• Open Access

Measurement and modelling of CO2 flux from a drained fen peatland cultivated with reed canary grass and spring barley

Authors


Correspondence: Tanka P. Kandel, tel. + 45 8715 4764, fax + 45 2343 1839, e-mail address: tanka.kandel@agrsci.dk

Abstract

Cultivation of bioenergy crops has been suggested as a promising option for reduction of greenhouse gas (GHG) emissions from arable organic soils (Histosols). Here, we report the annual net ecosystem exchange (NEE) fluxes of CO2 as measured with a dynamic closed chamber method at a drained fen peatland grown with reed canary grass (RCG) and spring barley (SB) in a plot experiment (= 3 for each cropping system). The CO2 flux was partitioned into gross photosynthesis (GP) and ecosystem respiration (RE). For the data analysis, simple yet useful GP and RE models were developed which introduce plot-scale ratio vegetation index as an active vegetation proxy. The GP model captures the effect of temperature and vegetation status, and the RE model estimates the proportion of foliar biomass dependent respiration (Rfb) in the total RE. Annual RE was 1887 ± 7 (mean ± standard error, = 3) and 1288 ± 19 g CO2-C m−2 in RCG and SB plots, respectively, with Rfb accounting for 32 and 22% respectively. Total estimated annual GP was −1818 ± 42 and −1329 ± 66 g CO2-C m−2 in RCG and SB plots leading to a NEE of 69 ± 36 g CO2-C m−2 yr−1 in RCG plots (i.e., a weak net source) and −41 ± 47 g CO2-C m−2 yr−1 in SB plots (i.e., a weak net sink). Standard errors related to spatial variation were small (as shown above), but more significant uncertainties were related to the modelling approach for establishment of annual budgets. In conclusion, the bioenergy cropping system was not more favourable than the food cropping system when looking at the atmospheric CO2 emissions during cultivation. However, in a broader GHG life-cycle perspective, the lower fertilizer N input and the higher biomass yield in bioenergy cropping systems could be beneficial.

Introduction

Natural peatlands are efficient ecosystems in storing carbon and serve as a net sink of atmospheric CO2 (Gorham, 1991; Joosten & Clarke, 2002; Bleuten et al., 2006). However, drainage and use of peatlands for agriculture and forestry may turn these natural ecosystems into net sources of CO2 as the peat degradation is accelerated due to processes such as increased soil aeration, fertilization and priming of soil organic carbon turn-over by root exudates (Maljanen et al., 2010). Thus, many studies have documented a high net emission of CO2 from drained peatlands used for annual arable crop production (Maljanen et al., 2001, 2004; Drösler et al., 2008).

As an alternative to annual arable cropping system, it has been suggested to grow high-yielding bioenergy crops to reduce the CO2 emission from cultivated peatlands (Shurpali et al., 2009, 2010; Mander et al., 2012). Hence, the higher gross photosynthesis (GP), as driven by longer growth periods of perennial energy crops, is expected to offset some of the emissions of CO2 from cultivated drained peatlands. Moreover, tillage-induced loss of CO2 (Rastogi et al., 2002; Jackson et al., 2003) is expected to be reduced significantly in perennial energy crop cultivation as there is no need of annual ploughing.

In the Nordic countries, one of the most suitable energy crops on peatlands is reed canary grass (RCG, Phalaris arundinaceae L.) as it may tolerate water logging due to aerenchymatous tissue in the roots (Zhou et al., 2011; Mander et al., 2012). Furthermore, biomass productivity of RCG is higher than for other perennial grasses in areas of colder climate (Lewandowski et al., 2003). In Denmark, commercial cultivation of RCG for energy production is only emerging, but there is a growing interest to cultivate RCG and other perennial grasses for biogas production as energy-rich supplements to animal slurry (Triolo et al., 2011).

Only few studies have yet reported on the CO2 emission of peatlands used for RCG cultivation, and the available studies have focused on ecosystems left after peat extraction (Shurpali et al., 2009, 2010; Mander et al., 2012). There is a current lack of knowledge on how a change from annual arable food crop to perennial energy crop cultivation in peatlands affects the net ecosystem exchange (NEE) of CO2 which in turn represents the first step in assessments of more complete carbon balances (Jacobs et al., 2007). Contributing to filling this gap, we measured CO2 fluxes with an advanced chamber method throughout a full year in a riparian fen peatland grown with RCG and spring barley (SB, Hordeum vulgare L.) in a replicated plot experiment. The objective was to estimate and compare the NEE of CO2 from the cultivation practices with either perennial energy crops or annual arable food crops. Furthermore, as NEE of CO2 was modelled and evaluated from the two component fluxes of GP and ecosystem respiration (RE), the study aimed to optimize the models used for extrapolating GP and RE by including (in both models) a simple ratio vegetation index (RVI) as a variable related to metabolically active plant material (cf. Kuzyakov & Gavrichkova, 2010; Wohlfahrt et al., 2010).

Materials and methods

Study site and cropping history

The study site was a fen peatland located in the Nørre Å river valley near Viborg, Denmark (56°44′N, 9°68′E). The long-term (1971–2000) average precipitation at the study site is 770 mm yr−1, mean annual temperature is 7.8 °C and mean incoming solar radiation is 3528 MJ m−2 yr−1. The peat soil at the study site has been drained to a depth of 60–70 cm and used for agricultural purposes for about a century. The site was cultivated with SB continuously from the 1970s to early 1980s. From 1985, the cropping system was changed to a 4-year rotation where SB was undersown with grass that then remained in the field after barley harvest for three consecutive years.

