Spatial and temporal variability of soil N2O and CH4 fluxes along a degradation gradient in a palm swamp peat forest in the Peruvian Amazon

Abstract Mauritia flexuosa palm swamp, the prevailing Peruvian Amazon peatland ecosystem, is extensively threatened by degradation. The unsustainable practice of cutting whole palms for fruit extraction modifies forest's structure and composition and eventually alters peat‐derived greenhouse gas (GHG) emissions. We evaluated the spatiotemporal variability of soil N2O and CH4 fluxes and environmental controls along a palm swamp degradation gradient formed by one undegraded site (Intact), one moderately degraded site (mDeg) and one heavily degraded site (hDeg). Microscale variability differentiated hummocks supporting live or cut palms from surrounding hollows. Macroscale analysis considered structural changes in vegetation and soil microtopography as impacted by degradation. Variables were monitored monthly over 3 years to evaluate intra‐ and inter‐annual variability. Degradation induced microscale changes in N2O and CH4 emission trends and controls. Site‐scale average annual CH4 emissions were similar along the degradation gradient (225.6 ± 50.7, 160.5 ± 65.9 and 169.4 ± 20.7 kg C ha−1 year−1 at the Intact, mDeg and hDeg sites, respectively). Site‐scale average annual N2O emissions (kg N ha−1 year−1) were lower at the mDeg site (0.5 ± 0.1) than at the Intact (1.3 ± 0.6) and hDeg sites (1.1 ± 0.4), but the difference seemed linked to heterogeneous fluctuations in soil water‐filled pore space (WFPS) along the forest complex rather than to degradation. Monthly and annual emissions were mainly controlled by variations in WFPS, water table level (WT) and net nitrification for N2O; WT, air temperature and net nitrification for CH4. Site‐scale N2O emissions remained steady over years, whereas CH4 emissions rose exponentially with increased precipitation. While the minor impact of degradation on palm swamp peatland N2O and CH4 fluxes should be tested elsewhere, the evidenced large and variable CH4 emissions and significant N2O emissions call for improved modeling of GHG dynamics in tropical peatlands to test their response to climate changes.

Peruvian Amazon, peatlands lie primarily in the Pastaza-Marañon Basin, where they cover 3.6 M ha, harbor peat deposits up to 7.5 m thick and store around 3.1 Pg C, that is, almost half the aboveground biomass C stock of Peruvian forests (Asner et al., 2014;Draper et al., 2014;Lähteenoja & Page, 2011). Lähteenoja and Page (2011) reported a substantial diversity of peatland ecosystems in the region, from rain-fed nutrient-poor ombrotrophic systems to river-fed nutrient-rich minerotrophic swamps. The majority (80%) support a growth of Mauritia flexuosa-dominated forests that are regularly flooded by dynamic rivers such as the Amazon and their tributaries (Draper et al., 2014).
Peruvian Amazonian peatlands have been partially protected from deforestation and fires in contrast to their Southeast Asian counterparts (Lilleskov et al., 2018;Roucoux et al., 2017), but have suffered intensive large-scale degradation over decades (Horn et al., 2018). In a 350,000 ha area in the Pastaza-Marañon Basin, 31% of M. flexuosa palm swamp peatlands were found to be strongly degraded, 42% moderately degraded and 27% had low levels of degradation . The fruits of M. flexuosa (locally named Aguaje) are highly demanded in the local market but are often collected by cutting down the entire palm. This degradation translates into changes in vegetation cover and composition, reductions in biomass C stocks and litter inputs, and could potentially result in destabilization of peat deposits (Bhomia et al., 2019;van Lent et al., 2019). Research on greenhouse gas (GHG) emissions from tropical peatlands under natural or disturbed conditions has mostly been conducted in Southeast Asia, and very little is known about fluxes in neotropical peatlands (e.g., Hoyos-Santillan et al., 2016;Teh et al., 2017;Wright et al., 2011Wright et al., , 2013 and their response to ecosystem degradation. In addition, due to the preponderance of CO 2 in drained peatlands GHG budgets, there has been continuing research emphasis on this gas despite the high global warming potential of CH 4 and nitrous oxide (N 2 O; Myhre et al., 2013) and the need for full GHG accounting for climate change predictions (Hergoualc'h & Verchot, 2014;Sjögersten et al., 2014). Soil-atmosphere exchanges of both CH 4 and N 2 O in peatland forests can be highly spatially variable because of microscale variations in topography. The peat surface consists of a mosaic of hummocks densely packed with roots around tree bases and sparsely vegetated hollows (Jauhiainen et al., 2005). Differences in hydrological and substrate conditions between these two microtopographies lead to large spatial variation in GHG fluxes (Jauhiainen et al., 2005). Degradation of palm swamp peatlands may alter CH 4 and N 2 O emissions in several ways. Changes in vegetation density and composition can prompt a modification in the hummock to hollow ratio and site-scale proportion of microtopography-derived emissions. Vegetation changes also induce alterations in litter quantity and quality (Hergoualc'h, Hendry, et al., 2017;van Lent et al., 2019), thereby influencing GHG fluxes.
