Spatial Variation of Soil CO2, CH4 and N2O Fluxes Across Topographical Positions in Tropical Forests of the Guiana Shield

The spatial variation of soil greenhouse gas fluxes (GHG; carbon dioxide—CO2, methane—CH4 and nitrous oxide—N2O) remains poorly understood in highly complex ecosystems such as tropical forests. We used 240 individual flux measurements of these three GHGs from different soil types, at three topographical positions and in two extreme hydric conditions in the tropical forests of the Guiana Shield (French Guiana, South America) to (1) test the effect of topographical positions on GHG fluxes and (2) identify the soil characteristics driving flux variation in these nutrient-poor tropical soils. Surprisingly, none of the three GHG flux rates differed with topographical position. CO2 effluxes covaried with soil pH, soil water content (SWC), available nitrogen and total phosphorus. The CH4 fluxes were best explained by variation in SWC, with soils acting as a sink under drier conditions and as a source under wetter conditions. Unexpectedly, our study areas were generally sinks for N2O and N2O fluxes were partly explained by total phosphorus and available nitrogen concentrations. This first study describing the spatial variation of soil fluxes of the three main GHGs measured simultaneously in forests of the Guiana Shield lays the foundation for specific studies of the processes underlying the observed patterns.


INTRODUCTION
The rise of greenhouse gas (GHG) concentrations in the atmosphere has been the main driver of recent climate warming on Earth. This increase is attributed mainly to anthropogenic activities, such as deforestation, agricultural practices and the burning of fossil fuels. Most of the atmospheric GHGs are, however, produced or consumed in natural ecosystems, particularly by soil processes. Carbon dioxide (CO 2 ), methane (CH 4 ) and nitrous oxide (N 2 O) are the three main GHGs in terms of radiative forcing. Special attention has historically been paid to CO 2 because it is present in higher quantities in the atmosphere and CO 2 effluxes are the easiest to measure. Nonetheless, the warming potentials of CH 4 and N 2 O molecules are 28 and 265 times greater than CO 2 , respectively (on a 100year basis), and even if their importance has been demonstrated worldwide, determining the magnitude of N 2 O and CH 4 fluxes and their driving forces in soils of various ecosystems has recently become one of the most pressing issues in the study of ecosystem GHG balances (Merbold and others 2015). The contribution of CH 4 and N 2 O exchange to the total GHG budgets of various ecosystems was estimated from relatively small (Merbold and others 2013-subalpine grassland; Peichl and others 2014-temperate pine plantation) to more than 50% (Hö rtnagl and Wohlfahrt 2014-meadow; Zona and others 2013-poplar plantation).
Soil physical, chemical and biological characteristics are linked to variation in GHGs emitted from soils which in turn usually display a high degree of spatial and temporal variability (Silver and others 1999;Arias-Navarro and others 2017). Soil GHG fluxes also vary with topography, either directly through dynamics of surface and subsurface water, nutrients and dissolved organic matter (Fang and others 2009), or indirectly, via differences in soil texture and vegetation (Luizao and others 2004). Although CO 2 effluxes from soils are always emissions, N 2 O and CH 4 fluxes can shift from sink to source depending on environmental conditions. For these two gases, the static chamber method, commonly used to measure net gas fluxes on the soil surface, cannot differentiate the simultaneously occurring production and consumption within the soil but gives the result of these co-occurring processes.
