CH4 and N2O emissions from smallholder agricultural systems on tropical peatlands in Southeast Asia

Abstract There are limited data for greenhouse gas (GHG) emissions from smallholder agricultural systems in tropical peatlands, with data for non‐CO2 emissions from human‐influenced tropical peatlands particularly scarce. The aim of this study was to quantify soil CH4 and N2O fluxes from smallholder agricultural systems on tropical peatlands in Southeast Asia and assess their environmental controls. The study was carried out in four regions in Malaysia and Indonesia. CH4 and N2O fluxes and environmental parameters were measured in cropland, oil palm plantation, tree plantation and forest. Annual CH4 emissions (in kg CH4 ha−1 year−1) were: 70.7 ± 29.5, 2.1 ± 1.2, 2.1 ± 0.6 and 6.2 ± 1.9 at the forest, tree plantation, oil palm and cropland land‐use classes, respectively. Annual N2O emissions (in kg N2O ha−1 year−1) were: 6.5 ± 2.8, 3.2 ± 1.2, 21.9 ± 11.4 and 33.6 ± 7.3 in the same order as above, respectively. Annual CH4 emissions were strongly determined by water table depth (WTD) and increased exponentially when annual WTD was above −25 cm. In contrast, annual N2O emissions were strongly correlated with mean total dissolved nitrogen (TDN) in soil water, following a sigmoidal relationship, up to an apparent threshold of 10 mg N L−1 beyond which TDN seemingly ceased to be limiting for N2O production. The new emissions data for CH4 and N2O presented here should help to develop more robust country level ‘emission factors’ for the quantification of national GHG inventory reporting. The impact of TDN on N2O emissions suggests that soil nutrient status strongly impacts emissions, and therefore, policies which reduce N‐fertilisation inputs might contribute to emissions mitigation from agricultural peat landscapes. However, the most important policy intervention for reducing emissions is one that reduces the conversion of peat swamp forest to agriculture on peatlands in the first place.


| INTRODUC TI ON
Tropical peatlands in their natural state are highly effective longterm carbon (C) stores (Cooper et al., 2020;Deshmukh et al., 2020;Mishra et al., 2021;Prananto et al., 2020). However, when they are converted to other land uses such as agriculture or plantations, they become significant sources of greenhouse gases (GHGs), especially CO 2 and N 2 O (Page et al., 2022 and references therein). While CO 2 is the main GHG emitted from drained tropical peatlands (Cooper et al., 2020;Deshmukh et al., 2021) and has been studied more intensively, peat surface CH 4 and N 2 O emissions have received comparatively little attention. Soil GHG emissions are controlled by a range of factors, with the main ones being groundwater level, temperature and organic matter quality (Couwenberg et al., 2010;Evans et al., 2021;Leifeld et al., 2012;Mishra et al., 2021). As these factors are strongly modified by conversion of peat swamp forest (PSF) to plantations (such as oil palm or Acacia) or other forms of agriculture, such changes have important implications for net GHG emissions. For example, emissions from tropical peatlands converted to agriculture contribute substantially to national emissions of GHGs in Indonesia and Malaysia (Cooper et al., 2020;Miettinen et al., 2017). These increases in GHG emissions are caused by loss of peat-forming swamp forest cover, drainage for cultivation, and use of fertilisers, which in turn change the soil C compound composition and microbial soil communities (Oktarita et al., 2017;Prananto et al., 2020).
To date, research on the impacts of peatland drainage and conversion to plantations and agriculture have been carried out both in large-scale plantations and smallholder systems (Deshmukh et al., 2020;Matysek et al., 2018;McCalmont et al., 2021;Miettinen et al., 2016). However, more research on the complex small-scale agricultural production system on converted peatlands is needed as it is an important land use in the region (Hadi et al., 2005;Inubushi et al., 2003;Swails et al., 2021). Small-scale agricultural production is responsible for around 44% of the drainage and conversion of peatlands, occupying a similar area of peatland to that incorporated into large-scale production (Miettinen et al., 2016;Wijedasa et al., 2018). In Indonesia, oil palm plantations cover around 14.6 Mha with approximately 41% of the total area identified as small-scale plantations and 3-4 Mha established on peatlands (BPS, 2020). Due to global market demands and shortage of available mineral soil for farming, it is expected that the expansion of smallholder agriculture on peatlands will increase (Euler et al., 2016;Jelsma et al., 2017), and it may become a regionally important source of emissions (Jauhiainen et al., 2012).