Experimental layout and crop management

The study was initiated in 2009 with nine plots of 18 × 24 m arranged in a Latin square design. Three of these plots were cultivated with RCG (cv. Bamse), three plots were cultivated with SB (cv. Keops) and three plots were cultivated with Salix. Only RCG and SB plots were used in this study and were each divided into two subplots (18 × 12 m). RCG was not harvested in the first year because of poor development of crop stand at the year of establishment, which is common for perennial grasses. In the second year (2010), both crops were harvested in early September. Then measurements of CO2 fluxes were started and continued for 1 year.

SB cultivation followed the general practices used by the farmers of the surrounding areas. This included use of mineral fertilizer (Table 1) and application of pesticides according to the crop requirements. For RCG, the standard management practice included fertilization before the growing season in each year (Table 1), but RCG plots were fertilized with half the amount of fertilizers applied for SB because of lower fertilizer requirement of perennial energy crops which can remobilize nutrients from the top to their root system at the end of growth season (Jørgensen, 1997). Table 1 summarizes the field operations during the study period as well as the preceding year.

Table 1. Overview of major field operations during 2009–2011. The applied amount of standard mineral fertilizer (NPK) is shown in brackets. RCG, reed canary grass; SB, spring barley. Measurements of CO2 fluxes started after the harvest in 2010
DateField operations
11 May, 2009Soil cultivation and sowing of RCG and SB
25 May, 2009Fertilization at RCG (39–34–109 kg N–P–K ha−1) and SB (120–26–155 kg N–P–K ha−1) plots
25 August, 2009Harvest of SB
21 April, 2010Fertilization at RCG plots (60–13–77 kg N–P–K ha−1)
26 May, 2010Soil cultivation and sowing at SB plots
16 June, 2010Fertilization at SB plots (118–25–152 kg N–P–K ha−1)
9 September, 2010Harvest of RCG and SB
6 April, 2011Fertilization at RCG plots (60–13–77 kg N–P–K ha−1)
9 May, 2011Soil cultivation and sowing at SB plots
30 May, 2011Fertilization at SB plots (118–25–152 kg N–P–K ha−1)
22 August, 2011Harvest of SB
22 September, 2011Harvest of RCG

Soil physical and chemical properties

Before starting the experiment, soil samples (0–30 cm) from each plot were collected on 15 May 2008 by pooling nine soil cores (diameter, 22 mm). Soil pH was measured following Danish standard DS-287 and total organic carbon and total nitrogen (N) were measured with a LECO CNS-1000 (LECO Corp., St. Joseph, MI. USA) according to ISO 10694 and ISO 13878 respectively. Peat soil at different depth (0–20, 20–50, 50–75 and 75–100 cm) was sampled from four corners of the experimental field to determine bulk density and degree of peat decomposition. Bulk density was determined following oven drying of volumetric soil samples at 80 °C to constant weight. The degree of peat decomposition was determined using the von Post method (e.g., Puustjärvi, 1970).

Shoot biomass and RVI

In RCG plots, biomass sampling started at the beginning of April 2011 when the RCG plants started to grow and sampling was done twice a month until the end of September. Similarly, biomass sampling started at the end of May in SB plots after crop emergence and continued until August. On each sampling date, aboveground biomass was harvested manually very close to the soil surface from a 1 m2 area in each subplot. A representative subsample from each harvest was oven dried at 70 °C to constant weight to determine the dry biomass.

The development of aboveground biomass in each subplot was also evaluated by nondestructive measurement of the RVI, as described by Christensen (1992) using a SpectroSense 2+ fitted with SKR1800 sensors (Skye Instruments, Powys, UK). The sensors measured the incident and reflected red (656 nm) and near-infrared (778 nm) radiation and the RVI was calculated as the ratio of red to near-infrared reflectance. The RVI was measured simultaneously with gas flux measurements except during three campaigns in December 2010 and January 2011 where the plots were fully or partially covered with ground frost. The average of four consecutive RVI measurements in each subplot was used in the further calculations.

Measurement of CO2 fluxes

Fluxes of CO2 were measured by a transparent chamber technique at each subplot from 22 September 2010 to 21 September 2011. In general, the CO2 flux was measured at 3 weeks intervals during winter (October–February) and at 2 weeks intervals during the growing season (March–September). One PVC collar with a basal area of 55 × 55 cm was placed in each of the 12 subplots after crop harvest in September 2010. The collars were inserted to a depth of 10 cm and had a flange parallel to ground surface for support of the flux chambers (Petersen et al., 2012). A PVC mat was placed in front of each collar to minimize disturbance of soil gas profiles during measurements. The collars and mats were removed from SB plots during seedbed preparation, but were subsequently re-installed at the same place.

Chambers used for CO2 measurements (60 × 60 × 45 cm) were temperature-controlled, transparent acrylic chambers as described in detail by Elsgaard et al. (2012). Briefly, a quantum sensor (190-SA; Li-Cor Inc., Lincoln, NE, USA) was placed inside the chamber to record photosynthetically active radiation (PAR). Shaded temperature sensors were placed both inside and outside the chamber and were used for control of a fan-driven cooling system that was switched on and off when the temperature inside and outside the chamber deviated by more than 0.2 °C. The chamber was also equipped with a fan running continuously to ensure constant air mixing. Air was circulated (in 4 mm tubing) to and from the chamber continuously and concentrations of CO2 and H2O were measured with a Li-Cor-840 infrared gas analyzer (Li-Cor Inc.) at 1 s intervals. A Campbell datalogger (CR-850; Campbell Scientific, Logan, UT, USA) was used to control and log all measurements. One or two PVC intersections, with similar dimension as the top chamber, were used when the vegetation height exceeded the chamber height during the crop growing season. Intersections were fitted with a separate fan as a single fan at the top chamber was insufficient to ensure proper air mixing with the increased volume of the chamber.