Finally, degradation may also be expected to modify microclimate conditions.
Understanding mechanisms and main factors driving soilatmosphere exchanges of CH 4 and N 2 O is essential for constraining non-CO 2 budgets and for predictions of climate and land-use changes (Aini et al., 2015;Hergoualc'h & Verchot, 2014). N 2 O fluxes from peat soils are governed by variables that limit the microbial processes of nitrification and denitrification, such as availability of mineral N and labile organic matter, soil moisture and aeration status, and soil temperature (Jauhiainen et al., 2012;Schlesinger, 2013). Soil N 2 O emissions often correlate well with the soil water-filled pore space (WFPS), with emission rates in the tropics found to be maximum around a WFPS of 60% and to remain high at 80% WFPS (van Lent et al., 2015). CH 4 dynamics in waterlogged soils result from the balance between CH 4 production and consumption by soil microorganisms (Bridgham et al., 2013). Production occurs with the anaerobic decay of organic material in the soil saturated zone under highly reduced conditions. Consumption, that is, the oxidation of CH 4 to CO 2 mostly occurs where oxygen is available and as CH 4 moves through less reduced zones in the peat (Horwath, 2007;Strack et al., 2008).
Therefore, CH 4 emissions in peatlands have been found to be particularly related to the water table level (WT), and several studies have reported a decline in net CH 4 flux accompanying a lowering of the WT (Bridgham et al., 2013;Hoyos-Santillana et al., 2016). Other general controls on wetlands CH 4 emissions include soil temperature and vegetation (Turetsky et al., 2014). In particular, the CH 4 production rate is influenced by C substrate quality (Hoyos-Santillana et al., 2016;Le Mer & Roger, 2001;Wright et al., 2011Wright et al., , 2013. Because the biosphere reacts to climate change, the importance of the response of GHG fluxes to temperature and precipitation has received attention in past decades (Singh et al., 2010). Disruption of GHG dynamics in palm swamp peatlands may occur under future changing climate (Wang et al., 2018). Projections in the northwestern Amazon indicate a warmer climate by the end of the 21st century and a trend toward increased precipitation, and a higher frequency of wet days and floods (Barichivich et al., 2018;Espinoza et al., 2016;Gloor et al., 2013;Marengo et al., 2018). Process-based modeling of the response of peat C accumulation rate to warmer and wetter climate suggests that palm swamp peatlands in the Peruvian Amazon may switch from a C sink to a source (Wang et al., 2018), but data uncertainty and scarcity leave doubts. For instance, CH 4 fluxes simulated by Wang et al. (2018) were not validated against field measurements, presumably owing to unavailability of long-term data. Also, the model focused on the C cycle disregarding any assessment of the response of N 2 O emissions to climate change.
This study investigated soil fluxes of N 2 O and CH 4 and their environmental controls along a gradient of degradation in a peat swamp forest in the Peruvian Amazon. The gradient consisted of one undegraded site and two sites with different levels of degradation. The measurements were conducted monthly over 3 years that include El Niño and La Niña episodes. We addressed the following research questions: (a) How do N 2 O and CH 4 fluxes vary spatially at the microscale and macroscale under undegraded and degraded conditions? (b) How do the fluxes vary intra-annually and inter-annually? (c) How do environmental variables control the spatial and temporal variation of the fluxes? Spatial variability at the microscale was assessed by differentiating hummocks supporting live or cut palms from surrounding hollows. The macroscale analysis considered structural changes in vegetation and soil microtopography as impacted by degradation.