However, small-scale production practices such as soil amendments and fertiliser applications are very diverse (Table S1), and they may differ from larger scale plantations in many critical respects (e.g. with regards to crop selection, soil management, nutrient application, drainage control and use of fire). A lack of research into smallholder production systems means that C losses and GHG emissions from such systems are poorly understood and emission factors (EFs, i.e. average emissions of each GHG associated with that land use; IPCC, 2006IPCC, , 2014 for these systems have not been quantified. Our lack of understanding of smallholder agricultural systems represents an important knowledge gap, and emissions data obtained from industrial plantations are unlikely to provide a reliable proxy given differences in management. In particular, there is a need for specific EFs for CH 4 and N 2 O that are both important GHGs in the context of conversion of tropical peat swamp forest to agriculture (Jauhiainen et al., 2012).
Ground water levels are a first order determinant of the functioning of peatlands; controlling GHG production and release (Couwenberg et al., 2010;Evans et al., 2021;Evers et al., 2017).
Under anaerobic conditions, methanogenic bacteria produce CH 4 . In contrast, CH 4 emissions from the peat surface are near-zero when groundwater levels are below −30 cm (Couwenberg et al., 2010;Evans et al., 2021). The effect of water table depth (WTD) on N 2 O emissions on tropical peatlands has, thus far, lacked extensive research attention. Combining data from a wide range of sites, Prananto et al. (2020) found higher N 2 O emissions from sites with deeper WTD. However, they did not disentangle the effects of drainage from fertiliser application practices. Indeed, drained agricultural sites are often managed more intensively and receive higher inputs of fertilisers, compared to undrained peatlands, confounding these two factors. Results from other locations and ecosystems have also shown contradictory results between soil moisture content and N 2 O emissions (Couwenberg et al., 2011;Pärn et al., 2018). This suggests that the groundwater level response is governed by site/system-specific factors, likely to be mainly determined by levels of N inputs, and that relationships between WTD and N 2 O emissions are not generic. In addition, degradation without drainage and fertiliser inputs also increases soil N 2 O emissions (Swails et al., 2021). The interaction of higher soil temperature, optimum soil moisture content and high mineral N concentrations, resulting from fertilisation and peat mineralisation, could convert tropical peatlands into hotspots of N 2 O emissions (Pärn et al., 2018).

| Study area and site selection
The study sites in Peninsular Malaysia and in Kalimantan (Indonesian Borneo) are in the humid tropical zone, characterised by heavy rainfall, high temperatures and relative humidity (Figure 1). The Malaysian sites experience two well-defined seasons, with peaks of rainfall in March-April and in October-November (Selangor State Forestry Department, 2014b). The climate at the Indonesian sites corresponds to an equatorial system (Aldrian & Dwi Susanto, 2003;Kuswanto et al., 2019). The wettest period typically extends from November to April with August being the driest month.
In Peninsular Malaysia, the study sites were located in two regions, North and South of Kuala Lumpur. The North Selangor region was located adjacent to the village of Raja Musa, within the North Selangor PSF, and the South Selangor region was located around the Kuala Langat peatlands complex, southeast of Banting.
