Soil fertility controls soil – atmosphere carbon dioxide and methane fluxes in a tropical landscape converted from lowland forest to rubber and oil palm plantations

Expansion of palm oil and rubber production, for which global demand is increasing, causes rapid deforestation in Sumatra, Indonesia, and is expected to continue in the next decades. Our study aimed to (1) quantify changes in soil CO2 and CH4 fluxes with land-use change and (2) determine their controlling factors. In Jambi Province, Sumatra, we selected two landscapes on heavily weathered soils that differ mainly in texture: loam and clay Acrisol soils. In each landscape, we investigated the reference land-use types (forest and secondary forest with regenerating rubber) and the converted land-use types (rubber, 7–17 years old, and oil palm plantations, 9–16 years old). We measured soil CO2 and CH4 fluxes monthly from December 2012 to December 2013. Annual soil CO2 fluxes from the reference land-use types were correlated with soil fertility: low extractable phosphorus (P) coincided with high annual CO2 fluxes from the loam Acrisol soil that had lower fertility than the clay Acrisol soil (P < 0.05). Soil CO2 fluxes from the oil palm (107.2 to 115.7 mg C m h) decreased compared to the other land-use types (between 178.7 and 195.9 mg C m h; P < 0.01). Across land-use types, annual CO2 fluxes were positively correlated with soil organic carbon (C) and negatively correlated with N signatures, extractable P and base saturation. This suggests that the reduced soil CO2 fluxes from oil palm were the result of strongly decomposed soil organic matter and reduced soil C stocks due to reduced litter input as well as being due to a possible reduction in C allocation to roots due to improved soil fertility from liming and P fertilization in these plantations. Soil CH4 uptake in the reference landuse types was negatively correlated with net nitrogen (N) mineralization and soil mineral N, suggesting N limitation of CH4 uptake, and positively correlated with exchangeable aluminum (Al), indicating a decrease in methanotrophic activity at high Al saturation. Reduction in soil CH4 uptake in the converted land-use types (ranging from−3.0 to−14.9 μg C m h) compared to the reference land-use types (ranging from −20.8 to −40.3 μg C m h; P < 0.01) was due to a decrease in soil N availability in the converted land-use types. Our study shows for the first time that differences in soil fertility control the soil–atmosphere exchange of CO2 and CH4 in a tropical landscape, a mechanism that we were able to detect by conducting this study on the landscape scale.


Introduction
Oil palm (Elaeis guineensis) and rubber (Hevea brasiliensis) are two of the fastest-expanding tree cash crops in the tropics (Clay, 2013).Global oil palm production has quintupled from 1990 to 2013 and is currently grown on an estimated area of 17 million hectare (Mha) (Food and Agricultural Organiza-E.Hassler et al.: Soil fertility controls soil-atmosphere carbon dioxide tion, 2014).Indonesia contributes nearly half of the global palm oil production (Food and Agricultural Organization, 2014) and is planning to double its production in the coming decade (Carlson et al., 2013).Similarly, rubber is grown on around 10 Mha globally, and Indonesia is the second largest rubber producer (Food and Agricultural Organization, 2014).Most Indonesian oil palm and rubber production is located in Sumatra (Indonesian Ministry of Agriculture, 2014), where the conversion of lowland rainforest to plantations has been widespread (Laumonier et al., 2010).It has been estimated that plantation establishment has caused a loss of 7.5 Mha of Sumatran natural forest in the last 2 decades (1990-2010) (Margono et al., 2012), and future expansion will probably be at the expense of large areas of tropical forest, unless a properly planned and spatially explicit development strategy is implemented (Koh and Ghazoul, 2010).
Although the majority of remaining lowland tropical forests are located on nutrient-poor, heavily weathered soils, these ecosystems are among the most productive worldwide and contain globally significant above-and belowground carbon stocks.The high ecosystem productivity is possible despite the nutrient-poor soils because of efficient cycling of rock-derived nutrients (phosphorus (P) and base cations) between vegetation and soil and also high soil nitrogen (N) availability caused by biological N fixation (Hedin et al., 2009).Conversion of tropical forest to agricultural land-use systems does not only decrease biodiversity and contribute to climate change (Danielsen et al., 2009) but also alters soil fertility and soil physical properties in the newly established land-use systems (Dechert et al., 2004;Klinge et al., 2004).Burning of slashed vegetation is typically part of forest conversion, releasing large amounts of nutrients previously bound in the vegetation.A considerable part of these nutrients ends up in the soil but is susceptible to losses (through leaching and gaseous emission), which are especially high in the earlier years of crop establishment and decrease with time (Klinge et al., 2004).Furthermore, forest conversion is often associated with increases in soil bulk density.These dynamic changes in soil fertility and soil bulk density following forest conversion do not only affect agricultural production but also the soil-atmosphere exchange of trace gases like carbon dioxide (CO 2 ) and methane (CH 4 ) since their production, consumption and exchange are directly related to soil fertility and soil bulk density (Keller et al., 1993;Veldkamp et al., 2008).
Globally, soils are the largest natural source of CO 2 (IPCC, 2007), which is released during respiration processes of microbial communities and roots (Raich and Schlesinger, 1992).While the important proximal controllers of soil CO 2 fluxes are soil temperature and moisture, it has been demonstrated that other distal regulators such as vegetation type and soil physical and biochemical properties (e.g., bulk density, texture, pH, carbon stocks) also affect soil CO 2 fluxes (Raich and Schlesinger, 1992).Soils also play a dominant role in the production and consumption of CH 4 , a greenhouse gas with a global warming potential 23 times that of CO 2 over a 100-year time period (IPCC, 2007).In soils, CH 4 can be produced during anaerobic decomposition by methanogenic archaea, while CH 4 can also be consumed by methanotrophic bacteria which are able to utilize CH 4 as an energy source.Whether net consumption or net emission of CH 4 occurs at the soil surface depends on the balance between production and consumption in the soil.For soil CH 4 fluxes, the proximal controllers are soil moisture, gas diffusivity and temperature, while other distal regulators include microbial activity, N availability and aluminum toxicity (Verchot et al., 2000;Tamai et al., 2003;Bodelier and Laanbroek, 2004;Veldkamp et al., 2013).
In tropical lowland forest landscapes with heavily weathered soils, much of the spatial variability in trace gas fluxes appears to be related to soil texture (Sotta et al., 2006).In the Brazilian Amazon, lowland forests on Acrisol and Ferralsol soils display high soil CO 2 emissions with large variations among sites that relate to soil texture: soils with sandy loam to sandy clay loam texture had 21-36 % higher CO 2 emissions than soils with clay texture (Keller et al., 2005;Sotta et al., 2006).Moreover, although well-drained soils in tropical lowland forests generally act as a sink for CH 4 (Keller and Reiners, 1994;Verchot et al., 2000;Veldkamp et al., 2013), their differences in CH 4 uptake are explicable by their differences in soil texture.In a review of 16 tropical lowland forests, the only factor correlating annual CH 4 fluxes with site characteristics was a significant positive correlation with clay contents, indicating that the higher the clay content the lower the CH 4 uptake (Veldkamp et al., 2013).
Since much of the original forest in our study area have been converted to oil palm and rubber plantations, the management practices in these land-use types added important factors that influence soil CO 2 and CH 4 fluxes from these converted landscapes.Earlier studies have shown that forest conversion to agricultural land-use types in the tropics lead to considerable changes in soil CO 2 fluxes, which were related to changes in belowground C allocation (Davidson et al., 2000;Salimon et al., 2004), carbon quality (Werner et al., 2006), living fine-root biomass and litter input (Ishizuka et al., 2002;Sheng et al., 2010).Conversion of tropical forest to agricultural land use causes a reduction in soil CH 4 uptake or even turns the soil into a source of CH 4 .Often this trend is explained by soil compaction, which leads to reduced gas diffusivity and accordingly limits aerobic CH 4 oxidation while enhancing anaerobic CH 4 production (Keller et al., 1993;Veldkamp et al., 2008).Changes in N availability may also play a role since CH 4 uptake may be N limited (Bodelier and Laanbroek, 2004;Veldkamp et al., 2013) and high concentrations of ammonium (NH + 4 , e.g., from fertilization) can inhibit CH 4 oxidation (Veldkamp et al., 2001;Werner et al., 2006).Finally, termites are known to produce CH 4 , and their presence may also affect the balance between production and consumption of CH 4 (Seiler et al., 1984).
Although Sumatra, Indonesia, represents a hot spot of land-use change, especially for the establishment of rubber and oil palm plantations, how this affects soil CO 2 and CH 4 fluxes remains highly uncertain for the following reasons: (1) most studies relating land-use change to trace gas emissions have been conducted in South and Central America (Keller and Reiners, 1994;Davidson et al., 2000;Verchot et al., 2000;Veldkamp et al., 2001;Salimon et al., 2004) and only few studies were conducted in southeast Asia (Ishizuka et al., 2002;Veldkamp et al., 2008); (2) most studies have focused on forest conversion to traditional land-use types, such as maize, pastures, slash-and-burn agriculture, cacao and coffee, and less on the rapidly expanding tree cash crops such as rubber and oil palm; (3) the few studies that reported CO 2 and CH 4 fluxes from oil palm plantations were conducted on peat soils (Melling et al., 2005a, b), whereas the studies conducted on mineral soils, where most of the rubber and oil palm plantations are located, were either conducted without spatial replication, covered only short periods of measurements (Ishizuka et al., 2002;Adachi et al., 2005;Werner et al., 2006) or measured only once (Ishizuka et al., 2005).It is imperative that better information becomes available on trace gas fluxes from these economically important and rapidly expanding rubber and oil palm plantations.Whether palmoil-based biofuel does indeed have environmental advantages compared to fossil fuel depends, among many factors, on the greenhouse gas balance during oil palm fruit production.
In the present study, our aims were to (1) quantify changes in soil-atmosphere fluxes of CO 2 and CH 4 with land-use change and (2) determine their controlling factors in a converted lowland landscape in Sumatra, Indonesia.Soilatmosphere fluxes of CO 2 and CH 4 were measured in forest and secondary forest with regenerating rubber (hereafter called jungle rubber, which is a more traditional rubber agroforestry system; Gouyon et al., 1993) as reference land-use types and the converted land-use types of monoculture rubber and oil palm plantations.Our study was designed to cover these four land-use types in each of the two landscapes on highly weathered soils that differed mainly in texture: clay and loam Acrisol soils.We tested the following hypotheses: (1) soil CO 2 emissions and CH 4 uptake will be higher in loam than in clay Acrisol soils, and (2) soil CO 2 fluxes and CH 4 uptake rates will be higher in the reference land-use types (forest and jungle rubber) than in the converted landuse types (rubber and oil palm plantations).Here, we present the first spatially replicated study with a full year of measurements that investigates soil CO 2 and CH 4 fluxes from conversion of forest or jungle rubber to rubber and oil palm plantations on mineral soils.We also evaluate the effect of management intensity since we compare rubber plantations without fertilizer inputs with fertilized oil palm plantations.Our results will be a critical contribution to trace gas lifecycle assessment of rubber and palm oil at the production stage.

