• Open Access

N2O fluxes of a bio-energy poplar plantation during a two years rotation period

Authors


Correspondence: Present address: Donatella Zona, Deptment of Animal and Plant Sciences, University of Sheffield, S10 2TN, UK, tel. + 323 265 2831, fax + 323 265 2271, e-mail: D.Zona@sheffield.ac.uk

Abstract

Nitrous oxide emissions are of critical importance for the assumed climate neutrality of bio-energy. In this study we report on the N2O fluxes from a bio-energy poplar plantation measured with eddy covariance for 2 years, after conversion of agricultural fields to few months after harvesting of the plantation. A pulse peak of N2O was detected after the land use change and in the wake of the first heavy rainfall. The N2O-N emission during just a single week was 2.7 kg N2O-N ha−1 which represented approximately 42% of the total N2O-N emitted during the 2 years of measurements. After this peak emission, N2O fluxes were constantly rather low, not increasing after rainfall events any longer. Lowest emissions (and even N2O sink) occurred mostly during the end of the second growing season with maximum canopy development, and water table deeper than 80 cm. Gross primary production (GPP) explained 68% of the monthly averaged variability in N2O emission from August to December 2011. Probably N uptake by vegetation during the peak of the second growing season limited N2O emission, which in fact increased again after the plantation was coppiced. For the majority of the measuring period, N2O fluxes did not present a well-defined diurnal pattern, with the exception of two periods: (1) from 19–22 August 2010 and (2) from September–November 2011. In both cases wind speed played a major role in controlling the diurnal pattern in these fluxes (explaining up to 80% of the diurnal variability in N2O fluxes on 19–22 August 2010), whereas at the end of the second growing season (September–November 2011), GPP explained 73% of the diurnal pattern in N2O fluxes.

Introduction

Nitrous oxide (N2O) is a major greenhouse gas with a global warming potential ~300 times higher than CO2, and with a role in the destruction of stratospheric ozone (IPCC , 2007). The initial conversion of pasturelands (Nikiema et al., 2012) or natural grassland (Gelfand et al., 2011) into bio-energy plantations causes a substantial initial N2O emission. This challenges the assumed environmental benefits and the carbon neutrality of bio-energy. N2O mainly originates from the biogenic processes of nitrification and denitrification (Davidson et al., 2000). The main driving factors on N2O emission are soil nitrogen (N) supply, water availability (connected to oxygen content), and carbon supply (Davidson, 1991; Davidson et al., 2000). However, large uncertainties remain on the estimates of N2O fluxes from the biosphere (Houghton et al., 1996; Kroeze et al., 1999; Schindlbacher et al., 2004; Neftel et al., 2007). These uncertainties arise from the complex interactions between the mechanisms and controls on production, consumption, and transport through the soil and on the release into the atmosphere (Davidson, 1991; Davidson et al., 2000; Smith et al., 2003; Schindlbacher et al., 2004).

The relationship between N2O emissions and soil water availability are very complex. Increases in soil water content initially usually increases N2O emission, but further increases have a negative effect on net N2O emissions. The soil water control on N2O fluxes is not yet fully understood (Davidson et al., 2000; Jungkunst et al., 2008; Castellano et al., 2010). Dry, well-aerated soils favor the oxidative process of nitrification (with transformation of inline image into inline image and NO emission), wet soils favor inline image and NO reduction, and N2O emission, and finally extremely wet soils favor the complete N2O reduction to N2 by denitrifiers (Davidson, 1991; Davidson et al., 2000). Overall, a water-filled pore space (WFPS) of ~60% frequently corresponds to the maximum N2O emission (Davidson, 1991). Previous studies highlighted the complexity and limitations in predicting N2O fluxes from environmental variables demonstrating that similar soil moisture could lead to very different emission rates (Schindlbacher et al., 2004; Castellano et al., 2010; Wu et al., 2010). Even long-term measurements (Hellebrand et al., 2003; Wagner-Riddle et al., 2007; Wu et al., 2010) were not able to identify with certainty the controls on N2O release. Consequently, modeling N2O emissions without site-specific calibration is still very challenging (Jungkunst et al., 2012). Generally, the rainfall pattern is crucial in influencing N2O emissions. Higher N2O emissions are expected with the same total rainfall, but longer dry periods (Rolston et al., 1982; Smith & Patrick, 1983).