Soil CO 2 efflux is the result of two main sources of CO 2 production in the soil: CO 2 respired by living roots and the rhizosphere, and CO 2 respired by heterotrophic microorganism activities during decomposition of fresh litter and soil organic matter (Hanson and others 2000;Janssens and others 2001). Soil temperature exerts a dominant control over the seasonal variation of CO 2 effluxes in temperate forests but is less important in tropical soils where temperatures are particularly high and stable year-round (Smith and others 2003). In contrast, in tropical soils, soil moisture and more generally the variation in precipitation are factors responsible for most of the seasonal variation in soil CO 2 effluxes (Davidson and others 2000), with generally decreasing fluxes under drier conditions (Davidson and others 2000;Bonal and others 2008;Rowland and others 2014;Meir and others 2015). Although microbial activity is often limited in dry conditions, very wet conditions can also inhibit CO 2 production in soils by limiting O 2 availability for decomposition processes and gas diffusivity (Davidson and others 2000). Although several studies have highlighted significantly higher CO 2 effluxes in lower topographical positions as compared to higher positions (Epron and others 2006;Brito and others 2009;Martin and Bolstad 2009;Riveros-Iregui and McGlynn 2009), such patterns are not always consistent (Arias-Navarro and others 2017). This could be due to high spatial heterogeneity in CO 2 effluxes, soil characteristics and nutrient availability in tropical forests.
CH 4 production is the result of the obligate relationships between fermentative bacteria and methanogenic archaea. This production primarily takes place in wetland soils (Bartlett and Harriss 1993) but can also occur in upland soils in anaerobic microsites (Silver and others 1999;Teh and others 2005;Liptzin and others 2011). In many parts of tropical forests, especially in lowlands, the soils can remain anoxic, for sometimes long periods, simulating wetland environments and enhancing methane production (Liptzin and others 2011). In well-aerated soils, CH 4 can be oxidized by methanotrophic microorganisms and CH 4 oxidation normally exceeds production, which results in a net CH 4 uptake (Von Hedin 2002, 2007). At an annual scale, the soils of tropical forests are generally a net sink for CH 4 (Dutaur and Verchot 2007). At fine temporal and spatial scales, the emission or consumption of CH 4 can vary depending on nutrient availability (Veldkamp and others 2013), soil oxygen availability (Silver and others 1999;Teh and others 2005;Liptzin and others 2011), bacterial community composition (Teh and others 2008), soil structure (Teh and Silver 2006) or soil water content (Davidson and others 2008). Similar to CO 2 , contrasting evidence suggests either an effect (CH 4 uptake in ridges and slopes and emissions in valleys) (Silver and others 1999) or lack of an effect (Reiners and others 1998;Wolf and others 2012) of topographical position on CH 4 fluxes.
The microbial processes of nitrification and denitrification are the dominant sources of N 2 O from the soil (Butterbach-Bahl and others 2013). This occurs during the conversion of NH 4 + -NO 3 − when ammonia is oxidized in nitrate in the presence of oxygen, and during the conversion of NO 3 − to N 2 O or N 2 when nitrate is used as an electron acceptor instead of oxygen Hall and others 2004). The relative importance of nitrification as a source of N 2 O may depend on the microbial populations involved in ammonia oxidation which in turn can vary with soil pH (Hink and others 2017;Prosser and Nicol 2008). N 2 O is emitted during these processes particularly when N availability is high. Production of N 2 O, controlled by N availability and soil moisture, is commonly found in tropical soils van Lent and others 2015). It is generally assumed that denitrification, the electron-transport-linked reduction in nitrogen oxides during respiration of heterotrophic bacteria, is responsible for N 2 O consumption (Bremner 1997). This also occurs in all biomes ranging from the tropics to the poles (Chapuis-Lardy and others 2007; Schlesinger 2013). In addition, nitrifiers also play a role in the consumption of N 2 O as they are able to produce N 2 from nitrite ðNO À 2 Þ. This pathway, called nitrifier denitrification, is now known to proceed from NO À 2 via nitric oxide (NO) and N 2 O-N 2 (Schmidt and others 2004). Spatial variation and temporal variation in N 2 O fluxes from the soil are notoriously high and require more understanding of the underlying soil characteristics that may drive N 2 O production.