Study area and experimental design
The study area is located in the lowlands (35-95 m above sea level) of Jambi province, Sumatra, Indonesia.In the past 2 decades, forest cover in Jambi province decreased by 1.14 Mha, which was about 40 % of the forest cover in 1990 (Margono et al., 2012).The climate is humid tropical with a mean annual air temperature of 26.7 ± 0.1 • C and a mean annual precipitation of 2235 ± 385 mm (1991-2011; data relating to Jambi Sultan Thaha Airport from the Indonesian Meteorological, Climatological and Geophysical Agency).The dry season is usually from May to September and the rainy season occurs from October to April.In 2013, during our study period, the wet season lasted slightly longer, while a drier period was detected between mid-June and the end of October.During this dry period, rainfall was reduced by 35-57 % compared to the wetter months during which rainfall was 333-362 mm per month.
We selected two landscapes on heavily weathered soils that differed mainly in texture: loam Acrisol soil (36 ± 6 % sand, 32 ± 4 % silt and 32 ± 2 % clay in the top 0.5 m) and clay Acrisol soil (26 ± 6 % sand, 29 ± 3 % silt and 45 ± 4 % clay in the top 0.5 m).This textural difference led to differences in soil fertility: forest sites in the clay Acrisol soil had a higher base saturation, Bray-extractable P and lower Al saturation compared to those in the loam Acrisol soil (P ≤ 0.01 to 0.04; Appendix Table A1; Allen et al., 2015).Detailed soil physical and biochemical characteristics from our study sites were measured by Allen et al. (2015) and are summarized in Table A1.Acrisol soils cover about 50 % of the land area in Sumatra and about one third of Indonesia (FAO et al., 2009).The clay Acrisol landscape was located about 160 km southwest of Jambi City between 01.94In each landscape, we studied four land-use types: lowland forest, jungle rubber, and smallholder monoculture plantations of rubber and oil palm.In Jambi province, the smallholder rubber and oil palm plantations were established after clearing and burning either the forest (often partially logged) or jungle rubber (based on interviews conducted by Euler et al.; unpublished data).Thus, in our study the lowland forest and jungle rubber served as the reference land-use types, representing the baseline conditions with which we compared the rubber and oil palm plantations.For each of the four land-use types within each landscape, we selected four replicate plots (50 m × 50 m each with a minimum distance of 200 m between plots), totalling 32 plots that were all located on relatively flat, well-drained positions in the landscape.Additional information on tree species composition, tree density, tree height, basal area, and plantation age of these plots is reported in Table A2.Within each plot, we established a 10 × 10 grid which was used to select four randomly nested subplots (5 m × 5 m each) that were at least 5 m from the plot's border.In each subplot, we placed one permanent chamber base at random to measure soil trace gas fluxes.
This approach of comparing soil CO 2 and CH 4 fluxes from the converted land-use types to that from the reference landuse types in order to assess the effects of land-cover change has the implicit assumption that before land-use conversion, soil characteristics were comparable.We tested this assumption by comparing the land-use-independent soil characteristics, i.e., clay content in 0.50-2.00m depth, among land-use types within each landscape.Since there were no significant differences in clay contents between the reference and converted land-use types at these depths (Table A1; Allen et al., 2015), we deduced that the sites within each landscape had previously similar soil characteristics and that differences in trace gas fluxes can be attributed to the changes in land-use type and its associated management practices.
Since all the plantations were managed by smallholders, management practices of rubber and oil palm were diverse.The following information on management practices was based on our interviews of the smallholders.In 2013, oil palm and rubber plantations were weeded once or twice a year either manually or using herbicide (most commonly 2-5 L Gramaxone ® or Roundup ® ha −1 yr −1 ).Oil palm plantations were fertilized, whereas rubber plantations were not.Oil palm plantations in the clay Acrisol landscape were fertilized only once during the rainy season, whereas those in the loam Acrisol landscape had a second fertilizer application in the dry season.The most commonly used fertilizers were NPK complete fertilizer (i.e., Phonska, Mahkota), potassium chloride (KCl) and urea.Plantations fertilized once received about 300 kg NPK fertilizer ha −1 yr −1 and plantations fertilized twice received about 550 kg NPK fertilizer ha −1 yr −1 .In terms of nutrient elements added, these rates were equivalent to 48-88 kg N ha −1 yr −1 , 21-38 kg P ha −1 yr −1 and 40-73 kg K ha −1 yr −1 .Additionally, three of the smallholders applied 157 and 143 kg KCl-K ha −1 yr −1 and two of the smallholders applied 138 kg urea-N ha −1 yr −1 .One of the smallholders also applied lime in 2013 at about 200 kg dolomite ha −1 yr −1 .Smallholders typically applied the fertilizer around the oil palm tree at about 0.8-1 m distance from the tree base.Additionally, senescing fronds were regularly cut and piled on the interrows of an oil palm plantation, typically in the middle of the 9 m distance between rows of oil palms.Oil palm fruits were harvested every 2 weeks, whereas the latex of the rubber and jungle rubber was collected weekly.
In addition to the monthly flux measurements conducted at the smallholder farms described above, we simulated fertilizations and conducted more frequent measurements (6 to 11 times) during 3 to 8 weeks following fertilization as the monthly sampling may have missed the short-term effect of fertilization on soil CO 2 and CH 4 fluxes.We chose three plots of oil palm plantations in each of the two landscapes, and in each plot we selected three trees separated by an interrow distance of 9 m (in total, 18 oil palm trees).At 0.8 m distance from the base of each tree, we applied the fertilizer manually to the area within 0.2 m around the tree using the same rate that smallholders applied to these oil palm plantations (i.e., equivalent to 2 kg fertilizer per tree, based on 300 kg NPK fertilizer ha −1 divided by 134-140 trees ha −1 ; Table A2).We used the same fertilizer forms that smallholders applied, i.e., NPK complete fertilizer in the clay Acrisol landscape and a combination of KCl, ammonium sulfate and NPK complete fertilizer in the loam Acrisol landscape.One chamber base was placed at 0.3 m distance from the tree base (chamber location a); another chamber base, to which fertilizer was applied, was placed at 0.8 m distance from the tree (chamber location b); and a third chamber was placed at 4-4.5 m distance from the tree and served as a reference chamber without direct fertilizer application (chamber location c).In the clay Acrisol landscape, measurements in the three oil palm plots were done from mid-October to mid-December 2013, mid-February to mid-March 2014, and mid-February to mid-April 2013.In the loam Acrisol landscape, measurements were done from the end of October 2013 to mid-December 2013, mid-January to mid-March 2014, and mid-March to the start of April 2014.Shorter intervals of sampling days (Appendix Fig. B1) were conducted right after the fertilizer application.

