Data were collected from Sabaju (latitude 003° 12′ N, longitude 113° 30′ E) and Sebungan (latitude 003° 09′ N, longitude 113° 21′ E) oil palm plantations in Sarawak, Malaysia. Sabaju and Sebungan Estates have been established on peat soils broadly classified as histosols (FAO, 2006). The peat depth was measured to be 3.0 m at Sabaju and 4.0 m at Sebungan. The mean annual precipitation was ~3,200 mm (MET Malaysia, 2017). The northeast monsoon from October to January has the most rainfall, with a slightly drier southwest monsoon between May and August (MET Malaysia, 2017). The two principal dry seasons fall between February to April and in September. The mean annual temperature was 26.5°C (MET Malaysia, 2017).

Prior to planting, the land use was a mixed species swamp forest, which had been heavily logged. The land was converted to a plantation in 2006 and the palms were on their first crop rotation. The palms were 8 years old when measurements began. The plantations were laid out systematically with ~35 ha blocks and drainage ditches every 28 m leading to a larger ditch running down the center of the block. Palms were planted every 8 m in rows that were 8 m apart, with a planting density of 160 palms per ha. Within the palm blocks, four different surface management microforms were present and two different drain types (Figure 1):

Six one hectare plots were established, three in Sabaju and three in Sebungan. Within each plot, three subplots were randomly placed. In each subplot, soil surface collars made from PVC plastic of dimensions 10 cm deep by 10.5 cm diameter were installed to 5 cm depth to measure from the harvest path, underneath a frond pile, in the cover plants and adjacent to three palms in order to take CO₂ and CH₄ measurements. Collars were installed 6 months prior to the commencement of sampling in order to avoid disturbance effects associated with chamber base installation (Varner et al., 2003).

A location was selected to sample the drain fluxes from both the field drain and the collection drain within each subplot. Soil surface collars of 10 cm depth and 10.5 cm diameter were installed into the base of each ditch to 5 cm, to prepare for sampling months when the drains were dry. Floating chambers were made for sampling when there was water in the drains (Figure 2A; Kent, 2018). Here two holes were drilled into plastic mixing bowls (30 cm rim diameter) and lengths of Tygon tubing (2.5 m long, 3 mm inner diameter) were attached to each hole using Swagelok fittings. Aluminum foil was taped over the bowls to prevent light from penetrating and foam cylinders were attached to the bowls to help them float.

Stem fluxes were only taken from one plot in Sabaju and one plot in Sebungan, due to time and resource limitations, adapting the chamber methodology used by Pangala et al. (2013) and Siegenthaler et al. (2016) to suit oil palms (Figure 2B; Figure S1; Manning, 2019). Here, five palms were selected from each plot, including three that had soil collars next to them. Permanent chamber bases were glued onto the palm surfaces—these were made out of 3 cm wide and 6 cm thick strips of neoprene, attached at 0.2 m and 0.7 m height. Expandable polyurethane (PU) foam was applied to the stem, filling any gaps between the neoprene strips and the palm surface that had been created by the uneven distribution of frond bases. Extra neoprene blocks were randomly placed inside the chamber base area to help maintain an even volume inside the chamber. The chamber itself consisted of a 2.5 m × 0.75 m × 0.003 m sheet of plastic, which was wrapped around the chamber base and fixed in place using two ratchet straps to attach the sheet to the neoprene strips, with a third ratchet strap around the middle of the chamber to keep the sheet closed. Prior to attachment, six fans (12 VDC 120 mm computer case fans; flow ~200 CFM) were distributed within the chamber base. After attaching the chamber, the chamber was wrapped first in plastic wrap to make it gas tight, and then in black plastic sheeting to block sunlight and prevent any photosynthesis. Furthermore, the palm stems were regularly cleaned of vegetation at least one week prior to flux measurement.

The soil chamber volume was determined by multiplying the height of the chamber (15 cm) with the surface area (radius 5.5 cm). Prior to each measurement, the depth of the collar was measured and included in the volume calculations. The floating drain chamber volume was 7 L. The volume inside the drain chamber did not change when the chamber was placed on the water.

The stem chamber volume needed to be calculated for each measurement due to the uneven surface of the palms and the chamber being flexible rather than rigid. Equation (1) was used to determine the stem chamber volume:

where V is the volume of the chamber, d is the mean depth of the chamber base (i.e., radially outwards from palm), h is the internal (vertical) height of each chamber, R is the radius of the outer chamber surface (i.e., that of the plastic sheet), and r is the radius of the inner chamber surface (i.e., the radius of the stem). Multiple measurements were made of d, h, R, and r for each palm and the mean values were used in this equation. More information can be found in Manning (2019).

Flux measurements were made using the static chamber approach (Livingston and Hutchinson, 1995). Soil and drain CO₂ and CH₄ fluxes were collected between June 2015 and May 2016 and stem fluxes were collected from May 2016 to May 2017. The fluxes were measured at different times due to limitations in resource availabilities. Soil, drain and stem fluxes were measured at the same time in May 2016 to ensure comparability. Furthermore, the different instruments were calibrated against each other to ensure compatibility (Manning, 2019).

R_tot measurements were made using an EGM-4 and SRC-1 chamber (PP Systems; Hansatech Instruments, Amesbury, USA). Two recordings were made in duplicate (one immediately after the other) and averaged during data processing because the EGM did not always save one of the recordings. For each replicate measurement, CO₂ concentrations were measured over a 2 min enclosure period, with concentrations recorded at 3 s intervals, or until an increase of 100 ppm CO₂ had been observed. Measurements were then made monthly for 12 months (per month n = 54 for by palm fluxes and n = 18 for each of the harvest path, frond pile and cover plants fluxes). Extra fluxes were measured next to the palm because of high variability in both R_tot, root biomass and R_a, and the desire to accurately capture plantation R_tot fluxes (Manning, 2019; Manning et al., in preparation).

Soil CH₄, drain CO₂, and drain CH₄ measurements were made using syringe sampling (soil CO₂ data were also obtained using this method but the EGM data are presented due to a larger dataset). Static chambers were placed on the soil collars and floating chambers were deployed on drains when there was water present, otherwise static chambers were used on the drain surface. These chambers were left in place for 40 min, and 30 ml samples were collected with syringes every 0, 10, 20, 30, and 40 min. These samples were transferred into 20 ml pre-evacuated vials, which had been sealed with chlorobutyl rubber septa and aluminum seals (Sigma-Aldrich, Dorset, UK), and the over-pressurized vials were shipped to the UK for analysis. Measurements were made in August 2016, September 2016, November 2016, January 2017, March 2017, and May 2017 providing per month n = 18 for each location. The air samples were sent to St. Andrews University for analysis where samples were manually injected into a Thermo TraceGC Ultra Gas Chromatograph, which was fitted with a FID and methanizer for determination of CO₂ (Thermo Electro Corporation). August 2016 samples were analyzed at the University of Aberdeen on an Agilent 6890 Gas Chromatograph Analyser, fitted with an FID (Agilent). The two different analysers were cross-calibrated with three different BOC standards, which comprised of mixes of CO₂ and CH₄ in different concentrations (Manning, 2019).