Soil temperature and moisture directly control production, consumption and transport of N 2 O through effects caused by the metabolic activity of microorganisms, soil aeration and diffusivity (Luo and others 2013). Moreover, other environmental parameters have an effect on N 2 O fluxes, such as oxygen availability, nutrient availability and pH of the soil (Weslien and others 2009;Rowlings and others 2012;Yang and others 2017). A recent meta-analysis (van Lent and others 2015) indicates that tropical N 2 O and NO fluxes can be expressed as a combination of nitrogen availability and waterfilled pore space, even though the predictive power for simulating overall N 2 O emissions was low (R 2 $ 0.4). Emissions of N 2 O from tropical soils are typically extremely variable in both space and time (Arias-Navarro and others 2017), with transient peaks associated with rain events (Nobre and others 2001;Petitjean and others 2015). N 2 O fluxes from tropical forests are generally greater during rainy seasons than during dry seasons, a result observed in Amazon (Davidson and others 2004), Central Africa (Rees and others 2006) and in south Chinese forests (Werner and others 2006). Studies reported thus far generally agree that topographical position does have an effect on N 2 O emissions with greater N 2 O emissions in the valley bottom positions than top hill or midslope positions (Pennock and others 1992;Corre and others 1996;Silver and others 1999; Arias-Navarro and others 2017), yet the soil properties driving these patterns are not clear.
French Guiana is located in the Guiana Shield, a part of the Amazonian rainforest that remains one of the largest undisturbed tropical forests in the world (Hansen and others 2013). This tropical region lays on a Precambrian geological substrate that is particularly low in P content compared to the generally younger, nutrient-rich soils of western Amazonia (Hammond 2005;Grau and others 2017). Although several studies have reported CO 2 effluxes from soils in the Guiana Shield (Janssens and others 1998;Epron and others 2006;Bonal and others 2008; Bré chet and others 2011), only one study recently measured N 2 O fluxes (Petitjean and others 2015) and none, to our knowledge, focused on CH 4 . Soil water content and nutrient availabilities vary with topographical position in French Guianese tropical forests (Epron and others 2006;Ferry and others 2010;Stahl and others 2011;Allié and others 2015). This local spatial variation (less than 200 m) creates specific habitats that differ in terms of soil characteristics, aboveground vegetation (Sabatier and others 1997), forest structure (Baraloto and others 2007; Allié and others 2015) and forest dynamics (Ferry and others 2010). Here, we simultaneously measured the three soil GHG fluxes together with soil chemistry and characteristics in two hydrologically contrasting periods (dry and wet conditions) along three topographical positions (top hill, middle slope and bottom slope). Our main objectives were (1) to test whether topographical position has an impact on GHG fluxes and (2) to identify the main environmental drivers of the three GHG fluxes in these nutrientpoor soils.

Description of the Sites
This study was conducted in French Guiana, which is part of the Guiana Shield, one of three South American cratons (Gibbs and Barron 1993). More specifically, we sampled two sites: the Nouragues research station (Bongers 2001) (04˚05´N, 52˚40´W) and the Paracou research station (Gourlet-Fleury and others 2004) (04˚15´N, 52˚55´W) ( Figure 1A). Both sites are covered by pristine tropical forest, receive similar mean annual quantities of rainfall (2990 and 3041 mm year −1 at Nouragues and Paracou, respectively) and have a mean annual air temperature near 25.7˚C (Bongers 2001;Gourlet-Fleury and others 2004). The tropical wet climate of French Guiana is highly seasonal due to the north/south movement of the Inter-Tropical Convergence Zone. This zone brings heavy rains from December to July (wet season) and a long dry period from August to November (dry season, Supplementary Figure 1). Precipitation during the dry period is typically less than 50 mm mo −1 . The Nouragues site has sandy soils of variable depth from a parental material of weathered granite (van der Meer and Bongers 1996). Study plots at the Paracou site are located on schist soils with veins of pegmatite along a Precambrian metamorphic formation called the Bonidoro series (Epron and others 2006). The soils at both sites are characterized as nutrient-poor Acrisols (FAO-ISRIC-ISSS 1998) (Nachtergaele and others 2000).