CO 2 and CH 4 flux measurement
Soil CO 2 and CH 4 fluxes were measured monthly from December 2012 to December 2013, using static vented chambers.We are aware that the use of static vented chambers for CO 2 and CH 4 flux measurements could have possible sources of error due to changes in diffusion gradients during chamber closure.We have however adapted the design of our chambers and the flux calculation to minimize, if not avoid, these possible errors (see below).A clear advantage of the static vented chamber method compared to other methods is that it remains the only operational method that can be used to measure trace gas fluxes at a large number of plots (in our case 32) spread over a large area with regular measurements within a year.Furthermore, it also yields information on short-distance spatial variability of trace gas fluxes, which cannot be quantified by more integrative tower-based methods.Our chamber bases were made of polyvinyl chloride (0.05 m 2 area) and inserted ∼ 0.03 m into the soil.We placed a permanent chamber base at random in each of the four subplots per replicate plot 1 month before the first measurement started.Since the area occupied by piled fronds or applied with fertilizer in oil palm plantations was relatively small, none of these randomly placed chamber bases (range of distance to the tree base was 1.8-5 m) happened to be located in such an area.During sampling, the chamber bases were covered with polyethylene hoods (0.27 m total chamber height and 12 L total volume) equipped with a Luer-lock sampling port and a vent for pressure equilibrium.Four gas samples (30 mL each) were removed at 1, 11, 21 and 31 min after chamber closure using a plastic syringe connected to the Luer-lock port.Immediately after sampling, gas samples were stored with overpressure in pre-evacuated 12 mL Labco Exetainers ® (Labco Limited, Lampeter, UK) with rubber septa that were only used once.Our group has tested these exetainers for an extended period of sample storage (e.g., up to 6 months) and air transport by storing and transporting standard gases of known concentrations in overpressure, and these exetainers were proven to be leak proof.Gas samples in exetainers were transported to Germany by air every 3-4 months and were analyzed upon arrival using a gas chromatograph (GC 6000 Vega Series 2, Carlo Erba Instruments, Milan, Italy, with an ASPEC autosampler, Gilson SAS, Villiers, Le Bel, France), equipped with a flame ionization detector and an electron capture detector.Three standard gases were used for calibration with concentrations from 350 to 5000 ppm for CO 2 and 1000 to 5000 ppb for CH 4 (Deuste Steininger GmbH, Mühlhausen, Germany).The software Probe66 version 1.3 (Messwert GmbH, Göttingen, Germany) was used to determine CH 4 and CO 2 sample concentrations by comparing integrated peak areas with those of the standard gases.Fluxes were calculated from the concentration change over the duration of chamber closure and adjusted with actual air temperature and pressure measured at the time of sampling.The linearity of increase in CO 2 concentrations with the duration of chamber closure (R 2 ≥ 0.98) was checked for each chamber measurement, and in a few cases where concentration curved at the last sampling time, we excluded the last data point and calculated the fluxes based on the linear increase in concentrations during the first three samplings.The majority of the measurements showed a linear change in CH 4 concentrations with the duration of chamber closure.There were a few measurements when changes in CH 4 concentrations with time of chamber closure were small, mostly when net CH 4 uptake was low; in such cases, the calculated CH 4 flux using linear regression was not significantly different from 0. These fluxes were, however, retained in the statistical analyses to avoid bias by excluding low CH 4 fluxes or by assuming that these fluxes were 0. Assuming constant flux rates per day, annual soil CO 2 and CH 4 fluxes from the monthly sampling and total fluxes during a fertilization event (see below) were calculated by adopting the trapezoidal rule on time intervals between measured flux rates, similar to the method we em-ployed in our earlier studies (e.g., Koehler et al., 2009;Veldkamp et al., 2013)