The NEE of CO2 was measured for 2 min at first with the transparent chamber without any shrouding. After that, the chamber was lifted from the collar and vented until the CO2 inside the chamber returned to ambient concentration, and possible vapour condensation at the wall of the chamber disappeared. After venting, the chamber was repositioned on the collar, covered with a shroud that blocked 50% of the incident PAR, and the resulting NEE was measured (2 min). Similarly, CO2 flux measurements were done with shroudings that blocked 75% and 100% of the PAR. In this way, four CO2 fluxes at varying levels of PAR intensity (i.e., including darkness) were measured at each collar in each campaign for subsequent modelling of the light response of GP (e.g., Burrows et al., 2005). The flux measurements were made on bright sunny days to characterize the light response of GP for a large range of PAR. Total RE was estimated as the dark CO2 flux (i.e., measured with 100% blocking of PAR). In few occasions when the plots were covered with ground frost in winter and when there was no vegetation in SB plots in a period between ploughing and germination, only RE was measured and GP was set to zero.

Environmental variables

Soil temperatures were measured manually at 10 cm depth with a high precision thermometer (GMH3710; Omega Newport, Deckenpfronn, Germany) nearby each collar during chamber deployment. In addition, continuous measurements of soil (10 cm depth) and air temperatures (2 m), precipitation and ground water table (GWT) were recorded continuously every hour at a climate station adjacent to the experimental site on the regularly cut border grass. Likewise, at the climate station, PAR was measured every second with a Li-Cor quantum sensor (190-SA) and average PAR was recorded every hour. A Campbell datalogger (CR-1000) was used to log the continuous measurements of environmental variables.

Calculation and modelling of CO2 fluxes

Ecosystem fluxes of CO2 were calculated using both linear and exponential regressions and best fit models were selected for each flux measurement using Akaike's Information Criterion (Burnham & Anderson, 2004). Calculations were done in the MATLAB® (MathWorks Inc., Natick, MA, USA) routine developed by Kutzbach et al. (2007) applying a water vapour correction algorithm and a dead-band of 10 s at the start and end of each 2 min run. This allowed the use of 100 data points for each flux calculation.

The CO2 flux was partitioned into light-dependent GP and light-independent RE. GP was estimated as GP = NEE−RE, i.e., following the atmospheric sign convention, where RE is always positive, GP is always negative and a net flux of CO2 to the atmosphere represents a positive NEE.

GP modelling

Using the resulting GP data from measurements at 0, 25, 50 and 100% of ambient PAR, we first modelled the light response of GP for individual measurement days using a rectangular hyperbolic saturation curve described by Thornley & Johnson (1990) and frequently applied in ecosystem analyses (Model 1):

display math(1)

where α (μg CO2 μmol per photon) is the initial slope of the photosynthetic light response and GPmax (μg CO2 m−2 s−1) is the theoretical maximum rate of photosynthesis at infinite PAR. Linearly interpolated time series of the estimated parameters (α and GPmax) were used to extrapolate GP between measurement days using continuous PAR measurements obtained from the climate station placed at the study site.

As an extension of the first approach, we secondly adapted the photosynthesis model of Long et al. (1985) to include the effect of biomass and temperature on GP as suggested by Mahadevan et al. (2008), but introducing plot-scale RVI data rather than satellite data as an active vegetation proxy (Model 2):

display math(2)

where Amax is the asymptotic maximum equivalent to GPmax and k (μmol photon m−2 s−1) is the half-saturation value. The parameter Tscale represents the temperature sensitivity of photosynthesis as defined by Raich et al. (1991):

display math(3)

where Tmin, Topt and Tmax are minimum, optimum and maximum air temperatures (°C) for photosynthesis respectively. Topt was set to 20 °C and Tmax was set to 40 °C for both crops (Mahadevan et al., 2008). Tmin was set to −2 °C as we had observed net uptake of CO2 down to this temperature during winter and other studies also have reported active photosynthesis at sub-zero temperature in Nordic countries (Sevanto et al., 2006; Lohila et al., 2007). If air temperature falls below Tmin, Tscale is set to be zero (Mahadevan et al., 2008).

For RCG, the flux data were grouped into two periods as preflowering (22 September 2010–14 June 2011) and postflowering (15 June 2011–21 September 2011) and the model was fitted separately for the two periods. The estimated parameters were used to extrapolate GP to the whole measurement period using continuous PAR and temperature data obtained from the climate station at the study site and linearly interpolated RVI measurements.

The total number of days with snow cover was recorded and GP in the snow covered period was set to zero. Similarly, GP in the SB plots was set to zero in the period between sowing and germination.

RE modelling

Initially, RE was modelled using a van't Hoff type first-order exponential temperature model (Model 3):

display math(4)

where R0 is the ecosystem respiration at 0 °C, b is the temperature sensitivity of respiration and T is the temperature (°C).

Second, an improved respiration model was developed (Model 4) inspired by the model of Larsen et al. (2007), but instead of including GP as a dynamic variable of ecosystem respiration we included RVI as a more stable vegetation parameter in our model:

display math(5)

where Rbase represents the basal soil respiration which is independent of foliar biomass, Rfb includes all direct and indirect contribution of foliar biomass in ecosystem respiration, (i.e., both aboveground and belowground respiration), R0 is the basal soil respiration at 0 °C (independent of foliar biomass), β is a scaling parameter between RVI and Rfb, b is the temperature sensitivity of respiration and T is the temperature (°C). To simplify the model, the temperature sensitivity of respiration (b) was assumed to be similar for basal soil respiration and foliar biomass dependent respiration. We fitted the model with air temperature, soil temperature (10 cm depth) and their average and found the best fit using average temperature, which is reported here. Monthly and annual estimates of RE were calculated following a similar extrapolation approach as for GP.