The North Selangor PSF has experienced logging in the past but is now protected and is subject to rewetting and restoration (Brown et al., 2018;Cooper et al., 2020;Parish et al., 2014). In Kalimantan, the study sites investigated were in the West and Central Kalimantan regions. The study sites in West Kalimantan were located in Teluk Empening, Kubu Raya Regency. Meanwhile, the study sites in Central Kalimantan were located in Kalampangan and Hampangen villages, south and north of Palangka Raya, respectively. The forest sites in Teluk Empening are heavily degraded and have experienced recurring fires. In contrast, the forest sites around Palangka Raya have been rewetted as old unmaintained drainage canals have gradually filled with fallen branches and leaf litter, and this has reduced the water discharge from the sites. Now the forests are in good condition albeit recovering from historical logging. The forest sites have hence all experienced human impacts in some way and show signs of degradation. Their condition ranges from areas with higher water tables and higher biomass to areas with low water tables and lowstanding biomass. We subsequently refer to these forested sites at different stages of degradation as 'forest'.
In each region, three to four land-use classes were selected according to local conditions. Land-use classes fell into four broad classes as classified by IPCC (2014) (Table S2). Soils at the study sites were classified as deep ombrotrophic lowland peats. Cropland sites in Selangor had shallow peat due to recurring fires, peat subsidence and wastage since agricultural conversion. By contrast, the forest sites had deeper peat. Sites in West Kalimantan had a mix of shallow and deep peat while the Central Kalimantan sites had in general deep peat (Table S2). The organic horizon depth across sites varied between 0.18 and 5.50 m. Annual vegetable crops were grown on raised beds (15-20 cm above ground) of variable lengths and approximately 1 m wide. Tree plantations, which were only present at the Indonesian study sites, were planted at a spacing of approximately 5 × 5 m. Oil palm plantations, of different age classes, were planted in a triangular spacing with distances between palms ranging between 7 to 9.5 m. Farmers at the youngest oil palm sites established an in- Cropland sites had crops planted in rows and used parallel raised bed (or strip) systems to facilitate management of water level, whereas the oil palm and tree plantation sites were planted in grids, at a regular spacing in each case. Therefore, within each replicate site, two sampling plots, at two different locations (close to the crops/stems and between crops/stems), were used to conduct the gas flux measurements. At the cropland sites, the 'near the crop' plot was installed within the raised bed or strip cropping area and the other plot was installed within the unplanted furrow between the raised beds. At the oil palm and tree plantation land-use classes, the 'near the stem' plot was established at around 30-50 cm from tree/palm trunks, and the 'far from the stem' plot at a point equidistant between the trees/palms. Although the same approach was followed at the forest sites, the 'far from the stem' plot was only at around 1.5 m from the nearest tree due to the dense vegetation growing at the study sites. Each plot consisted of a measuring area of 40 × 40 cm in which the exact location for gas flux measurements was marked with four wooden sticks. All ground vegetation (e.g. non-vascular plants, grasses and tree seedlings) were carefully removed at the beginning of the experiments and on a monthly basis thereafter. Distance between the plots was around 2-3 m at the forest sites and around 1 m at the other sites. Next to each plot, a perforated PVC dip-well (2 m long) was inserted 1.5 m into the soil. A PVC cap was used to close the dip-well to prevent ingress of rainfall and other debris. Plots were established between March and May 2018. There were 96 sample plots in total (4 regions × 12 sites × 2 locations). During sampling, the chambers were carefully inserted into the surface of the peat to create a seal against the ground. After the sampling, the chamber was removed to avoid water logging of the surface of the sampling plots, and for the next sampling occasion the same sampling plots were used. The chamber headspace was 11.5 L and the inner diameter was 28 cm. The top of each chamber was perforated with a 15 mm drill bit and a 19 mm rubber Suba-Seal Septa was inserted through the hole. The joints around the septas were sealed with silicone to prevent gas leakage. Each measurement consisted of four 25 mL gas samples taken at 4 minute intervals through the septa using plastic syringes and hypodermic needles.

| Gas sampling and ancillary measurements
After each sampling event, 5 mL of gas was flushed out and the remaining 20 mL of sample was injected into 12 mL pre-evacuated glass vials (Exetainer®; Labco Ltd., UK). Samples were stored at room temperature for up to 6 months prior to analysis. Measurements at the paired plots (near vs. far locations) were conducted in parallel using two chambers, with 1 min delay between the sampling events.