| Site description
The study was carried out in the province of Loreto, southwest of Iquitos, in the Northern Peruvian Amazon. Based on 1948Based on -1994 weather data, the climate of the region is warm and humid with mean annual temperature of 27°C and mean annual rainfall of 3,087 mm (Marengo, 1998). Most months (66%) exhibit precipitation rates in the range 100-300 mm; while months with precipitation either <100 mm (7%) or >400 mm (10%) are infrequent. High precipitation months (>300 mm) are infrequent between June and September, and more frequent between January and April.
This research was conducted in an area of peat swamp forest dominated by M. flexuosa palms, located near the Itaya River, one of the tributaries of the Amazon River . Peat deposits up to 5 m deep have been reported in this forest (Lähteenoja, Ruokoleinen, Schulman, & Oinonen, 2009). Following successional ecological stages, the M. flexuosa swamp palm forest which persists today started to form about 1,000 years ago, with its current vegetation community established c. 400 years ago (Roucoux et al., 2013).
The vegetation development over time has been highly influenced by the flooding regime. The swamp is quasi-permanently waterlogged and occasionally floods, such as in 1998 (30 cm) and 2012 (100 cm; Roucoux et al., 2013). The peat was classified as minerotrophic (nutrient-rich;Lähteenoja, Ruokolainen, Schulman, Alvarez, et al., 2009;van Lent et al., 2019) and likely receives nutrients during flooding events, as well as during the annual flood pulses of the Amazon River.
Along this swamp forest complex, we selected a site that was undegraded, thereafter referred to as "Intact" (S 03°49.949′ W 073°18.851′), a site moderately degraded ("mDeg," S 03°50.364′ W 073°19.501′), and a heavily degraded site ("hDeg," S 03°48.539′ W 073°18.428′). The Intact site was located within the protected Quistococha Regional Reserve. The mDeg and hDeg sites were adjacent to the reserve and used by local communities for extraction of M. flexuosa fruits and timber harvesting. The Intact site exhibited a closed canopy; the mDeg site had reduced canopy closure, and the hDeg site had a very open canopy with few standing trees . M. flexuosa was present at all sites and was the most important species at the Intact and mDeg sites according to IVI (Importance Value Index), while the hDeg site was dominated by Cecropia membranacea, a pioneer tree species (Bhomia et al., 2019).
The density (individuals with a diameter at breast height ≥10 cm) of M. flexuosa palms was 170, 164 and 16 ha −1 at the Intact, mDeg and hDeg sites, respectively, and that of dicotyledonous trees was 1,496, 700 and 679 ha −1 , respectively. Peat depth and pH of stagnant water at the Intact, mDeg and hDeg sites were 2.2, >2.7, 1.0 m and 5.9, 5.9, 6.6, respectively (Bhomia et al., 2019). (c) Layout of the two subplots within a plot at the medium degraded (mDeg) and highly degraded (hDeg) sites, including a standing live palm and a cut palm. Chambers were permanently installed on hummocks and hollows to measure soil CH 4 and N 2 O fluxes. Source of the Mauritia flexuosa scan in (b) and (c) is from Caballero (2017  and CH 4 at ambient concentration and chamber deployment time.

| Experimental design and GHG flux measurement
Mean and coefficient of variation of 35 N 2 O and CH 4 ambient samples taken at random from monthly GC control procedure datasets amounted to, respectively, 330 ppb and 9% for N 2 O and 1,522 ppb and 1% for CH 4 . The detection limit was 6.8 g N ha −1 day −1 for N 2 O and 2 g C ha −1 day −1 for CH 4 . Following recommendations by Parkin et al. (2012) and Gilbert (1987), we report the actual measured flux values even if they fall within the detection limit band.

| Soil properties
In all, 24 samples per site (12 from hollows and 12 from hummocks) were collected from the upper 5 cm of the soil profile using a metal ring (radius = 4.5 cm; van Lent et al., 2019). Eighteen of them (nine from hollows and nine from hummocks) were oven-dried at 60°C until constant mass and their bulk density was determined from the dry mass per ring volume. The remaining six samples (three per microtopography) were each ground, homogenized and analyzed for total C and N content by the induction furnace method (Costech EA C-N Analyzer). Their exchangeable cations, cation exchange capacity and base saturation were determined using the standard ammonium acetate at pH 7 method (Pansu et al., 2001). Copper (Cu), manganese (Mn), zinc (Zn) and phosphorus (P) were determined by the Mehlich 3 method (Ziadi & Sen Tran, 2007). All analyses were performed by the University of Hawaii-Hilo.