Stem CO₂ and CH₄ fluxes were measured using a Los Gatos UGGA (Ultraportable Greenhouse Gas Analyzer; Los Gatos Research Inc, San Jose, California, USA). This UGGA was attached to the chamber by 2 m long tubing and data were continuously measured for 10 min at a frequency of 1 Hz. Data were stored on the field laptop and the dry air ppm were reported.

Ancillary measurements were made during and just after the completion of flux measurements. Ambient air temperature was measured at the same time as all flux measurements, using a thermometer (LCD Digital Thermometer, ATP Instrumentation, Leicestershire, UK; precision: ±1°C). Soil temperature and soil moisture measurements at 0–10 cm depth were taken following the completion of soil CO₂ and CH₄ sampling, adjacent to the soil collars, with soil moisture measured using an ML3-probe and HH2 moisture meter (Delta-T, Cambridge, UK; precision for soil moisture: ±1%). Soil moisture measurements at 0–10, 10–20, 20–30, and 30–40 cm intervals were made next to a subset of soil collars following each measurement, namely at one subplot in each plot, using a PR2-probe and HH2 moisture meter (Delta-T, Cambridge, UK; precision: ±4%). If soil collars were being used to measure drain fluxes, soil temperature and soil moisture at 0–10 cm were also sampled from the peat at the base of the drain following the completion of the flux measurement. When floating chambers were being used to measure drain fluxes, water temperature was measured following the completion of the flux measurement, using the same thermometer as used for air and soil temperature. Stem temperature was measured using a Fluke 62 Max Infrared Thermometer (MEA, Dubai, United Arab Emirates; precision: ±2°C). Throughout the study climate data were collected from two weather stations (Davis Vantage Pro2 Plus, Hayward, California, USA), one on each plantation. Measurements from these weather stations include precipitation, air temperature, UV index, and air humidity (precisions: ±4%, ±0.3°C, ±5 and ±2%, respectively).

Water table depth (WTD) and the depth of the water in the drains (drain water depth) were both measured after the flux measurements were completed. To prepare for WTD measurements from the soil, PVC pipes of 3 m length and 5 cm diameter had holes drilled into them at 10 cm intervals along the pipe, in all four directions. These pipes were then installed into the peat to depths of 2.5 m 1 month prior to the commencement of flux measurement. WTD was then measured by inserting a 2.5 cm diameter pipe with a measuring tape attached to it into the PVC pipe. Drain water depth was measured using a pipe with a measuring tape attached to it, which was lowered into the drain until it reached the bottom of the drain. If the drain was dry when CO₂ and CH₄ fluxes had been measured from the drain surface, the WTD was measured by digging a hole 0.5 m away from the flux collar from the drain surface to the water table and measuring this distance.

Fluxes were calculated using the HMR package in R version 2.15.1 (http://www.R-project.org; Pedersen et al., 2010). Here linear and non-linear regressions were applied to the fluxes in accordance with the HMR methodology, whereby both options were calculated and the fit with the lowest standard error was used. All of the data were included for the flux calculations when the data were collected by the EGM. The vial CO₂ data were first plotted in conjunction with CH₄ to determine whether any of the vials had leaked, before the HMR function was applied, with the leaked vials being excluded from the analysis. The units of the HMR outputs are μL m⁻² s⁻¹. These units were converted to Mg CO₂-C ha⁻¹ yr⁻¹ or kg CH₄-C ha⁻¹ yr⁻¹ using Equation (2):

where F is the C flux, Δ_C is the change in CO₂ or CH₄ concentration over the measurements period (ppm), Δ_t is the duration of the measurement period (s), P is pressure (mb), T is temperature (K), V is volume (m³), A is surface area of the soil (m²), R is the Universal Gas Constant 8.31432 J mol⁻¹ K⁻¹, M is the relative molecular mass of CO₂ or CH₄ and Y is the conversion to upscale the flux to annual emissions.

Annual flux estimates were upscaled using spatial and temporal weighting. Fluxes from the soil and drain microforms were spatially weighted by multiplying the mean CO₂ or CH₄ flux by the proportional area of the microform (Tables S1, S2). Stem fluxes were first upscaled to the plot level fluxes (F′), expressed as Mg CO₂-C ha⁻¹ or kg CH₄-C ha⁻¹ soil—first the flux (F) was multiplied by the surface area (A) of each palm to calculate the mean stem flux (A⁻¹) and then A⁻¹ was then multiplied by the planting density (166 palm ha⁻¹) to get the per hectare stem carbon flux. Second, each spatially weighted flux (F′) was linearly interpolated between months using Equation (3):

where t₁ and t₂ are the timings of the measurements in days. These temporally weighted fluxes F^* were subsequently summed over one year to produce annual flux estimates (F_tot). The microform annual estimates were summed to provide component and plantation flux estimates.

The rate of peat oxidation (i.e., R_h) was calculated by sampling root-free soil 4 m away from the palms, based on prior work by Manning et al. (in preparation), where roots and R_tot were measured in transects at 0, 0.25, 0.5, 0.75, 1, 1.5, 2, 2.5, 3, 3.5, and 4 m distance from the palm, going into the harvest path, frond piles and cover plants, to 0.3 m depth. Root biomass was found to be negligible further than 1.0 m away from the palm and R_tot did not vary with increasing distance more than 0.75 m distance from the palm (Manning et al., in preparation). Therefore, at this site the contribution of R_a to R_tot was considered to be negligible more than 1 m from the palm and sites were chosen for R_h measurements exactly 4 m away from the palms (halfway between two palms).

A range of R_h values were proposed, assuming: (1) the harvest path only provided true R_h estimates, (2) the harvest path and frond pile gave R_h estimates, and (3) assuming that surface R_tot fluxes measured from the harvest path, frond pile and cover plants only contributed to R_h. We estimated R_h using this range of methods because peat oxidation occurs at all of these different microforms and potentially at different rates due to variations between the microclimates. These estimates of R_h were extrapolated to the plantation level by area-weighting each flux by the proportional surface area of its microform. For methods (1) and (2) the flux measured from the harvest path was used to represent R_h from the microforms that were not included in the specific R_h calculation, when spatially upscaling. In all three methods the harvest path flux was also assumed to represent the rate of R_h for the area next to the palm.

All statistical analyses were done in R (version 3.5.1). Kruskal-Wallis tests were used to test whether there were significant variations in the CO₂ and CH₄ fluxes between the two different plantations, the three plots within each plantation, the three subplots within each plot and the different surface microforms. Kruskal-Wallis tests were used to test for significant variation in the CO₂ and CH₄ fluxes between the months. Post-hoc multiple comparisons tests were performed using the R package pgirmess (Giraudoux et al., 2018).