Study Plots
For this study, we picked three topographical positions: (1) top of hills (top hill), (2) the middle of the slopes at intermediate elevation (middle slope) and (3) bottom end of the slopes, at low elevation, just above the creek (bottom slope) ( Table 1). At each site, four plots of 20 × 20 m per topographical position were established (distances between plots of 10-200 m) in the vicinity of long-term undisturbed plots that have been monitored at both sites for 30 years ( Figure 1). The sand content was higher, and the clay content was lower in the bottom slope plots than in the top hill and middle slope plots at both sites (Table 1). In each plot, five polyvinyl chloride (PVC) collars 20 cm in diameter were inserted into the soil 6 months prior to the first measurement to an average depth of 3.8 cm (± 0.5 cm) in the 20 × 20 m plot ( Figure 1B), for a total of 120 sampling points (2 sites, 3 hill positions, 4 plots per position, 5 points per plot).

Soil Characteristics
Soil samples for chemical and physical characterization were collected on the same day as fluxes. A composite soil sample of three soil cores around each sampling point (that is collars) was analyzed. Soil water content (SWC) at a depth of 10 cm and air surface temperature (average of three measurements) were recorded around each collar and on each sampling occasion using a time-domain reflectometer (IMKO-HD2 portable meter fitted with a PICO64 probe, Ettlingen, Germany). Soil bulk density to a depth of 5 cm was measured with 100-cm 3 cylinders. All other soil characteristics were measured on 0-15 cm top soil at each sampling point (once in the dry period and once in the wet period). Soil pH (KCl) was measured by mixing 10 g of moist soil with 1 M KCl in a 1:2.5 ratio. The resulting slurry was stirred for 1 h and then allowed to sit for another hour before pH was measured using a pH probe. The amount of total phosphorus (P in ppm) was determined through microwave digestion (CEM, MARS-5) of pulverized soils, previously dried for 48 h at 70˚C, in trace-metal-free acid (mixture of concentrated HNO 3 , HCl and HF). P concentrations in digests were measured on ICP-MS (Optima 4300 DV, Perkin-Elmer, Waltham, USA). The amount of available P (in ppm) was measured by Bray-P acid fluoride extraction (Bray and Kurtz 1945) of soil dried at 60˚C for 48 h, and the resulting solution was analyzed on an iCAP 6300 Duo ICP optical emission spectrometer (Thermo Fisher Scientific). The amount of available nitrogen (N in ppm) was measured by extracting moist soil with 1 M KCl, after which the concentrations of NH 4 + and NO 3 − were determined colorimetrically on a SAN++ continuous flow analyzer (Skalar Inc, Breda, the Netherlands). The concentrations of NH 4 + and NO 3 − were summed and treated as a single value during analyses and are reported hereafter as available N. Soil C and N concentrations (in percent) were determined by combustion (Elemental Analyzer, CE Instruments/ Thermo Electron, Milan, Italy) coupled with gas chromatography/mass spectrometry (Delta V Advantage, Thermo Fisher Scientific, Cambridge, USA). Soil C:N ratios were obtained by dividing C concentrations (percent) by N concentrations (percent).

Soil Fluxes
Two single-survey sampling campaigns were conducted at each site, one during the dry period (October 2015) and one during the wet period (May 2016) corresponding to the two extreme conditions in soil water content (Supplementary Figure 1) to capture the full range of local environment difference. We measured the soil fluxes of CO 2 , CH 4 and N 2 O during each campaign using opaque (no light allowed) static soil PVC chambers (volume, 0.0026 m 3 ; area, 0.029 m 2 , designed in Antwerp laboratory) between 10 a.m. and 2 p.m. to Values are means ± standard errors (N = 4). Percentages of sand and clay contents were determined for a composite sample from each topographical position at each site, so only one value is available for each topographical position.
