Temporal variation in nitrous oxide (N2O) fluxes from an oil palm plantation in Indonesia: An ecosystem‐scale analysis

The rapidly growing areal extent of oil palm (Elaeis guineensis Jacq.) plantations and their high fertilizer input raises concerns about their role as substantial N2O sources. In this study, we present the first eddy covariance (EC) measurements of ecosystem‐scale N2O fluxes in an oil palm plantation and combine them with vented soil chamber measurements of point‐scale soil N2O fluxes. Based on EC measurements during the period August 2017 to April 2019, the studied oil palm plantation in the tropical lowlands of Jambi Province (Sumatra, Indonesia) is a high source of N2O, with average emission of 0.32 ± 0.003 g N2O‐N m−2 year−1 (149.85 ± 1.40 g CO2‐equivalent m−2 year−1). Compared to the EC‐based N2O flux, average chamber‐based soil N2O fluxes (0.16 ± 0.047 g N2O‐N m−2 year−1, 74.93 ± 23.41 g CO2‐equivalent m−2 year−1) are significantly (~49%, p < 0.05) lower, suggesting that important N2O pathways are not covered by the chamber measurements. Conventional chamber‐based N2O emission estimates from oil palm up‐scaled to ecosystem level might therefore be substantially underestimated. We show that the dynamic gas exchange of the oil palm canopy with the atmosphere and the oil palms' response to meteorological and soil conditions may play an important but yet widely unexplored role in the N2O budget of oil palm plantations. Diel pattern of N2O fluxes showed strong causal relationships with photosynthesis‐related variables, i.e. latent heat flux, incoming photosynthetically active radiation and gross primary productivity during day time, and ecosystem respiration and soil temperature during night time. At longer time scales (>2 days), soil temperature and water‐filled pore space gained importance on N2O flux variation. These results suggest a plant‐mediated N2O transport, providing important input for modelling approaches and strategies to mitigate the negative impact of N2O emissions from oil palm cultivation through appropriate site selection and management.

(149.85 ± 1.40 g CO 2 -equivalent m −2 year −1 ). Compared to the EC-based N 2 O flux, average chamber-based soil N 2 O fluxes (0.16 ± 0.047 g N 2 O-N m −2 year −1 , 74.93 ± 23.41 g CO 2 -equivalent m −2 year −1 ) are significantly (~49%, p < 0.05) lower, suggesting that important N 2 O pathways are not covered by the chamber measurements. Conventional chamber-based N 2 O emission estimates from oil palm up-scaled to ecosystem level might therefore be substantially underestimated. We show that the dynamic gas exchange of the oil palm canopy with the atmosphere and the oil palms' response to meteorological and soil conditions may play an important but yet widely unexplored role in the N 2 O budget of oil palm plantations. Diel pattern of N 2 O fluxes showed strong causal relationships with photosynthesis-related variables, i.e. latent heat flux, incoming photosynthetically active radiation and gross primary productivity during day time, and ecosystem respiration and soil temperature during night time. At longer time scales (>2 days), soil temperature and water-filled pore space gained importance on N 2 O flux variation. These results suggest a plant-mediated N 2 O transport, providing important input for modelling approaches and strategies to mitigate the negative impact of N 2 O emissions from oil palm cultivation through appropriate site selection and management.