| Environmental parameters
Daily precipitation rates were measured using a rain gauge with a tipping bucket (0.2 mm resolution, Delta Ohm HD2013R) equipped with a thermometer for hourly recording of air temperature. This climate station was situated between 1 and 3 km from the sites. at 60°C until reaching constant mass and soil moisture was determined on a dry mass basis. The soil WFPS was calculated based on the formula by Linn and Doran (1984) as the product of gravimetric soil moisture and bulk density divided with the porosity, assuming a particle density of 1.4 g/cm 3 (Driessen & Rochimah, 1976

| Statistics
The potential effects of disturbance level (i.e., site: Intact, mDeg and hDeg), status (live and cut palms), microtopography (hummock and hollow) and temporal variability (40 months; i.e., overlapping period across sites) on the fluxes of N 2 O and CH 4 were examined using a mixed-effects ANOVA model with four factors. Sampling month was treated as a random factor, whereas disturbance level, status and microtopography were treated as fixed effects. The design was asymmetric because not all status levels were present at all sites (there were no cut palms at the intact site), and it was unbalanced because not all combinations of treatments had nine replicates per sampling month (due to missing data). Considering these limitations, adequate degrees of freedom, mean squares and F-ratio were calculated using the PERMANOVA routine in PRIMER v7. Although this method was specially designed for multivariate data, it can be used for univariate data using Euclidean distance (Anderson, 2017). Its advantage is that it properly handles asymmetrical and unbalanced designs, like this one. Additionally, as residuals were in general not normally distributed, the ANOVA using the traditional tabulated F values would be inappropriate. Therefore, the probability associated to each empirical F-ratio was estimated using 9,999 permutations of the residuals under the reduced null model. To verify the assumptions for a correct interpretation of each ANOVA, graphical and quantitative analyses were performed ( Figure S1). Also, the Levene's test for homogeneity of variances was used to assess the equality of variances when significant interactions were found in the ANOVA. When main effects or interactions effects were statistically significant, pairwise t tests based on permutation were applied. A probability level of 5% was used to test the significance of effects.
The temporal and spatial changes of the environmental variables were evaluated by applying the same multifactorial ANOVA model used for N 2 O and CH 4 emissions to each variable. For the first significant interaction that included degradation level, status or microtopography, a paired t test with combined variances was applied after weighting by n and with probability correction for multiple comparisons using the Holm procedure (Holm, 1979). For each of the paired comparisons, probability values were obtained after the Holm adjustment. Soil properties and inorganic N were compared between sites within a microtopography, and between microtopographies within a site, using nonparametric tests as their residuals were not normally distributed according to the Shapiro-Wilks test.
Relationships between the fluxes and environmental variables were examined using monthly averages, annual averages and sitescale annual values. Relationships with monthly averages were investigated site specifically at the microscale and by aggregating palm status and/or microtopography. They were also investigated across sites using classes for the independent variables (e.g., 10% intervals for the WFPS). Relationships with annual averages were tested across sites considering palm status and microtopography.
Detection of emission hotspots was performed using boxplots analysis. A chamber qualified as a hotspot when it displayed over the monitored period at least three values higher than three times the interquartile range from the upper edge of the 50% percentile (i.e.,

| Soil properties and environmental variables
Edaphic properties were homogeneous among sites and microtopographies (Table S1). Bulk densities were low, and in hollows they were higher as degradation increased (from 0.08 to 0.11 g d.m./cm 3 ; p = .0003). Soil macronutrients differed between sites only in hollows, with higher levels of Ca at the hDeg site than at the mDeg site and a higher concentration of Mn at the hDeg site than at the Intact site.
Over the whole measurement period, inorganic N pools were dominated by NH 4 were quite low (<4 mg N/kg d.m.; Table 2). The absolute value of net mineralization rate (>18 mg N kg −1 d.m. day −1 ) was also much higher than net nitrification rate (<1 mg N kg −1 d.m. day −1 ). Soil NH 4 + content was lower in hollows than in hummocks at the hDeg site (p < .0001). Likewise, net mineralization was lower in hollows than in hummocks at the Intact site (p < .0001), and in general (p = .05) corresponding to higher soil moisture levels (p < .0001). NH 4 + content was lower at the hDeg site than at the other sites while net mineralization increased with degradation (p < .0001 for both). Soil mineral N content and dynamics did not follow a consistent temporal pattern but in April 2017 overall NH 4 + contents, net mineralization and net nitrification were at their maximum, while NO 3 − contents were at their minimum (p < .0001; Figure S2d,e). Monthly NH 4 + content decreased when soil respiration raised in hummocks at the hDeg site (R 2 = .55, p = .03) and increased when the water level went up in hollows at the hDeg site (R 2 = .57, p = .03). Monthly net nitrification rate decreased when the soil WFPS rose in hollows at the mDeg site (R 2 = .55, p = .03).