CO₂ and CH₄ measurements were modeled as a function of environmental variables using Gaussian linear mixed effect models from the nlme package in R (Pinheiro et al., 2017). Log or square root transformations of CO₂ and CH₄ measurements were sometimes necessary to normalize model residuals. The random effect structure and correlation structure, the latter used to take into account seasonality, were determined by comparing model residual plots. Fixed effect variables were chosen by top down selection, with the non-significant variables removed one by one from models using the ANOVA function (see Data Sheets 1–5, for the selection processes for the fixed effects in each model). The following five models were used for this study, with ^* used to denote significant interactions between variables:

Precipitation, relative humidity and air temperature were measured from weather stations in Sabaju (November 2015–March 2017) and Sebungan (June 2015–March 2017). Annual means for climate variables can be found in Table 1. Annual precipitation was recorded at 2,015 mm in Sabaju, using an average of moving annual sums of the data available. Precipitation records from Sebungan were unreliable due to equipment failure. The dry season consisted of February, March and May—all of which recorded <100 mm rain in Sabaju. Collectively, the rainy season had significantly more rainfall than the dry season, averaging 216 ± 25 mm/month in the rainy season and 90 ± 46 mm/month in the dry season (Kruskal-Wallis: chi-squared = 4.5; d.f. = 1; p = 0.03). Relative humidity was significantly higher in Sabaju than in Sebungan in both measurement periods (2015–2016: Kruskal-Wallis: chi-squared = 6.2; d.f. = 1; p = 0.01; 2016–2017: Kruskal-Wallis: chi-squared = 12.8; d.f. = 1; p < 0.0001). Using averages taken between November 2015 and January 2017, where there were complete records of relative humidity from both plantations, relative humidity was significantly higher in the rainy season (88.0 ± 0.5%) than in the dry season (87.3 ± 1.1%; Kruskal-Wallis: chi-squared = 6.5; df = 1; p = 0.01). Annual mean air temperatures measured from the weather stations were the same for both plantations, in both measurement periods. Air temperature measured at the time and location of the soil flux sampling was significantly higher in Sabaju than in Sebungan (Sabaju: 30.7 ± 0.1°C; Sebungan: 29.2 ± 0.1°C; Kruskal-Wallis: chi-squared = 109.7; d.f. = 1; p < 0.0001). Air temperature measured at the same time as the stem fluxes was not significantly different between the two plantations. Air temperature did not vary significantly between the rainy and dry seasons.

WTD varied between the two plantations during 2015–2016 but not during 2016–2017 (Table 1; Table S3). During 2015–2016, when the soil and drain measurements were made, mean WTD was −0.49 ± 0.0041 m in Sabaju and −0.76 ± 0.0072 m in Sebungan, which were significantly different (Kruskal-Wallis: chi-squared = 789.2; d.f. = 1; p < 0.0001). During 2016–2017, when the stem measurements were made, monthly mean WTD was −0.46 ± 0.16 m for Sabaju and −0.43 ± 0.16 m for Sebungan, which were not significantly different. Seasonal variation was observed in WTD; WTD was significantly higher in the dry season than in the rainy season in both measurement periods (2015–2016: dry season: −0.72 ± 0.00048 m; rainy season: −0.61 ± 0.00046 m; Kruskal-Wallis: chi-squared = 50.11; d.f. = 1; p < 0.001; 2016–2017: dry season: −0.49 ± 0.02 m; rainy season: −0.42 ± 0.02 m; Kruskal-Wallis: chi-squared = 5.68; d.f. = 1; p = 0.017).

In addition to climate and WTD measurements, different environmental measurements were made at the time of flux sampling, with annual averages reported in Table 1. Of particular interest are soil temperature and soil moisture, which varied significantly between the two plantations and between the different surface microforms. Soil temperature was significantly higher in Sabaju than in Sebungan (Kruskal-Wallis: chi-squared = 185.9; d.f. = 1; p < 0.0001) and varied significantly between the different surface microforms (Kruskal-Wallis: chi-squared = 253.7; d.f. = 3; p < 0.0001). Soil temperature was highest next to the palm and lowest beneath the frond piles, decreasing in order by palm > harvest path > cover plants > frond pile. Soil temperature was also significantly higher in the dry season (29.4 ± 0.05°C) than the rainy season (28.1 ± 0.05°C; Kruskal-Wallis: chi-squared = 29.2; d.f. = 1; p < 0.001). This seasonality was measured in both plantations (Sabaju Kruskal-Wallis: chi-squared = 10.60; d.f. = 1; p = 0.0011; Sebungan Kruskal-Wallis: chi-squared = 30.31; d.f. = 1; p < 0.001).

Soil moisture was significantly greater in Sebungan than in Sabaju (Kruskal-Wallis: chi-squared = 15.8; d.f. = 1; p < 0.0001). Soil moisture varied significantly between the different surface microforms (Kruskal-Wallis: chi-squared = 840.0; d.f. = 3; p < 0.0001). Soil moisture was consistently drier next to the palm than in the other microforms, decreasing in order harvest path > frond pile > cover plants > by palm, although soil moisture was highest in the frond pile in Sabaju and the harvest path in Sebungan. Soil moisture was significantly greater in the rainy season than in the dry season, with a mean of 39.4 ± 1.1% in the rainy season and 32.4 ± 1.0% in the dry season (Kruskal-Wallis: chi-squared = 23.8; d.f. = 1; p < 0.001). This relationship was seen in the different microforms, at the different plantations and in the different microforms at each plantation.

The CO₂ fluxes are summarized in Table 2. Mean R_tot was recorded as 0.83 ± 0.048 g CO₂-C m⁻² hr⁻¹. Drain CO₂ averaged 0.14 ± 0.0012 g CO₂-C m⁻² hr⁻¹. Mean stem respiration was 0.23 g ± 0.012 CO₂-C m⁻² hr⁻¹. There was no significant difference in the overall mean CO₂ flux between the plantations.

Significant differences were seen in the magnitude of the CO₂ fluxes between the different surface microforms (Figure 3). Considering the plantations together, R_tot measured next to the palm gave the highest CO₂ flux from the plantation (Kruskal-Wallis: chi-squared = 1,555.5; d.f. = 6; p < 0.0001). Within each plantation, we found differences in the magnitude of CO₂ flux from each microform (Sabaju: Kruskal-Wallis: chi-squared = 572.6; d.f. = 6; p < 0.0001; Sebungan: Kruskal-Wallis: chi-squared = 1,078.6; d.f. = 6; p < 0.0001). For example, in Sabaju, the harvest path R_tot was significantly higher than R_tot measured from the frond piles and cover plants—although they themselves did not differ (multiple comparison test for Kruskal-Wallis: p ≤ 0.05). Drain fluxes were lower than soil fluxes but not significantly different to stem CO₂ (multiple comparison test for Kruskal-Wallis: p ≤ 0.05). Stem CO₂ fluxes were significantly lower than R_tot from next to the palm, the harvest path and beneath the frond piles, but not the cover plants (multiple comparison test for Kruskal-Wallis: p ≤ 0.05). In contrast, in Sebungan, frond pile R_tot was significantly higher than the field drain CO₂ flux, but otherwise the soil, drain and stem fluxes did not differ significantly, apart from the high R_tot fluxes measured next to the palms (multiple comparison test for Kruskal-Wallis: p ≤ 0.05).

Summaries of the CH₄ fluxes are presented in Table 2. The mean soil CH₄ flux was 0.29 ± 0.047 mg CH₄-C m⁻² hr⁻¹. The mean drain CH₄ flux was 4.24 ± 1.64 mg CH₄-C m⁻² hr⁻¹. The stem CH₄ flux averaged 0.043 ± 0.0056 mg CH₄-C m⁻² hr⁻¹. We did not observe a significant difference in the overall CH₄ flux when comparing all the CH₄ fluxes from the two plantations.

