature was determined daily for each measurement cycle and per chamber, by means of a thermometer inserted at 5 cm depth and approximately 10 cm from each chamber.

Soil-atmosphere exchange
12 h after the installation of the collars, the chambers were closed for a period of 1 h, and samples of 20 mL were taken with disposable syringes from the headspace air of the chambers every 20 minutes: T1 = 0, T2 = 20, T3 = 40 and T4 = 60 min; T1 or 145 time-zero indicates the sample taken immediately after the chamber was closed. Moreover, prior to each sample collection, the syringe was flushed twice with air of the chamber to mix the chamber headspace and to avoid any possible stratification of them.
The 20 mL samples were injected in pre-evacuated 12 mL exetainer vials (over-pressurized), and once the sampling campaign 150 was over, the samples were sent to Belgium for analysis by gas chromatography at Ghent University. For CH4 and CO2 analysis a gas chromatograph (Finnigan Trace GC Ultra; Thermo Electron Corporation, Milan, Italy) equipped with a flame ionization detector (FID) and a thermal conductivity detector (TCD) was used, respectively. For N2O, another gas chromatograph equipped with an electron capture detector (ECD) (Shimadzu GC-14B; Shimadzu Corporation, Tokyo, Japan) was used.

N2O bulk isotopic composition 155
Two extra gas samples were taken for stable isotope analysis at the start (T1) and at the end (T4) of a chamber closure. These samples were taken only once per site and in only one of the chambers. For this, in addition to the small exetainer vials, preevacuated big serum vials (110 mL) were used to inject gas samples of 180 mL (over-pressured). At the end of the field campaign, the samples were transported to Switzerland (ETH Zurich) and analyzed for bulk 15 N measurement of N2O (δ

Data analysis
All statistical analyses were conducted in R Studio, version 3.5.2 (The R Core Team, 2019), and the statistical significance was reported at 95% confidence level (P ≤ 0.05), unless otherwise stated.

165
Mean values with standard deviations (SD) per site and depth were calculated for the physicochemical soil properties. The fluxes for each gas (CO2, CH4 and N2O) were calculated by means of linear regressions using the four consecutive measurements of each measurement cycle. The slope of the regressions represented the flux. Thus, following the ideal gas law, and considering the head space volume of the chamber and the chamber area, the net gas flux was calculated by Eq. (2) (Collier et al., 2014;Kutzbach et al., 2007): 170 where Fc corresponds to the net gas flux (CO2, CH4 or N2O), ∆c/t is the rate of change of the gas concentration within the chamber or the slope of the regression line [ppm min -1 or µL L -1 min -1 ], P/RT corresponds to the ideal gas law used to convert 175 concentration from volumetric to mass at normal temperature and pressure: P = absolute pressure (1 atm), R = gas law constant (0.08206 L atm mol-1 k-1), and T = temperature (293 K); MW is the molecular weight of the gas (CO2-C and CH4-C: 12.01 g mol-1, N2O-N: 14.01 g mol-1), Vch is the headspace volume of the chamber, and Ach the area of the chamber. The goodnessof-fit was evaluated for every linear regression using the adjusted coefficient of determination (R 2 ), and time series (concentration vs time) with R 2 < 0.60 were excluded from further analysis. 180 For the purpose of evaluating if there were any statistical differences at each site between the fluxes obtained in August and September, a one-way ANOVA was performed per site; verifying for each case the respective assumptions (i.e. equality of variances and normality). Moreover, a linear model was fitted using the sites as a factorial explanatory variable per gas flux measurement to assess differences across sites, to estimate the effect sizes of the net fluxes, and to determine to which extent 185 the variability of the net fluxes could be explained by these explanatories. For this, the validity of the model was evaluated through verification of assumptions of linearity, homoscedasticity (or equality of variances), and normality of the error terms; and due to the nature of the data in the measurements of N2O, the N2O fluxes were log-transformed to homogenize variances.
In order to determine the physicochemical soil characteristics able to explain to the greatest extent the net fluxes of CO2, CH4 190 and N2O (response variable), a preliminary stepwise multiple linear regression was done for each soil gas. However, due to the low amount of data points (number of plots per altitude) and the small variability within plots, we averaged the measurements out, rather than to explicitly use separate measurements in a more complex linear mixed effect model. Therefore, we started with a full model that included all variables measured at 5 cm depth (predictors), and the average per plot and per https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c Author(s) 2020. CC BY 4.0 License. day of the fluxes measured only during the first month (August) (the soil samples for physicochemical soil properties were 195 only taken in August). Then, using the "step" function of the "stats" package in R (The R Core Team, 2019) and the Akaike Information Criterion (AIC), a model was selected for each gas. Subsequently, the variance inflation factor (VIF) was calculated to avoid multicollinearity problems between predictors, and by means of the "vif" function from the "car" package in R (The R Core Team, 2019), along with a VIF threshold of < 3, the predictors with a higher VIF were excluded one by one.
Finally, a simple linear regression was carried out for each flux with the retained predictors, verifying in each case the validity 200 of the model and the respective assumptions (i.e. linearity, homoscedasticity and normality of the error terms).
The soil isotopic signature of N2O was calculated using a two-source mixing model. Based on the conservation of mass Finally, in order to compare the fluxes of the three GHGs measured (i.e. CO2, CH4 and N2O) and to determine the overall budget of the soils, the CO2-eq emissions for each gas were calculated by means of Eq. (6), and by using a global warming potential (GWP) of 1, 28 and 265 for CO2, CH4 and N2O, respectively (IPCC, 2014a). Moreover, the total GHG budget at each 215 site was obtained by summing the CO2-eq emissions of each gas (Myhre et al., 2013).
where i refers to CO2, CH4 or N2O.