| INTRODUCTION
During the past decades, rising global demand for cheap oils and fats has promoted the expansion of oil palm (Elaeis guineensis Jacq.) plantations in tropical regions due to the plant's high yield (FAO, 2016, FAOSTAT;Wahid et al., 2005). Indonesia and Malaysia are, by far, the largest producers of palm oil. In 2016 and 2017, Indonesia contributed 56% and Malaysia 30% to the global supply of palm oil (USDA, 2018). In 2015, the combined area of oil palm plantations in the two countries covered 17 million ha (Chong et al., 2017). In order to meet the growing demand for food and biofuel, the area of oil palm plantation is expected to further increase in the near future (Miettinen et al., 2012;Pirker et al., 2016;Sumarga & Hein, 2016).
Oil palm expansion affects ecosystem properties and functions such as biodiversity (Barnes et al., 2014), land surface temperature, (Sabajo et al., 2017), microclimate (Meijide et al., 2018), hydrology (Merten et al., 2016(Merten et al., , 2020, carbon pools (Guillaume et al., 2015;Lamade & Bouillet, 2005;van Straaten et al., 2015) and greenhouse gas (GHG) balances (Kusin et al., 2017;Meijide et al., 2020). Economically effective oil palm management requires site-specific knowledge of environmental factors and management practices to understand nutrient availability and distribution (Castanheira et al., 2014;Drewer et al., 2021;Iddris et al., 2023;Pardon, Bessou, Saint-Geours, et al., 2016;Webb et al., 2011) and to reduce damage to the environment and climate. Deficits in soil nutrients are commonly compensated with fertilization, a common practice since oil palm requires large quantities of nitrogen (N), potassium, phosphorus, magnesium, and boron to generate economic profits (Woittiez et al., 2017). However, high fertilizer levels or inappropriate timing may result in eutrophication, soil acidification, and a disproportional increase in atmospheric N 2 O emissions (Ito et al., 2018;Signor & Cerri, 2013;Smith, 2017;Tian & Niu, 2015;Webb et al., 2011).  showed that emissions in oil palm are, on average 3.4 kg N 2 O-N ha −1 year −1 , with highest emissions observed at juvenile plantation stage. Sisbudi et al. (2012) found that in smallholder and company oil palm plantations, N 2 O emissions can account for 58% and 31%, respectively, of the total GHG emissions, with lower emission rates in the latter due to the use of organic fertilizers. Therefore, optimal fertilizer management needs to ensure high nutrient efficiency while reducing negative environmental impacts. The quantification of ecosystem-scale N 2 O emissions, and an improved understanding of the complex spatial and temporal variation in N 2 O fluxes and their controls is crucial for assessing how climate impacts oil palm cultivation.
Fluxes of N 2 O are typically measured using manual static chambers (Hassler et al., 2017) or the eddy covariance (EC) technique (Aubinet et al., 2012), which complement each other with regard to their spatial and temporal resolution. Manual soil chamber measurements can accurately capture the spatial heterogeneity of GHG fluxes that results from small-scale differences in the natural source strength and management. However, chambers can only cover soil emissions and may therefore miss other ecosystem compartments that may contribute to the total emissions, for example N 2 O emissions from organic soils in the cavity of canopy epiphytes soil (Allen et al., 2018). Further, the manual use of chambers is laborious and the temporal resolution of the generated data set is therefore often coarse. In contrast, EC measurements provide highfrequency measurements at ecosystem scale. However, the spatially aggregated flux signal cannot resolve potentially varying source strengths, as they may result from varying fertilizer application rates or soil conditions leading to hot and cold spots of C and N dynamics. The coordinated combination of both measurement techniques can provide a better understanding of the temporal dynamics and spatial patterns of N 2 O fluxes and thereby reduce uncertainties in extrapolating N 2 O fluxes at the ecosystem or regional scale (Levy et al., 2017;Shurpali et al., 2016). So far, such a combined approach has not been carried out in oil palm plantations.
Previous research reports on N 2 O flux pattern in various ecosystems were largely attributed to abiotic drivers such as soil moisture and soil temperature (Imer et al., 2013;Kroon et al., 2010;Levy et al., 2017;Liang et al., 2018). The erratic nature of N 2 O emissions and hysteresis of N 2 O fluxes with environmental factors (Butterbach-Bahl et al., 2013;Harris et al., 2021) complicate finding statistically significant correlations to constrain the underlying cause-and-effect relationships. Moreover, Shurpali et al. (2016) have shown, for the first time, that vegetation may play an important role in the processes associated with ecosystem N 2 O exchange. In oil palm plantations, little is known about the overall N 2 O balance at the ecosystem-scale and potential vegetation-mediated N 2 O emissions and, to our knowledge, no analysis approach has systematically targeted the complex temporal patterns and cause-and effect relationships of N 2 O emissions in these highly relevant land-use types.
In this study, we combine continuous ecosystem-scale measurements using the EC approach with spatially highly resolved closed chamber measurements to provide a comprehensive perspective on N 2 O fluxes in an oil palm plantation in the tropical lowlands of Jambi province (Sumatra, Indonesia) during the period August 2017 to April 2019. We aim to (1) quantify the overall ecosystemscale N 2 O emissions and emission factors, derived from both measurement approaches, (2) determine the spatial heterogeneity and temporal variation of N 2 O fluxes, with particular emphasis on diel and multi-day emission dynamics, and (3) assess the underlying environmental and meteorological drivers using wavelet analysis and crosscorrelation function (CCF) to identify drivers of N 2 O fluxes at different time scales.

| Study site
We conducted our measurements in a mature (~16 years old) commercial oil palm plantation (1°41′35.0′′ S, 103°23′29.0′′ E, 76 m a.s.l.) in tropical lowland Jambi province (Sumatra, Indonesia) ( Figure 1) . Long-term temperature and precipitation data (1991-2020 baseline period) collected by the Indonesian Meteorological Service (BMKG) at Sultan Thaha Airport in Jambi city, located approximately 29 km northeast of our study site, showed mean annual air temperature of 26.8°C (±0.2°C standard deviation [SD]) and mean annual precipitation of 2128 mm year −1 (±415 mm SD), with a dry season from June to September and bimodal rainy seasons around March and December. The terrain around our study site is generally flat, with small elevation variations of approximately ±15 m. The area is dominated by highly weathered loam Acrisol soils (Allen et al., 2015). In the plantation, the mean soil carbon and nitrogen content is generally low, reaching 1.12% (±0.34% SD) and 0.08% (±0.02% SD), respectively.
The plantation covers a total area of 2185 ha. Our measurement site is located in the north-western part of the plantation where oil palm seedlings were planted in 2002. In the entire plantation, oil palms were planted in a triangular array, with 9 m × 9 m horizontal density and 142 palms per ha ( Figure 2). Leaf area index was estimated to be 3.64 m 2 m −2 at our study site, based on an average palm height of 12 m, with 35-45 expanded leaves per palm (Fan et al., 2015). Understory vegetation is scarce due to regular application of Glyphosate herbicide and occasional mowing while epiphytes, such as flowering plants (Melastomataceae, Orchidaceae) or ferns (Polypodiophyta) densely cover the stumps of pruned oil palm leaves.
The oil palm plantation is fertilized manually with "Pamafert" nitrogen-phosphorus-potassium granular (174.4 kg N, 47.6 kg P and 190.6 kg K ha −1 year −1 ), and dolomite (426 kg ha −1 year −1 , based on average soil pH of 5.6) in top-dress application (Darras et al., 2019;Iddris et al., 2023;Meijide et al., 2020). According to the plantation owner, fertilization follows a defined routine and time schedule in which the plantation is divided into management micro-divisions. In each micro-division, fertilizers are applied in a 2 m radius from the oil palm base twice per year, that is toward the beginning and the end of the wet season (October and April, respectively). The footprint of our EC measurement tower and the location of the vented soil chambers covers several of these micro-divisions.