Auxiliary measurements
Soil temperature, moisture and mineral N content were measured with each trace gas measurement.Soil temperature was determined in the top 0.05 m depth using a GMH 1170 digital thermometer (Greisinger electronic GmbH, Regenstauf, Germany).Soil samples for mineral N content were also taken from the top 0.05 m at approximately 1 m distance from each chamber with four soil samples per plot.These four soil samples were pooled to represent each replicate plot on each day of measurement, and roots, leaves and twigs were manually removed.For the fertilization experiment, soil samples taken near each chamber location (a, b and c) were processed separately.Since even short storage of disturbed soil samples can affect mineral N concentrations (Arnold et al., 2008), extraction was done in the field immediately following soil sampling.A soil sample was added to a prepared 250 mL plastic bottle containing 150 mL of 0.5 mol L −1 K 2 SO 4 (approximately 1 : 3 ratio of fresh soil to extractant volume) and transported to the field station.At the field station, samples were shaken for 1 h, filtered through prewashed (with 0.5 mol L −1 K 2 SO 4 ) filter papers (Whatman, GE Healthcare Life Sciences, 4 µm nominal pore size), and the filtrate was immediately stored in a freezer.The remaining field-moist soil samples were stored in plastic bags and gravimetric moisture content was determined at the field station, for which 50-100 g of fresh soil was dried at 105 • C for at least 1 day.The gravimetric moisture content was used to calculate the dry mass of the soil extracted for mineral N. Frozen extracts were transported by air to Germany to ensure that they stayed frozen throughout the transport until analysis.At our laboratory at the University of Göttingen, Germany, NH + 4 and NO − 3 concentrations in the extracts were analyzed using continuous flow injection colorimetry (SEAL Analytical AA3, SEAL Analytical GmbH, Norderstedt, Germany).NH + 4 was determined by salicylate and dicloroisocyanuric acid reaction (Autoanalyzer Method G-102-93), and NO − 3 was determined by the cadmium reduction method with an NH 4 Cl buffer (Autoanalyzer Method G-254-02).Soil water content was expressed as water-filled pore space (WFPS), calculated using a particle density of 2.65 g cm −3 for mineral soil and the measured bulk densities at our study sites (Table A1).

Statistical analysis
All statistical analyses of the monthly measurements of soil CO 2 and CH 4 fluxes were conducted using the means of the four chambers (or subplots) that represent each replicate plot on a given sampling day.Data were checked for normal distribution (using Shapiro-Wilk tests), and if necessary, a logarithmic (for CO 2 , CH 4 , and mineral N) or square www.biogeosciences.net/12/5831/2015/Biogeosciences, 12, 5831-5852, 2015 root (for WFPS) transformation was used.For our first objective, we conducted comparisons of the reference land-use types between the two landscapes in order to test the first hypothesis.Then we carried out comparisons among land-use types within each landscape to test our second hypotheses.Linear mixed-effect models (LMEs) were applied (Crawley, 2009) with either landscape (i.e., comparing landscapes for each reference land use) or land use (i.e., comparing land-use types within each landscape) as the fixed effect and replicate plots and sampling days as the random effects.For the fertilization experiment, we tested differences in soil CO 2 and CH 4 fluxes between chamber locations within each oil palm plantation plot, using LMEs with chamber location as the fixed effect and palm trees and sampling days as the random effects.We extended the LME model to include either (1) a variance function that allows different variances of the fixed effect and/or (2) a first-order temporal autoregressive process, which assumes that correlation between measurements decreases with increasing time difference if this improved the relative goodness of the model fit based on the Akaike information criterion.Fixed effect was considered significant based on the analysis of variance at P ≤ 0.05, and differences between landscapes or land-use types (or chamber locations for the fertilization experiment) were assessed using Fisher's least significant difference (LSD) test (P ≤ 0.05).
For our second objective, we assess how soil factors influence the seasonal variations in soil CO 2 and CH 4 fluxes, using Pearson's correlation tests with soil temperature, WFPS, NO − 3 , NH + 4 and total mineral N.This assessment of seasonal controls of trace gas fluxes was conducted for each land use type within each landscape using the means of the four replicate plots (as spatial representation) on each of the 12 monthly measurements in order to focus the analysis on temporal variation.Lastly, we assessed the influence of soil physical and biochemical characteristics (Table A1) on the spatial variations in soil annual CO 2 and CH 4 fluxes first on the reference land-use types across landscapes (16 plots) and second across land-use types within each landscape (16 plots), using Spearman's rank correlation test.The first was to assess the spatial controls of trace gas fluxes from the reference land-use types, and the second was to evaluate which soil factors drive the spatial variation in trace gas fluxes across landuse types.We used the annual fluxes as temporal representation in order to focus this analysis on the spatial variation.For all correlation tests, statistical significance was taken at P ≤ 0.05, except in a few cases for which a marginal significance at P ≤ 0.09 was considered because our experimental design encompassed the inherent spatial variability in the studied landscapes.All statistical analyses were conducted using R 3.0.2(R Development Core Team, 2013).

Reference land-use types (forest and jungle rubber): comparison between landscapes
WFPS in the jungle rubber was higher in the clay than loam Acrisol soils (P < 0.01), but, in the forest, WFPS did not differ between landscapes (P = 0.56; Fig. 1a, b).In both landscapes, we detected lower WFPS during the drier period (mean WFPS ranged from 38 to 80 % between mid-June and October) compared to the wetter period (mean WFPS ranged from 50 to 96 %; P < 0.01; Fig. 1a, b).Soil temperatures in the forest were lower in the clay than loam Acrisol soils (P = 0.02; Fig. 1c, d), which was probably due to the difference in the time of the day when measurements were conducted.Soil temperatures in the jungle rubber did not differ between landscapes (P = 0.17).
In both landscapes, NH + 4 was the dominant form of mineral N (Table 1).Soil NH + 4 contents in the jungle rubber were higher in the clay than in loam Acrisol soils (P = 0.02), but in the forest soil NH + 4 contents did not differ between landscapes (P = 0.90; Table 1).Soil NO − 3 contents in the forest were higher in the clay than in loam Acrisol soils (P < 0.01), whereas soil NO − 3 contents in the jungle rubber were higher in the loam than in clay Acrisol soils (P = 0.02; Table 1).Total mineral N contents in both reference land-use types did not differ between landscapes (P = 0.11-0.19;Table 1).
We detected no differences in soil CO 2 fluxes between landscapes for the reference land-use types (P = 0.63-0.69;Table 2; Fig. 2a, b).Similarly, soil CH 4 fluxes from both reference land-use types were also comparable between the two landscapes (P = 0.26-0.27;Table 2; Fig. 2c, d).However, in the loam Acrisol soil, two of the four forest sites displayed net CH 4 emissions in 26 % of the measurements, which resulted in its largest variation among plots (as indicated by its largest standard error around the mean; Table 2).