Physicochemical soil properties
Soils are Andosols and the soil texture was classified (USDA) between loam and sandy loam at all sites (WRB; Table 2). All sites had a relatively acidic soil; pHwater ranged from strong to medium acidic (4.61 -5.69), with an increase in acidity with depth, except at P_3010 ( Table 2). The most acidic soil was found at S_400 at 5 cm, although not significantly different from Except for P_3010, NO3-N concentrations were 2-4 times higher at 5 cm compared to 20 cm depth; the highest variability was observed at S_400, and in comparison to the other sites, P_3010 seems to be depleted in NO3-N at both depths (0.8 -3.6 µg g-1). In contrast, the highest concentration of NH4-N was obtained at P_3010 at 20 cm, followed by S_400 at 5 cm. However, at all sites, NH4-N concentrations at 5 cm were not significantly different from each other. Such as NO3-N, NH4-N also decreased with depth, except at P_3010 where the increase at 20 cm with respect to 5 cm was almost doubling. Higher N 230 contents were measured at 5 cm compared to 20 cm depth at all sites; and S_400 exhibited the highest content at both depths, 1.3-1.4 times higher than any other N percentage at the same depth, and even 4 times higher than any other N percentage at 20 cm depth. The C content showed a general decrease with depth at all sites, with the highest percentage at S_400 at 5 cm, and the lowest one at M_1100 at 20 cm. Higher δ 15 N signatures were obtained at 20 cm compared to 5 cm depth; at S_400 the soil was most enriched in 15 N, and P_3010 showed the most depleted one. 235 Soil temperature decreased with altitude with a gradient of -4.2 °C per 1000 m, with no statistical difference between months.

Greenhouse gas fluxes
In general, all sites were sources of CO2 (Fig. 1a, Table 3). Except for P_3010, mean CO2 emissions were higher in September compared to August, but due to the high variability in the measurements, there was no significant difference between months at M_1100 and P_3010 (P > 0.05). The lowest and highest emissions were observed at C_2200 and P_3010, respectively, in both months, and all sites were significant predictors and able to explain 56% of the variability of CO2 emissions during the 245 field campaign (Fig. 1a).
All mean CH4 fluxes were negative, indicating a net flux from the atmosphere to the soil (Fig. 1b, Table 3). Although the mean CH4 fluxes (except for P_3010) were higher in September compared to August, there was no significant difference (P > 0.05) between months at any site. Moreover, the linear model performed for CH4 only explained 3% of the variability. 250 Finally, the mean N2O fluxes showed a general negative trend with increasing altitude (Fig. 1c). A marked net N2O consumption was observed at the sites located at 2200 and 3010 m a.s.l., and besides these sites, M_1100 also acted as a net sink in September; however, there was no significant difference (P > 0.05) in any plot between months. The highest consumption was observed in August at P_3010, while the highest emission was in September at M_1100 (Table 3). On the 255 other hand, the fitted linear model explained 65% of the variability of the N2O fluxes during the field campaign.
Although only monthly average fluxes will be discussed, the large variability observed with most of the gas fluxes (Table 3 and Fig. 1) are the result of the spatial (i.e. differences in GHG fluxes between chambers) and temporal (i.e. differences in GHG fluxes per day) variability within each site. 260