| EC measurements and processing
We performed EC flux measurements of N 2 O, H 2 O and CO 2 net ecosystem exchange (NEE) during the period August 2017 to April 2019 with a LI7500A open-path CO 2 / H 2 O infrared gas analyser (LI-COR, Inc.), a closed-path N 2 O/CO/H 2 O laser spectrometer (LGR), model 913-0014 (Los Gatos Research), and two Metek uSonic-3 Scientific sonic anemometers (Metek). The sample frequency of all EC sensors is 10 Hz. The open-path CO 2 /H 2 O analyser, the tube inlet of the N 2 O/CO/H 2 O sensor and one of the two anemometers were placed above the oil palm canopy at the top of a 22 m high steel framework tower. The N 2 O/ CO/H 2 O sensor was placed in a small air-conditioned (27°C) cabin next to the tower. The air samples were drawn from the intake point on the tower into the analyser optical cell through a 30.95 m long polythene tube (inner diameter of 10 mm) at a flow rate of 58.6 L min −1 . Concentrations of N 2 O and H 2 O were recorded as dry mole fractions and we logged internal cell temperature and cell pressure of the N 2 O/CO/H 2 O sensor, inlet flow and pressure. Industrial filter papers (glass-microfiber discs MGA; Ahlstrom-Munksjö), with a particle retention of 1.6 μm, were placed in the sampling tube near the analyser and were replaced weekly. During extensive dry periods, filters were replaced daily due to high aerosol concentration in the sample air. The N 2 O/CO/H 2 O sensor was calibrated every 6 months with N 2 O reference gas and H 2 O measurements were calibrated using a dew point generator. The second sonic anemometer was installed at the tower at 2.4 m height to measure below-canopy wind speed and wind direction.
We used the software package EddyPro version 6.2.0 (LI-COR, Inc.) for EC data processing. Raw data screening and correction, statistical tests and processing of CO 2 , H 2 O and energy fluxes followed standard procedures described in Meijide et al. (2017). Data processing of N 2 O fluxes follows the procedure recommended by Nemitz et al. (2018), which comprises raw data screening, statistical tests, and raw data processing such as despiking and block-averaging, time lag compensation and high-and low-frequency spectral correction. Raw data quality assessment included removal of N 2 O concentration measurements from the subsequent flux calculation when the cell pressure and cell temperature were outside of the sensorspecific range (0.1% of available 30-min measurements) or when N 2 O mole fraction exceeded 1.0 μmol mol −1 (0.2% of available 30-min measurements).
We derived N 2 O flux from the covariance of the vertical wind velocity and N 2 O dry mole fraction. The source area of our N 2 O flux measurements, that is the EC footprint area as derived from Kljun et al. (2004), stretched evenly distributed around the tower, with 90% and 50% of the fluxes originating within a distance of ~237 and 113 m, respectively ( Figure 1). Flux sources beyond the plantation boundaries were excluded . Situations of low-turbulence conditions as well as possible atmospheric decoupling between above-and below-canopy air and related flux advection were identified based on two different criteria: (i) the site-specific relationship between nocturnal abovecanopy friction velocity (u*) and CO 2 respiration and (ii) above-and below-canopy SD of vertical wind (σ w ) (Jocher et al., 2018) in relation with N 2 O fluxes. We (i) performed a segmented linear regression model that estimates breakpoints and abrupt changes of the dependent variable by fitting regression relationships over user-defined segment sizes, using the R-package "strucchange" (Zeileis et al., 2002). At low nocturnal abovecanopy turbulence, CO 2 respiration increased with increasing u* but levels off at u* > 0.15 m s −1 ( Figure S1). We defined this value of u* as a threshold for low turbulence and decoupling and removed fluxes of N 2 O during such periods. Overall, 45% of N 2 O fluxes fell in low-turbulence periods, 75% of which occurred during nighttime.
Implausibly high N 2 O fluxes, with both positive and negative sign, could not fully be omitted by u* filtering but were related with σ w . We therefore (ii) applied a σ w threshold (Jocher et al., 2017) discarding situations with σ w < 0.2 m s −1 . This approach omitted another 5% of N 2 O fluxes. Based on the two filtering approaches, 36% of the original N 2 O flux measurements met our stringent criteria for u* and σ w . Filtering removed 75.5% of nocturnal and 51.9% of daytime N 2 O flux measurements. Finally, we calculated global warming potential as g N 2 O m −2 year −1 × 298 CO 2 -eq. on a 100-years time horizon (Forster et al., 2007).
To validate instrument performance, we compared H 2 O fluxes from the open-path LI7500A infrared gas analyser with the closed-path N 2 O/CO/H 2 O laser spectrometer (LGR) ( Figure S2). H 2 O fluxes from both sensors agree well (R 2 = 0.88), with only 4% higher fluxes measured with the open-path than the closed-path system ( Figure S2a), similar to other studies (Haslwanter et al., 2009;Holl et al., 2019;Polonik et al., 2019). Differences in measured H 2 O fluxes between the two sensors were most pronounced at values of ~3 mmol m −2 s −1 measured with the LI7500A ( Figure S2b). Mean diel cycles show similar trends and intensities, with an H 2 O flux peak in the early afternoon ( Figure S2c,d). We use the term "diel" to express the full 24 h cycle, "diurnal" for local daytime (06:30-18:00 h local time) and "nocturnal" for local nighttime (18:30-06:00 h local time) periods. Since H 2 O fluxes of the open-and closed-path system reveal such similarities, with only minor flux deviations at low H 2 O fluxes, it increases confidence in the performance of the closed-path system (Eugster et al., 2007).