N2O release is characterized by short peak emissions connected to fertilization, precipitation events (Eugster et al., 2007; Wagner-Riddle et al., 2007; Jungkunst et al., 2008; Neftel et al., 2010), and freeze-thaw cycle (Papen & Butterbach-Bahl, 1999; Butterbach-Bahl et al., 2002; Wu et al., 2010). These peak N2O–N emission events could be extremely relevant to the total annual budget (Yamulki et al., 1995; Butterbach-Bahl et al., 2002; Barton et al., 2008; Scheer et al., 2008), and may account for 24–73% (Papen & Butterbach-Bahl, 1999; Wu et al., 2010), and even up to 94% of annual emission (Guckland et al., 2010). Therefore, it is extremely important to predict, simulate, and identify driving factor for these peak events. The short-term nature of the N2O release and the difficulties in modeling N2O emission generate the need for continuous monitoring to improve estimates of N emission from ecosystems. Unfortunately, most studies had to be restricted to discontinuous measurements mainly performed with chambers (e.g. duration of weeks to months) (Jungkunst et al., 2004; Neftel et al., 2007, 2010; Kavdir et al., 2008; Mammarella et al., 2010).

Diurnal patterns in N2O fluxes have been observed sporadically (e.g. Christensen, 1983; Kroon et al., 2010; Neftel et al., 2010), but the discontinuous nature of most measurements also limits the understanding of the underlying processes controlling the occurrence of these diurnal patterns. Other relevant and largely unknown processes are linked to the net N2O uptakes observed from different ecosystems, such as grasslands (Glatzel & Stahr, 2001; Neftel et al., 2007) and forests (Cavigelli & Robertson, 2001; Butterbach-Bahl et al., 2002). Although N2O uptake has been connected to anaerobic microbial denitrification (Zumft, 1997), the processes leading to N2O consumption are still largely unknown (Chapuis-Lardy et al., 2007). Therefore, continuous measurements with intense temporal sampling are critical to understand the controls on the N2O sink.

In the present study we investigated the environmental controls of the N2O emission from a bio-energy poplar plantation during the first rotation cycle (2 years) from land conversion from agriculture, until few months after the copping of the plantation. We investigated the overall importance of the different phases of the plantation establishment and management, and of peak emission events for the final budget of N2O–N emission during these 2 years. We also examined if N2O fluxes presented a diurnal pattern and explored the controls on these patterns.

Materials and methods

Site description

The research site is a bio-energy poplar plantation located in Lochristi, Belgium (51°06′44″ N, 3°51′02″ E), 11 km from the city of Ghent at an altitude of 6.25 m above sea level, with annual temperature of 9.5 °C and average annual precipitation of 726 mm (Royal Meteorological Institute of Belgium). The first 30–60 cm of soil are sandy (clay content ~10%), and deeper layers (~75 cm) are loamy sandy. The plantation (planted with Populus deltoides, P. maximowiczii, P. nigra, and P. trichocarpa, and interspecific hybrids) was established on 7–10 April 2010 on 18.4 ha of former agricultural land with regular fertilizer use (Broeckx et al., 2012), and coppiced (harvested) on 2–3 February 2012. The previous land uses were pasture and cropland (ryegrass, wheat, potatoes, beets, and most recently monoculture corn with regular fertilization, 200–300 kg N ha−1 yr−1 liquid animal manure and chemical fertilizers). The plantation was not fertilized or irrigated during this experiment, and the weeding was mostly mechanical with sporadic herbicide applications (Broeckx et al., 2012). More details on the plantation establishment and site management are provided in Broeckx et al. (2012).

Environmental variables

Environmental variables were recorded continuously from June 2010 to the end of July. Soil water content was monitored diagonally and vertically in the soil layers of 0–30, 0–20, 0–10 cm at different locations, and horizontally across a vertical transect (at 1 m, 60 cm, 40 cm, 30 cm, and 20 cm below the surface) in the proximity of the eddy covariance tower using time domain reflectometry (TDR model CS616; Campbell Scientific, Logan, UT, USA) moisture probes. Soil temperature (soil T) was recorded by temperature probes that averaged the temperature of a soil layer of 0–8 cm depth (model TCAV-L averaging thermocouples; Campbell Scientific). Surface temperature (surf T) was recorded using an infrared sensor (Apogee Instruments, Inc., Logan UT, USA); air temperature (air T) and relative humidity both were recorded on the eddy covariance tower using a Vaisala probe (model HMP45C; Vaisala, Helsinki, Finland) at a height of 5.4 m above the ground surface. An air temperature profile was also measured at a meteorological tower at different heights (50 cm, 1 m, 2 m, and 4 m above the ground) with type-T thermocouples. Precipitation was recorded using a tipping bucket rain gauge (model 3665R; Spectrum Technologies Inc., Plainfield, IL, USA); water table was recorded with a pressure transducer (model PDCR1830; Campbell Scientific) installed in a pipe inserted into the ground to 1.85 m depth. All these meteorological sensors were connected to two data loggers (model CR5000 and model CR1000; Campbell Scientific) and then output to a computer. More details on the experimental set-up of these sensors are provided in Zona et al. (2011) Vegetation height was measured weekly with a graduate rod permanently inserted in proximity of the eddy covariance tower (16 m). This single-point measurement was compared with the average height computed from almost 5000 measurements across the plantation (Broeckx et al., 2012).