avoid diurnal variability (Teh and others 2014; Yan and others 2014; Brechet and others 2011). In total, 240 flux estimations were made during the two-sampling campaign (60 sampling points in each forest in the two periods and the two sites). Each sampling point was therefore only measured once in each period. After the chambers were sealed onto the soil collars, four headspace gas samples were collected, one immediately after closure and then at subsequent 10-min intervals. Air samples were taken with a 12-ml syringe whose needle was inserted through a septum in the chamber and then injected into pre-evacuated 12ml vials (Labco Limited, Ceredigion, UK). To mix the air in the chamber headspace prior to sampling, air was flushed five times with the syringe volume prior to the second, third and fourth air samples in each chamber. The CO 2 , CH 4 and N 2 O mole fractions in the gas samples were determined by gas chromatography (Trace GC Ultra, Thermo Fisher Scientific, Vienna, Austria) equipped with a vacuum dosing system (S + H Analytics, Germany). Gas samples were filled into a 1 ml sample loop at a pressure of 600 mbar. We used a flame ionization detector (FID) with a methanizer for CO 2 and CH 4 detection and a pulsed-discharge detector for N 2 O detection.
The limits of detection (LoD) of CH 4 and N 2 O were calculated from least-square regression data according to the ICH (International Council for Harmonization). We used the "residual standard deviation of the regression line" method between the peak area and concentration of certified standards that were in the range of the detection limit (0.29, 0.54 and 0.91 ppm N 2 O and 1.01, 2.16 and 4.17 ppm CH 4 ). We also checked our data for homoscedasticity (that is, independence of the peak area dispersion from the concentration of the standards). The LoD of the samples was then calculated as 3 times the standard deviation of the residuals divided by the slope of the linear regression of the calibration. For comparison, we also used the "standard deviation of the intercept" method, which yielded LoD values in the same range than the residual method. By using the "residual standard deviation of the regression line" method, we determined the LoD to be 0.116 ppm for N 2 O and 0.202 ppm for CH 4 over all standards (n = 54). Between different sample batches, the LoD varied only slightly between 0.115 and 0.130 ppm for N 2 O and between 0.166 and 0.247 ppm for CH 4 , respectively.
For 17 fluxes measurements (10 in wet period and 7 in dry period), data points were missing either due to problems in the field (that is, leaks) or during laboratory analyses and these estimations were removed for the three gases. All other fluxes were computed using the HMR package (Pedersen and others 2010) for the three gases using linear regression (LR), or revised Hutchinson/Mosier (HMR) methods following recommendations from Pedersen and others (2010) (70% HMR and 30% LR for CO 2 ; 55% HMR and 45% LR for CH 4 ; 65% HMR and 35% LR for N 2 O). Gas mixing ratios (ppm) were converted by using the ideal gas law to solve for the quantity of gas in the headspace (on a mole or mass basis), normalized by the surface area of each static flux chamber.
Calculation of minimum detectable flux (MDF) of N 2 O was made with the methodology developed by Parkin and others (2012). The mean concentration (that is, ambient concentration) of 0.300 ppm of N 2 O and the coefficient of variation of 0.05 was defined from values of N 2 O at sampling time 0. Minimum detectable flux was ± 8.3 μg N m −2 h −1 and 27 fluxes (11%) were below the minimum detection limit (MDL) and included in the analysis as null fluxes.

Statistical Analyses
To test the effect of topographical position on GHG fluxes, we performed a linear mixed-effects model (LMM) using topographical position (top hill, middle slope or bottom slope) as fixed explanatory variables and period (dry or wet season conditions) and site (Paracou or Nouragues) as random factor. Variables were transformed as follows: log (flux) for CO 2 and log (flux − min (flux) + 1) for CH 4 and N 2 O for this analysis. Significance was tested by comparing the null model to a model including topographical positions as fixed factor with a Chisquare test.
The variations of soil characteristics were assessed with a principal component analysis (PCA) using eight soil variables (total P, C:N ratio, SWC, air soil surface temperature-temperature, soil density, soil pH, available P and available N). The spatial segregations of the measurements across period (dry or wet period), site (Paracou or Nouragues) and topographical position (top hill, middle slope or bottom slope) along the two main PCA axes were tested using permutational MANOVA (multivariate analysis of variance).