| Meteorological measurements
We measured meteorological variables, that is air temperature and air humidity with hygro-thermo transmitters (type 1.1025.55.000; Thies Clima) protected inside weather and thermal radiation shields at 22, 16.3, 12.3, 8.1, 2.3 and 0.9 m heights, wind direction with a wind direction transmitter (Thies Clima) at 15.4 m height, wind speed with a 3-cup anemometer (Thies Clima) at 18.5, 15.4, 13 and 2.3 m height, air pressure with a baro transmitter (Thies Clima) at 22 m height, precipitation with two precipitation gauges (Thies Clima) at 11.5 m height, and radiation components at 22 m height, that is incoming and reflected photosynthetically active radiation (PAR) with PQS1-PARQuantum sensors (Kipp & Zonen), incoming and outgoing short-and long-wave radiation with a CNR4 net radiometer (Kipp & Zonen), global and diffuse radiation and sunshine duration with a BF5 sunshine sensor (Delta T). Soil moisture and soil temperature were measured with Trime-Pico 32 soil sensors (Imko) along three profiles at 0.05, 0.3, 0.6 and 1 m depth, respectively. Based on a particle density of 2.65 g cm −3 (Hassler et al., 2017) and measured soil bulk density of 1.19 g cm −3 using the core method, we calculated water filled pore space (WFPS). To investigate the availability of soil water, we calculated relative extractable soil water (REW), a simple drought stress index based on daily minimum and maximum soil water content (Bréda et al., 2006;Vilhar, 2016), with average amplitudes of 7.8% and 11.2% during the dry and wet season, respectively. All meteorological variables were measured every 15 s, stored as 10 min statistics (mean, minimum and maximum) in a DL16 Pro data logger (Thies Clima), and finally averaged to 30-min values.

| Soil N 2 O flux measurements
In addition to the EC measurements of N 2 O fluxes, we conducted monthly measurements of soil N 2 O fluxes during the period July 2017 to March 2018 using vented, static chambers (Hassler et al., 2017) within the footprint area of the EC tower ( Figure 2). A total of 16 chamber bases (polyvinyl chloride, 0.05 m 2 , inserted ~0.03 m into the soil) were deployed along four transects. Each transect captures different management zones along a fertilization gradient with increasing distance from the oil palm base. Management zones comprise: (i) fertilized palmcircle area within 0.8 m from the palm, (ii) unfertilized inter-row at 2.8 and 4.8 m from the palm, (iii) and frondstacked area where the pruned oil palm leaves are piled on every other inter-row. Area-weighting factors of the management zones were 18% (palm-circle), 33.5% (2.8 m inter-rows), 33.5% (4.8 m inter-rows), and 15% (frondstacked area). For the flux measurements, vented static, polyethylene enclosures (surface area: 0.05 m 2 , total volume: 12 L) were placed over the chamber bases and gas samples were taken 1, 11, 21 and 31 min after closure from each chamber (total 160 measurements) through a Luer lock. Sampling was performed between 9 a.m. and 3 p.m. local time. The gas samples were stored in pre-evacuated 12 mL Labco exetainers (Labco Limited) for 4 months, transported to Germany via air freight and analysed with a gas chromatograph equipped with electron capture detector (SRI 8610C; SRI Instruments Europe GmbH). In addition, standard gases were stored in the same exetainers for the same duration as the gas samples to check for possible leakage which we did not observe (Hassler et al., 2017). Soil N 2 O fluxes for each chamber surface were calculated from the linear increase of N 2 O concentration over the chamber closure period. We aggregated soil N 2 O fluxes to annual fluxes by linear interpolation between monthly measurements (Hassler et al., 2015(Hassler et al., , 2017. Emission factors were calculated based on the fraction of N applied as fertilizer and N emitted as N 2 O. Since no data on the exact timing and amount of N application were available, we used the annual rate of N fertilizer applied (174.4 kg N ha −1 year −1 ).