Linear regressions with physicochemical soil characteristics
Changes in pHwater and NO3-N at 5 cm depth explained 83% of the temporal (daily measurements) and spatial (four sites) variability of CO2 fluxes in August (Table 4). Both predictors were positively correlated with CO2 emissions, thus, pHwater (P < 0.001) and NO3-N content (P = 0.51) were positively related to CO2 fluxes. For CH4 consumption, it was not possible to obtain a model since none of the predictors were retained during the stepwise selection. In case of N2O fluxes, bulk density (P 265 = 0.07) and pHwater (P = 0.06) at 5 cm depth explained 36% of the temporal and spatial variability in August. Although both predictors were close to the threshold (P = 0.05) and considered as non-significant, they indicate a negative correlation where every increase in bulk density and pHwater lead to a decrease in N2O fluxes.

15
ranged from -13.08 to 11.54‰, with the lowest and highest isotopic signature observed at S_400 (September) (Fig.  270 2, Table S1). During both months, 15 values of M_1100, C_2200 and P_3010 exhibited 15 N enrichments, and all of them reflected chambers where negative fluxes were obtained i.e. consumption of N2O from the atmosphere to the soil (Table   S1).

Soil GHG Budget
The average soil GHG balance indicates that all plots during August and September are considered as sources of GHGs, largely 275 driven by CO2 emissions (Fig. 3a), and with the highest compensation from CH4 consumption (Fig. 3b). During both months, the highest and lowest CO2-eq emissions were obtained at P_3010 (August: 499.6, September: 450.6 mg CO2-eq m -2 h -1 ) and at C_2200 (August: 200.3, September: 248.3 mg CO2-eq m -2 h -1 ), respectively; in both cases, N2O and CH4 consumption resulted in a ~1% offset of the total CO2-eq emissions. Across our study sites, P_3010 exhibited the highest soil CO2 emissions ( Fig. 1a and Table 3). Even though an increase in temperature (up to an optimum of ca. 50°C; Luo and Zhou, 2006a;Oertel et al., 2016), moisture-WFPS (up to an optimum 60%; Luo and Zhou, 2006b)  pHwater in the overall CO2 budget is also apparent across our study range. Nonetheless, shifts in C allocation could also give rise to shifts in CO2 emissions, and thus support the increase observed in the site located at the highest altitude (P_3010). As such, an increase in fine root biomass is expected in tropical mountain forests compared to lowland forests. In fact, a study carried out in the South Ecuadorian Andes from 1050 to 3060 m a.s.l., indicates a positive linear correlation (R 2 = 0.87, P = 0.01) between fine root biomass and altitude, arguing that imbalances or limitations in resource (water and/or nutrients) 295 availability at higher altitudes may be the cause (Leuschner et al., 2007). Consequently, the observed increase in CO2 emissions at high altitude might be further driven by an increase in root biomass as the latter has been shown to be positively correlated with soil respiration (Han et al., 2007;Luo and Zhou, 2006a;Reth et al., 2005;Silver et al., 2005).
In contrast to P_3010, the low emissions observed at C_2200 could be attributed to (1) the lower WFPS, (2) the lower contents 300 of C and N, and (3) the higher bulk density. Indeed, the lowest content of water was observed at this site in August at 5 cm depth, and exactly in this month, the lowest emissions of CO2 were obtained. The low contents of C and N exhibited in C_2200 (indeed, the lowest from all the sites), could have hampered the CO2 emissions, since an increase in C content normally leads to higher levels of respiration, and N itself is required for plants and soil microorganisms to grow Luo and Zhou, 2006a;Oertel et al., 2016). In additions, this site also had the highest soil bulk density (i.e. lowest porosity), which 305 could have led to a decrease in soil respiration either by a physical impediment for root growth or by a decrease in soil aeration for microbial activities (Dilustro et al., 2005;Zhou, 2006c, 2006a). For instance, Kowalenko et al. (1978) reported an almost double CO2 emission from a clayey loamy soil (7.04 mg CO2-C m -2 h -1 ) than from a sandy soil (3.75 mg CO2-C m -2 h -1 ), and Dilustro et al. (Dilustro et al., 2005) reported an overall 1.5 increase in CO2 emissions from a clayey soil (171.07 mg CO2-C m -2 h -1 ) with respect to a sandy one (117.07 mg CO2-C m -2 h -1 ). 310 Finally, our measurements are enveloped by earlier work when framed in a broader and pantropical context (Table S3, Fig. 4).
However, although a quantitative comparison is difficult to make due to differences in e.g. sampling durations (single seasons, whole year cycles or only specific dates), sampling frequencies (sub-daily, daily or monthly), measuring methods, intrinsic site properties, etc., our fluxes are well below most of the CO2 fluxes reported for South America, except for a study performed 315 in Brazil at 130 m a.s.l. (17.4 mg CO2-C m -2 h -1 ) (Verchot et al., 2000). Moreover, although it is not common to obtain higher fluxes at higher altitudes, only one study performed in Peru, between 2811 -2962 m a.s.l., showed a flux similar in magnitude (dry season: 120 mg CO2-C m -2 h -1 , wet season: 170 mg CO2-C m -2 h -1 ) (Jones et al., 2016) to our flux obtained at P_3010. https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c Author(s) 2020. CC BY 4.0 License.