Statistics, uncertainties and model evaluation

Results from subplots were averaged on the plot level for statistical analysis and the standard error (SE) of the measured RVI, biomass yield and CO2 fluxes therefore denote the spatial variation at the plot scale (= 3). No anova of raw data is presented because the temporal effects were obvious.

Fitting of GP and RE models was done using the dynamic fit curve function available in SigmaPlot 2011 (Systat Inc., Chicago, IL, USA) which is based on the Levenberg–Marquardt algorithm. The R2 was used as a measure of the goodness-of-fit (Zar, 1996). In extrapolations, all models were run for each individual collar, i.e., at the subplot level (cf. Larsen et al., 2007) thus allowing NEE to be calculated for each subplot. Results were subsequently averaged by plot and the uncertainty associated with spatial variation in annual estimates of GP, RE and NEE was assessed as described above, i.e., as SE at the plot scale (= 3).

In addition to the uncertainty associated with spatial variation, the fitted GP and RE model parameters were derived with associated SE signifying the magnitude of uncertainty related to modelling. No consensus exists, however, on how to propagate this parameter-uncertainty into a representative modelling uncertainty in the extrapolated annual estimates. Here, we used an approach based on extrapolations made with lower and upper values of the model parameters ± SE to derive a SE for annual GP and RE at each subplot (cf. Drösler, 2005; Elsgaard et al., 2012). This approach produced a slightly asymmetrical interval around the annual estimates, and the largest difference from the mean was used as the resulting SE estimate. Such modelling uncertainties for GP and RE was derived for each individual subplot, and SE for NEE at each subplot (i.e., NEE = GP + RE) was then derived using the quadrature rule of error propagation. Then, for both GP, RE and NEE, similar error propagation was done first to the plot level and then for the mean of the three plots to assess the overall modelling uncertainty for the mean annual fluxes of GP, RE and NEE.

Model performance was evaluated using the Nash-Sutcliffe modelling efficiency (ME):

display math(6)

where Mesi and Modi are the corresponding measured and modelled values, and inline image is the mean of the measured values. The ME can vary from −∞ to 1, but values closer to 1 indicate higher prediction ability of the model (Haefner, 2005).

Results

Environmental conditions and soil characteristics

The average annual temperature at the experimental site (22 September 2010–21 September 2011) was 7.3 °C, being slightly lower than the long-term average of 7.8 °C (Fig. 1a). Total precipitation in the experimental year (750 mm) was similar to the long-term average of 770 mm (Fig. 1b). The first snow fell on 22 November 2010 and melted only after 20 days. The soil temperature at 10 cm depth was close to zero continuously from January to April and freeze-thaw events were observed during January and February. The GWT was generally 30–40 cm below the surface during winter, but it was about 50 cm below the surface during the growth period (Fig. 1b).

Figure 1.

(a) Hourly air (2 m) and soil (10 cm depth) temperatures measured at the study site during the experimental period. (b) Daily recordings of precipitation and depth of ground water table (GWT) during the experimental period.

The average peat thickness at the study site was more than 1 m. The peat was only slightly decomposed across the depth corresponding to values of H3 – H4 on the von Post scale. Bulk density of the peat soil decreased gradually from 0.29 g cm−3 at the surface to 0.12 g cm−3 at 1 m depth. Total organic C content in the peat ranged from 27% to 40%, while total N ranged from 2.2% to 3.1% resulting in peat C : N ratios of 11 to 13. The soil pH ranged from 6.1 to 7.1.

Biomass yield and RVI

Biomass development in RCG started in April and increased to a maximum level of 12 Mg ha−1 that was stable from August until harvest in late September (Fig. 2a). There was already 4 Mg ha−1 standing biomass in RCG plots when SB emerged at the end of May and, during the growing season, the standing dry biomass per unit area was always significantly higher for RCG than for SB. Yet, after emergence of the barley plants, the SB biomass yield increased rapidly until harvest in August and reached a peak level of 10 Mg ha−1 (Fig. 2a).

Figure 2.

(a) Accumulated aboveground dry biomass of spring barley (SB) and reed canary grass (RCG) during the growing season until harvest. (b) Canopy growth pattern of vegetation at RCG and SB plots measured as ratio vegetation index (RVI) during the experimental period. (c) Gross photosynthesis (GP), (d) ecosystem respiration (RE) and (e) net ecosystem exchange (NEE) of CO2 measured at midday from 10:00 hours to 14:00 hours NEE was measured with fully transparent chambers and RE was measured with fully opaque chambers. GP was calculated as NEE - RE. Error bars represent the standard error of the mean of results from three individual plots (= 3).

The patterns of RVI in SB and RCG plots are shown in Fig. 2b. RVI was always higher in SB than in RCG plots during the cold season. In April and May, a sharp increase in RVI was observed in RCG plots, while RVI in SB plots decreased to nearly zero after ploughing for seedbed preparation. However, when RVI peaked during the vegetative growth it was 17% higher in SB plots than in RCG plots. RVI sharply declined in SB plots at the stage of crop maturity, but increased again after harvest as a result of new growth of volunteer grasses. The maturity-induced decrease of RVI in RCG plots was more gradual and without subsequent increases (Fig. 2b).