CH₄ fluxes varied significantly among the different surface microforms (Figure 3; Kruskal-Wallis: chi-squared = 258.79; d.f. = 6; p < 0.0001). Multiple comparisons tests indicate that the CH₄ fluxes from different microforms fell into one of two groups; the first (lower emission) group consisted of the harvest path, frond pile, cover plants and stem fluxes, while the second (higher emission) group consisted of the by palm, collection drain and field drain fluxes (multiple comparison test for Kruskal-Wallis: p ≤ 0.05). This trend was the same in both plantations.

Seasonal trends were observed in trace gas fluxes between the rainy and dry seasons. CO₂ fluxes varied between the rainy and dry seasons, but this trend was not statistically significant. The rainy season flux averaged 0.74 ± 0.054 g CO₂-C m⁻² hr⁻¹ and the dry season flux averaged 0.94 ± 0.081 g CO₂-C m⁻² hr⁻¹.

Disaggregating the data by microform showed that each microform had different trends with respect to CO₂ fluxes (Figure 4). R_tot measured next to the palm was significantly higher in the dry season than in the rainy season (Kruskal-Wallis: chi-squared = 25.6; d.f. = 11; p < 0.0001). R_tot measured from the harvest path, frond piles and cover plants did not vary significantly between the rainy and dry seasons. Significantly more CO₂ was measured from the surfaces of the drainage ditches in the rainy season than in the dry season (Kruskal-Wallis: chi-squared = 11.54; d.f. = 1; p < 0.0001). Splitting the drains, the larger collection drains had significant seasonal variation in CO₂ fluxes (Kruskal-Wallis: chi-squared = 8.1; d.f. = 1; p = 0.004), with higher CO₂ fluxes in the rainy season than the dry season (multiple comparison test for Kruskal-Wallis: p ≤ 0.05) and the smaller field drains did not. Stem CO₂ did not show significant variation between the rainy and dry seasons.

Collectively, CH₄ fluxes did not vary significantly between rainy and dry seasons, although the rainy season flux averaged 0.76 ± 0.17 mg CH₄-C m⁻² hr⁻¹ and the dry season flux averaged 2.27 ± 1.14 mg CH₄-C m⁻² hr⁻¹. Soil and stem CH₄ fluxes did not show significant variation between the rainy and dry seasons at the different plantations. Drain CH₄ fluxes were significantly higher in the rainy season than in the dry season (Figure 4; Kruskal-Wallis: chi-squared = 4.33; d.f. = 1; p = 0.04). Considering the plantations separately, there was no significant difference between the rainy and dry seasons in the different microforms or locations.

Annual estimates of CO₂ fluxes were produced for the plantations both together and separately (Table 3). Combining R_tot and drain CO₂ fluxes to give the total CO₂ flux from the ground surfaces gave an average of 23.57 ± 0.50 Mg CO₂-C ha⁻¹ yr⁻¹. Sabaju and Sebungan differed in their annual combined R_tot and drain CO₂ flux. Sabaju was estimated to produce 29.54 ± 0.47 Mg CO₂-C ha⁻¹ yr⁻¹ and Sebungan was estimated to produce 17.62 ± 0.53 Mg CO₂-C ha⁻¹ yr⁻¹. The plantations differed in this CO₂ flux because of the contribution from the harvest path at each plantation; Sebungan averaged 9.21 ± 1.40 Mg CO₂-C ha⁻¹ yr⁻¹ from the harvest path, while Sabaju emitted 20.19 ± 1.33 Mg CO₂-C ha⁻¹ yr⁻¹.

Annual estimates of CH₄ fluxes were calculated, collectively for both plantations and individually, with very little difference between the total CH₄ flux (Table 4). Combined, the annual CH₄ flux from oil palm plantations, including soil, drain and stem fluxes, was 61.02 ± 17.78 kg CH₄-C ha⁻¹ yr⁻¹. In Sabaju, the annual CH₄ flux was 60.84 ± 4.35 kg CH₄-C ha⁻¹ yr⁻¹ and in Sebungan the annual CH₄ flux was 61.19 ± 25.28 kg CH₄-C ha⁻¹ yr⁻¹.

R_tot was partitioned into R_h and R_a at both plantations (Table 5). Collectively, R_h ranged from 17.61 to 17.89 Mg CO₂-C ha⁻¹ yr⁻¹ and R_a ranged from 4.19 to 4.47 Mg CO₂-C ha⁻¹ yr⁻¹. Plantation R_a increased from 4.71 to 4.99 Mg CO₂-C ha⁻¹ yr⁻¹ when stem respiration was included. R_h varied between the two plantations; R_h in Sabaju ranged from 23.89 to 24.39 Mg CO₂-C ha⁻¹ yr⁻¹, while in Sebungan it ranged from 11.13 to 11.81 Mg CO₂-C ha⁻¹ yr⁻¹. Considering the plantations separately, R_a ranged from 3.10 to 3.60 Mg CO₂-C ha⁻¹ yr⁻¹ in Sabaju and from 4.36 to 5.04 Mg CO₂-C ha⁻¹ yr⁻¹ in Sebungan. When stem respiration was combined with soil R_a, plantation R_a increased to range from 3.52 to 4.02 Mg CO₂-C ha⁻¹ yr⁻¹ in Sabaju and from 4.98 to 5.66 Mg CO₂-C ha⁻¹ yr⁻¹ in Sebungan.

Plantation peat carbon losses were comprised of R_h, drain CO₂, soil CH₄, drain CH₄, and stem CH₄ components and were estimated in CO_2−eq. Overall plantation carbon losses ranged from 72.35 ± 0.14 to 73.38 ± 0.16 Mg CO_2−eq ha⁻¹ yr⁻¹. Plantation carbon losses varied between the two plantations. In Sabaju, the plantation carbon losses ranged from 95.58 ± 0.14 to 97.41 ± 0.17 Mg CO_2−eq ha⁻¹ yr⁻¹. In Sebungan, the plantation carbon losses ranged from 48.37 ± 0.16 to 50.86 ± 0.19 Mg CO_2−eq ha⁻¹ yr⁻¹.

Breaking the carbon losses down by component showed that 92% of the CO₂ produced through R_h and drain CO₂ pathways came from soils and 8% came from the drains (Table 6). This was similar for both plantations. In Sabaju, 94% of the CO₂ produced through R_h or drain CO₂ came from R_h and 6% came from drain CO₂. In Sebungan, 89% of the CO₂ produced through R_h or drain CO₂ came from R_h and 11% came from drain CO₂.

The CH₄ losses showed different component contributions to CO₂ (Table 6). Overall, 30% of CH₄ fluxes came from the soil and 70% of plantation CH₄ fluxes came from the drainage ditches. Stem CH₄ fluxes gave negligible contributions. The CH₄ pathways varied between the two plantations. In Sabaju, 50% of the plantation CH₄ flux was measured from the soil surface and 50% was recorded from the drainage ditches. In Sebungan, 11% of the plantation CH₄ flux came from the soil and 89% came from the drainage ditches.