Monthly and annual precipitation during the measurement period are presented in Figure 2.  (Figure 3b left chart). Soil temperature over the 3 years was higher at the Intact site than at the degraded sites (Figure 3b right chart).
The WT was higher in hollows than in hummocks at all sites and  Left charts displays means per month (n = 9 per palm status by spatial position at each site) and site-scale annual means per year. Right charts display 3-year means per palm status by spatial position (n = 308-324) and at site scale. Letters indicate significant differences in means between microtopographies within a site and palm status. Numbers indicate significant differences in means between sites the soil surface. Neither that month nor the previous month had the rainfall been higher than usual (Figure 2).
The soil WFPS was similar among microtopographies at the Intact site ( Figure 5 right chart). It was higher in hummocks than in hollows at the mDeg site for both live and cut palms, while the opposite occurred at the hDeg site. Over the 3 years, the WFPS was higher at the mDeg site than at the Intact and hDeg sites, in hummocks and hollows and for both live and cut palms ( Figure 5 right chart). Sitescale annual WFPS followed the order Intact < hDeg < mDeg over Site-scale annual emissions were similar among sites in year 1 when the water level was high (>6 cm above ground). They were higher at the intact site than at the degraded sites in year 2, and followed F I G U R E 6 Monthly mean N 2 O emissions in hummocks and hollows around live and cut palms at the Intact, moderately (mDeg) and heavily (hDeg) degraded sites. Error bars are SE. Left charts displays means per month (n = 9 per palm status by spatial position); right charts display 3-year means (n = 316-320 per palm status by spatial position). In left charts, note different scale between top (Intact), bottom (hDeg) and middle (mDeg) panels. Letters indicate significant differences in means between microtopographies within a site and palm status. Numbers indicate significant differences in means between sites within a microtopography and palm status TA B L E 3 Relationships between monthly average N 2 O fluxes and environmental parameters at the Intact, moderately (mDeg) and heavily (hDeg) degraded sites according to palm status (live, cut) and microtopography (hummock, hollow). Site-specific relationships aggregated by palm status and/or microtopography are also presented is in kg C ha -1 day -1 .
the order hDeg > Intact > mDeg in year 3. Site-scale annual emissions averaged over the 3 years were lower at the mDeg site than at the other two sites. As mentioned above, these differences seemed driven by inherent WFPS variations at the sites rather than to degradation.

| Soil CH 4 fluxes
Mean monthly CH 4 emissions were higher in hummocks than in hollows at both degraded sites around live and cut palms, while it was the opposite at the Intact site (Figure 8 right charts). The emissions F I G U R E 7 Relationship between monthly average (a, b) or annual average (c) N 2 O emissions and the water-filled pore space (WFPS), the water table level (WT) and net nitrification rate. (a, b) present monthly emissions averaged within the respective WFPS (10% interval) and WT (10 cm interval) classes, disaggregated by site in left charts and all sites combined in right charts. In (c), annual averages were disaggregated by microtopography per site (excludes year 1 when mineral N data were not monitored). Black dots or bubbles and bars present average values and SE; dashed gray lines show the models. The models are presented with coefficients (SE) and their level of significance (*p < .05, **p < .01, ***p < .001). The size of the bubbles is relative to the sample size in (a) and (b); it is relative to net nitrification rate in (c) (range −0.3 to 1.8 mg N kg −1 day −1 )  Greek letters indicate significant differences in site-scale cumulative annual rate between years within a site. Number indicate significant differences between sites in site-scale cumulative annual rate. No letters or numbers are displayed in the absence of a significant difference.
were lower at the Intact site than at the degraded sites in hummocks supporting live palms. Conversely, emissions were higher at the Intact site than at the degraded sites in hollows surrounding live palms.