Simultaneously with each gas sampling event, air temperature was measured on top of the chamber. Once the gas sampling was concluded, soil temperature was measured vertically at depths of 5 and 10 cm next to each plot. Additionally, WTD was measured inside the dip-wells using a laser distance measurer and a thin piece of polystyrene, attached to a nylon string, floating on the water surface inside Central Kalimantan (CK). Notice that three replicates per vegetation group were selected (see Table S1). flooding). Thereafter, around 150 mL of water was taken from each dip-well using a PVC bailer and a measuring jug. The collected water samples were combined for each site and stored in polypropylene bottles of 120 mL (Malaysian sites) and 250 mL (Indonesian sites) capacity. Equipment and bottles were gently rinsed with sample water in triplicate before being filled. Once in the laboratory, water samples were filtered through 0.45 μm cellulose nitrate membrane filters using a vacuum pump. Around 60 mL of filtered water was stored in polypropylene centrifuge tubes and stored in the dark at around 4°C until further analysis. Total dissolved nitrogen ( This was complementary to the vial sampling that was applied in parallel at the four different main study areas. For the Los Gatos measurements, the chamber characteristics were the same as the static chamber except that instead of having a rubber Suba-Seal Septa on top, this chamber was equipped with two push-fit connectors that acted as an inlet and outlet allowing the gas to flow from the chamber to the analyser and then back into the chamber.

Land-use class
After the chamber was placed on the measuring point, it was gently pushed into a 2 cm deep groove in the peat to create an airtight headspace between the ground surface and the chamber walls.
The soil CH 4 concentration (in ppm) increment inside the chamber was measured during 6-10 min, with data automatically recorded at every 20 s. In addition, real time measurements were streamed on a mobile phone using VNC Viewer and Los Gatos Wi-Fi capabilities.
If the CH 4 concentration increase over time was considered poor (e.g. plotted data with no apparent linear fit) or if pulses of CH 4 and / or a gas leakage were detected, the measurement was stopped (i.e. lifting the chamber from the soil) and a new measurement was taken after CH 4 readings on the mobile phone had stabilised (1 min approximately). As with the static chamber method, once measurements were concluded, soil temperatures at 5 and 10 cm depths and WTD were measured. Around 40% of the soil CH4 fluxes were measured with Los Gatos.
Soil CH 4 fluxes calculated using Los Gatos and GC analysis were not statistically different (p = .76; Figure S1) and therefore, fluxes from both methods were combined to calculate mean hourly and annual fluxes. Further investigation at the plot level showed that, at 3 out of the 84 sampling plots, the CH 4 fluxes calculated with Los Gatos and GC analysis were significantly different ( Figure S2).
However, these differences were not consistent across the three sampling plots indicating that no systematic error was linked to the analysis approach.

| Data analysis and calculation of emissions factors
Trace gas concentrations, determined by either GC analysis or by Los Gatos, were converted to mass units using the ideal gas law interval) were required to calculate the fluxes. If some flux measurements contained less data points due to system malfunctioning (e.g. flat battery and system overheating), the fluxes were still calculated but a warning message was generated and exported with the data.
In this instance, the flux calculation was checked manually to ensure erroneous data was not included in the statistical evaluation of the data. Positive fluxes indicated GHG emissions from the soil into the atmosphere and negative fluxes indicated soil uptake of atmospheric GHG.
(1) n = PV RT , where, n is the number moles of trace gas (mol L −1 ), P is the atmospheric pressure (Pa), V is the volume of trace gas per litre of air (L L −1 ), R is the gas law constant (8314.46 L·Pa·K −1 ·mol −1 ), and T is the temperature (K).