| Gap filling
We constrain our detailed analysis on N 2 O flux dynamics to the period from December 20, 2018 to April 30, 2019, when the EC N 2 O sensor operated without any major disruptions. Existing gaps in the N 2 O flux record were filled using the approach suggested by Nemitz et al. (2018): Gaps equal to or shorter than 2 h were linearly interpolated. Gaps longer than 2 h were filled using the diel average value of the corresponding 30-min flux. If the number of missing fluxes exceeded 17 per day, we did not perform N 2 O flux gap filling for that specific day. Gaps in the time series of CO 2 NEE and evapotranspiration were filled with marginal distribution sampling (Reichstein et al., 2005). Further, NEE was partitioned into its component fluxes gross primary productivity (GPP) and ecosystem respiration (R eco ) using the environmental response functions proposed by Falge et al. (2001). Both flux processing approaches are implemented in the REddyProc online EC data processing tool (Falge et al., 2001;Reichstein et al., 2005;Wutzler et al., 2018).

| Statistical analysis
All statistical analyses and graphing were performed with R (version 4.2.2, R Core Team, 2022). Pearson's correlation test was applied to study the influence of environmental and meteorological variables on N 2 O fluxes. Differences of chamber based N 2 O fluxes between the different management zones were evaluated using Turkey's Honest Significant Difference test. For all analyses, we considered statistical significance level of 0.05.
We investigated the temporal dynamics of N 2 O emissions using wavelet analysis that separately describes the time and frequency components of a time series (Chavez & Cazelles, 2019;Grinsted et al., 2004). Further, we applied wavelet coherence to address correlations between N 2 O emissions and possible drivers in the time-frequency domain (Cazelles et al., 2008). Wavelet analysis was performed in R using the package "biwavelet" (Gouhier et al., 2021). We tested for the correlation between N 2 O fluxes with ( To select valid correlations in the time-frequency domain, we filtered for significance using a >95% confidence level against red noise (Gouhier et al., 2021). For visualization, the filtered correlation patterns mapped in the time-frequency domain were stacked as described in Koebsch et al. (2015).
To study diel behaviour of N 2 O fluxes and their hysteresis to environmental variables, we used the complete wet season 2018/2019 data set and applied the following approach: First, we used CCF (Vio & Wamsteker, 2001) to assess the strength of correlation of diel N 2 O flux with ecosystem variables (GPP, R eco , LE), meteorological variables (PAR in , VPD, air temperature, H), soil variables (G, REW, WFPS, soil temperature) and atmospheric conditions [u*, TKE, (z − d)/L]. Since CCF is not able to answer whether there are possible causal relationships between time series (Dean & Dunsmuir, 2016), we performed, in a next step, Granger causality test (Granger, 1969), which provides additional information on the causality of the found correlations. Causality is here defined as correlation with a defined time lag between the assumed cause (meteorological or environmental) and effect (N 2 O flux) variable (Amornbunchornvej et al., 2021). This approach has been applied to explore ecosystem-atmosphere interactions from EC measurements (Cicuéndez et al., 2023;Dai et al., 2019;Detto et al., 2012;Díaz et al., 2022). Here, we use the term "Granger-causing" or "Granger-causality" referring to causal relationships defined by Granger (1969) and Amornbunchornvej et al. (2021). We evaluated variable time delays of the Granger causality tests' underlying vector autoregressive (VAR) models with Akaike information criterion (AIC) scores (Akaike, 1973) and selected the respective VAR with lowest AIC as input model for the Granger causality test (Bruns & Stern, 2019). For the respective hysteresis, correlation and Granger causality analyses, we used the R-packages "tseries" (Trapletti et al., 2020), "lmtest" (Hothorn et al., 2020) and "vars" (Pfaff & Stigler, 2018). For the analysis of wavelet coherence, CCF and Granger causality we used soil properties (WFPS, REW, soil temperature, soil moisture) at 5 cm depth.

| Meteorological and environmental conditions
The overall meteorological conditions at the study site were characterized less by seasonal than by stronger day-to-day variations (Figure 3a-d) exhibiting typical diel behaviour (Figure 7a,c). Over the entire study period, the mean daily air temperature was 26.5°C, with mean daily minimum and maximum air temperature of 22.8 and 29.8°C, respectively ( Figure 3a). VPD varied considerably between 0 and 1.33 kPa, with an average of 0.52 kPa (Figure 3a) and showed a strong correlation with air temperature (R 2 = 0.86, p < 0.01). Precipitation follows a distinct seasonal pattern (Figure 3b), with a dry season (May-October) and a wet season (November-April). Monthly accumulated precipitation <100 mm occurred only during F I G U R E 3 Meteorological and environmental conditions during the study period. (a) Daily average (semi-transparent) and 7-day moving average (solid line) of above-canopy air temperature and above-canopy vapor pressure deficit (VPD); (b) daily and monthly accumulated precipitation (no precipitation data for January 2019); (c) daily average (semi-transparent) and 7-day moving average (solid line) water filled pore space (WFPS) and relative extractable soil water (REW), and (d) above-canopy daytime average (6:30-18 h local time) incoming photosynthetically active radiation (PAR in ). Shaded areas in (a-d) indicate wet season. the dry season, with a minimum in July 2017 and 2018 (22.5 and 48.0 mm, respectively). November, however, was the wettest month (499.5 and 415.5 mm, in 2017 and 2018, respectively). Average potential evapotranspiration at the study site was slightly higher (0.30 mm h −1 ) during the dry season compared to the wet season (0.28 mm h −1 ). The higher frequency and intensity of precipitation during the wet season is only partially reflected in soil water conditions (Figure 3c), with slightly higher WFPS of 57.0 ± 4.0% compared to the dry season (54.0 ± 3.9%) and higher REW of 0.52 ± 0.11 during the wet season compared to the dry season (0.49 ± 0.14). Increased cloudiness over the wet season resulted in lower daytime average incoming PAR (791.6 ± 584.4 μmol m −2 s −1 ) compared to the dry season (856.6 ± 597.6 μmol m −2 s −1 ) (Table 1; Figure 3d).