Measured CO2 fluxes

The observed GP rates showed an expected seasonal pattern for both crops (Fig. 2c). At the first measurement campaign, 2 weeks after crop harvest, GP were −200 and −83 CO2 μg m−2 s−1 in SB and RCG plots respectively. This difference was due to volunteer grass growth in the SB plots after crop harvest. The volunteer grass survived throughout the winter in SB plots and contributed to higher GP rates in SB plots during the cold season. After April 2011, the GP rates increased sharply in RCG plots as the plants started growing and reached a peak of −1920 μg CO2 m−2 s−1 at the middle of May. The RCG started to produce panicle at the middle of June and the leaves started to turn yellow, which caused a decline in the GP rate. Yet, the plants produced new branches after panicle initiation and photosynthesis from new leaves on these branches maintained a relatively stable GP rate throughout the rest of the growth season (Fig. 2c). In SB plots, GP (from volunteer grass) increased during March and April until it was stopped by ploughing for seedbed preparation. Then, after germination of new SB plants, a rapid increase in GP was observed until July. Midday rates of GP in SB plots peaked on 6 July with −2888 μg CO2 m−2 s−1 followed by a rapid decrease during the stage of plant senescence. After SB harvest, volunteer grass growth was again observed which contributed to a slightly increasing GP in the SB plots after harvest.

Ecosystem respiration also followed a strong seasonal pattern (Fig. 2d). At the first measurement, and during the entire cold season, RE was low and rather similar in SB and RCG plots (<170 μg CO2 m−2 s−1). Overall, RE decreased continuously during the winter months reaching almost zero in the beginning of March when the peat at the experimental plots was still frozen. Unlike GP, however, RE was not lower in RCG than in SB plots during winter although the RCG plots were not covered with green biomass; this indicated a higher soil respiration in the RCG plots. A steady increase in RE was observed in both cropping systems after March which coincided with higher temperature and the start of growth of foliar biomass. RE continuously increased in RCG plots until 27 June reaching a peak of 1003 μg CO2 m−2 s−1. In SB plots, a similar increase in RE occurred, but a slight decline was observed after ploughing for seedbed preparation. The peak of RE in SB plots was 970 μg CO2 m−2 s−1 measured on 27 June 2011. After the peak period, RE was lower in SB plots as compared to RCG towards the end of the growth period probably as a result of lower respiration from the senesced SB plants compared to the relatively green RCG plants.

The midday measured NEE of CO2 was negative during a large part of the year indicating a net CO2 uptake in both cropping system during the day (Fig. 2e). Thus, although a large RE was observed during the peak growth period of both crops, the midday GP rate was even higher making both ecosystems net sinks of CO2. A net emission of CO2 was observed only in SB plots in a brief period after ploughing and sowing.

GP models

For GP modelling, the light response was initially fitted for individual campaigns and collars resulting in a total of 96 response curves for each crop - as exemplified in Fig. 3 for each meteorological season. The rectangular hyperbola function (Model 1) fitted well for individual responses with R2 > 0.95 for all curves. Overall, the curves showed a strong dependency of GP to PAR throughout the year and a significantly higher light response (α) and GPmax was observed in SB plots than in RCG plots throughout the cold season (Supplementary Material, Table S1). The GPmax value extended beyond the normal PAR range during the growth season; this indicated that shaded leaves in the lower canopy of tall and dense vegetation could not get enough PAR to reach saturation.

Figure 3.

Responses of gross photosynthesis (GP) to incident photosynthetically active radiation (PAR) at four meteorological seasons (measurement dates shown in upper right corners). Four flux measurements at different level of light intensity in each collar were used to develop the individual light response curves. Fitted lines and parameters for the rectangular hyperbola function (Model 1) are shown in the graphs both for spring barley (SB; closed symbols) and reed canary grass (RCG; open symbols). GPmx is the maximum rate of gross photosynthesis (μg CO2 m−2 s−1) and α is the PAR use efficiency (μg CO2 μmol per photon). Note that the scales are different in all graphs.

Using the second (integrated) GP model (Model 2), integrating the effect of RVI and temperature, the associations between GP and PAR could be analysed in one model run for SB and two for RCG (pre- and postflowering). The R2 obtained for this integrated model was 0.91 for SB and 0.79–0.92 for RCG as shown in Table 2 along with the model parameters obtained (Amax and k). Modelling efficiencies calculated for the integrated model were 0.89 and 0.87 for SB and RCG respectively. Monthly time series of modelled GP fluxes derived by the two model approaches (Model 1 and Model 2) were rather similar as shown in Fig. 4. For the further cropping system analyses, the results of the integrated modelling approach were used.

Figure 4.

Comparison of estimated monthly rates (September 2010 to August 2011) of gross photosynthesis (GP) in each collar using the hyperbolic light response model (Model 1) and the integrated model (Model 2) including ratio vegetation index (RVI) and temperature as variables. Correlation coefficients (r) of modelled GP fluxes derived by the two model approaches are shown for both crops (RCG, reed canary grass; SB, spring barley). The diagonal line indicates 1 : 1 agreement.

Table 2. Fitted parameters (Amax: μg CO2 m−2 s−1, k: μmol photon m−2 s−1) from model estimation of gross photosynthesis (GP) according to the model integrating the ratio vegetation index (RVI) and temperature as variables (Model 2). Uncertainties (shown in parentheses) are standard error (SE) of the parameter estimates. Coefficients of determination (R2) and significance (P) of regression equations are shown. SB, spring barley; RCG, reed canary grass
Crop A max k R 2 P
SB−330 (24)1880 (210)0.91<0.001
RCG (preflowering)−217 (15)1297 (155)0.92<0.001
RCG (postflowering)−355 (35)780 (158)0.79<0.001

RE models

Temperature was identified as a major driver of RE as the classical temperature model (Model 3) showed a R2 value of 0.71 for RCG and 0.81 for SB (Table 3). Yet, including RVI as a proxy for active foliar biomass in total respiration modelling improved the classical model for both crops as seen from the increased R2 (Table 3). Furthermore, as compared to the classical model, the RVI model increased the ME from 0.79 to 0.90 for RCG plots and from 0.81 to 0.88 for SB plots (Fig. 5).