| Statistical analysis
The relationship between WTD and CH 4 and N 2 O fluxes was inves- were included. The same statistical non-parametric method was used to assess potential differences in CH 4 and N 2 O fluxes measured at the locations (e.g. far vs. near). All non-linear regression analyses were conducted using SigmaPlot 13 (systat Software Inc. USA) and figures were produced using R (R Core Team, 2013) and the package (Wickham, 2009). The non-parametric Wilcoxon test was performed using the R package ggpubr (Kassambara, 2021). All the statistical tests were realised at the p = .05 significance level.

| CH 4 and N 2 O fluxes and environmental factors
Mean soil temperature was lowest at the forest land-use class (26.5 ± 0.1°C) and highest at the cropland land-use class (28.7 ± 0.1°C). Temperatures at these two land-use classes were significantly lower and higher, respectively, than soil temperatures at the other land-use classes. By contrast, soil temperatures were similar (between 27.7 ± 0.1 and 27.1 ± 0.1°C) at the oil palm forest and tree plantation land-use classes (Table S3). Groundwater level had a clear temporal variation related to precipitation. Lowest and highest WTDs were recorded in August-September (usually the driest months) and November-January (usually the wettest months), respectively ( Figure 2). Across all study sites, mean WTD varied between −75 ± 6 cm (at the oil palm sites in Central Kalimantan) and −2 ± 2 cm (at the forest sites in Central Kalimantan) (Tables S3 and S4). Although mean WTD was deepest at the oil palm land-use class (−45 ± 4 cm) it was not significantly different from mean WTD at the tree plantation (−41 ± 6 cm) or cropland (−38 ± 4 cm) land-use classes. Mean WTD was shallowest at the forest land-use class (−25 ± 1 cm). Sites from the cropland land-use class had the highest mean TDN concentrations ( Figure 3). The pineapple sites had the greatest mean TDN (37.3 ± 18.0 mg L −1 ), followed by the turmeric sites (17.4 ± 8.0 mg L −1 ), in the North and South Selangor regions, respectively (Table S3). The lowest mean TDN concentration was measured at the oil palm sites in Central Kalimantan (1.1 ± 0.1 mg L −1 ), followed by the forest and jelutung sites in Central Kalimantan (both with 1.1 ± 0.1 mg L −1 ). On average, the cropland and forest land-use classes had the highest and the lowest mean TDN (7.5 ± 2.0 and 2.1 ± 0.1 mg L −1 ), respectively.
Soil CH 4 fluxes were generally low across all sites except at the forest sites in Central Kalimantan and North Selangor regions ( Figure 4). Mean hourly CH 4 fluxes varied between −0.05 and 3.60 mg m −2 h −1 across all study sites (Table S3). Overall, the forest land use class had the highest mean hourly CH 4 fluxes (0.78 ± 0.1 mg m −2 h −1 ). By contrast, the tree plantation and oil palm land-use classes, which had the deepest WTDs, had the lowest mean hourly CH 4 fluxes (both with 0.02 ± 0.01 mg m −2 h −1 ; Table 2; Tables S3 and S4).

F I G U R E 2
Seasonal variation of water table depth in the four land-use classes, grey-forest, blue-tree plantations, green-oil palm and orange and pink-cropland. The specific oil palm age, tree plantation species and crop types differed among regions so different plantation and crop types were measured in each location. Note that the forest condition differs substantially among the four regions. Each boxplot represents data from three replicates of each vegetation group.
Methane and N 2 O fluxes were strongly determined by WTD and vegetation group at a specific site ( Table 2). Highest CH 4 emissions were recorded at the wettest sites in the forest land-use class. In contrast, the lowest CH 4 fluxes were found at the driest sites in the oil palm and tree plantations. Fluxes of CH 4 followed a clear seasonal variation, related to WTD, with highest emissions measured during the wettest months (October to January with another peak in CH 4 emissions in May) (Figures 2 and 4). In addition, the combination of location (i.e. near vs. far from the crop/stem), WTD and vegetation group had a significant effect on CH 4 emissions (Table 2; Figure S3). N 2 O fluxes were lowest at the forest and tree plantation sites, and they were greatest at the oil palm and cropland sites ( Figure 5; Table 2) with mean hourly N 2 O fluxes varying between <0.001 ± 0.005 and 1.23 ± 0.54 mg m −2 h −1 across all sites (Table S3).