We used generalized additive model (GAM) for each gas to identify the best set of variables explaining the variation in the gas fluxes (candidates variables: total P, C:N ratio, SWC, air soil surface temperature-temperature, soil density, soil pH, available P and available N). GAMs are semiparametric extensions of generalized linear models allowing nonlinear and non-monotonic relationships between a response and a set of explanatory variables. The form of the predictor function is the principal originality of this method (Fewster and others 2000). The best model (that is, explaining most of the variation) is selected using Akaikes's information criterion. Once the model best fitting the data is selected, a second step using recursive feature elimination allows for the ranking of variables using their importance in the model. For this, variables were backward selected using recursive feature elimination. Variable importance was then defined as the total reduction in the validation statistic (generalized cross-validation estimate of error) when each predictor's feature is added to the model.
All data were processed with R statistical software (R Development Core Team, 2010) using the package HMR (Pedersen and others 2010) for flux computation, the packages ade4 (Dray and Dufour 2007) and Momocs (Bonhomme and others 2014) for PCA analysis and permutational MANOVA, respectively, the package lme4 (Bates and others 2014) for LMM, the packages mgcv (Wood and Wood 2007) for GAM, and the package caret (Kuhn 2008) for recursive feature elimination of variables.

Variation in GHG Fluxes
Mean soil CO 2 efflux ranged from 131.9 ± 65.8 mg C m −2 h −1 in the dry period to 156.4 ± 69.2 mg C m −2 h −1 in the wet period (Figure 2, Supplementary Table 1). Soils tended to be sinks of CH 4 in the dry campaign and sources of CH 4 in the wet campaign (mean fluxes of − 27.7 ± 129.8 µg C m −2 h −1 in the dry period and 12.0 ± 125.2 µg C m −2 h −1 in the wet period, Figure Table 1). Overall, 98% of the sampling points in the dry season and 95% in the wet season were sinks of N 2 O.

Topographical Positions and Soil Characteristics
Topographical position (top hill, middle slope, bottom slope) had no effect on any of the three gas fluxes (LMM; CO 2 χ 2 = 1.19 p = 0.55; CH 4 χ 2 = 3.48 p = 0.18; N 2 O χ 2 = 0.16 p = 0.92). The PCA classified the environmental variables over all sites into two principal components (PCs), which accounted for 51.4% of variation in the soil characteristics. PC1 ( Figure 3A) explained 28.5% of the variance and was correlated positively with soil bulk density and temperature and negatively with available N and total P and. PC2 (Figure 2A) explained 22.9% of the variance and was correlated positively with SWC and negatively with C:N. Soil characteristics in the dry and wet periods are differentiated using these two axes (MANOVA, F = 121.54, p < 0.001), mostly along the second axis ( Figure 3B). Sites are also differentiated using these two axes (MANOVA, F = 33.14, p < 0.001), mostly along the first axis ( Figure 3C). Despite the large overlapping areas in PCA ( Figure 3D), MANOVA also indicated that topographical position can be distinguished using this set of variables (MANOVA, F = 34.08, p < 0.001).

Spatial Variation of Soil GHG Fluxes
The spatial variation in CO 2 efflux was significantly explained by four variables (GAM, SWC > soil pH > available N > total P, Table 2, Figure 4). These four variables in combination explained 20.7% of the variation. The variation in the CH 4 flux was explained by only one variable (GAM, SWC, Table 2, Figure 4), which explained 4.5% of the variation. The CH 4 fluxes were higher in wetter soils (Figure 4). The variation in N 2 O flux was ex- plained by two variables (GAM, CN > total P, Table 2, Figure 4), which explained 9.2% of the variation.