Eddy covariance measurements

In June 2010, an eddy covariance (EC) system was installed in the north-east corner of the plantation. The system included a sonic anemometer (model CSAT3; Campbell Scientific), and several fast response analyzers including a CO2/H2O infrared analyzer (model LI-7000; LI-COR, Lincoln, NE, USA) and a N2O/CO analyzer (model 908-0014; Los Gatos Research, Mountain View, CA, USA). The height of the sonic anemometer and the inlets of the sampling lines were placed at 5.8 m above the soil surface during the first measuring period (1 June 2010 to end of August 2011), at 6.6 m (from end of August 2011 until 22 February 2012), at 2.5 m (from 22 February until 6 July 2012), and at 4.9 m from 6 July 2012 thereafter. The data collection of the sonic, and the CO2 analyzer started on 1 June 2010, whereas the data collection from the N2O analyzer started on 4 August 2010. More details on the set-up, calibration, and maintenance of the EC system are provided in Zona et al. (2011).

Postprocessing of the eddy covariance data

EdiRe (R. Clement, University of Edinburgh, UK) was used to calculate fluxes of CO2 (also alternatively indicates as net ecosystem exchange, NEE), N2O fluxes, and latent heat (LE), averaged over 30 min periods. A two-component rotation was applied to set mean lateral and vertical velocity components to zero. Time delay between scalar and vertical wind velocity fluctuations was determined by cross-correlation optimization (Rebmann et al., 2012). The correction applied to the fluxes were as follows: frequency response correction (Horst, 1997) and water vapor correction term of the Webb, Pearman, & Leuning correction (WPL, Webb et al., 1980; Ibrom et al., 2007). The data removal of the N2O fluxes was performed according to the following criteria: when the internal pressure was not stable, during maintenance of the analyzer (such as mirror cleaning and filter change), when the wind direction was between 50° and 250° (outside of the acceptable footprint), when the minimum half-hourly averaged N2O concentration was negative, when diagnostics of the CSAT-3D reported an error of the sonic anemometer, when the quality flags were equal to nine (Foken & Wichura,1996; Foken et al., 2004). After data filtering the percentage in data coverage was 35% in 2010, 40% in 2011, and 38% in 2012.

Gap filling and flux partitioning (to estimate gross primary production, GPP) of the CO2 fluxes (NEE) were done according to the standard methodology used in Fluxnet (http://www.bgc-jena.mpg.de/~MDIwork/eddyproc/, Reichstein et al., 2005; Lasslop et al., 2010). Gap filling of N2O fluxes was performed using the linear extrapolation (inline image where F is the gap-filled fluxes, n is the total number of periods with original flux values, N is the total number of averaging periods in the gap-filled fluxes, and fi are the original flux values, Mishurov & Kiely, 2011). The gap filling was performed for monthly intervals (with similar rates of emission), with the exception of the peak emission of 2010 (which was gap filled separately, by summing the daily averages during the first four and the last three peak emission days). As the N2O analyzer started collecting data from 4 August 2010, we gap filled the month of June and July using the average N2O emission rates observed before the peak emission in August 2010. The periods used for the gap filling presented similar environmental conditions (e.g. water table constantly lower than 80 cm and fairly low precipitation from 1 June to 14 August 2010, Fig. 1). We compared this gap filling performed using linear extrapolation with a gap filling based on multiple imputation (MI) (Hui et al., 2004). These two methods provided similar results (D. Zona, unpublished data).

Figure 1.

Daily precipitation and a daily average air temperature (a), water table (WT) and canopy height (b), daily averaged N2O fluxes, and daily averaged soil temperature at 20 cm depth (c), for the period from 1 June 2010 to 29 July 2012.