In order to isolate which pathways were dominating the plantation peat carbon losses, the fluxes were compared in CO_2−eq (Table 6). Considering the soil component in isolation, 99% of soil peat carbon losses were attributable to R_h and 1% to CH₄. Considering the drain carbon fluxes, 78% of the carbon losses came through drain CO₂ and 22% of the carbon measured from the drain surfaces came from drain CH₄.

Overall, 89% of plantation peat carbon losses came from R_h, 8% came from drain CO₂, 1% from soil CH₄, 2% from drain CH₄, and 0% from stem CH₄; 90% of plantation carbon losses came from the soil and 10% from drainage ditches (Table 6). In Sabaju, 92% of plantation peat carbon losses were attributable to R_h, 6% to drain CO₂, 1% to soil CH₄, 1% to drain CH₄, and 0% to stem CH₄; 93% of peat carbon losses came from the soil and 7% from the drainage ditches. In Sebungan, 84–85% of plantation carbon losses were from R_h, 10–11% were from drain CO₂, 0% from soil CH₄, 4% from drain CH₄, and 0% from stem CH₄; 85–86% came from the soil and 14–15% from the drainage ditches.

R_h, the most dominant component in the peat carbon losses, was controlled by variations in air temperature, WTD, soil temperature, and soil moisture (Table S4). R_h increased as air temperature increased (Figure 5). Whilst the interaction between air temperature and plantation was not significant, the rate of increase in R_h as air temperature increased was predicted to be almost three times greater in Sabaju than in Sebungan (Table S5). Unexpectedly, R_h increased as WTD rose, with a modeled increase of 0.75 Mg CO₂-C ha⁻¹ yr⁻¹ for every 10 cm decrease in WTD. The interaction between WTD and plantation was significant, with opposite relationships seen in both plantations (Figure 5). In Sabaju, R_h increased by 0.85 Mg CO₂-C ha⁻¹ yr⁻¹ for every 10 cm rise in WTD. R_h increased in Sebungan as the WTD was lowered, by a modeled rate of −0.29 Mg CO₂-C ha⁻¹ yr⁻¹ for every 10 cm drawdown of WTD.

Overall, R_h increased as soil temperature increased and R_h increased as soil moisture decreased (Tables S4, S5). Soil temperature and soil moisture had significant relationships with R_h when their individual interactions with plantation and microform were included in the model. Considering the plantations together and the microforms separately, the harvest path increased in R_h as soil temperature increased, and the frond pile and cover plants decreased in R_h as soil temperature increased. This pattern was seen in Sebungan but not in Sabaju when the plantations were considered separately. In Sabaju, R_h increased when soil temperature increased throughout the microforms. Increasing soil moisture decreased the rate of R_h, with the greatest effect beneath the frond pile, followed by the harvest path and then the cover plants. This trend was also observed when the microforms in Sebungan were considered in isolation. In Sabaju, rates of R_h decreased as soil moisture increased in the cover plants but not in the harvest path and frond piles.

Soil CH₄ was controlled by soil moisture between 0 and 10 cm, soil moisture between 30 and 40 cm, WTD and soil temperature (Table S6). Soil CH₄ increased as soil moisture between 0 and 10 cm increased (Table S5). The interaction between plantation, microform, WTD and soil moisture between 30 and 40 cm also significantly explained the variation in soil CH₄. Here the expected relationship of soil CH₄ increasing as WTD decreased was observed in the harvest path and next to the palm. This trend was seen when considering the plantations together or in Sabaju. In Sebungan, soil CH₄ was modeled to increase as WTD decreased in the frond piles and cover plants. The opposite relationship was seen in the other microforms. Similarly, variation was seen in the relationship between soil CH₄ and soil moisture between 30 and 40 cm. Soil CH₄ increased as soil moisture between 30 and 40 cm increased in the harvest path, frond pile and next to the palm, with the opposite relationship in the cover plants. In Sabaju, soil CH₄ increased as soil moisture between 30 and 40 cm increased in the harvest path and by the palm. In Sebungan, soil CH₄ increased as soil moisture between 30 and 40 cm increased in the frond pile and cover plants. Soil CH₄ increased as soil temperature decreased. The interaction between soil temperature and plantation was significant, with the rate of soil CH₄ increasing twice as quickly in Sabaju than in Sebungan when soil temperature decreased.

Drain CO₂ and drain CH₄ had different environmental relationships (Tables S5, S7). Drain CO₂ did not have a significant relationship with any of the environmental variables measured in this study, but did vary significantly depending on the size of the microform. Overall, the field drain had a greater drain CO₂ flux. Drain CH₄ increased significantly as air temperature increased. This was seen at both plantations.

Stem CH₄ was explained by variations in monthly relative humidity, WTD, soil moisture, air temperature and soil temperature (Table S8). Stem CH₄ decreased as monthly relative humidity increased (Table S5). This relationship was seen on both plantations and the interaction between the individual palms within each plantation with monthly relative humidity was significant. Stem CH₄ increased as WTD increased. This was also seen at both plantations. For soil moisture, air temperature and soil temperature, opposite trends were seen between the two plantations. Stem CH₄ increased as soil moisture increased, this trend was the same when the plantations were considered together and also in Sebungan when the plantations were considered separately, but not in Sabaju. Stem CH₄ increased as air temperature and soil temperature increased, when both plantations were considered together. These trends were seen in Sebungan but not in Sabaju when the plantations were considered separately.

CO₂ and CH₄ fluxes varied significantly when measured from different surface microforms in two oil palm plantations on peat soil. The highest soil CO₂ and CH₄ fluxes were measured next to the palms. This pattern was attributed to the high density of roots in the palm rhizosphere; root biomass was greatest within the immediate 1 m radius around each palm, and decreased with increasing distance from the palm (Farmer, 2013; Dariah et al., 2014; Manning, 2019; Manning et al., in preparation). In these studies R_tot and soil respiration were measured at consistent intervals from the palm into the harvest path (Farmer, 2013; Dariah et al., 2014), frond pile and cover plants (Manning, 2019; Manning et al., in preparation) from mature oil palm plantations in Jambi, Indonesia (Farmer, 2013; Dariah et al., 2014) and this plantation (Manning, 2019; Manning et al., in preparation). Respiration from root biomass has been shown to drive high R_tot rates next to the palm, with a progressive decline in R_tot with increasing distance from each palm (Farmer, 2013; Dariah et al., 2014; Manning, 2019; Manning et al., in preparation). Likewise, high CH₄ fluxes from the rhizosphere could be due to a number of causes, including accelerated organic matter turnover facilitated by the metabolism of root exudates (Girkin et al., 2018a,b), or transport of CH₄ from deeper in the profile via porous tissues in the roots.

Outside the palm rhizosphere (i.e., >1 m distance from the palm), different soil surface management practices appeared to influence R_tot but not CH₄. R_tot was significantly higher in the harvest path than from either the frond piles or beneath the cover plants. There was no significant difference in CH₄ measured from these microforms. Considering the plantations separately gave different trends for R_tot, but did not change the pattern of CH₄. For example, in Sebungan, there was no significant difference in R_tot between the non-rhizosphere fluxes. In contrast, in Sabaju the harvest path had significantly higher R_tot than the frond pile and cover plants. These differences in R_tot among microforms are noteworthy because few other studies consider the effects of surface management practices on R_tot, with the exception of work by Arifin et al. (2015), who also found that R_tot was significantly lower when measured from beneath cover plants than from the harvest path. These differences in R_tot among microforms are significant because they have direct bearing not only predicting and upscaling land-atmosphere fluxes, but also because they provide insight into how soil surface management practices are affecting soil CO₂ and CH₄ dynamics.