Converted land-use types (rubber and oil palm plantations): comparisons to the reference land-use types within each landscape
While in both landscapes WFPS did not differ among landuse types (P = 0.12-0.26;Fig. 1a, b), soil temperatures were slightly higher in the plantations (27.2 ± 0.1 • C) compared to the reference land-use types (25.9 ± 0.1 • C) in each landscape (both P < 0.01; Fig. 1c, d).Soil NH + 4 contents in rubber were lower than in all other land-use types in the clay Acrisol soil (P = 0.05), and soil NH + 4 contents in both rubber and oil palm were also lower than in the reference landuse types in the loam Acrisol soil (P = 0.03; Table 1).In the clay Acrisol soil, NO − 3 contents in rubber and oil palm were lower than in forest (P < 0.01), and in the loam Acrisol soil NO − 3 contents in rubber were lowest, whereas these were intermediary in oil palm (P < 0.01; Table 1).The latter was related to fertilizer application by the owner of one of the
Soil CO 2 fluxes from oil palm were 45 % lower in the clay Acrisol soil and 38 % lower in the loam Acrisol soil compared to the forest (both P < 0.01; Table 2; Fig. 2a, b    oil palm base (chamber locations a and b) were on average 2.3 ± 0.2 times higher than those 4-4.5 m from the tree base (chamber location c) in both landscapes (all P < 0.01; Table A3).However, this area within 1 m distance to the tree base is only 3 m 2 per tree or 4 % on a hectare basis.Even if we were to include these elevated soil CO 2 fluxes into annual estimates using area-weighted contributions, the high fluxes within 1 m distance to the tree base would account for less than the standard errors (7-9 %) of the mean annual fluxes.Soil CH 4 uptake in the plantations was 84 % (oil palm) and 93 % (rubber) lower compared to the forest and 69 % (oil palm) and 86 % (rubber) lower compared to the jungle rubber in the clay Acrisol soil (P < 0.01; Table 2; Fig. 2c, d).Also in the loam Acrisol soil, CH 4 uptake in the plantations was lower (reduction of 64 % for rubber and 44 % for oil palm) compared to jungle rubber (P = 0.02; Table 2; Fig. 2c, d).However, CH 4 uptake in the forest on loam Acrisol soils deviated from the differences detected in the clay Acrisol soils because of the two forest sites that displayed net CH 4 emissions (see above; Table 2; Fig. 2d).From the fertilization experiment, soil CH 4 uptake from the area of fertilizer application (chamber location b) was 2.6 ± 0.2 times lower than the unfertilized chamber locations a and c (respectively 0.3 m and 4-4.5 m from the oil palm base) (P ≤ 0.01-0.05;Table A3), with the exception of plot 3 in the clay Acrisol soil (P = 0.45; Table A3).In most cases, CH 4 uptake in chamber location b was reduced immediately following fertilizer application and was restored to prefertilization values after about 6 weeks (Fig. B1).Even if we included these reduced soil CH 4 fluxes into annual estimates using area-weighted contributions, the low area coverage (∼1.3 % of 1 ha) and time duration (12 % of 1 year) would render a negligible effect.

Seasonal controls of CO 2 and CH 4 fluxes from each land-use type
In the clay Acrisol soil, CH 4 fluxes were positively correlated with WFPS (Table 3) in each of the four land-use types, signifying the higher CH 4 uptake in the dry than in the wet season (P ≤ 0.01-0.03;Fig. 2c).Soil CH 4 fluxes correlated negatively with NO − 3 contents in forest and with soil CO 2 fluxes in rubber (Table 3).Across all land-use types, soil CH 4 uptake was negatively correlated with total mineral N content (R = −0.47,P ≤ 0.01, n = 41; Fig. 3a) and NO − 3 content (R = −0.73,P ≤ 0.01, n = 41).Some correlations in Table 3 were possibly spurious: in oil palm, soil CO 2 fluxes were positively correlated with soil temperatures (Table 3) even if the temperate range was small (25.5-28.8• C) such that this correlation was likely caused by the differences in sampling time during the day rather than by a seasonal temperature pattern.In jungle rubber, plotting soil CO 2 fluxes against soil NO − 3 contents showed that their correlation (Table 3) was apparently caused by a group of many low NO − 3 contents against one high NO − 3 value and this correlation became insignificant when the one high value was removed.Also, in rubber, the marginal negative correlation between soil CH 4 fluxes with NO − 3 contents was due to the correlation between WFPS and NO − 3 contents.
In the loam Acrisol soil, seasonal variations in soil CO 2 fluxes were positively correlated with WFPS in jungle rubber and negatively correlated with WFPS in rubber (Table 3).Some correlations in Table 3 were also spurious, caused by differences in sampling time during the day rather than by a seasonal pattern: correlation between soil CO 2 fluxes and soil temperature in forest and oil palm despite narrow temperate ranges (24.8-27.2• C in forest and 25.8-29.4• C in oil palm).The negative correlation between soil CO 2 fluxes and NO − 3 contents in jungle rubber was driven by the negative correlation between WFPS and NO − 3 (Table 3).As was observed in the clay Acrisol soil, seasonal variation in soil CH 4 fluxes from the loam Acrisol soil was positively correlated with WFPS in each of the three land-use types (Table 3) with the exception of the forest which was caused by one extreme CH 4 emission from a single chamber (656.47 µg C m −2 h −1 ).When this one value was excluded, a positive correlation between soil CH 4 fluxes and WFPS was also detected for forest (R = 0.60, P ≤ 0.01, n = 12).Soil CH 4 fluxes also correlated positively with soil CO 2 fluxes in jungle rubber, whereas this correlation was negative in rubber (Table 3).In the jungle rubber, soil CH 4 fluxes correlated negatively with soil NO − 3 contents (Table 3).Across land-use types, we also observed negative correlations of soil CH 4 uptake with total mineral N content (R = −0.52,P ≤ 0.01, n = 38; Fig. 3b) and NO − 3 content (R = −0.75,P ≤ 0.01, n = 38).

Spatial controls of annual CO 2 and CH 4 fluxes across land-use types within each landscape
For these correlation analyses, we used all soil physical and biochemical characteristics which are reported in Table A1.
Apart from the correlations reported here, there were no other significant correlations with any of the tested soil physical and biochemical characteristics.First, analyzing both reference land-use types (forest and jungle rubber) across landscapes, the only significant correlation between annual soil CO 2 emissions and soil parameters was with sand content (Spearman's ρ = −0.51,P = 0.08, n = 16).However, analyzing for each landscape separately, annual soil CO  A1).