Soil nitrogen measurements

Six PVC tubes (5 cm diameter) were inserted up to a 2 m depth in the soil in July 2011 in six different plots across the field site. From August 2011 to July 2012, groundwater samples were collected in each of these six plots, with about a monthly sampling intensity, and the water table depth was recorded during each sampling dates. A 2 m plastic tube connected to a glass bottle was used to collect these samples by applying a vacuum of 60 KPa; 200 ml groundwater was collected each time, 100 ml for analyses of NH3 and inline image and 100 ml for nitrates (inline image) and nitrites(inline image). In addition, 20 suction cups were installed in several locations across the field site in December 2010 to sample the soil pore water at 60 cm depth. These suction caps consist of conical porous glass cups (diameter 16–22 mm) made from borosilicate glass to avoid sorption of dissolved organic matter (DOM) by the samplers. A pressure of 60 KPa for 24 h was applied to collect the soil water. Due to sandy soil and the dry conditions during most of the growing season 2011, only two samples were collected in September and December 2011. After collection, both the groundwater and soil water samples were stored at 4 °C in the laboratory (SGS, Antwerp, Belgium) where the analysis was performed within the next 24 h after collection. Nitrate and nitrite concentrations were determined by spectrophotometric detection (Aquakem 250, discrete analyzer; Labmedics, Heaton Lane, Stockport, UK), whereas organic nitrogen concentrations (NH3 and inline image and other organic nitrogen) were determined following the Kjeldahl method (SKALAR continuous flow analyzer; Breda, the Netherlands).

Statistical analyses

General linear modeling (GLM) was used to investigate the most important controls on N2O fluxes (Systat version 13; Systat Software Inc., Chicago, IL, USA, 2002). A forward stepwise multiple regression approach was used to discriminate among and rank the most important variables (air and surface temperature, soil temperature and soil moisture in the different soil profiles at 0–8, 20, 30, 40, and 60 cm, water table depth, wind speed, NEE, and GPP) in explaining the variability in N2O fluxes. Models were applied to the half-hour averaged N2O fluxes during the peak emission days (for 19–25 August 2010, and separately for 19–22 August and 23–25 August 2010), and to the daily averaged N2O fluxes during the entire measurement period. To reduce the model complexity we tried to maximize the explanatory power while limiting the number of variables included in the model. A GLM was also applied to investigate the environmental drivers on diurnal patterns in the N2O fluxes during selected periods (19–22 August and 23–25 August 2010, and September, October, and November 2011) with the most defined diurnal pattern: to reduce the noise in the data, fluxes were averaged over corresponding half hours during 19–22 August, 23–25 August 2010, and September, October, and November 2011. Linear regression (Least-squared mean) was used to test the correlation between water table level and nitrogen content (inline image,inline image,inline image and NH3) in the groundwater.

Results

Canopy development, environmental conditions, and groundwater N content

The canopy height measured with the two sampling schemes previously described provided similar results: at the end of the growing season 2010 the maximum canopy height reported by Broeckx et al. (2012) (n = 4921) was 240 cm and was 221 cm according to the measurements reported here; at the end of the second growing season Broeckx et al. (2012) reported an maximum average canopy height of 445 cm (n = 4921), and from the measurements reported here maximum average canopy height was 470 cm. The initial period of measurement (June–July 2010) was characterized by fairly dry conditions. The low rainfall (Fig. 1a) led to a fairly deep water table until mid-August 2010 (~136 cm below the surface and below 80 cm for the entire summer season 2010, Fig. 1). On the 16–17 August 2010, an extreme precipitation event (total precipitation in 48 h was 81 mm) steeply increased the water table up to the surface (Fig. 1b), causing flooding of the field site. After this peak precipitation event, frequent precipitation maintained fairly wet conditions during autumn and winter 2010, until spring 2011. The spring 2011 was characterized by fairly low precipitation: the precipitation in March was 18.3 mm, in April 9 mm, and May 16.8 mm, far below the long-term monthly averages (39.0, 32.1, 42.1 mm for these 3 months, respectively). The low precipitation caused the water table to progressively decrease during the growing season 2011 (Fig. 1a and b). This dry period ended in July–August 2011 when precipitation increased again (Fig. 1a), causing the rise of the water table from 145 cm below the surface to about 80 cm below the surface. Another intense precipitation event occurred in December 2011, which led to the increase in the water table up to the surface over a few days (12–14 December 2011, Fig. 1b). The spring 2012 was characterized by much wetter conditions than the spring of 2011, with much higher water table (Fig. 1b), and generally lower air temperature, with temperature below freezing in February 2012 (Fig. 1a). The average air T in April 2012 (8.4 °C) was 2.4 °C degrees lower than the long-term mean (10.8 °C).

The nitrogen concentration in the groundwater fluctuated substantially during the measuring period, but maintained fairly high values (Table 1). inline image was slightly influenced by the water table, even if the correlation was not significant (R2 = 31% and P = 0.072), whereas [NH3]and inline image (and other organic N) were not significantly correlated with water table. inline image was very low or below the detection limit for most of the measuring period (Table 1). The inline image in the suction cups was only collected on 19 September 2011 and 12 December 2011 and in both dates it was <0.10 mg l−1. inline image and [NH3] (and other organic N) were 2.4 and 1.7 mg l−1 on 19 September 2011 and 12 December 2011, respectively. Overall, even if the N content in the groundwater was very high (Table 1), the N content in the shallower soil water (the few times it was possible to extract soil water with the suction caps) presented very low values.