| N 2 O flux characteristics
Annual emissions derived from EC measurements amounted to 0.32 ± 0.003 g N 2 O-N m −2 year −1 , which corresponds to a global warming potential of 149.85 ± 1.40 g CO 2 -equivalent m −2 year −1 ( Table 2). This resulted in an annual emission factor (emitted N 2 O-N to applied fertilizer-N) of 1.8%. During the period July 2017 to March 2018, when EC-and chamber measurements were conducted simultaneously, EC-based N 2 O emissions were 0.22 ± 0.008 g N 2 O-N m −2 year −1 (103.02 ± 3.75 CO 2equivalent m −2 year −1 ), with an emission factor of 1.3%. Chamber-based soil N 2 O emissions during the same period were ~27% lower compared to EC-derived N 2 O emissions (p < 0.05) with, on average, 0.16 ± 0.047 g N 2 O-N m −2 year −1 (74.93 ± 23.41 CO 2 -equivalent m −2 year −1 ), which resulted in an emission factor of 0.92% (Table 3). In the palm-circle area, emissions showed a small peak in December 2017 while in the inter-rows, emissions showed a strong peak in October 2017. Yet over the entire study period, soil N 2 O fluxes of the different management zones (palm-circle at 0.8 m, inter-row at 2.8 and 4.8 m distance from the palm, and frond-stacked area) were not significantly different (Tukey-test at the 5% level of significance) (Figure 4).

| Spectral analysis of N 2 O fluxes
The temporal variation in N 2 O fluxes is characterized by a relatively uniform distribution of spectral energy over the time domain under consideration (1 h to 40 days, Figure 5). Hence, the spectral features of N 2 O fluxes are less pronounced in comparison to the studied environmental variables, in particular the fluxes of LE, GPP and R eco . The marked spectral energy of N 2 O fluxes at the high frequency range (i.e. at the time scale of few hours) corresponds to the spectral pattern of turbulence variables such as (z − d)/L, TKE and u*. Spectral peaks of N 2 O fluxes occur on a scale of 24 h and 10 days, approximately. The observed peak in the spectra of N 2 O fluxes on a scale of 24 h corresponds well with most meteorological and environmental variables while the 12-and 6-h peak in the spectra of N 2 O fluxes, in turn, may relate to a method-inherent artefact that smears spectral energy into smaller time scales. High N 2 O flux variation at a scale of approximate 10-12 days is generally not clearly reflected in the variations of the other studied variables but soil temperature and WFPS exhibit an increase in temporal variation at scales of 2 days, and larger, that could be related with the relatively high variation of N 2 O fluxes at these temporal frequencies.
Throughout most of the 2018/19 wet season, short-term N 2 O flux fluctuations at ~1-2 days were in phase with most of the studied environmental variables ( Figure 6). Hence, from this analysis we could not assign any particular variable as the most dominant control for N 2 O fluxes at the diel scale. Further, at a diel scale, variations in N 2 O fluxes and the studied variables were intermittent and limited to Note: Mean (±SE) and annual soil N 2 O fluxes for different management zones (n = 4 transects, Figure 2) located within the EC footprint. Mean soil N 2 O fluxes span the period from June 2017 to March 2018 while annual emissions were derived using linear interpolation between monthly measurements. Global warming potential is calculated as g N 2 O m −2 year −1 × 298 CO 2 -eq. on a 100-years time horizon (Forster et al., 2007). Area-weighting factors of the management zones were 18% (palm-circle), 33.5% (2.8 m inter-rows), 33.5% (4.8 m inter-rows) and 15% (frond-stacked area).
Abbreviation: EC, eddy covariance.  Table 2). During the day, we observed moderate to high correlation (CCF >0.65) of N 2 O fluxes with GPP, H, G, u*, LE, PAR in , and TKE ( Figure 8). The latter four (Figure 8c,d,l,m) showed high Granger causality (p < 0.05) indicating a significant correlation between cause (LE, PAR in , TKE, u*) and effect (N 2 O flux) and relatively short response lag of N 2 O fluxes. For GPP and G, we also observed relatively small hysteresis area with N 2 O fluxes but the correlation was less pronounced and Granger causality was not significant (p > 0.1) (Figure 8a,h). During the night, only R eco and soil temperature showed moderate correlation with N 2 O fluxes (CCF = −0.56 and −0.60, respectively) and high Granger causality (p = 0.05 and 0.13, respectively) (Figure 8b,k). The high CCF and high Granger causality of N 2 O fluxes with daytime GPP, LE, and PAR in , and nocturnal soil temperature and R eco suggests that oil palm photosynthesis and respiration may be closely connected with the observed N 2 O emissions. In addition, N 2 O fluxes are also impacted by the strength of daytime atmospheric turbulent transport (u*, TKE).