Drainage ditches made a substantial contribution to land-atmosphere CO₂ and CH₄ fluxes, particularly with respect to CH₄. Drain CH₄ was significantly higher than CH₄ fluxes from all soil microforms, with the exception of the CH₄ flux measured next to the palm. Comparing our drain data against other datasets, we found that our CH₄ fluxes fell within the range observed in drained peatlands and drained Acacia sp. plantations in Kalimantan (Jauhiainen and Silvennoinen, 2012). These high drain CH₄ fluxes could be attributable to the photochemical or microbial breakdown of DOC, or lateral transport of dissolved CH₄ produced in the peat into surface waters (Billett and Moore, 2008; Teh et al., 2011; Cory et al., 2014; Logue et al., 2016). These data are significant because there are currently no published data on drain CO₂ and CH₄ fluxes from oil palm plantation drainage ditches, and they have implications for extrapolating both plot-level CO₂ and CH₄ fluxes to larger spatial scales. In particular, the estimates of CO₂ and CH₄ fluxes based on straight mean averaging or non-spatially explicit sampling may tend to overestimate the net release of CO₂ to the atmosphere and underestimate the net release of CH₄ from tropical peatlands, given that drainage ditches form a small but significant portion of the landscape.

Stem respiration fell within a similar range to drain CO₂ and the R_tot fluxes when excluding the high fluxes next to the palm. Once again, disaggregation of the dataset by plantation revealed local differences in stem respiration relative to other CO₂-emitting processes. Stem respiration was lower in Sabaju than in Sebungan. Furthermore, in Sabaju, stem respiration was significantly lower than R_tot measurements taken from next to the palm, the harvest path and the frond pile. In contrast, in Sebungan, stem respiration was similar to all other soil fluxes with the exception of R_tot measured next to the palm. We believe that these local differences in stem respiration may reflect differences in productivity between the two plantations. Generally, stem respiration correlates with photosynthetic rate (Yang et al., 2016); hence, the higher stem respiration in Sebungan may reflect that this plantation shows higher rates of CO₂ uptake and growth than Sabaju, evidenced by higher rates of net primary productivity, larger fronds and formation of a denser, more closed canopy.

Stem CH₄ fluxes were recorded from the palms in both plantations, with lower CH₄ fluxes than those reported from the soil or drain surfaces, by factors of 22 and 200, respectively. Stem CH₄ has been recorded from the stems of tropical trees growing in swamp forests in Brunei and the Amazon (Pangala et al., 2013, 2017). Stem fluxes from Brunei ranged between similar magnitudes to the stem fluxes measured at this site (here: −0.016–0.440 mg CH₄-C m⁻² hr⁻¹; Brunei: 0.013–0.139 mg CH₄-C m⁻² hr⁻¹) when the fluxes were considered per m⁻² of stem surface area (Pangala et al., 2013). Stem fluxes in the Amazon were considerably higher than the stem fluxes in Borneo (0.248–435.75 mg CH₄-C m⁻² hr⁻¹; Pangala et al., 2017). Stem CH₄ has been shown to correlate with rates of evapotranspiration (Pangala et al., 2014). We propose that the similar rates of relative humidity between Sarawak and Brunei may contribute to the similarities in range of fluxes from the stems of plants at these sites. The CH₄ oxidation reported from this site was measured four times and only when relative humidity exceeded 93% (Manning, 2019). We suggest that this may be due to methanotrophs on the palm surface, that are normally masked from stem CH₄ fluxes, but observable when high relative humidity prevents transpiration (Raghoebarsing et al., 2005).

Both CO₂ and CH₄ fluxes were greater in the dry season than the rainy season, but these differences were not statistically significant. In this region, two rainy seasons occur, with the main rainy season between October and January, and a secondary rainy season between May and August. Dry season falls from February to April and in September.

Considering the surface microforms separately, significant seasonal variation was seen from R_tot measured next to the palm, with higher fluxes in the dry season than in the rainy season. R_tot next to the palm was predominately attributed to R_a (Farmer, 2013; Dariah et al., 2014; Manning et al., in preparation). Stem respiration gave similar overall trends to by palm R_tot, but differences were not statistically significant. Overall these fluxes, dominated by autotrophic processes, suggest strong seasonal growth from the palms, with more photosynthetic activity in the dry season than in the rainy season. Seasonality has been recorded from R_tot next to the palm in other plantations (Comeau, 2016; Hergoualc'h et al., 2017), as well as from R_a sampled from the edge of the canopy (Melling et al., 2013).

In contrast, R_tot measured from the harvest path, frond pile and cover plants did not show significant seasonal variation. R_tot fluxes measured from these locations were dominated by R_h. R_h was shown to have significant seasonality in a rubber plantation in Kalimantan, Indonesia growing on peat soil, where the seasonality in R_h was driven by seasonal changes in WTD (Wakhid et al., 2017). Similar to this study, those on other oil palm plantations have not found significant seasonality in R_h (Comeau, 2016; Hergoualc'h et al., 2017). The lack of apparent seasonality in R_h from oil palm plantations could be due to reduced variability in the environmental variables driving R_h. For example, Melling et al. (2013) found that R_h varied in the forest due to variations in WTD but not in the oil palm plantation where the WTD was more consistent.

Drain CO₂ and CH₄ fluxes showed seasonality; drain CO₂ fluxes were significantly higher in the rainy season than the dry season and drain CH₄ fluxes were significantly higher in the dry season than in the rainy season. Cook et al. (2018) found that drain discharge varied seasonally but total organic carbon (DOC plus particulate organic carbon) concentrations in the drains did not vary seasonally from Sabaju and Sebungan. Billett and Moore (2008) found that drain CO₂ increased when flow rate increased, whilst DOC concentrations remained consistent regardless of flow rate.

R_h was estimated to range from 17.61 ± 1.65 to 17.89 ± 0.85 Mg CO₂-C ha⁻¹ yr⁻¹ (mean: 17.75 ± 1.54 Mg CO₂-C ha⁻¹ yr⁻¹). For these calculations, R_h was partitioned from R_tot using the distance from palm approach, as this was deemed sufficient in these plantations (Manning, 2019; Manning et al., in preparation). We present a range of estimates that include (1) only the harvest path—comprising of bare soil and often the only microform measured in R_h studies; (2) the harvest path and the frond pile—the latter being bare soil covered in dead fronds; (3) the harvest path, frond pile and cover plants—with the cover plants being included, despite the presence of cover plant roots, due to the increased R_h being potentially from priming, as opposed to R_a or decomposition from the cover plants (Manning et al., in preparation). In this study the lowest R_h estimate was obtained from when the harvest path and frond pile were both included in the calculations and the highest estimate came from when the cover plants were included as well. However, the estimates did not vary significantly regardless of which microforms were included.