Table 1. Concentrations of inline image, inline image, inline image and NH3 (mg l−1) in the groundwater (average ± standard deviation = 6) and corresponding water table in each of the sampling days
  inline image inline image inline image & NH3 and other organic NWater table
  1. a

    Measurements (n = 6) performed in the water of the ditches surrounding the field (Zona et al., 2011).

29/10/2010a3.0 ± 0.70.032 ± 0.0040.2 ± 0.013−37 ± 1.4
18/8/201127.4 ± 17.00.052 ± 0.0593.9 ± 3.5−135 ± 22
31/8/201136.9 ± 38.00.022 ± 0.0161.5 ± 0.1−130 ± 23
19/9/201137.4 ± 27.80.032 ± 0.0311.3 ± 0.3−121 ± 24
19/12/20111.8 ± 2.90.035 ± 0.0211.9 ± 1.0−11 ± 14
13/2/20122.4 ± 4.70.014 ± 0.0092.2 ± 0.4−114 ± 31
6/3/20123.7 ± 5.9<0.012.1 ± 0.7−56 ± 28
6/4/20122.8 ± 4.8<0.012.3 ± 0.8−110 ± 22
10/5/20127.2 ± 9.3<0.013.6 ± 2.2−45 ± 27
5/6/20123.8 ± 5.80.012 ± 0.0052.3 ± 0.5−116 ± 24
2/7/20123.1 ± 4.6<0.012.4 ± 0.6−119 ± 23

Seasonal pattern in N2O fluxes

In 2010, the first large rainfall event (on 16–17 August) after a prolonged dry period (Fig. 1) abruptly increased the water table, leading to a steep increase in N2O emission from the plantation (Fig. 1b and c). This large N2O emission was not observed during the high precipitation event or when the site was flooded, but after few days, when the water table progressively decreased (Fig. 1b). The total N2O–N emitted during just a week was about 2.7 kg N2O–N ha−1, and it was 3.1 kg N2O–N ha−1 for the entire month of August 2010 (Fig. 2). Importantly, precipitation events that occurred after the peak rain fall of August 2010 led to similar increases and decreases in water table, but did not lead to N2O emissions of the same magnitude as the one observed on 19–25 August 2010 (Figs 1c and 2). The total N2O–N emission for the first 6 months after conversion from agriculture and the plantation establishment was 4.2 kg N2O–N ha−1 (Fig. 2), but importantly 74% of the entire emission during these 6 months occurred just in the month of August (Fig. 2). After this peak emission event in August 2010, the N2O fluxes presented fairly low values for the entire 2011 (Fig. 1). In 2011 the total N2O–N emission was only 0.9 kg N2O–N ha−1, and eventually an N2O sink was observed at the end of the growing season (September–October 2011) corresponding to the maximum plant development (as shown by the largest vegetation height, Fig. 1b). In November 2011 after the plants lost their leaves, the N2O emission increased in correspondence also of a higher water table (Fig. 1). In 2012, N2O emissions increased again particularly right after the coppicing of the plantation (2–3 February 2012, Fig. 1), and again after a sudden increase in temperature and water table in May 2012 (Fig. 1a and b). A sporadic N2O sink was again observed at the end of July 2012, in coincidence of a sudden drop in water table, and increase in canopy development (Fig. 1b). The N2O–N emitted in February 2012 (the month when harvesting of the plantation occurred) was 0.21 kg N2O–N ha−1 and in May 2012 was 0.34 kg N2O–N ha−1. The total N2O–N emission from January to July 2012 was 1.25 kg N2O–N ha−1. Overall, the N2O fluxes were the highest during the initial measuring period (during the establishment of the plantation after the large rainfall), and after coppicing in winter and spring 2012, whereas they were the lowest the entire year 2011 (particularly at the end of the growing season, Fig. 1c).

Figure 2.

Monthly total N2O–N emission (a) and total cumulative N2O–N emission (b) for the period from 1 June 2010 to 29 July 2012.