Monthly N 2 O fluxes at the cropland and oil palm land-use classes followed a seasonal variation, also closely related to WTD, across all regions except at the South Selangor sites where relatively high fluxes were recorded at all sampling events ( Figure 5). In addition, the forest site in West Kalimantan followed a similar seasonal pattern to the cropland site, which in West Kalimantan was ginger. The  Table 2; Table S4). This effect was more evident at the turmeric, water spinach and recently planted oil palm sites, with higher TDN concentrations (Figure 6e,f) and significantly higher N 2 O fluxes as well. However, responses show some variation among vegetation groups at specific sites over the duration of the experiment ( Figure S4).

F I G U R E 3
Seasonal variation of total dissolved nitrogen (TDN) in the four land-use classes, grey-forest, blue-tree plantations, green-oil palm and orange and pink-cropland. The specific oil palm age, tree plantation species and crop types differed among regions so different plantation and crop types were measured in each location. Note that the forest condition differs substantially among the four regions. Each boxplot represents data from three replicates of each vegetation group.

| Modelling of CH 4 and N 2 O fluxes
Based on the linear mixed model analysis, the effects of WTD and TDN on CH 4 and N 2 O fluxes were investigated through non-linear regression analysis.
Methane fluxes were strongly dependent on WTD (Figure 7a,b).
Water table depth explained 37% and 95% of the hourly and annual variation in CH 4 fluxes, respectively ( Table 3) Table 3). No other significant predictive models were found between CH 4 , N 2 O and the other studied variables.

| Environmental drivers of CH 4 and N 2 O emissions
Soil CH 4 emissions were strongly dependent on WTD (as also reported by Couwenberg et al., 2010;Deshmukh et al., 2020;Hergoualc'h et al., 2020;Prananto et al., 2020). Hourly CH 4 fluxes increased exponentially with WTD values higher than ≈ −30 cm. A slightly higher WTD threshold for CH 4 oxidation/production was found by Jauhiainen et al. (2008) at a PSF affected by drainage and at a deforested and burnt peatland site in Central Kalimantan (i.e. -22 and −12 cm, respectively). Sakabe et al. (2018) found that, in the same PSF in Central Kalimantan, this WTD threshold was −11 cm.
Furthermore, similar relationships between WTD and net ecosystem F I G U R E 4 Seasonal variation of hourly CH 4 fluxes in the four land-use classes, grey-forest, blue-tree plantations, green-oil palm and orange and pink-cropland. The specific oil palm age, tree plantation species and crop types differed among regions so different plantation and crop types were measured in each location. Note that the forest condition differs substantially among the four regions. Each boxplot represents data from three replicates of each vegetation group. Data from the static and dynamic chambers are combined. CH 4 exchange have been found in another recent study, where the production of CH 4 increased exponentially when WTD was less than ≈ −20 cm (Deshmukh et al., 2020). These results were supported in a meta-analysis of GHG fluxes from SE Asia, in which a WTD of −20 cm was identified as the threshold for peat soil CH 4 oxidation/production and deeper values of WTD were associated with very low CH 4 emissions (Couwenberg et al., 2010). Under flood conditions, methanogenesis dominates biological activity in the soil (see Bridgham et al., 2013 and references therein) and CH 4 production potential increases with the length of flooding and anaerobic conditions. The combination of WTD, duration of waterlogging events, together with the quantity and quality of available organic matter, may help explain why approximately 32% of the CH 4 fluxes used in the empirical modelling (Figure 7a) had WTD values shallower than −30 cm, but CH 4 fluxes were smaller than 200 μg m −2 h −1 . Flux measurements taken a few days after heavy rain events would have showed a temporarily high WTD but the short duration of these waterlogging events (as a consequence of the drainage systems present at most of the sites) may have prevented the proliferation of methanogens (Sakabe et al., 2018). These combined effects could also help explain why different studies have found different WTD thresholds for soil CH 4 oxidation/production in tropical peatlands. While WTD thresholds have formerly been suggested for specific land uses (Deshmukh et al., 2020;Jauhiainen et al., 2008;Sakabe et al., 2018), our study has incorporated a broader range of land-use classes and conditions. A comparable approach by Evans et al. (2021) based on annual CH 4 fluxes from multiple land-use classes on temperate peatlands reported a similar WTD threshold (around −30 cm) as the one found in our study.