Measured flux values were 4% within the [−2; 2] g C ha −1 day −1 detection limit band. These low fluxes occurred when the WT was on average low. Monthly fluxes fluctuated over a wide range, from large uptake (−2,685 g C ha −1 day −1 ) to high emissions (6,021 g C ha −1 day −1 ; Site-scale annual CH 4 emissions along the degradation gradient ranged from 79.2 ± 5.5 to 319.4 ± 77.5 kg C ha −1 year −1 (Table 6) and were 1.5-4 times higher in the first year (El Niño) than in the third year (ENSO neutral; Figure 9d). They were higher at the Intact and mDeg sites than at the hDeg site in year 1, in the order Intact > hDeg > m Deg in year 2 and displayed no difference between sites in the last year. Over the 3 years, annual CH 4 emissions were on average similar among sites.

F I G U R E 8
Monthly mean CH 4 emissions in hummocks and hollows around live and cut palms at the Intact, moderately (mDeg) and heavily (hDeg) degraded sites. Error bars are SE. Left charts displays means per month (n = 9 per palm status by spatial position); right charts display 3-year means (n = 312-318 per palm status by spatial position). In left charts, letters indicate significant differences in means between microtopographies within a site and palm status. Numbers indicate significant differences in means between sites within a microtopography and palm status TA B L E 5 Relationships between monthly average CH 4 fluxes and environmental parameters at the Intact, moderately (mDeg) and heavily (hDeg) degraded sites according to palm status (live, cut) and microtopography (hummock, hollow)   Table 2) is typical of damp peat soils (e.g. Oktarita et al., 2017;Swails et al., 2017) and reflects an inhibition of nitrification due to limitation in O 2 supply. This inhibition is evidenced by the low net nitrification rates, especially at the Intact site. Nitrification can also be inhibited by polyphenolic compounds (Zeller et al., 2007) which have been found to be important in Southeast Asian peat swamp forests (Yule et al., 2016).
The lower NH 4 + content at the hDeg site compared to concentrations at the other sites may be due to enhanced immobilization "in situ". Cheng et al. (2014) found that litter decomposition in acidic forest soils could induce a shift from net N mineralization to net N immobilization. Therefore, the faster litter decomposition rate and lower litterfall C/N ratio at the hDeg than at the other sites (van Lent et al., 2019, in preparation) could partly explain differences in NH 4 + contents between sites. The higher "in vitro" net mineralization rate with increasing degradation was not linked to soil moisture and cannot be attributed to soil properties which were homogeneous among sites as mentioned previously. A potential explanation and avenue for further research would be an alteration of microbial communities prompted by changes in botanical composition associated with degradation (Girkin et al., 2019;Zhou et al., 2017).
The sites showed some differences in WT and soil moisture especially in hummocks which represent a small share (<15%) of site-scale surface. At this microtopography, the lower WT level at the Intact site than at the degraded sites and the lower WFPS at the Intact and highly degraded sites than at the site of medium degradation were consistent TA B L E 6 Cumulative annual and mean cumulative annual ± SE of CH 4 emissions (kg C ha −1 year −1 ) at the sites of the degradation gradient according to palm status (live, cut) and microtopography (hummock, hollow). Site-scale annual values were computed using relative proportions from Greek letters indicate significant differences in site-scale cumulative annual rate between years within a site. Numbers indicate significant differences between sites in site-scale cumulative annual rate. No letters or numbers are displayed in the absence of a significant difference. over time. Hollows displayed a similar WT level among sites (Figure 4 right charts) and a WFPS in the order mDeg > hDeg > Intact on average

| Spatiotemporal variations of soil N 2 O fluxes along the degradation gradient and their controls
The high spatiotemporal variability of N 2 O fluxes found along the degradation gradient concurs with results by, for example, Inubushi et al. (2003) or Melling et al. (2007) in tropical peatlands. Variability of fluxes has been attributed to several factors, but especially to variations in WT and soil inorganic N (Martikainen et al., 1993;Melling et al., 2007;Oktarita et al., 2017). Spatial variability at the microscale was characterized by a general trend toward higher N 2 O emissions from hummocks than from hollows ( Figure 6 left charts).