R_h estimates from Sebungan were found to range from 11.13 ± 1.69 to 11.81 ± 0.87 Mg CO₂-C ha⁻¹ yr⁻¹ (mean: 11.43 ± 1.37 Mg CO₂-C ha⁻¹ yr⁻¹) and R_h from Sabaju was measured ranging between 23.89 ± 1.07 and 24.39 ± 1.61 Mg CO₂-C ha⁻¹ yr⁻¹ (mean: 24.08 ± 1.42 Mg CO₂-C ha⁻¹ yr⁻¹). In these estimates R_h was lowest from Sebungan when the harvest path only was measured, and highest when the cover plants were included. In Sabaju the lowest R_h estimate used the harvest path and frond pile data and the highest R_h estimate only used the harvest path results. The Sebungan R_h estimate fell within the reported range of R_h from oil palm plantations (4.1 to 22.9 Mg CO₂-C ha⁻¹ yr⁻¹, but the Sabaju estimate was greater than this range (Farmer, 2013; Melling et al., 2013; Dariah et al., 2014; Husnain et al., 2014; Comeau, 2016; Comeau et al., 2016; Hergoualc'h et al., 2017; Ishikura et al., 2018; Matysek et al., 2018).

These data are relevant to policy because they suggest that the IPCC emissions factor for R_h is underestimated. For example, whilst R_h estimates from Sebungan have been shown to be similar to the IPCC (2014) emissions factor of 11 Mg CO₂-C ha⁻¹ yr⁻¹, R_h from Sabaju is at least twice as large. This has implications when upscaling—estimating the R_h emissions from industrial oil palm plantations using the areal extent of plantations given in Miettinen et al. (2016) increases predicted R_h from 34.1 Tg CO₂-C yr⁻¹ if the IPCC emission factor is used to 55.8 Tg CO₂-C yr⁻¹ if the mean R_h from this study is used.

Another important policy-relevant finding is that area-weighted CH₄ fluxes from these managed systems are similar to swamp forests in SE Asia, suggesting that drainage has not diminished the CH₄ emissions potential of these systems. Annual estimates of CH₄ were 61.02 ± 17.78 kg CH₄-C ha⁻¹ yr⁻¹ including soil, drain and stem fluxes. Sabaju and Sebungan produced similar rates of CH₄ of 60.84 ± 4.35 kg CH₄-C ha⁻¹ yr⁻¹ and 61.19 ± 25.28 kg CH₄-C ha⁻¹ yr⁻¹, respectively. Annual estimates from forest CH₄ have been shown to range between 0.2 and 72 kg CH₄-C ha⁻¹ yr⁻¹ from Malaysian, Indonesian and Brunei peatlands (Inubushi et al., 1998, 2003; Furukawa et al., 2005; Hadi et al., 2005; Jauhiainen et al., 2005; Melling et al., 2005b; Hirano et al., 2009; Pangala et al., 2013). The results from this study are potentially controversial because current conceptual ideas of peatland drainage predict a net reduction in CH₄ emission linked to peatland drainage (Jauhiainen et al., 2005, 2008; Lai, 2009; Couwenberg et al., 2010). This is reflected in the IPCC (2014) emissions factors for oil palm plantations that are 0 kg CH₄-C ha⁻¹ yr⁻¹ from the peat soil and 45.18 kg CH₄-C ha⁻¹ yr⁻¹ from the drainage ditches. The results from this study show that draining peat does not reduce CH₄ fluxes. The high CH₄ fluxes from the drains suggest that either the drain sediments have increased rates of CH₄ production, or that the CH₄ transported in the water may be carried to the drains before being egressed, rather than diffusing through the soil and being oxidized by methanotrophs (Müller et al., 2015; Evans et al., 2016).

Combining rates of R_h, drain CO₂, soil CH₄, drain CH₄, and stem CH₄ in CO_2−eq gave overall plantation peat carbon losses. Here, net peat carbon losses averaged across plantations ranged between 72.35 ± 0.14 and 73.38 ± 0.16 Mg CO_2−eq ha⁻¹ yr⁻¹. Of this flux, 89% was attributed to R_h. Drain CO₂ was the next largest driver, contributing 8% of the plantation carbon losses. CH₄ contributed 2, 1, and 0% to plantation carbon losses for drain, soil and stem fluxes, respectively. We suggest this particularly low contribution from palm stems is due to two factors: (1) a strong gradient for CH₄ transport to the atmosphere through the drainage ditches, and (2) the low density of palms growing on the soil. As previously mentioned, the measured stem CH₄ fluxes were in the same range as the stem fluxes from trees in a swamp forest in Brunei (Pangala et al., 2013). Following upscaling the stem CH₄ fluxes from Brunei were 10–23 times greater than the stem CH₄ fluxes measured here (Pangala et al., 2013). However, the swamp forest trees were taller and more densely populated than oil palms, allowing for a greater overall stem surface area and therefore a greater total plot-scale stem CH₄. Similarly, the total contribution of tree stem CH₄ to plot-scale CH₄ was 62–87% of the total ecosystem flux—but there were no drainage ditches in the forest plot, meaning that stem fluxes had a greater overall representation (Pangala et al., 2013). Excluding the drainage ditches from the estimates in this study increased the stem contribution to 0.5%.

Sabaju and Sebungan differed in their overall net peat carbon losses. Sabaju produced double the amount of peat carbon losses compared to Sebungan, with peat carbon losses ranging between 95.58 ± 0.14 and 97.41 ± 0.17 Mg CO_2−eq ha⁻¹ yr⁻¹ in Sabaju. In Sebungan, net peat carbon losses fell within the range of 48.37 ± 0.16 to 50.86 ± 0.19 Mg CO_2−eq ha⁻¹ yr⁻¹. Sabaju and Sebungan peat carbon losses were made up from slightly different proportions of CO₂ and CH₄. R_h dominated the proportional losses of peat carbon in Sabaju; 92% of the total flux was from R_h, 6% from drain CO₂, 1% from soil CH₄, 1% from drain CH₄, and 0% from stem CH₄. In Sebungan, R_h was still the predominant peat carbon loss pathway, but the drain CO₂ and CH₄ fluxes doubled in contribution compared to Sabaju; 84–85% of plantation carbon losses in Sebungan was made up from R_h, 10–11% was from drain CO₂, 4% was from drain CH₄, and 0% was from both soil and stem CH₄.

Air temperature, soil temperature, WTD and soil moisture controlled rates of R_h. WTD, soil moisture at 0–10 cm, soil moisture at 30–40 cm and soil temperature controlled soil CH₄ fluxes. Drain CO₂ did not correlate with the environmental variables measured in this study. Drain CH₄ was controlled by air temperature. Stem CH₄ had significant relationships with relative humidity, WTD, air temperature, soil temperature and soil moisture.