CO 2 Effluxes did not Differ Among Topographical Positions but were Partly Explained by Soil Physical and Chemical Characteristics
The soil CO 2 effluxes measured in this study were within the ranges of previous studies in French Guianese forests (Janssens and others 1998;Epron and others 2006;Bonal and others 2008;Bré chet and others 2011;Rowland and others 2014). Contrary to previous studies in tropical forest (Brito and others 2009;Martin and Bolstad 2009;Riveros-Iregui and McGlynn 2009), even though one conducted in French Guiana (Epron and others 2006), soil CO 2 fluxes were not related to topographical position. However, soil characteristics represented by the PCA analysis also did not differ as much at the different topographical positions defined in this study as in other studies, which might explain this contrast. Moreover, in Epron and others (2006), all measurements were taken in September, during the dry season only, which might explain the discrepancy between our study and their results. A recent study conducted in tropical areas (Arias-  Navarro and others 2017) also did not detect any effect of topographical position on CO 2 effluxes despite a very good spatial coverage of the study site. This result is explained by the very large diversity of tree species in tropical forests (as in our two tropical pristine forests, Table 1), leading to heterogeneity in the chemical, structural and functional traits of roots and leaves affecting biogeochemical processes (Hä ttenschwiler and others 2008; Townsend and others 2008) and resulting in a high spatial heterogeneity in soil respiration. SWC and soil pH were the main drivers of soil CO 2 effluxes across topographical positions even if only 20.7% of the variation was explained by the model. In agreement with previous studies conducted in French Guiana (Bonal and others 2008;Rowland and others 2014), we found a positive relationship between SWC and CO 2 effluxes. Soil pH was negatively correlated with soil CO 2 effluxes as found in a previous study (Epron and others 2006), whereas in temperate areas several studies have highlighted the opposite pattern (Reth and others 2005;Chen and others 2015). This could be due to the relatively low pH (3.9 ± 0.2) of these French Guianese soils. The effect of pH alone is unknown on soil CO 2 effluxes, but some studies suggest a combined effect of pH with some others components like clay content and soil organic matter (Epron and others 2006) or via decreased root respiration (Chen and others 2015). Nutrients availability also appears to have an influence on CO 2 effluxes as CO 2 effluxes were positively correlated with available nitrogen and negatively with total phosphorus.

CH 4 Fluxes Shifted from Source Under Wet Conditions to Sink Under Dry Conditions
As previously shown for other tropical sites (Reiners and others 1998; Wolf and others 2012), we found no impact of topographical positions on CH 4 fluxes. Soil CH 4 fluxes were correlated with SWC with only 4.5% of the variation explained by this variable. CH 4 fluxes shifted from uptake under dryer conditions to moderate emissions under wet conditions, similar to other tropical forest sites (Keller and Reiners 1994;Keller and others 2005;Davidson and others 2008;Teh and others 2014). Increased precipitation is likely to decrease rates of O 2 diffusion into the soil (Silver and others 1999;Teh and others 2005;Liptzin and others 2011), decreasing CH 4 oxidation and increasing CH 4 emissions. Because no other studies are currently available in the surrounding area and as the tem-poral coverage of this study is not sufficient to estimate annual budgets from forests of the Guiana Shield, further investigations on soil CH 4 fluxes need to be conducted.

N 2 O Uptake is Linked to C:N Ratio and Total Phosphorus
Surprisingly, most of the N 2 O fluxes measured in this study were negative across the topographical positions and between hydric conditions (98% of the total samples). This was unexpected in light of the N-rich, wet conditions of these forests and based on previous results from French Guiana (Petitjean and others 2015) and other tropical soils others 2013, 2014;van Lent and others 2015;Arias-Navarro and others 2017). The spatial resolution in our study is in line with an extensive survey of N 2 O fluxes in Africa (Arias-Navarro and others 2017) which suggest that measurements at 78 ± 5 locations per hectare are needed to obtain an estimate of N 2 O fluxes within 10% of the true mean (in our study, 5 sampling locations per 400 m 2 correspond to 125 locations per hectare). We did not detect effects of topographical position on the magnitude of the uptake. The mean flux of − 32 µg N m −2 h −1 measured in our study falls within the range of negative N 2 O fluxes of − 1.4 ng N m −2 h −1 to − 484 µg N m −2 h −1 reported in the review by Chapuis-Lardy and others (2007). More data at a broader spatial and temporal scale are urgently needed to confirm the generality of this pattern and estimate the annual sink capacity of French Guianan soils.