Figure 5.

Comparison of modelled and measured ecosystem respiration (RE) in each collar using (a) a classical exponential temperature model (Model 3) and (b) a modified temperature model including also ratio vegetation index (RVI) as a controlling factor of ecosystem respiration (Model 4). Modelling efficiency (ME) of each model is shown for the two crops (RCG, reed canary grass; SB, spring barley). The diagonal lines indicate 1 : 1 agreement.

Table 3. Fitted parameters (R0: μg CO2 m−2 s−1, b: temperature sensitivity parameter, β: scaling parameter) from model estimation of ecosystem respiration (RE) using a classical van't Hoff model (Model 3) and a modified model (Model 4) including a dependency of RE on foliar biomass measured as ratio vegetation index (RVI). Uncertainties (shown in parentheses) are standard error (SE) of the parameter estimates. Coefficients of determination (R2) and significance (P) of regression equations are shown and apparent temperature coefficients (Q10) are indicated. SB, spring barley; RCG, reed canary grass
CropModel R 0 b Q 10 a β R 2 P
  1. a

    Calculated as Q10 = e10b.

SB3 (classical)18.13 (4.01)0.177 (0.011)5.880.81<0.001
 4 (with RVI)29.90 (4.78)0.135 (0.009)3.871.37 (0.35)0.88<0.001
RCG3 (classical)66.32 (9.95)0.116 (0.008)3.190.71<0.001
 4 (with RVI)58.94 (6.31)0.099 (0.005)2.684.83 (0.67)0.90<0.001

In the modified RE model, β was significant (< 0.001) for both cropping systems showing a significant contribution of foliar biomass to ecosystem respiration (Table 3). As the increased respiration during the crop growth season was partly explained by increased foliar biomass in the modified model, the apparent sensitivity of respiration to temperature (b) was lowered significantly in both crops. As a result, the annual Q10 of respiration decreased from 5.88 to 3.87 in SB plots and from 3.19 to 2.68 in RCG plots, i.e., from rather high to more realistic values (Table 3).

Comparison of modelled CO2 fluxes in RCG and SB cropping systems

The modelled monthly estimates of GP in the cold period from October to March were in total 5% and 11% of the annual uptake in RCG and SB plots respectively (Fig. 6a). Estimated monthly GP peaked at −448 g CO2-C m−2 in RCG and −387 g CO2-C m−2 in SB plots in July which corresponded to a 25% and 29% share of annual uptake respectively.

Figure 6.

Estimated monthly ecosystem fluxes: (a) Gross photosynthesis (GP), (b) ecosystem respiration (RE) with the shaded part representing the contribution of foliar biomass in total respiration (Rfb), and (c) net ecosystem exchange of CO2 (NEE). Error bars represent the spatial variation at plot scale (standard error, = 3).

The modelled monthly estimates of RE showed that 22% and 18% of annual respiration took place from October to March in RCG and SB plots respectively (Fig. 6b). Estimated monthly respiration peaked in June in RCG plots with 350 g CO2-C m−2 and in July in SB plots with 300 g CO2-C m−2 which were 19% and 17% of annual emissions respectively. The contribution of foliar biomass dependent respiration (Rfb) to RE was small throughout the winter in both cropping systems, but it comprised up to 50% during peak biomass growth. The proportion of Rfb in RE was more fluctuating in SB plots than in RCG plots. Field operations (basically tillage) and the shorter growth season of SB were apparently the reasons for this fluctuation.

Reed canary grass plots were significantly larger net source of CO2 than SB plots during the cold season (Fig. 6c). During the growth period, higher respiration in the RCG plots was offset by higher rates of photosynthesis making the ecosystem a total sink of carbon from May to August with the highest net sink capacity in July (−126 g CO2-C m−2). SB plots acted as a net sink of carbon from March to July except in May, where a fallow period of almost 2 weeks occurred after ploughing the field for seedbed preparation. SB plots acted as a large source of CO2 in August at the senescence stage of plants, and then continued as a weak source of CO2 during the cold season.

Modelled annual estimates of GP and RE, respectively, were 37 and 47% larger for RCG plots than for SB plots (Fig. 7). Increased RE in RCG plots was mainly contributed by Rfb (114% relative increase) and partly by higher Rbase (27% relative increase). The annual NEE of CO2 was close to zero, but negative in SB plots (−41 g CO2-C m−2 yr−1) and positive in RCG plots (69 g CO2-C m−2 yr−1).

Figure 7.

Estimated annual gross photosynthesis (GP), ecosystem respiration (RE) with the shaded part representing contribution of foliar biomass in total respiration (Rfb), and net ecosystem exchange of CO2 (NEE) for the cropping systems with spring barley (SB) and reed canary grass (RCG). Error bars represent the spatial variation at plot scale (standard error, = 3).