Air and soil temperatures have been shown to control the rate of R_h, both on these plantations and in other studies (Jauhiainen et al., 2012; Farmer, 2013; Hergoualc'h et al., 2017). Temperature has been shown to increase the rate of R_tot and R_h due to increased activation energy for the chemical reactions (Lloyd and Taylor, 1994). Air temperature increased the rate of R_h in both plantations and the model predicted it was at a three times greater rate in Sabaju than in Sebungan. This steeper gradient in Sabaju might explain why Sabaju was only 1.5°C warmer than Sebungan but rates of R_h were twice as high. R_h also increased as soil temperature increased in both plantations—with differences seen in the relationships between the microforms. As soil temperature increased, R_h was predicted to increase in the harvest path and in the cover plants in Sabaju, which also did not always have a closed canopy, exposing the soil directly to the Sun. This relationship with temperature was not seen in the frond piles, potentially due to extra shading, a different microclimate and the interaction of other variables, such as moisture.

Hydrology also controlled rates of R_h, with soil moisture and WTD acting as proxies for redox potential. Increasing soil moisture lowered the rate of R_h, as supported by Farmer (2013), Comeau (2016) and Hergoualc'h et al. (2017). This was seen in every microform in Sebungan, particularly in the frond pile, and in the frond pile and cover plants in Sabaju. R_h increased as soil moisture increased in Sabaju, but the effect size was negligible. WTD had opposite relationships with R_h at both plantations. Sebungan gave the expected relationship with R_h increasing as the WTD lowered. This trend has also been seen by Hergoualc'h et al. (2017). However, Sabaju had the opposite relationship with WTD, with increasing WTD decreasing rates of R_h. The WTD measurements made in Sebungan were deeper and had a wider range than those made in Sabaju. Within the WTD range sampled, the more labile carbon may have already been oxidized in Sabaju, with fresher labile carbon exposed at the deeper WTDs in Sebungan, explaining the variation in relationship with WTD (Hooijer et al., 2012).

Soil CH₄ was controlled by soil temperature, soil moisture between 0 and 10 cm, soil moisture between 30 and 40 cm and WTD. Melling et al. (2005b) also found that soil temperature, water filled pore space and WTD controlled soil CH₄ fluxes. Here, soil CH₄ increased as soil moisture between 0 and 10 cm increased. This result would be expected, as CH₄ is more likely to be produced in anoxic conditions. The interaction between soil CH₄, soil moisture between 30 and 40 cm, plantation and location was significant. In the different plantations and at different microforms, the expected relationship of soil CH₄ increasing as WTD increased and soil moisture between 30 and 40 cm increased was seen, but not in every microform. It would be expected that soil CH₄ increased as WTD decreased and soil moisture between 30 and 40 cm increased due to a larger volume of anoxic conditions for methanogens to break the peat down, and a lower volume of oxic conditions for methanotrophs to break down the CH₄ (Iiyama et al., 2012; Carlson et al., 2015). Soil moisture at 0–10 cm did not have a significant interaction with soil moisture between 30 and 40 cm or WTD. Furthermore, Manning (2019) found that surface soil moisture correlated with climatic trends and soil moisture between 30 and 40 cm was determined by WTD. We propose that redox potential nearer the surface of the peat is more important than WTD for soil CH₄ fluxes, with CH₄ oxidized to CO₂ regardless of WTD, if the surface of the peat does not inhibit methanotrophs. Finally, soil CH₄ increased as soil temperature decreased. This may be due to seasonality and soil temperatures being cooler when the peat was wetter.

Drain CO₂ and drain CH₄ were controlled by different variables. Drain CO₂, produced from DOC, did not have a significant relationship with the environmental variables but did have a significant relationship with drain type, being higher in the smaller field drains than the larger collection drains. DOC was greater in Sebungan than Sabaju during this measurement period and this was associated with the increased WTD at time of measurement (Cook et al., 2018). This same relationship was not observed in this study for drain CO₂. The smaller field drains may therefore have a higher drain CO₂ flux than the collection drains due to being the first drain that DOC reaches from the soil—after all the field drains feed into the collection drains. Drain CH₄ was controlled by the rate of air temperature in both plantations. This was presumably due to the increase in rate of diffusion for the CH₄ (Billett and Moore, 2008).

Stem CH₄ was controlled by relative humidity, WTD, air temperature, soil temperature and soil moisture. Stem CH₄ reduced as relative humidity increased. We hypothesize that this is due to reduced rates of evapotranspiration at higher relative humidities, from a reduced moisture gradient. Evapotraspiration has been shown to control the rates of stem CH₄ (Pangala et al., 2014). Slowing the rate of water through the xylem would reduce the speed of transport for dissolved CH₄ and thus reduce the rate of stem CH₄. Increasing WTD increased the rate of stem CH₄. This has been seen in a manipulation experiment by Pangala et al. (2014), who found that if there is soil volume between the WTD and plant roots, CH₄ is oxidized before it reaches the roots. Air temperature, soil temperature and soil moisture gave opposite relationships with stem CH₄ at each plantation, with stronger relationships in Sebungan where increasing air temperature, soil temperature and soil moisture all increased stem CH₄. Increasing temperatures would be expected to increase the rate of methanogenesis, whilst increasing soil moisture would reduce the redox potential, also increasing methanogenesis or transport of stem CH₄ to the roots (Pangala et al., 2014). These relationships were negative in Sabaju but had very small effects.

The most effective way to reduce carbon losses from oil palm plantations on peat soil is to reduce the rate of R_h. The strong influence of temperature and WTD on R_h suggest that means of controlling soil surface temperatures and WTD are the best means of mitigating carbon losses from the peat. For example, the impact of temperature could be reduced by providing better coverage of the soil surface—particularly when the plantation canopy has not closed. Here, in Sebungan the canopy was closed, soil and air temperatures were lower and the R_h flux was half the rate of Sabaju. In Sabaju, the canopy was open and air temperature increased the rate of R_h at three times the extent in Sebungan. Covering the soil in Sabaju with frond piles lowered the mean soil temperature (harvest path: 28.6 ± 0.7°C; frond pile: 27.8 ± 0.6°C) and increased mean soil moisture (harvest path: 57.8 ± 0.9%; frond pile 61.6 ± 1.0%); both changes in environmental conditions have been shown to reduce the R_h flux. Jauhiainen et al. (2014) found that R_h could be reduced by 30% if the tropical peat was shaded. Annual R_tot was 30% lower beneath the frond pile in Sabaju than from the harvest path—extending the shade could therefore reduce Sabaju R_h to range between 16.9 ± 1.61 and 17.4 ± 0.82 Mg CO₂-C ha⁻¹ yr⁻¹, reducing the plantation net carbon losses to between 69.59 ± 0.62 and 71.75 ± 0.65 Mg CO_2−eq ha⁻¹ yr⁻¹.

Raising the water table is also an effective way of suppressing R_h. This was particularly apparent in Sebungan, where the canopy was closed and temperatures were lower. Sebungan had a lower mean WTD than the RSPO recommendations at the time of measurement and thus fresh peat may have been exposed to heterotrophic bacteria, increasing the rate of R_h from Sebungan (Lim et al., 2012; Mishra et al., 2014; Carlson et al., 2015).

Furthermore, temperature control may be more important than WTD control. Sebungan had a lower WTD than Sabaju but Sabaju had higher rates of R_h. It would be expected that R_h decreased with WTD (Carlson et al., 2015). However, the result from Sabaju did not fit the trend. We propose this is due to the open canopy in Sabaju and thus there being no barrier for the Sun to heat up the peat and increase the rate of its decomposition.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.