This trend is consistent with drier conditions (see WT in Figure 4), higher soil organic matter decomposition (van Lent et al., in preparation) and larger soil net N mineralization rates in hummocks than in hollows ( Table 2). The infrequent, irregular but very large N 2 O emissions (with maximum as high as 333 g N ha −1 day −1 ) and the presence of one hotspot at the sites could not be clearly associated with the environmental variables. Such high flux pulses have also been reported for other tropical peat forests (Jauhiainen et al., 2012;Takakai et al., 2006) and have been attributed to a series of plant and soil factors that govern oxygen diffusion and the fate of inorganic N (Oktarita et al., 2017). Most fluxes of N 2 O (84%) were within the detection limit band; as also reported by several studies conducted in undrained peatlands (Jordan et al., 2016;Wilson et al., 2013Wilson et al., , 2016. The [−6.8; 6.8] g N ha −1 day −1 detection limit band was much lower than the 11-88 g N ha −1 day −1 limit for the method implemented by Jordan et al. (2016) in Swedish peatlands and similar to that of 6.1 g N ha −1 day −1 for the research by Oktarita et al. (2017) in Indonesian peatlands.
Despite the large spatiotemporal variability described above, the ranges for the 3-year averages along the degradation gradient (1.3-7.4 g N ha −1 day −1 ) or for annual site-scale rates (0.5-2.6 kg N ha −1 year −1 ; Table 4) were rather narrow. The average rates are in the same order of magnitude as the annual means of 0.6 and 2.0 g N ha −1 day −1 reported by Jauhiainen et al. (2012) and Melling et al. (2007) for undrained peat swamp forests in Southeast Asia but 200 to 1,000-fold higher than the very low mean was frequent at all sites, particularly in hollows (Figure 6 left charts) and was registered over a wide range of soil moisture. N 2 O uptake has been detected in Southeast Asian peatlands (e.g., Jauhiainen et al., 2012;Takakai et al., 2006) and is usually favored by low nitrate availability, high WFPS and more generally conditions impeding N 2 O diffusion in the soil (Chapuis-Lardy et al., 2007). These conditions were fulfilled at the sites where the soil was often water-logged and nitrate-limited (Table 2).
N 2 O fluxes were on average lower at the mDeg site than at the other sites; however, this difference seems more closely related to spatial variation in soil moisture fluctuation along the forest complex than to degradation (Figure 7a). When WT levels were high in year 1 emissions were similar among sites (Table 4) (Table 3, Equations 7, 9 and 11-13). On the other hand, soil net mineralization rate which diminished with degradation was not related to fluxes at the degraded sites (Table 3) while together soil NO 3 − content it exerted a strong control on emissions at the Intact site. Emissions in hummocks and hollows increased as "in situ" NO 3 − contents got higher (Table 3, Equations 2-4) but decreased as "in vitro" net nitrification shifted from small negative to small positive rates (Table 3, Equation 6). This suggests that processes others than denitrification such as, for example, nitrifier denitrification (Wrage et al., 2001)  in the tropics and similar to the 50% soil moisture for the peak of the bell-shaped curve found by Pärn et al. (2018). This mid-point WFPS corresponds to optimal oxygen conditions for N 2 O production; while denitrification produces gradually more nitric oxide as the WFPS goes down and reduces N 2 O into N 2 as the WFPS goes up (Davidson et al., 2000). The logarithmic decrease in N 2 O emissions as the WT rose ( Figure 7b) is similar to the response found for peatlands in Southeast Asia (Hergoualc'h & Verchot, 2014) and denotes decreased oxygen supply for N 2 O production as sites get flooded.
The best fit biogeochemical model included WT level and net nitrification rate and explained 81% of annual flux variation over years and across sites and microtopographies (Figure 7c). This relationship coincides with the positive and negative correlations between N 2 O fluxes and, respectively, nitrification potential and WT found by Regina et al. (1996)

| Spatiotemporal variations of soil CH 4 emissions along the degradation gradient and their controls
Soil-atmosphere CH 4 exchange results from complex interactions between several processes controlling production, consumption, transport and release of the gas, and the dominance of any particular process at one site may not occur elsewhere, causing high spatial variability (Bartlett & Harris, 1993 (Bridgham et al., 2013).
Spatial variation was also exemplified by the presence of one hotspot in a hollow at the Intact site. The hotspot acted either as a strong source in the months following the El Niño-induced flooding when litterfall rates peaked (van Lent et al., in preparation) or as a strong sink. High CH 4 uptakes have also been reported in peatlands of Panama (up to 1,152 g C ha −1 day −1 , Wright et al., 2011). The [−2; 2] g C ha −1 day −1 detection limit band of our method was similar to limits reported by Bartlett et al. (1988;0.8 g C ha −1 day −1 ) for Amazonian floodplains and Smith and Lewis (1992; 1 g C ha −1 day −1 ) for temperate wetlands and in the low range of the 1.5-9 g C ha −1 day −1 limit evaluated by Christensen (1993) for Arctic tundra. Given the large magnitude of the fluxes at the sites, only 4% of them fell within the detection limit band.