CH4 fluxes
In general, all sites acted as net sinks for CH4 (i.e. uptake of atmospheric CH4 by soils). During the whole field campaign only 320 one chamber at one site (S_400) and a specific date (08/09/2018) showed a net source of CH4 (45.82 µg CH4-C m-2 h -1 ).
However, there were no statistical differences between months. The mean CH4 fluxes were quite similar even between sites, and there was no significant linear regression with the physicochemical soil characteristics able to explain the CH4 fluxes. All sites exhibited indeed a high temporal and spatial variability ( Fig. 1b and Table 3 Even so, except for one study carried out in Ecuador at 400 m a.s.l. (19.37 µg CH4-C m -2 h -1 ), all fluxes depicted in Fig. 5 and measured during a dry season (DS), represent a sink of CH4, which is supported by the fact that humid tropical forests are 335 responsible of 10% to 20% of the global soil sink for atmospheric CH4 (Verchot et al., 2000).

N2O fluxes
Only S_400 (both months) and M_1100 (September) (i.e. plots located at the lower locations) acted as net sources of N2O ( Fig. 1c, Table 3). In contrast, the lowest flux was observed in the plot located at the highest stratum (P_3010) during August and September, which showed a general net consumption. The high emissions obtained at the lowest strata corroborate with 340 literature data on lowland tropical forests (Butterbach-Bahl et al., 2004, 2013Koehler et al., 2009) and are mainly attributed to (1) soil water content, (2) temperature, and (3) N availability.
Firstly, N2O emissions in tropical forest soils are predominantly governed by WFPS, which influences microbial activity, soil aeration and thus diffusion of N2O out of the soil (Davidson et al., 2006;Werner et al., 2007). Ideally, the highest emissions 345 via denitrification are observed between 60-80% WFPS, but they can vary between 50-80% or 60-90% depending on the soil physical properties (Butterbach-Bahl et al., 2013;Davidson et al., 2006;Oertel et al., 2016). At lower percentages (optimum: 20%), nitrification takes place and although N2O can be produced as well, it yields a higher potential for NO production (Davidson et al., 2006;Oertel et al., 2016). As a second main driver for N2O emissions, and in comparison to CO2 emissions, denitrification is very sensitive to rising temperatures (Oertel et al., 2016). An increase in temperature leads to an increase in soil respiration and thus to a depletion of O2 concentrations, which is indeed a major driver in N2O emissions.
Moreover, rising temperatures lead to a positive feedback in microbial metabolism; the stimulation of mineralization and nitrification processes induces an increase in the availability of substrates for denitrification, and thus to an increase in N2O emissions (Butterbach-Bahl et al., 2013;Sousa Neto et al., 2011). Hence, the high emissions observed at S_400 can also be supported by the high temperatures observed at this site; which are indeed the highest from all sites (Fig. S1). Finally, the 355 dependency of N2O emissions on WFPS and temperature is affected by substrate availability (NO3) which was also the highest at S_400 (almost 2.5 times higher than the second highest content observed). High contents of NO3give an indication of an open or "leaky" N cycle with higher rates of mineralization, nitrification and thus N2O emissions (Davidson et al., 2006).
Moreover, NO3is normally preferred as an electron acceptor over N2O and it can also inhibit the rate of N2O consumption to N2 . 360 In contrast to the low elevation sites where net N2O emissions were observed, P_3010 presented the highest consumption (negative values, i.e. fluxes from the atmosphere to the soil), followed by C_2200. From 37 valid measurements only 1 resulted in net emission at P_3010 (range: -12.91 to 1.33 µg N2O-N m -2 h -1 ), whereas from 36 measurements, 20 resulted in net emissions at C_2200 (range: -11.10 to -0.31 µg N2O-N m -2 h -1 ). Net N2O consumption is often related to N-limited ecosystems. 365 Indeed, at low NO3concentrations, atmospheric or gaseous N2O may be the only electron acceptor left for denitrification (Chapui-Lardy et al., 2007;Goossens et al., 2001). Studies performed by Teh et al. (Teh et al., 2014) and Müller et al. (Müller et al., 2015) in the Southern Peruvian and Ecuadorian Andes, respectively, related the decrease in N2O emissions -and thus the potential for N2 production in soils -at high elevations to differences in NO3availability.