For the entire measuring period (with the inclusion of the peak emission of August 2010), a multivariable model that included soil T and soil moisture (both at 1 m depth) presented an explanatory power of about 5% (P < 0.001) of the daily averaged N2O fluxes. If the August 2010 peak emission was removed from the analysis, the multivariable model with the highest explanatory power (which included soil T at 20 and 0–8 cm and soil moisture at 1 m depth) explained 12.4% of the variability of the N2O fluxes (P < 0.001). In both cases, the addition of other variables to the model either did not increase or only increased the explanatory power of the model by a few percent. In addition, as N2O sink was mostly observed at the end of the growing season 2011, when maximum canopy development occurred, we tested if the net CO2 fluxes (NEE) and GPP influenced N2O emission. The monthly total GPP explained 68% (P = 0.086) of the monthly total N2O fluxes for the period August to December 2011.

Diurnal pattern in N2O fluxes

For the majority of the 2 years of measurements, N2O fluxes did not present a well-defined diurnal pattern, with the exception of few periods. To understand the controls on these diurnal patterns we investigated these periods in details. The most pronounced diurnal pattern in N2O emission occurred during the first 3 days of the peak N2O emission in August 2010 (19–22 August). During these 4 days, N2O emission followed the increase in soil temperature and wind speed (Fig. 3). From 23–25 August 2010—when the wind speed was constantly >2 m s−1—N2O emissions no longer presented the same pronounced diurnal pattern, even if N2O emission slightly increased during daytime (Fig. 3). Particularly during the first four days (19–22 August 2010), the diurnal pattern in N2O fluxes was the same as the one of the GPP (Fig. 3). When N2O fluxes were averaged over corresponding half hours and the periods 19–22 August and 23–25 August 2010 were modeled in the GLM separately, wind speed explained 80% of the variability in N2O fluxes (P < 0.001) from the 19–22 August, and wind speed and net CO2 fluxes (i.e. NEE) explained 44% of the variability in N2O fluxes (P < 0.001) from 23–25 August 2010. If the entire period (19–25 August 2010) was modeled together surf T and soil T (at 20 cm) were the most important variables and together explained 68% of the variability in N2O fluxes (P < 0.001). If noise in the data was further reduced by performing a 5-points running mean, wind speed remained the most important control and explained 89% (P < 0.001) of the variability in N2O fluxes from 19–22 August, and 69% (P < 0.001) from 23–25 August 2010. If these two periods were modeled together surf T and soil T (at 20 cm) explained 76% of the variability in N2O fluxes (P < 0.001). Overall, the smoothing did not majorly change the ranking of the drivers on the diurnal pattern in N2O emissions during the peak emission event of August 2010.

Figure 3.

Wind speed (m s−1), surface temperature (surf T), and soil temperature (soil T) at 0–8 and 20 cm depth in the soil (°C), CO2 fluxes (mg CO2 m−2 s−1), and N2O fluxes averaged over corresponding half hours for the indicated periods in 2010.

During the end of the growing season 2011 (from September to November) another fairly pronounced diurnal pattern in N2O fluxes was observed, this time with a daytime N2O uptake (September and October 2011) or a lower daytime N2O emission (November 2011) (Fig. 4), and higher N2O emission at night. During this period several environmental variables (surf T and soil T at 0–8 and 20 cm, NEE, GPP, and wind speed) were again tested in the GLM to quantify the controls on N2O fluxes (data averaged in corresponding half hours in each of these months). If these 3 months were modeled together, the most important variable in explaining the N2O fluxes was surf T (which explained 52% of the variability in N2O fluxes P < 0.001). When these 3 months were modeled separately wind speed was the most important variable, and explained 63% of the variability in N2O fluxes (P < 0.001) in September, 45% of the variability in N2O fluxes in October (P < 0.001). Wind speed lost its importance in November, when NEE explained 38% of the variability in N2O fluxes (P < 0.001). If noise in the data was further reduced by performing a 5-points running mean, GPP explained 90% of the diurnal variability in N2O fluxes (P < 0.001) in September 2011, 81% (P < 0.001) in October, and 83% (P < 0.001) in November 2011. If these 3 months were modeled together GPP was still the most important variable and explained 73% of the variability in N2O fluxes (P < 0.001).

Figure 4.

Wind speed (m s−1), surface temperature (surf T), and soil temperature (soil T) at 0–8 and 20 cm depth in the soil (°C), CO2 fluxes (mg CO2 m−2 s−1), and N2O fluxes averaged over corresponding half hours for the indicated months (September–November 2011).