Emissions of CH 4 were extremely variable over time (Figure 8 left charts). The 3-year average emission rates (from 357 to 870 g C ha −1 day −1 , Figure 8 right chart) are comparable to averages measured by Griffis et al. (2020) by eddy covariance at ecosystem scale at the Intact site (548-658 g C ha −1 day −1 ) and reported by Teh et al. (2017) for M. flexuosa palm swamp peatlands in the Peruvian Amazon (255-534 g C ha −1 day −1 ) but lower than values recorded in forested peatlands of Panama (384-3,024 g C ha −1 day −1 ; Wright et al., 2013). Mean annual site-scale emissions (161-226 kg C ha −1 year −1 , Table 6) are about seven times higher than the average reported for intact peat swamp forests in Southeast Asia (29 kg C ha −1 year −1 ; Hergoualc'h & Verchot, 2014). High CH 4 emissions in comparison with those of Southeast Asia could be associated with a higher quality of organic matter (Swails et al., 2017) containing lower levels of lignin (Hatano et al., 2016). They could also be related to higher WT at our sites ( Degradation did not affect site-scale annual emissions (Table 6) but seem to have impacted microscale fluxes (Figure 8 right charts). In hummocks, the lower average emission rate at the Intact than at the degraded sites could be related to differences in WT level among sites (Figure 4 right charts) though CH 4 -WT relationships at this spatial position explained only 36% of flux variation on average (Table 5, Equations 16, 17, 21, 24 and 25). As earlier mentioned, alteration of litterfall quality at the degraded sites could have stimulated hummock emissions; possibly explaining the strong CH 4 -NO 3 relationship at the hDeg site (Table 5,   Equation 23). In hollows where the WT was even among sites ( Figure 4 right charts), emissions were higher at the Intact site than at the degraded sites which could be linked to weakening of plant-mediated emissions following degradation. We also found strong relationships between air temperature, NH 4 + content, net mineralization rate and CH 4 fluxes at the mDeg site (  (Figure 9a). WT fluctuations influence aerobic and anaerobic decomposition by displacing, respectively, the oxic and anoxic layers where CH 4 is oxidized and produced. The depression of CH 4 oxidation in response to raised WT level has been documented in numerous studies (Bridgham et al., 2013). To a lower extent, monthly variations were also influenced by air temperature ( Figure 9b). Temperature is recognized as an important control over methanogenesis while it affects less methanotrophy (Le Mer & Roger, 2001). Annual average emissions over years and across sites and microtopographies increased when net nitrification rate went higher (Figure 9c). Increased mineral N content is generally believed to have the potential to enhance CH 4 emissions due to its inhibitory effect on methanotrophy; however, few studies have been conducted in natural wetlands (Bodelier & Laanbroek, 2004).
Using a long-term experiment in a boreal mire, Eriksson et al. (2010) demonstrated that N deposition simulated by adding ammonium nitrate caused increased CH 4 production but did not affect CH 4 oxidation. Finally, annual site-scale CH 4 emissions were positively related to precipitation (Figure 9d). This relationship suggests that climate change predictions of higher precipitations in the northwest Amazon (Malhi et al., 2008;Marengo et al., 2008) may imply higher CH 4 emissions from these palm swamp peatlands.

| CON CLUS ION
Tropical peatlands are of global and regional importance but remain critically understudied especially outside of Southeast Asia. Very little is known about their state of conservation and their contribution to mitigate and/or exacerbate climate change. Our study showed that degradation of palm swamp peatlands in the Peruvian Amazon modifies soil-atmospheric exchanges of N 2 O and CH 4 at the microscale but not significantly at the macroscale. Some of the main environmental drivers of emissions for both gases (soil WFPS, WT) were moisture-related suggesting that future changes in rainfall patterns in the region may substantially alter N 2 O and CH 4 emissions.
Our study also provided evidence of the significant magnitude of the emissions, particularly for CH 4 ; highlighting the critical need for their inclusion in biogeochemical modeling of peatlands. support. Finally, four referees and the editor of the journal did their utmost to improve this manuscript; their contribution is very much appreciated.