CO 2 and CH 4 fluxes from the reference land-use types
Mean soil CO 2 fluxes from our forest sites (Table 2) were within the range of reported fluxes (123-228 mg C m −2 h −1 ) from tropical rainforests in Asia (Adachi et al., 2005;Ohashi et al., 2008) and Latin America (Davidson et al., 2000;Schwendenmann et al., 2003;Keller et al., 2005;Sotta et al., 2006;Koehler et al., 2009).Compared to measurements conducted in Indonesia, our lowland forests had higher soil CO 2 fluxes than a montane forest in Sulawesi at 1000 m elevation with similar spatially replicated and temporally intensive measurements (127 mg C m −2 h −1 ; van Straaten et al., 2011); fluxes were also higher than in the seven partially logged forest sites in Jambi, where measurements were only made once (162 mg C m −2 h −1 ; Ishizuka et al., 2005).While the difference with this last study may be caused by their one-off sampling, the only other study that measured CO 2 fluxes from the same region (which conducted nine measurements spread over 1 year at three plots) reported values that were as low as 33 to 50% of our measured soil CO 2 fluxes (63-94 mg C m −2 h −1 ; Ishizuka et al., 2002).Such values are hard to reconcile with our measurements and those of others in tropical lowland forests, including the measurements by Ishizuka et al. (2005).Seasonal variation in soil CO 2 fluxes from the reference land-use types was driven by changes in soil water content, as suggested by the positive correlation with WFPS in jungle rubber on the loam Acrisol soil (Table 3).Other studies conducted in tropical rainforests have shown that seasonal changes in soil CO 2 fluxes are often caused by changes in soil water content (e.g., Sotta et al., 2007;Koehler et al., 2009;van Straaten et al., 2011), sometimes in combination with a reduction in solar irradiation caused by clouds during the wet season (Schwendenmann et al., 2003).In tropical forest soils, the relationship of soil CO 2 flux with soil water content is curvilinear, with the highest fluxes typically at field capacity (pF ∼ 2 or WFPS between 50 and 55 %; Sotta et al., 2007;Koehler et al., 2009;van Straaten et al., 2011), which explains why WFPS did not show correlation in forests in both landscapes where WFPS (mostly ≥ 60-80 %; Fig. 1a,  b) fluctuated at the top curve of this curvilinear relationship.
In contrast to our first hypothesis, soil texture was not the proximal factor controlling annual soil CO 2 fluxes, but instead sand content indirectly affected soil fertility (e.g., Brayextractable P), which, in turn, influenced soil CO 2 fluxes.In the reference land-use types, the negative correlation of annual soil CO 2 fluxes with the sand contents contrasted with results in the Amazon Basin where sandy Ferralsol soil had higher soil CO 2 fluxes than the clay Ferralsol soil (Sotta et al., 2006).In the study by Sotta et al. (2006), annual CO 2 emissions were negatively correlated with total soil P content.In our loam Acrisol soil, which had lower soil fertility (i.e., lower Bray-extractable P and base saturation and Table 3. Pearson correlation coefficients (n = 12) between soil CO 2 flux (mg C m −2 h −1 ), soil CH 4 flux (µg C m −2 h −1 ), soil temperature ( • C, top 0.05 m depth), water-filled pore space (WFPS) (%, top 0.05 m depth) and extractable mineral nitrogen (mg N kg −1 , top 0.05 m depth), using the monthly-measurement means of the four replicate plots per land-use type between December 2012 and December 2013.higher Al saturation) than the clay Acrisol soil (Table A1; Allen et al., 2015), there may be strong competition for P such that trees have to allocate more C to their root or rootmycorrhizal system to obtain this nutrient.At the same study sites, reference land-use types on loam Acrisol soil showed a lower P concentration in the fine roots in the top 0.2 m of soil than reference land-use types on clay Acrisol soil (Sah-ner et al., 2015).This strategy of high belowground C investment was reflected in the negative correlation of annual soil CO 2 fluxes from the reference land-use types with Brayextractable P contents in the loam Acrisol soil.Mean soil CH 4 fluxes from our forest sites (  Veldkamp et al., 2013); however, our measured CH 4 uptake rates were at the upper end (towards more negative values) of these reported rates and were also higher than the CH 4 uptake rates reported for old-growth forests in Jambi, Indonesia (−21.3 to +4.2 µg CH 4 -C m −2 h −1 ; Ishizuka et al., 2002).Seasonal variation in soil CH 4 fluxes was strongly controlled by soil water content with higher uptake in the dry season (Fig. 1a, b), as shown by the strong positive correlations with WFPS in all land-use types in both landscapes (Table 3).Such seasonal changes reflect diffusional limitation on the supply of CH 4 to methanotrophs at high WFPS (Keller and Reiners, 1994) and the possible occurrence of anaerobic decomposition, producing CH 4 , which may partially offset CH 4 consumption (Keller and Reiners, 1994;Verchot et al., 2000).Since we measured occasional net CH 4 emissions from some reference land-use types (Fig. 2d), we cannot exclude this anaerobic CH 4 production.High microbial and root activity consume oxygen in the soil, which may contribute to the creation of anaerobic microsites where CH 4 can be produced.This may have occurred in the jungle rubber on the loam Acrisol soil, where we detected a positive correlation of soil CO 2 fluxes with soil CH 4 fluxes (Table 3).Positive correlations of soil CO 2 fluxes and CH 4 fluxes have been reported also for tropical forests (Verchot et al., 2000).In addition to WFPS, soil mineral N dynamics also influenced the seasonal variation in soil CH 4 fluxes.The negative correlations of soil CH 4 fluxes with soil NO − 3 contents in the forest on the clay Acrisol soil and in the jungle rubber on the loam Acrisol soil (Table 3) imply that some of the observed seasonal variability may have been caused by temporal N limitation of CH 4 oxidation (Bodelier and Laanbroek, 2004;Veldkamp et al. 2013).
We found strong indications that CH 4 uptake in this converted tropical lowland was both N-limited and affected by high, potentially toxic, exchangeable Al concentrations in the soil.We interpreted negative correlations of annual soil CH 4 fluxes from the reference land-use types with net N mineralization rates (see Sect. 3.4) across landscapes as evidence for N-limited CH 4 uptake.Indications of N-limited CH 4 uptake have been reported for tropical forests in Panama (Veldkamp et al., 2013) and Ecuador (Wolf et al., 2012), but this is the first time that it was observed on a landscape scale in the tropics.Furthermore, the positive correlations of annual soil CH 4 fluxes from the reference land-use types with exchangeable Al within each landscape reflected the lower CH 4 uptake measured at sites with more exchangeable Al in the soil.Soil Al saturation in our reference land-use types was high (mean values ranged from 61 to 80 %; Appendix Table A1).High Al 3+ concentrations in the soil solution and higher exchangeable Al in the soil are known to be toxic for plants, the root growth of which may be inhibited (Ma et al., 2001).Dissolved Al 3+ can also be toxic for soil microorganisms, and it has been shown that high dissolved Al concentrations in the soil inhibited CH 4 uptake in a temperate forest soil in Japan (Tamai et al., 2003).We are not aware of any study reporting such a relationship for tropical ecosystems, which is not surprising since in most trace gas studies exchangeable Al in the soil is either not measured or does not reach such high levels as at our sites.
In summary, seasonal variation in soil CO 2 fluxes from the reference land-use types were related to soil water content (i.e., jungle rubber in loam Acrisol soil), while the spatial control of annual soil CO 2 fluxes across landscapes was related to soil fertility: low Bray-extractable P concentrations coincided with high annual soil CO 2 fluxes from the loam Acrisol soil, which had lower soil fertility than the clay Acrisol soil.Seasonal variation in CH 4 fluxes from the reference land-use types were mainly explained by soil water content, although we found indications that temporal N limitation may also have played a role.Spatial controls of annual soil CH 4 fluxes across landscapes were also related to soil fertility, as shown by their negative correlation with soil N availability, suggesting N limitation on CH 4 uptake, and their positive correlation with exchangeable Al, suggesting Al toxicity towards methanotrophs, which has not yet been reported for tropical ecosystems.These results are in contrast to our first hypothesis -soil texture was not the proximal factor controlling soil CO 2 and CH 4 fluxes but only did so indirectly through its influence on soil fertility, which, in turn, controlled the spatial variations in these trace gases across our studied landscapes.