Discussion

The emission rates of N2O differed dramatically between the week following the first extreme rain event and the rest of the 2 years of measurements. Low N2O emission was observed at the beginning of August 2010 (before the large rainfall on 16–17 August 2010) and during most of the growing season of 2011. These low N2O fluxes could be related to the fact that well-aerated sandy-loam soils are less likely to develop anaerobic microsites necessary for N2O production by denitrification (Skiba et al., 1993). The large rain event after a long dry period induced the release of substantial amounts of N2O in August 2010. Maximum N2O emissions (Fig. 1) observed after the large rain fall were higher than what has been usually observed in forest ecosystems (Davidson et al., 2000; Pihlatie et al., 2005; Pilegaard et al., 2006), and comparable to reported peak N2O emissions in grassland (3200 μg N2O–N m−2 h−1, Kroon et al., 2008), in forest soils (a maximum of about 800 μg N2O–N m−2 h−1, Wu et al., 2010; and almost 3000 μg N2O–N m−2 h−1, Jungkunst et al., 2008), in another bio-energy poplar plantation (up to 900 μg N2O–N m−2 h−1, Hellebrand et al., 2003). Importantly, during only a weeklong period (19–25 August 2010) the total N2O–N emission was 2.7 kg N2O–N ha−1 (~48% of the emission from agricultural soils in European countries, estimated to be 5.6 kg N2O–N ha−1y−1; Boeckx & Cleemput, 2001), and 42% of the total N2O–N emission over the 2 years of measurements. Peak N2O emissions after rewetting of dry soils had been observed in several ecosystems (Sexstone et al., 1985; Wagner-Riddle et al., 1996, 2007; Hsieh et al., 2005). The importance of these peak emission events on the annual budget of N2O fluxes highlights the urgent need for more widespread continuous measurement to accurately estimate the N loss from agro-ecosystems.

The observed substantial N2O emission during the initial phase of the plantation establishment is in agreement with the reported large N2O loss after conversion of pasturelands (Nikiema et al., 2012), or natural grassland (Gelfand et al., 2011). In fact, the soil disturbance associated with tillage and cultivation accelerates soil organic matter (SOM) turnover, freeing inorganic N, utilized by the microbial community of nitrifiers and denitrifiers, leading to N2O release (Pinto et al., 2004; Grandy & Robertson, 2006). The N cycle is the most affected by land-use change, which may increase N2O emission of two orders of magnitudes during the initial phase of establishment of willow and poplar plantation (Nikiema et al., 2012). The importance of this initial disturbance on N2O emission is confirmed by the absence of peak N2O emissions of similar magnitude for the remaining of the measuring period, even after large rainfall events and the similar increases in water table. The lack of N2O emission after successive rainfall events has been explained as nitrate or carbon limitation (Sexstone et al., 1985; Wagner-Riddle et al., 1996) because a very specific combination of soil water content and nutrient availability is necessary to cause peak N2O fluxes (Grundmann et al., 1988). In addition, the previous land use, and the substantial fertilizer use (200–300 kg N ha−1 y−1) may have also contributed to this peak N2O emission.

Wind speed, surf T, soil T were important controls on the diurnal N2O pattern, together with GPP. Surprisingly, the same environmental variables that were controlling the diurnal increase in N2O loss during the peak emission period in August 2010 were also controlling the diurnal trend in N2O fluxes (and daytime N2O sink) observed in September–October 2011. The processes controlling the N2O sink in soils are still poorly understood, and N2O sink has been observed under very contrasting conditions (Butterbach-Bahl et al., 2002; Longoria Ramírez et al., 2003; Xu & Baldocchi, 2004; Chapuis-Lardy et al., 2007; Jørgensen & Elberling, 2012). N2O diffusion within the soil profile and its dissolution in the groundwater may play an important role in leading to N2O sink (Chapuis-Lardy et al., 2007), explaining the importance of wind speed on the observed diurnal patterns in N2O fluxes.

N2O sink has been mostly connected to N-poor environments (Glatzel & Stahr, 2001), as inline image is favored as electron acceptor over N2O (Schlegel, 1985). The low N content in the shallower soil water (when it was possible to extract it) suggests the possible N limitation in part of the soil profiles, even in a N-rich environment, such as this study site. It is possible that the vegetation uptake of N together with inline image leaching (e.g. inline image is very mobile) ultimately reduced the availability of N to the microbes, reducing N2O emission, and eventually leading to N2O uptake. The strong link between N2O fluxes and GPP on the monthly scale and diurnal scale would support the important role of the vegetation on the N2O fluxes. Overall, the combination of drier condition of the growing season 2011, and the competition plant microbes to acquire N in the shallower soil layer (together with the substantial inline image leaching) most probably explained the N2O sink at the end of the growing season 2011. This is consistent with the observed lowest N2O emission in 2011 in correspondence to the maximum canopy development (Fig. 1), with the higher leaf N content in 2011 than in 2010 (in peak season 2010, 2.9 ± 0.5%, and in the peak season 2011, 3.5 ± 0.6%, Broeckx et al., 2012). The importance of vegetation development in reducing N2O emission is further supported by the observed increase in N2O emission in November 2011, after the plants lost their leaves, even under a similar water table and comparable temperatures to previous periods (Fig. 1). The coppicing of the plantation in February 2012, in coincidence with a sudden temperature decrease, also increased N2O emission. This result is in agreement with the sensitivity of the N cycle to disturbance (Pinto et al., 2004; Grandy & Robertson, 2006; Nikiema et al., 2012), and with the influence of freeze-thaw cycles on N2O release (Papen & Butterbach-Bahl, 1999; Butterbach-Bahl et al., 2002; Wu et al., 2010).