We contend, therefore, that our model, which is based on 48 different sites from several land-use classes across Malaysia and Indonesia, may provide a robust basis for estimating peat CH 4 fluxes at both site and regional levels for similar land-use classes and peatland systems.
Although some incubation experiments have found that under anaerobic conditions, soil temperature has a significant and strong positive exponential effect on soil peat CH 4 emissions , no clear effect of soil temperature on CH 4 emissions has been reported under field experiments on tropical peatlands, probably due to the small variation of temperature in the tropics (Deshmukh et al., 2020;Jauhiainen et al., 2014;Luta et al., 2021). Similarly, we did not find a predictive relationship between in situ temperature and CH 4 emissions. One possible explanation for these discrepancies between incubation and field experiments could be the low variation in TA B L E 2 Results from the mixed linear models, with repeated measurements, explaining the variations in CH 4 and N 2 O fluxes. Fixed terms include vegetation (refers to the specific vegetation groups as described in Table 1), location (near the crop/stem vs. far from the crop/stem where the flux measurements were conducted), T10 (soil temperature at 10 cm depth), WTD (water table depth) and TDN (total dissolved nitrogen). Significant (p < .05) and near significant (p < .1) effects are shown. The full output table describing also the nonsignificant interactions from the mixed model including 'vegetation group' as a fixed effect in the analysis are shown in Table S4.  (Deshmukh et al., 2020). For the cropland and young oil palm sites, the temperature range was greater but the deeper WTDs, limited duration of waterlogging events and reduced inputs of fresh litterfall and root exudates resulted in low CH 4 production (Girkin et al., 2018(Girkin et al., , 2020Guillaume et al., 2016;Pulunggono et al., 2019;Van Noordwijk et al., 1997).
Soil N 2 O emissions showed large spatial and temporal variability across all sites but these were in the range reported previously in the region (Hergoualc'h et al., 2020;Oktarita et al., 2017;Swails et al., 2021). Most of this variation was explained by the concentration of TDN in water samples, together with WTD. No data on the amounts of N fertiliser applied by farmers or the N content of the peat soils were obtained. However, TDN is a good indicator of available nitrogen (Hu et al., 2013;Wang et al., 2021) and it could be considered a proxy for the sum of N inputs from fertiliser application and peat mineralisation. Given the much higher TDN concentrations at cropland sites versus drained forest sites with similar WTDs, we infer that differences in TDN were mainly determined by fertiliser application rate. The high TDN concentration found for many of the sampling events at the cropland and oil palm sites suggests local soil treatment practices and in particular, fertiliser application rates with a level of N that exceeds crop demand (Kennedy et al., 2020), are consistent with smallholder agricultural practices reported in Southeast Asia (and elsewhere in Asia) (Poudel et al., 1998;Qiao & Huang, 2021;Zikria & Damayanti, 2019). It is likely that the excess N was available for denitrifying bacteria, a process that produces N 2 O if the denitrification is incomplete (Too et al., 2021;Wolf & Russow, 2000;Xu et al., 2016) (Pärn et al., 2018;Rubol et al., 2012), in line with the high N 2 O F I G U R E 5 Seasonal variation of hourly N 2 O fluxes in the four land-use classes, grey-forest, blue-tree plantations, green-oil palm and orange and pink-cropland. The specific oil palm age, tree plantation species and crop types differed among regions so different plantation and crop types were measured in each location. Note that the forest condition differs substantially among the four regions. Each boxplot represents data from three replicates of each vegetation group.   (Equation 4). Models were developed using data from all land-use classes and study sites. Average annual CH 4 emissions from the forest in Central Kalimantan (i.e. 121 ± 98 kg CH 4 ha −1 year −1 ) were at the higher end of previously reported CH 4 fluxes from PSF in Kalimantan (between −4 ± 1 and 89 ± 120 kg CH 4 ha −1 year −1 ) (Hirano et al., 2009;Ishikura et al., 2018;Jauhiainen et al., 2008;Sakabe et al., 2018). Our calculated emissions are based on daytime measurements. However, diurnal fluctuation of CH 4 fluxes (with higher emissions during the day than during the night) have been identified in continuous eddy covariance flux measurements in a PSF in Indonesia (91 ± 9 kg CH 4 ha −1 year −1 ) and Acacia plantation (47 ± 15 kg CH 4 ha −1 year −1 ) (Deshmukh et al., 2020). Although our daytime measurement may overestimate emissions, Deshmukh et al. (2020) reported that this diurnal effect was not significant in a drained Acacia plantation and therefore, we suggest that this would also be the case at the drained tree plantation, oil palm and cropland land uses in our study. Also, the same authors attributed the high CH 4 emissions at the Acacia plantations to fluxes from the water surfaces of nearby ditches and canals.

| Effect of land use on annual CH 4 and N 2 O emissions
During the wet season, CH 4 emissions from stem fluxes and pneumatophores of some tree species can represent a large fraction of the net ecosystem CH 4 emissions (Pangala et al., 2013(Pangala et al., , 2017Sjögersten et al., 2020). However, our annual CH 4 fluxes only included peat surface CH 4 fluxes and therefore, it is possible that the annual CH 4 fluxes at the forest and tree plantation sites may underestimate total net emissions. Methane production in wetlands is limited by labile organic substrates which are precursors of the substrates needed by methanogens (Bridgham et al., 2013;King et al., 2002;Whiting & Chanton, 1993). Therefore, the regular addition of labile C, nitrogen and phosphorus into the soil from fresh litter and root exudates from the forest vegetation would have also led to higher CH 4 fluxes by increasing microbial activity and decomposition of recalcitrant organic matter (Girkin et al., 2018;Hoyos-Santillan et al., 2016;Jauhiainen et al., 2014;Könönen et al., 2016;Sakabe et al., 2018).
In contrast, degraded peatlands, with recalcitrant organic matter as a consequence of land-use conversion (e.g. from degraded forests to cropland), with much drier conditions, changes in vegetation and amount and quality of organic inputs into the soil, can explain the low CH 4 emissions measured at the other land-use classes. This study presents CH 4 and N 2 O fluxes that would be useful for updating Tier 1 EFs for cropland land use in tropical peatlands.
In addition, the development of Tier 2 EFs for this land-use class in Malaysia and Indonesia would improve the accuracy of countrywide GHG emissions and inventories. Going forward, one of the main challenges of developing N 2 O EFs is to accurately determine the nitrogen inputs from fertilisers in order to separate soil N 2 O emissions caused by peat oxidation from those caused by applied fertilisers (note that these fluxes are reported separately according to IPCC methodology (IPCC, 2006(IPCC, , 2014). However, the strong dependence of N 2 O emissions on local agricultural management offers the prospect that a substantial proportion of emissions could be mitigated by improved agronomic and water management. Given that over-fertilisation of smallholder farming systems represents a significant cost to the farmers and the environment for little or no additional yield benefit (Good & Beatty, 2011;Hendricks et al., 2019;Zhang et al., 2018) and that over-drainage leads to accelerated soil loss (Evans et al., 2019) and potentially also to reduced crop yields, improved management of smallholder agricultural systems offers the potential both to mitigate GHG emissions and to improve farm incomes and livelihoods.

ACK N O WLE D G E M ENTS
We acknowledge research support from the United Kingdom Research