The magnitude of uncertainty related to spatial variation was low for both cropping systems (Figs. 6 and 7), but a more significant uncertainty was related to the modelling approach. Thus, considering the uncertainty related to modelling, the estimated GP ranged from −1672 to −1964 and −1238 to (−1420 g CO2-C m−2 yr−1 in RCG and SB plots respectively. Similarly, RE ranged from 1699 to 2075 and 1116 to 1460 g CO2-C m−2 yr−1 in RCG and SB plots respectively. With NEE calculated as the (small) difference between RE and GP, the propagation of uncertainty in RE and GP resulted in a high relative uncertainty of NEE fluxes thus ranging from −169 to 307 and −236 to 154 g CO2-C m−2 yr−1 in RCG and SB plots respectively.

Discussion

Annual CO2 exchange

In this study, we observed an expected higher GP by RCG than by SB cultivation, but also RE was increased by RCG cultivation. Higher GP in RCG plots was favoured by a longer growth period, but on the other hand the peak GP from SB was significantly higher than from RCG. In terms of annual NEE of CO2, both cropping systems showed a resulting emission close to zero and no advantage of the energy cropping system was observed.

Annual CO2 exchange from cropping systems with RCG and SB on organic soils has not previously (to our knowledge) been compared in side-by-side plots. A few Finnish studies exist, however, where the cropping systems have been studied separately. Annual GP of RCG reported in the present study (−1818 g CO2-C m−2) was substantially higher than reported for the same crop in Finland (−690 g CO2-C m−2) by Shurpali et al. (2009) where RCG was cultivated at an active peat extraction site. Similarly, annual estimated GP of SB in our study (−1329 g CO2-C m−2) was higher than from SB cultivation in Finland (−950 g CO2-C m−2) as reported by Maljanen et al. (2001). Larger aboveground biomass production at our study site (1200 g dry mass m−2 yr−1 for RCG and 1040 g dry mass m−2 yr−1 for SB) corroborates the higher GP as in the Finnish studies maximum biomass production in the best year was only 470 g dry mass m−2 yr−1 for RCG and 440 g dry mass m−2 yr−1 for SB (Maljanen et al., 2001; Shurpali et al., 2009). Higher annual PAR and temperature, as well as a longer growing season, in Denmark compared to Finland are all factors favouring a higher GP and concomitant biomass production.

Our estimated annual RE (1887 g CO2-C m−2 for RCG and 1288 g CO2-C m−2 for SB) were also much higher than the RE (555 g CO2-C m−2) reported for RCG by Shurpali et al. (2009) and for SB (830 g CO2-Cm−2) by Maljanen et al. (2004). There might be several reasons for this higher RE. First, higher biomass production must have contributed to larger plant respiration. Second, annual temperature, which is the most important factor of RE, at our experimental site (7.3 °C) was more than three times higher than at the Finnish study sites (2.1 to 2.4 °C). Third, the depth of the GWT, which can also be an important determining factor for RE in peatlands (Alm et al., 2007), may not have limited soil respiration from the upper peat layer in our experiment as the GWT was always below 50 cm during the growth season (Fig. 1b). The importance of such climatic differences in the magnitude of annual RE was substantiated by other recent annual estimates of RE in agricultural peatlands in Denmark which ranged from ca. 1200 to 3300 g CO2-C m−2 at five sites with arable crop rotation and from 2500 to 2900 g CO2-C m−2 at permanent grassland sites (Elsgaard et al., 2012).

Uncertainties of the modelled annual fluxes were estimated for both spatial variation and as uncertainty related to modelling (Laine et al., 2009). The uncertainty related to spatial variation was generally small both in measured (Fig. 2) and modelled (Figs. 6 and 7) fluxes. The uncertainty related to modelling was more important at least for the relative uncertainty in NEE. However, the significance of the estimated modelling uncertainty is difficult to evaluate. Different approaches have previously been applied for estimating the uncertainty in modelled annual fluxes based on chamber measurements. These methods include parameter estimation by boot-strapping techniques (Laine et al., 2009), Taylor series approximations of response functions (Bubier et al., 1999) and derivation of SE from models run with parameter estimates ± SE (Drösler, 2005). However, no consensus seems to exist on best-available uncertainty estimates related modelled CO2 fluxes. A comprehensive uncertainty analysis is beyond the scope of this article, but we considered the present modelling uncertainty estimates as indicative of similar levels of uncertainty in the two cropping systems, and for the data interpretation, we put more emphasis on the uncertainties related to spatial variation.

Application of modified CO2 flux models

This study presents simple yet useful GP and RE models which introduce plot-scale RVI data as an active vegetation proxy. The models confirmed that biomass is one of the major drivers of both GP and RE at cultivated peatlands together with environmental variables such as temperature and PAR (Lohila et al., 2004; Maljanen et al., 2004).

Using the integrated GP model (Model 2), interpolations between derived model parameters are avoided as opposed to the Model 1 approach. Thus, the integrated GP model, with simple mathematical structure and small numbers of parameters, can be a robust method to simulate both diurnal and seasonal variations in GP as it incorporates both weather (temperature) and crop development parameters (RVI) as drivers of photosynthesis. Although the term Tscale did not enhance the model performance considerably, goodness-of-fit between observed and modelled values was increased in winter months when photosynthesis was limited by cold weather. A reduced model without Tscale predicted photosynthesis in winter even below Tmin which eventually projected unrealistically high wintertime photosynthesis; therefore the term Tscale remained in the model. Comparing the resulting monthly GP from the two model approaches showed a Pearson's correlation coefficient (r) of 0.97 and 0.95 for SB and RCG respectively. Such model comparison and consistency has not previously been presented to our knowledge, but importantly serves to substantiate the use of either approach at least in the present ecosystem analyses.