Effects of land-use change on CO 2 and CH 4 fluxes
Mean soil CO 2 fluxes from our rubber plantations (Table 2) were on the same order of magnitude as those reported for seven rubber plantations in Jambi (Indonesia) where fluxes were measured once (171 mg C m −2 h −1 ; Ishizuka et al., 2005), while soil CO 2 fluxes from a rubber plantation in a sandy clay loam Nitisol soil in Malaysia with one measurement were lower (123 mg C m −2 h −1 ; Adachi et al., 2005).Some other studies reported soil CO 2 fluxes that are much lower than our measured fluxes: at a rubber plantation on a heavily weathered silty clay soil in China (35 mg C m −2 h −1 ; Werner et al., 2006) and a rubber plantation in Jambi (Indonesia) with nine measurements (75 mg C m −2 h −1 ; Ishizuka et al., 2002).Since this last study also reported 33-50 % lower soil CO 2 fluxes from forests (see Sect. 4.1) as well as 50 % lower soil CO 2 fluxes from oil palm (51 mg C m −2 h −1 ; Ishizuka et al., 2002) than our measured fluxes from the same region (Table 2), we suspect some methodological issues in this study.Mean soil CO 2 fluxes from our oil palm sites were comparable with other reported fluxes from five oil palm plantations in Jambi (Indonesia) where fluxes were measured once (98 mg C m −2 h −1 ; Ishizuka et al., 2005).Lastly, soil CO 2 fluxes from an oil palm plantation that were more than double our measured fluxes were reported from a one-time measurement in a sandy clay loam Nitisol soil in Malaysia (222 mg C m −2 h −1 ; Adachi et al., 2005).
Seasonal variation in soil CO 2 fluxes from oil palm was not as pronounced as that in fluxes from rubber (Fig. 2a,  b).In rubber plantations in loam Acrisol soil, where WFPS were all above 55 % (Fig. 1b), the seasonal variation in soil CO 2 fluxes reflected the curvilinear relationship of soil CO 2 fluxes with WFPS, whereby soil CO 2 fluxes typically decline at WFPS larger than about 55 % (e.g., Sotta et al., 2007;Koehler et al., 2009), hence the negative correlation with WFPS (Table 3).
The spatial variation in annual soil CO 2 fluxes across landuse types reflected the changes in soil organic-matter quality and quantity with changes in land use, as indicated by the negative correlations with soil 15 N natural abundance signatures (see Sect. 3.4) and the positive correlation with soil organic C content.Soil 15 N signatures in our studied oil palm plantations were significantly higher than in the reference land-use types (i.e., loam Acrisol soil; Table A1; Allen et al., 2015), which we interpreted as an indication of the degree of decomposition of soil organic matter.The more decomposed the soil organic matter, the higher the soil 15 N signature, as illustrated by increasing soil 15 N signatures with increasing depth of tropical forest soils (Sotta et al., 2008;Baldos et al., 2015).The more decomposed soil organic matter in the oil palm plantations was probably caused by reduced litter input and lower fine-and coarse-root production compared to the forests (measured at the same sites by Kotowska et al., 2015).The low litter input and root production influenced not only the degree of decomposition of the soil organic matter but also led to strong reductions in soil organic C stocks of oil palm and rubber plantations (measured in the same study region by van Straaten et al., 2015).This may have contributed to the low CO 2 emissions, since we also detected a positive correlation of soil CO 2 emissions with soil organic C content.Similar findings were reported from the conversion of forest to tree plantations on Acrisol soil in subtropical southern China, where decreases in soil CO 2 fluxes were also explained by decreases in annual litterfall and root biomass (Sheng et al., 2010).In addition to changes in soil organic-matter quality, spatial variation in annual soil CO 2 fluxes across land-use types in the loam Acrisol soil was also controlled by changes in soil fertility with land-use change, as shown by the fluxes' negative correlations with base saturation and Bray-extractable P (see Sect. 3.4).Conversion of forest or jungle rubber to rubber and oil palm plantations was accompanied by the burning of slashed vegetation, whereby considerable amounts of bases and P could be released from the plant biomass to ashes (Klinge et al., 2004).Input of these nutrients to the soil from the ashes, combined with P fertilization and liming (particularly in the oil palm plantations), significantly increased soil pH in both rubber and oil palm as well as base saturation and Bray-extractable P in oil palm (Table A1; Allen et al., 2015).The negative correlations of annual soil CO 2 fluxes with base saturation and Bray-extractable P across land-use types suggest that C allocation to its root-mycorrhizal system may have decreased with increased base cations and P availability, contributing to the observed decrease in soil CO 2 fluxes from oil palm compared to the other land-use types (Table 2).In contrast, we could not support the speculation by Ishizuka et al. (2005) that low soil CO 2 fluxes from oil palm plantations could be explained by higher soil bulk densities related to intensive management practices since soil bulk densities in these converted land-use types were comparable to the reference landuse types (Table A1; Allen et al., 2015).
As was the case for the reference land-use types, seasonal variation in soil CH 4 fluxes from the converted landuse types was also controlled by WFPS (Table 3), and the possible mechanisms were the same (see Sect. 4.1).Moreover, strong negative correlations of soil CH 4 uptake with total mineral N (Fig. 3) and NO − 3 contents across all landuse types (see Sect. 3.3), with total mineral N being lowest in the converted land-use types (Table 1), also suggest temporal N limitation on methanotrophic activity (Veldkamp et al., 2013) that may have contributed to the decrease in CH 4 uptake in the converted land-use types (Fig. 2c, d; Table 2).
The negative correlations of annual soil CH 4 fluxes with net N mineralization rates across land-use types further suggest N limitation on CH 4 uptake, as indicated by the lowest CH 4 uptake in the converted land-use types (Table 2) that had the lowest (i.e., rubber with no N fertilization) to intermediate (i.e., oil palm with N fertilization) net N mineralization rates (see Sect. 3.4).The results from the fertilization experiment in the oil palm sites that showed inhibition of CH 4 uptake in the fertilized spot (chamber location b; Table A3) within 6 weeks following fertilizer application (Fig. B1) were probably caused by salt effect, as has been observed in a fertilization experiment in tropical pastures (Veldkamp et al., 2001).However, this CH 4 inhibition following fertilizer application did not influence our annual flux estimates because of the negligible area coverage of the fertilized spots (∼ 1.3 % of 1 ha) and its short-term effect (12 % of 1 year).
In summary, soil CO 2 fluxes decreased only in oil palm and not in rubber, which partly supports our second hypothesis.These converted land-use types showed a decrease in soil CH 4 uptake, which supports our second hypothesis.Seasonality of soil CO 2 and CH 4 fluxes in the converted land-use types appeared to be controlled by the same factors as in the reference land-use types.The strong decrease in soil CO 2 fluxes from the oil palm was probably caused by a combination of strongly decomposed soil organic matter and low soil carbon stocks (caused by the low input of litterfall and low fine-root production) and possibly by low C allocation to the root or the root-mycorrhizal system (due to the improved base cations and P availability from liming and P fertilization).Reduction in annual CH 4 uptake in the converted land-use types was primarily caused by the decrease in soil N availability in these converted land-use types.