| N 2 O flux dynamics and pathways
We observed a substantial difference between chamberand EC-based N 2 O emissions (Tables 2 and 3 measurements are comparable with those reported from other studies in oil palm (Akhir et al., 2015;Melling et al., 2007), we observed only one N 2 O emission peak in October 2017 (Figure 2b) and therefore may have missed pulse N 2 O emissions following N-fertilizer application (Oktarita et al., 2017;Veldkamp et al., 1998). Important soil-and climate-related processes contributing to N 2 O emissions may also remain undetected (Butterbach-Bahl et al., 2013). At our study site, Hassler et al. (2017) showed peak N 2 O fluxes with 5-21 days after fertilization. Nevertheless, the same authors conclude that such fertilizer-induced N 2 O emissions contribute only 6% to the annual fluxes of our study site. The observed difference between chamber-and EC-based N 2 O emissions may therefore not only be related to uncertainties in chamberbased N 2 O emissions, suggesting that chamber-based soil N 2 O measurements may miss substantial emission pathways, for example via plant-mediated transport via the oil palm.
The detailed analysis of N 2 O fluxes revealed a clear diel pattern, with peak fluxes at noon. Such patterns have also been reported in other ecosystems: Das et al. (2012) and Liang et al. (2018) relate daytime N 2 O emission patterns of pastures to changes in soil temperature, while Keane et al. (2017) found strong correlation of daytime N 2 O emissions with PAR for oilseed rape. It has also been shown from other environments that N 2 O produced in the soil is taken up by roots, transported into above-ground plant tissues and finally released into the atmosphere via lenticels or stomata, or N 2 O is consumed along the stem of trees (Iddris et al., 2020;Machacova et al., 2016;Wen et al., 2017). Our analysis in the time-frequency domain showed that correlations shifted over the course of the day, with variables related to photosynthesis, such as GPP, LE and PAR in , and TKE being associated with N 2 O fluxes at daytime (Figure 8a,c,d,m) and R eco and soil temperature being associated with N 2 O fluxes at night (Figure 8b,k). Medium emission states of N 2 O at daytime were also related to high  (Koebsch et al., 2015). The thin grey line represents the cone of influence that defines the region beyond which results are influenced by edge effects. GPP, gross primary productivity; LE, latent heat; PAR in , incoming photosynthetically active radiation; R eco , ecosystem respiration; VPD, vapour pressure deficit.
values of LE (Figure 8c). The observed strong causal relationships of daytime N 2 O fluxes with variables related to NEE indicate that N 2 O emissions might be coupled with the diel variation in oil palm physiological activity. The relatively small hysteresis of N 2 O fluxes with GPP, LE, and PAR in and the peak of diurnal N 2 O emission around noon further support this hypothesis (Figure 8a,c,d).
During night, N 2 O emissions correlated with respiration and soil temperature variation, though patterns were less pronounced due to the cessation of emissions and stomatal transport. There is evidence that oil palm and other plant species refill their stem water reservoir during night to sustain high rates of transpiration during daytime photosynthesis Sperling et al., 2015;Yang et al., 2012). Nonetheless, this nocturnal refilling mechanism might contribute to soil N 2 O uptake and short-term storage of N 2 O in the oil palm and therefore, direct nocturnal N 2 O emission via the soil could be small.
Soil moisture content at our study site was generally low ( Table 1). The WFPS range of 50%-60% (Figure 3c) favors for N 2 O production from both nitrification and denitrification (Davidson et al., 2000). In addition, we observe a distinct dry and wet season, as well as synoptic-scale weather changes that could impose low-frequency oscillations of N 2 O fluxes on the scale of several days to weeks. This is also supported by the significant synchronicity between WFPS and N 2 O fluxes ( Figure 5c). Oil palm, with its high photosynthetic efficiency (Apichatmeta et al., 2017), shows distinct responses to drought, such as partial stomata closure, decreased maximum rate of photosynthesis, and increased water use efficiency (Bakoumé et al., 2013;Paterson et al., 2013;Stiegler et al., 2019). During dry periods, a reduction in GPP and stomatal water loss may therefore decrease N 2 O emissions. Contrary, when there is no pronounced limitation in water, these favorable conditions for soil bacteria and oil palm photosynthesis may contribute to the observed higher N 2 O emissions compared to the dry season as N 2 O production in the soil diffuses as dissolved gas in water and is carried by transpirational pull along the stem. Also, short-term increase in soil moisture after a dry period may favor the burst of soil microbial activity, stimulating N 2 O emissions (Donoso et al., 1993;Drewer et al., 2020). We did not observe pulse N 2 O emissions after precipitation, probably because the ecosystem-scale EC measurements cannot resolve the subtle differences in N 2 O source strength found in this study ( Figure S3) or the temporal resolution of the soil chamber measurements of 1 month was too coarse. Also, the seasonal change in soil water content was relatively small and extensive dry periods did not occur during our study period. Further, the emission of N 2 O during soil wetting may be hampered by the oil palm via its nocturnal water refilling metabolism and related short-term storage of N 2 O in the oil palm.