Ecosystem respiration showed a good seasonal correlation to RVI (= 0.75 in RCG and = 0.70 in SB plots). Therefore, we expanded Model 3 to include RVI as a proxy of foliar biomass in RE as a modification of the approach of Larsen et al. (2007), who used GP as an instantaneous driver of RE in a temperate heath ecosystem. RVI is a rather stable vegetation parameter and was preferred because an instantaneous coupling of photosynthesis with respiration is uncertain at least for tall-statured vegetation (Kuzyakov & Gavrichkova, 2010) and would imply a strong sensitivity of RE with diurnal changes in PAR.

Expectedly, including other forms of vegetation measurements, such as leaf area index (LAI), plant index (PI) and amount of biomass (Lohila et al., 2004; Maljanen et al., 2004; Burrows et al., 2005) would also improve our model fits, but canopy reflectance (such as RVI) may be preferable as it to some extent integrates the amount of vegetation and the photosynthetic activity. For example, studies have documented a sharp decline in dark respiration during plant maturity stages (Constable & Rawson, 1980; Marron et al., 2008) which may not be reflected in the biomass, but which was manifested by a sharp and concurrent decline in RVI and RE in our study (Fig 2b and d).

Including RVI in the RE model improved the model fit significantly, but also made it possible to estimate the contribution of green foliar biomass to RE. This regression analysis technique can give an approximate estimation of plant contribution to total ecosystem respiration which is very difficult (and may become biased) with field experiments (Kuzyakov, 2006). The model estimated a total annual contribution of Rfb to RE of ca. 22% and 32% in SB and RCG plots respectively. Longer growing period of RCG with higher amount of standing biomass compared to SB must be the reason for increased contribution of Rfb to RE. It was notable that Rbase was also higher in RCG plots than in SB plots. The possible reason for such higher Rbase may be related to increased priming of soil organic carbon turnover by more root exudates (Tufekcioglu et al., 2001) as total GP was significantly higher in RCG plots. Furthermore, potentially influencing the lower Rbase in SB plots, we did not observe an expected tillage-induced CO2 emission peak from SB plots after ploughing; this loss may be relatively short-lived (Jackson et al., 2003) and could have been missed due to the sampling frequency.

Potentials to improve NEE and GHG balance in RCG cultivation

Net ecosystem exchange of CO2 was near to zero for both cropping systems – yet with the RCG system acting as a weak net source of CO2. This result was to some extent surprising as better CO2 balance was expected in perennial energy crop cultivation. However, there might be a number of management options of RCG cultivation to reduce the CO2 emissions. (i) RCG is an aquatic plant and it can tolerate water logging conditions for several weeks (Shurpali et al., 2010; Zhou et al., 2011). Raising the water level of drained peatlands when cultivating RCG may reduce the CO2 emissions from the soil significantly without affecting the carbon uptake by the plants. However, elevated GWT – at least to levels closer than 20–30 cm to the soil surface – might increase CH4 emissions as a result of the formation of anaerobic conditions simultaneously conducive of methanogenesis and limiting the activity of methane oxidizing soil bacteria (Lai, 2009; Schäfer et al., 2012). (ii) RCG being an invasive species has not been subjected to intensive breeding for higher photosynthesis rate and biomass yield. Therefore, there is a certain potential to improve the CO2 uptake by increased biomass yield from varietal improvement. (iii) Studies on better agronomic practices like nutrient management, and management of harvesting time and frequency are important issues that also may increase CO2 uptake and biomass yield in a peat ecosystem used for RCG cultivation.

In addition to CO2 fluxes from the field, the role of carbon balances, GHG balances and potential use of biomass for replacement of fossil fuels should be considered when evaluating the overall emissions from the two cropping systems. In the present study, we compared the RCG and SB cropping systems in terms of NEE of CO2 from an atmospheric perspective. Therefore, the annual fluxes presented in Fig. 7 does not account for the C removal from the field by harvesting of biomass. The amount of C removed as biomass was 470 and 540 g C m−2 in SB and RCG plots, respectively, assuming a 45% C content in biomass (cf. Fig. 2a). Considering these losses, both cropping systems acted as net sources of carbon from a soil ecosystem perspective, still with the RCG cropping system representing a larger C loss (609 g C m−2) than the SB (429 g C m−2). However, the higher biomass yield from RCG plots could potentially replace a higher amount of fossil fuel when the biomass is utilized for bioenergy production. Therefore, a more detailed life-cycle assessment of annual food crop and perennial energy crop cultivation on peatland is needed. This could also take into account the GHG emissions associated with the cultivation step from such factors as energy needed for field operations and N2O emissions associated with production and use of mineral fertilizers (Cherubini, 2010; Kasimir-Klemedtsson & Smith, 2011). Here, the lower intensity (e.g., lower N-fertilizer rate) in the RCG bioenergy cropping system would contribute to a lower GHG emission profile than for the SB cropping system. For soil N2O flux, this was partly confirmed by measurements of annual N2O emission in the experimental plots, although the emissions were rather low from both cropping systems, i.e., 1.5 and 2.1 kg N2O ha−1 yr−1 from RCG and SB plots, respectively, corresponding to 12 and 17 g CO2-C equivalents m−2 yr−1 (S. Karki, personal communication). These results further served to substantiate that NEE of CO2 and carbon balances were the most important components of GHG balances for the ecosystem emissions at the drained peatland used for agricultural purposes.

Acknowledgements

The study was supported by funding from the European Regional Development Fund as a part of the projects ENERCOAST (http://enercoast.net/) and BioM (http://www.biom-kask.eu/). The authors like to express their gratitude for excellent technical assistance of Jørgen M. Nielsen, Stig T. Rasmussen, Karin Dyrberg, Bodil B. Christensen and Jens B. Kjeldsen.

Ancillary