370
In this case, P_3010 had the lowest content of NO3 -, along with the lowest soil δ 15 N, which clearly reflects the shift towards a more closed N cycle at higher elevations (Bauters et al., 2017b). Moreover, this was the site with 1) the highest content of clay and hence more microsites for N2O reduction, 2) the lowest soil water content (% of WFPS) and hence more diffusion of atmospheric N2O into the soil, 3) the highest pH-value, which could have alleviated inhibition of the nitrous oxide reductase at low pH, and 4) the highest CO2 emissions observed, which could have indeed led to the development of anaerobic microsites 375 where denitrification could have occurred (Chapui-Lardy et al., 2007). This is also valid for M_1100, where N2O consumption was observed in August (3 out of 7 valid measurements; range: -10.48 to 9.18 µg N2O-N m -2 h -1 ), whereas N2O emission in September (5 out of 16 valid measurements; range: -9.67 to 94.42 µg N2O-N m -2 h -1 ). Although samples of soils were not taken in September, this month represents the transition or the beginning of the wet season in the region. Thus, it is expected to have an increase in available soil N due to the accumulation of litter during the dry season and the following but rapid mineralization 380 after soil rewetting at the beginning of the rainy season, which could have indeed led to pulses of higher N2O emissions (Werner et al., 2007). https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c Author(s) 2020. CC BY 4.0 License.
Contrary to CO2 emissions, pHwater negatively affected N2O emissions (R 2 = 0.36). This relation has been already reported previously (Baggs et al., 2010;Chapui-Lardy et al., 2007;Wrage et al., 2001), and pH has been considered as a "master 385 variable" to predict N transformations (Baggs et al., 2010). As a second predictor, bulk density also exhibited a negative correlation with N2O emissions. Reflecting the fact that higher values of bulk density are translated into less oxygen diffusion and thus, more anaerobic sites ideal for the reduction of N2O to N2. Likewise, Klefoth et al.  (1) N2O studies in the tropics are biased toward lowland forests (Müller et al., 2015;Purbopuspito et al., 2006), and (2) our results 400 along with the low fluxes (and even negative data points) reported especially in the Andes, highlight the importance of a probably unaccounted sink of N2O at high altitudes; keeping in mind that tropical montane forests represent 11% of the world's tropical forests (Müller et al., 2015).

Isotopic signature of N2O ( ) 405
Previous studies have indicated that during the reduction of N2O to N2, N2O-reductase fractionates against 15 N (Barford et al., 1999;Butterbach-Bahl et al., 2013;Menyailo and Hungate, 2006;Pérez et al., 2000). Consequently, complete denitrification i.e. consumption of N2O, leads to a 15 N enrichment of the residual N2O, and thus to higher δ 15 Ns Bulk values (Park et al., 2011) relative to the atmospheric bulk N2O composition (6.3‰; (Harris et al., 2017)).This indeed is reflected in the enriched δ 15 Ns Bulk values measured during N2O consumption while the N2O bulk signature during N2O production was highly depleted compared 410 to that of atmospheric N2O (two samples taken in September at S_400) ( Fig. 2; Table S1). This is also in line with Park et al. (2011) andPérez et al. (2000) who have attributed δ 15 Ns Bulk values between -22 and 2‰ in natural tropical forest soils to denitrification. Therefore, in addition to the soil isotope signatures, the bulk N2O isotope signatures confirm the net N2O consumption at higher altitudes, and net N2O emission at lower altitudes, and rule out that our net consumption rates are due to sampling artefacts.