| N 2 O emission in oil palm
Given the few studies conducted, N 2 O fluxes in oil palm plantations are highly variable in space and time, and dependent on plantation management practices (e.g., amounts of fertilizers; Hassler et al., 2017), plantation age Pardon, Bessou, Saint-Geours, et al., 2016) (e.g., as it loses or attains steady low organic matter levels with age, with highest N 2 O emissions observed at juvenile plantation stage), plantation topography and adjacent land-use type (e.g., movement of surface and sub-surface water and related nutrient transport) (Drewer et al., 2021;Sakata et al., 2016), soil type (Sakata et al., 2015), temperature and precipitation (Pardon, Bessou, Saint-Geours, et al., 2016). Our study site, with its mineral soil and relatively low soil water content, has similar N 2 O emissions compared to other oil palm plantations located on mineral soils (Sakata et al., 2015) but substantially lower N 2 O emissions and N 2 O emission factors compared to oil palm plantations established on drained peat soils and wetlands (Chaddy et al., 2019;Oktarita et al., 2017;Rahman et al., 2019;Sakata et al., 2015). This highlights significant mitigation potential of oil crop production systems through appropriate site selection confined to mineral soils. Based on ECmeasurements, the N 2 O emission factor accounts for 1.8%, which was in good agreement with the IPCC Tier 1 factor of 1.6% (IPCC, 2019; Mathivanan et al., 2021). When estimated with chamber measurements, the N 2 O emission factor accounts for only 0.93%, which was well below the IPCC reference. N 2 O fluxes from our chamber-based measurements are smaller to previous chamber-based N 2 O flux measurements in the same large-scale plantation (Hassler et al., 2017) but higher compared to those reported from smallholder plantations in Jambi province and from commercial large-scale plantations in Malaysia (Aini et al., 2015;Fowler et al., 2011;Melling et al., 2007). Compared to natural forest systems in Jambi province (Hassler et al., 2017;Ishizuka et al., 2005), the studied oil palm plantation showed up to 77% higher N 2 O emissions, making it an important anthropogenic source of N 2 O. Adequate plantation management including the optimization of the N fertilization regime are therefore crucial to reduce the plantations' global warming potential. In the same oil palm plantation, it was shown that reducing fertilizer inputs, combined with mechanical weeding, promote biodiversity, ecosystem multifunctionality, maintain high yield and increase profit as compared to the conventional high fertilizer inputs with herbicide application (Iddris et al., 2023).

| Fertilizer effects
For both EC-and chamber-based N 2 O flux measurements, monthly average N 2 O fluxes, their temporal development and respective maxima in October 2017 (chamber measurements) and October 2018 (EC measurements) are in good coherence with the reported fertilization schedule of the plantation owner. However, we did not observe any fertilizer-induced pulse emissions of N 2 O as they might occur directly after fertilizer application (Chaddy et al., 2019;Cowan et al., 2020), neither with the closed chamber nor with the EC approach. Although the footprint area of the EC-tower covers the spatially different management zones (Figure 2; Table 3), the EC-system cannot resolve the subtle differences in N 2 O source strength found in this study ( Figure S3). Ecosystem-scale N 2 O flux represent a composite flux of soil and aboveground vegetation, and time lag responses of N 2 O flux peaks. Chamberbased N 2 O flux measurements were not performed during or shortly after fertilization events but they showed emission peaks in the oil palm inter-rows.

| OUTLOOK AND CONCLUSION
N 2 O emissions from agricultural systems such as oil palm, contribute considerably to the global N 2 O emissions (Ussiri & Lal, 2013). Given the high fertilization rates in industrial oil palm plantations (five times higher than smallholder oil palm plantations; Hassler et al., 2017;Iddris et al., 2023), and the current and predicted future land transformations related to oil palm expansion in tropical regions around the globe, it is crucial to determine oil palm N 2 O GHG exchange, biogeochemical, meteorological and environmental drivers of N 2 O, fluxes, as well as adaptation strategies to reduce N 2 O emissions. In our study, the combination of point-scale chamber and ecosystem-scale EC N 2 O flux measurements indicate that oil palms and its dynamic gas exchange with the atmosphere may play an important role. However, this role in the N 2 O budget of oil palm plantations is yet widely unexplored. In this study, we showed that N 2 O emissions in an oil palm plantation might have been underestimated if derived only from chamber measurements. Additional research is needed to provide further insights on key factors controlling N 2 O production and N 2 O emissions in these systems to derive accurate N 2 O budgets. Chamberand EC-based N 2 O flux measurements in oil palm are still scarce and to our best knowledge, no long-term measurements yet exist. With regard to the long rotation cycles of oil palm plantations (~25 years), N 2 O fluxes may vary substantially over such time spans. N 2 O emissions from our study site on mineral soils are lower compared to organic soils, highlighting appropriate site selection to mitigate negative effects of oil palm cultivation. It is therefore crucial to promote research on N 2 O emissions from oil palm at different soil types, ages and over longer time spans. Modelling, for example CLM (Fan et al., 2015), can benefit from such measurements by incorporating them with meteorological, environmental and physiological measurements to strengthen estimates of N 2 O emissions and the overall understanding of biogeochemical cycles of N (Liang et al., 2018).