The observed N2O sink at the end of the growing season 2011 occurred under similar conditions to the N2O sink reported by Jørgensen & Elberling (2012) (e.g. water table almost 1 m deep). However, in our study site water table position alone was not enough to explain N2O sink, as this sink was not observed during the rest of the growing season 2011, even with a similar water table (Fig. 1). This further illustrates the complexity of the processes controlling N2O fluxes. Previous studies already highlighted (Butterbach-Bahl et al., 2001; Groffman et al., 2009; Stolk et al., 2011) the challenges in modeling N2O fluxes based on environmental variables. In this study, only a small percentage (about 10%) of the daily averaged N2O fluxes over the entire 2 years of measurements was explained by environmental variables (mostly by soil moisture and soil T), but this is far from a satisfactory predictive model.

Overall, N2O emission, after the initial substantial loss, presented fairly low rates. The cumulative emission (0.9 kg N2O–N ha−1) in the second year of measurement (2011) was much lower than the annual emission from agricultural crops (Boeckx & Cleemput, 2001), in agreement with what was reported for other bio-energy crops, such as Miscathus and willow (Jørgensen et al., 1997; Drewer et al., 2012). On the other hand, this emission was higher than the one reported from forest and wetlands (Pilegaard et al., 2006; Skiba et al., 2009). N2O emissions from grassland ecosystems span a large range; reported average N2O emission vary between 0.16 and 12.7 kg N2O–N ha−1 (Skiba et al., 2009), and between 0.032 and 2.89 kg N2O–N ha−1 (Jones et al., 2011), also depending on the measurement technique used (Jones et al., 2011). Overall, to understand the impact of land use on N2O emission, different ecosystems with similar N input (e.g. similar N-deposition rates) should be compared. Therefore, more widespread, longer term datasets (with better temporal coverage) are urgently needed to increase our knowledge of the processes controlling the N2O emission and consumption, evaluate the importance of the initial N loss after land conversion, and ultimately more accurately estimate the environmental impact of bio-energy.

Overall, during the 2 years of measurements, N2O emissions varied greatly. The same conditions (large rainfall event) that led to the release of a substantial amount of N2O at the beginning of measuring period after land-use change did not lead to emissions of the same magnitude afterward. The vegetation development, probably competing with the microbial community for N uptake, may have played an important role in reducing N2O fluxes, which was the lowest with the highest canopy development. An important link between GPP and N2O fluxes was observed both at the monthly scale (August–December 2011) and at the diurnal scale (September–December 2011). Notably, the use of eddy covariance allowed capturing the importance of wind speed in driving both N2O emissions and N2O uptake, and the occasional existence of pronounced diurnal pattern (with daytime release or uptake). Overall, only a small percentage (about 10%) of the daily average N2O fluxes was explained by soil moisture and soil T. These results highlight the large uncertainty in the processes controlling N2O fluxes and confirm the complexity in modeling N2O emission and the need for additional continuous studies.

Acknowledgements

This research was funded by the European Commission's Seventh Framework Programme (FP7/2007-2013) as a European Research Council Advanced Grant (n 233 366, POPFULL) as well as by the Flemish Hercules Foundation as Infrastructure contract ZW09-06. D. Zona is supported by the Marie Curie Reintegration grant (PIRG07-GA-2010-268 257), S. Vicca is a postdoctoral research associate of the Flemish FWO. We thank J. Cools and K. Mouton for logistic support, K. Butterbach-Bahl for suggestion on the data analysis, F. Miglietta, P. Toscano, G. Alberti, and A. Zaldei for help with the set-up, R. Clement, G. Burba, G. Fratini, and T. De Groote for help with data processing, A.L. Ballesteros for field assistance, the Royal Meteorological Institute of Belgium and the Flemish Environment Agency (VMM) for providing climate data, F. Fierens and the ECMWF for boundary layer data, W. Babel and T. Foken for help with footprint analysis. The COST Action ES0804 provided funding for a Short Term Scientific Mission (STSM) to discuss data analysis.

Ancillary