Soil GHG Budget
The differences in fluxes for each GHG are clearly visualized in Fig. 3. The high CO2 emissions observed at P_3010 give rise to the highest CO2-eq emissions, and in terms of non-CO2 GHG, this plot also exhibited the highest sink due to CH4 and N2O consumption. However, it is important to mention that the calculated CO2-eq emissions for CO2 reflect only the impact of soil emissions (heterotrophic and autotrophic respiration) on the soil GHG budget, excluding photosynthesis and aboveground 420 respiration. Therefore, based on the fluxes here obtained, the upland soils from our study clearly show a marked sink of non-CO2 GHG. Even so, besides this and the known potential for C sequestration in tropical forests, the marked CH4 and N2O sinks observed in this case, enhances the importance and protection of tropical forests, not only in terms of biodiversity and ecosystem services, but also as means of mitigation options to curb global warming.

Conclusions 425
Overall, we found that our CO2 fluxes are well below most of the CO2 fluxes reported in literature for tropical forest soils, with an unusual but marked increase at the highest altitude, mainly explained by soil pH and root biomass. Moreover, our CH4 uptake fluxes are among the highest in the tropics (i.e. highest consumption of atmospheric CH4) and reiterate the role of humid tropical forest soils as a known CH4 sink. Contrary to the net N2O emissions observed in the lowest strata, the net consumption at higher elevation seems to be quite unique and reflects (1) the worldwide bias of N2O studies toward lowland 430 forests, (2) the need of coupling environmental and physicochemical soil variables with microbial analyses, and (3) the importance of conserving upland forest for N2O consumption. This net N2O uptake was confirmed independently by soil and N2O 15 N isotope signatures. Finally, although an altitudinal gradient was selected to evaluate the potential effect of temperature -and other factors that co-vary with altitude -on the GHG budget of the forest soils, our results for CO2, CH4 and N2O fluxes clearly reflect the "complex" interplay of different environmental controls and physicochemical soil characteristics rather than 435 a generalized trend related to altitude.   Table S1. 15 values and N2O fluxes. Van Groenigen, J. W., Huygens, D., Boeckx, P., Kuyper, T. W., Lubbers, I. M., Rütting, T. and Groffman, P. M.: The soil n cycle: New insights and key challenges, Soil, 1(1), 235-256, doi:10.5194/soil-1-235-2015, 2015. 510 Han, G., Zhou, G., Xu, Z., Yang, Y., Liu, J. and Shi, K.: Biotic and abiotic factors controlling the spatial and temporal variation of soil respiration in an agricultural ecosystem, Soil Biol. Biochem., 39, 418-425, doi:10.1016Biochem., 39, 418-425, doi:10. /j.soilbio.2006Biochem., 39, 418-425, doi:10. .08.009, 2007 Harris, E., Henne, S., Hüglin, C., Zellweger, C., Tuzson, B., Ibraim, E., Emmenegger, L. and Mohn, J.: Tracking nitrous oxide https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c Author(s) 2020. CC BY 4.0 License. Note: the coordinates were taken at the center of the plots. Note: flux values represent the mean of 5 chambers per site and per measurement week using the four-point time series and considering the constraint set to evaluate linearity in each measurement cycle (R 2 > 0.60). 5 https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c Author(s) 2020. CC BY 4.0 License. Table 4. Retained predictors of multiple linear regressions for each flux gas (CO2, CH4 and N2O) in August. ρb stands for bulk density and "_5" for properties measured at 5 cm depth. R 2 is the adjusted coefficient of determination, and P the significance level of the model. In each case, the R 2 and P values reflect the result of the multiple regressions done for each flux and considering all the retained predictors i.e. considering the predictors that are even not significant. 'ns', '*', '**', and '***' represent the significant levels of each estimate at P > 0.05 (non-significant), 0.01 < P ≤ 0.05, 0.001 < P ≤ 0.01, and P ≤ 0.001, 10 respectively. https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c Author(s) 2020. CC BY 4.0 License. https://doi.org/10.5194/bg-2020-105 Preprint. Discussion started: 3 April 2020 c Author(s) 2020. CC BY 4.0 License.