The flux of N 2 O measured with soil chambers at the soil/atmosphere interface is the result of dynamic production and consumption processes in the soil. This capacity of the soil to act as an N 2 O sink is linked to the abundance and phylogenetic diversity of a group of N 2 O-reducing microbes (Jones and others 2014). The community composition of these microorganisms is influenced by soil properties and in particular by soil stoichiometry (total P and C:N ratio) (Butterbach-Bahl and others 2013). Although N 2 O fluxes were significantly correlated with soil nutrients, most of the variation in N 2 O fluxes remained unexplained in our study (only 9.2% of the variation explained by the model).
In contrast to other studies in tropical forest soils (Butterbach-Bahl and others 2004; Tang and others 2006; Luo and others 2013), SWC was not linked with N 2 O fluxes. This could be mostly due to the fact that we almost only measured N 2 O consumption in our study (98% of all sampling points), whereas other studies report mostly N 2 O emission. Moreover, soils with a low water-retention capacity (which is the case for most of the French Guianan soils with high sand content) have limited anaerobic conditions. In addition, competition for NO 3 − between different nitrate-reducing processes under varying moisture conditions (for example, dissimilatory nitrate reduction to ammonium versus denitrification) may obscure the relationship between soil moisture content and N 2 O flux (Morley and Baggs 2010;Morley and others 2008). N 2 O emissions from tropical soils can be sporadic and transient, for example, after heavy rains (Breuer and others 2000;Nobre and others 2001), and are characterized by short pulses of emissions associated with higher nitrogen inputs (Bai and others 2014) or high precipitation events (Breuer and others 2000;Nobre and others 2001;Geng and others 2017). Nevertheless, we detected only one high N 2 O emission event which could not be linked to a heavy rain, but this study was not designed to capture such a fine level of temporal variation. These transient strong peaks of N 2 O emission are likely responsible for the net emission budget of tropical soils highlighted in another study (Nobre and others 2001); a high frequency of N 2 O flux measurements is urgently needed to confirm this relationship.

Conclusion and Perspectives
Our study has shed new light on the drivers of the three main GHG fluxes in tropical soils of the Guiana Shield. The fluxes of the three gases did not differ among topographical positions. Nevertheless, we highlighted the significant role of soil physical (SWC) and chemical (pH, available N, C:N ratio and total P) properties as drivers of soil GHG fluxes even if most of the variation in the three GHG fluxes remains unexplained highlighting again their high spatial variability. Proximate controls on soil fluxes are complex and difficult to elucidate from field measurements alone and more experimental studies are also needed to disentangle the effect of different soil characteristics and microbial processes, which result in a modification of soil GHG fluxes. Nitrogen deposition is expected to increase in tropical areas (Penuelas and others 2013) which can induce soil acidification (Tian and Niu 2015) and changes in soil stoichiometry and could impact GHG emissions in the future. Tropical forest soils contribute an estimated 28% to the global CH 4 uptake (Dutaur and Verchot 2007), which is large enough to alter the accumulation of CH 4 in the atmosphere if large changes occur in this sink due to climate change such as increased drought (Brumme and others 1999).
Specific studies should be conducted to assess the influence of climatic changes (especially changes in precipitation and an increased occurrence or severity of drought) combined with modifications to soil stoichiometry (mainly by N deposition) on the magnitude of soil GHG fluxes in these poornutrient tropical soils over a long period to determine (1) the relative frequencies of consumption and emission and (2) the net GHG balance at annual scales. thank the subject-matter editor Dr Butterbach-Bahl, Dr Teh and one anonymous reviewer for their comments on previous versions of this manuscript. We thank Nicola Arriga, Jan Segers and Fred Kockelbergh for building the chambers and for advice on the field measurements. We are grateful to Stanislas Talaga, Jé rô me Levy-Valensky and Jean-Pierre Robert for their help in the field, to Oriol Grau and Vincent Freycon for the identification and characterization of each topographical position and to Margarethe Watzka for the gas analyses.