Consequences of land-use change for soil trace gas fluxes
Our study shows that land-use change had a profound effect on the soil-atmosphere fluxes of the trace gases CO 2 and CH 4 , with reduced soil CO 2 fluxes from oil palm plantations and reduced soil CH 4 uptake in both rubber and oil palm plantations.The reduced soil CO 2 fluxes in the oil palm should not be interpreted as reduced net ecosystem emissions because we did not measure the net CO 2 uptake by the vegetation and the changes in soil and vegetation carbon stocks.
Rather the strong decrease in soil CO 2 fluxes from oil palm is a reflection of the present belowground carbon dynamics in this land-use type.Due to decreases in litterfall and fineroot production (Kotowska et al., 2015) as well as frond management practice (stacking fronds in interrows) that reduced fresh litter input in the whole area, soil organic C stocks in these oil palm plantations decreased over time (van Straaten et al., 2015), reflecting the reductions in soil CO 2 emissions.
Our estimate of the decrease in CH 4 uptake resulting from conversion of forest or jungle rubber to rubber and oil palm in these landscapes was about 2 kg CH 4 -C ha −1 yr −1 (based on the average of the values in Table 2).If we multiply this with 0.52 Mha, the increase in areal coverage of oil palm and rubber plantations in Jambi from 1996 to 2011 (BPS, 2012), this suggests that the capacity of the province of Jambi to remove this potent greenhouse gas from the atmosphere has decreased by about 1040 Mg CH 4 -C yr −1 as a result of this land-use conversion.This calculation does not take into account land-use changes that occurred in the peatlands.Finally, we detected important soil fertility controls on trace gas exchange in this converted tropical landscape on highly weathered Acrisol soils, including the controls of base cation and P availability on annual soil CO 2 fluxes and N availability and Al toxicity on annual soil CH 4 fluxes.Such controls on the landscape scale have not yet been reported, and thus we stress the importance of conducting landscape-scale studies as field studies on a few small plots or laboratory-based studies may not be able to detect such important controls.
Table A3.Mean (±SE, n = 3 oil palm trees) soil CO 2 and CH 4 fluxes from three different chamber locations during a fertilization in three oil palm plantations within each soil landscape, measured 6 to 11 times during 3-8.5 weeks following fertilization.Means followed by different letters indicate significant differences among chamber locations within each oil palm plantation site (linear mixed-effect models with Fisher's LSD test at P ≤ 0.05).Chamber locations a, b and c were placed at 0.3, 0.8, and 4-4.5 m, respectively, from each of the three trees at each oil palm plantation site.Smallholders fertilized the area around the base of each tree about 0.8-1 m from the tree base, and thus chamber location b was on this fertilized area and chamber location c serves as the reference chamber not receiving any fertilizer.The same fertilization rate and form as applied by the smallholders were used in these studied oil palm plantations, described in Sect.2.2 "CO 2 and CH 4 flux measurement".

Oil palm
Chamber

Figure 1 .
Figure 1.Mean (±SE, n = 4) soil water-filled pore space (WFPS) and soil temperature in the top 0.05 m depth under forest ( ), jungle rubber (♦), rubber ( ) and oil palm ( ) in the clay Acrisol soil (a and c) and the loam Acrisol soil (b and d) in Jambi, Sumatra, Indonesia, measured monthly from December 2012 to December 2013.Grey shadings mark the dry season.

Table 1 .
Mean (±SE, n = 4) soil-extractable mineral nitrogen in the top 0.05 m depth for different land-use types within each soil landscape in Jambi, Sumatra, Indonesia, measured monthly from December 2012 to December 2013.Means followed by different lowercase letters indicate significant differences among land-use types within a soil landscape and different capital letters indicate significant differences between soil landscapes within a land-use type (linear mixed-effect models with Fisher's least significant difference (LSD) test at P ≤ 0.05).
).From the fertilization experiment, soil CO 2 fluxes within 1 m of the www

Figure 2 .
Figure 2. Mean (±SE, n = 4) soil CO 2 fluxes and soil CH 4 fluxes from forest ( ), jungle rubber (♦), rubber ( ) and oil palm ( ) on the clay Acrisol soil (a and c) and the loam Acrisol soil (b and d) in Jambi, Sumatra, Indonesia, measured monthly from December 2012 to December 2013.Grey shadings mark the dry season.

Table 2 .
Mean (±SE, n = 4) soil CO 2 and CH 4 fluxes and annual soil CO 2 and CH 4 fluxes from different land-use types within each soil landscape in Jambi, Sumatra, Indonesia, measured monthly from December 2012 to December 2013.Means followed by different lowercase letters indicate significant differences among land-use types within a soil landscape and different capital letters indicate significant differences between soil landscapes within a land-use type (linear mixed-effect models with Fisher's LSD test at P ≤ 0.05).Annual soil CO 2 and CH 4 fluxes were not statistically tested for differences between landscapes or land-use types since these annual values are trapezoidal extrapolations.For the loam Acrisol soil, CH 4 fluxes in parenthesis included only the two forest sites that had a dominantly net CH 4 uptake, and comparison among land-use types was conducted between jungle rubber, rubber and oil palm, which all showed net CH 4 uptake.Land-use type CO 2 fluxes Annual CO 2 fluxes CH 4 fluxes Annual CH 4 fluxes

Table 2 )
fall within the range for tropical lowland forests reported by other studies (−6.3 to −55.9 µg CH 4 -C m −2 h −1 ; summarized by