Effects of drought conditions on the carbon dioxide dynamics in a temperate peatland

Drought is arguably the most important regulator of inter-annual variation in net ecosystem CO2 exchange (NEE) in peatlands. This study investigates effects of drought periods on NEE and its components, gross primary production (GPP) and ecosystem respiration (Reco), on the basis of eddy covariance measurements of land–atmosphere exchange of CO2 in 2006–2009 in a south Swedish nutrient-poor peatland. Two drought periods had dissimilar effects on the CO2 exchange. In 2006, there was a short but severe drought period in the middle of the growing season resulting in increased Reco rates, but no detectable effect on GPP rates. In contrast, in 2008 the drought period began early in the growing season and lasted for a longer period of time, resulting in reduced GPP rates, suggesting that GPP is most sensitive to drought during leaf out and canopy development compared with the full canopy stage. Both in 2006 and in 2008 the peatland acted as an annual source of atmospheric CO2, while in 2007 and 2009, when there were no drought periods, the peatland constituted a CO2 sink. It was concluded that the timing, severity and duration of drought periods regulate the effects on peatland GPP, Reco and NEE.


Introduction
Northern peatlands are important in the context of biospheric feedback effects on the climate system, as they store vast amounts of organic carbon (C) in their soils. Estimates of the amount of C stored in peatlands range 270-450 Pg C (Gorham 1991, Tolonen and Turunen 1996, Turunen et al 2002, which equals about one-half of the current atmospheric C pool. Hydrological conditions exert a strong control on both peatland formation and functioning; and there is a growing consensus that drought is the most important controller of inter-annual variation in net ecosystem CO 2 exchange (NEE) in peatlands (Limpens et al 2008, Lafleur 2009). Drought implies a lowering of the ground water table which potentially enhances soil respiration and decreases photosynthesis, which may turn the ecosystem into a source of CO 2 to the atmosphere (Lafleur 2009). Earlier studies have found that dry conditions affect CO 2 fluxes in peatlands (Shurpali et al 1995, Lafleur et al 1997, Alm et al 1999, Arneth et al 2002, Lafleur et al 2003, Aurela et al 2007, Lund et al 2007; however, the underlying causes and dynamics affecting ecosystem respiration (R eco ) and gross primary production (GPP) are not fully understood.
During a drought period, vascular plants respond physiologically and structurally to prevent excessive water loss. Physiological responses include reduced enzymatic activity and stomatal closure, while structural responses include reductions in leaf area due to leaf senescence and leaf shed (van der Molen et al 2011). Sphagnum mosses, which are an integral part of most peatlands due to their ability to intercept airborne nutrients (Malmer and Wallén 2005) and their recalcitrant litter (Aerts et al 1999), lack xylem to transport water but do absorb water through leaves and stems. During drought conditions, Sphagnum mosses suffer from various degree of desiccation resulting in reduced photosynthetic rates (Schipperges and Rydin 1998).
Ecosystem respiration includes both autotrophic and heterotrophic respiration. Autotrophic respiration, which has been reported to account for 30-90% of R eco in peatlands (Frolking et al 2002, Kurbatova et al 2009, St-Hilaire et al 2010, is related to plant biomass and activity (Flanagan and Johnson 2005). Heterotrophic respiration is also related to vegetation processes through its dependency on supply and quality of substrates provided by plants (Giardina and Ryan 2000). In peatlands, drought conditions will decrease soil wetness and lower water tables, resulting in thicker aerobic layers and increased potential for aerobic respiration (Lafleur 2009).
In this study we present land-atmosphere exchange measurements of CO 2 between 2006 and 2009 from a temperate, nutrient-poor peatland (bog) in southern Sweden. The main objective was to investigate the effects of drought conditions, identified through water table depth (WTD) measurements, on the CO 2 dynamics and CO 2 balance of the peatland. Measurements of NEE were conducted using the eddy covariance technique, allowing for landscape scale interpretation of whole ecosystem response to droughts.

Site description
The study site, Fäjemyr, is a temperate, ombrotrophic peatland (raised bog) located in southern Sweden (56 • 15 N, 13 • 33 E). The long-term (1961-90) mean annual temperature is 6.2 • C with July being warmest month (15.1 • C) and January coldest (−2.4 • C). Long-term mean annual precipitation is 700 mm. The water table fluctuates 0-20 cm below the surface, and the topographical pattern is dominated by hummocks, lawns and carpets, while hollows and open pools are scarce. The vegetation is dominated by dwarf shrubs (Calluna vulgaris, Erica tetralix), sedges (Eriophorum vaginatum) and Sphagnum mosses (S. magellanicum, S. rubellum). The depth of accumulated peat is approximately 5 m. More detailed site descriptions can be found in Lund et al (2007Lund et al ( , 2009).

Measurements
Measurements of the land-atmosphere net ecosystem exchange (NEE) of CO 2 were performed using an eddy covariance system, consisting of a closed-path infrared gas analyser (Li-Cor 6262, Li-Cor Inc., USA) and a three-dimensional sonic anemometer (Gill R3, Gill Instruments, UK). The anemometer was mounted on a mast at a height of 3.4 m above ground. In all directions surrounding the mast, there were at least 290 m of homogeneous surface properties. According to previous footprint modelling, more than 90% of the flux emanated from this area for atmospheric conditions ranging from weakly stable to unstable (Lund et al 2007). Additional environmental variables such as air (EMS 33, EMS Brno, Czech Republic) and soil temperatures (Type T thermocouples), precipitation (ARG100, Skye Instruments, UK), water table depth (WTD; SKPS 1830, Skye Instruments, UK) and photosynthetic photon flux density (PPFD; JYP 1000, SDEC, France) were monitored continuously at the site. Water table depth was measured in a lawn community type, i.e. the zero datum was set at an intermediate topographical level.
Normalized difference vegetation index (NDVI) data was acquired from the Terra/MODIS vegetation index product MOD13Q1 (16 day temporal resolution, 250 m spatial resolution). The selected single grid cell subset included the eddy covariance mast and covered only peatland area, excluding surrounding forest. Due to the prevailing south-westerly winds at Fäjemyr, a cell where the eddy covariance tower was situated in the eastern part was found to largely overlap the flux footprint (Schubert et al 2010).

Data handling
Raw data from the eddy covariance system was collected at a frequency of 20 Hz and processed using the Ecoflux software, producing half-hourly fluxes according to Fluxnet methodology (Aubinet et al 1999). The software corrects for sensor separation by covariance optimization, air density fluctuations and frequency response losses. The storage term was calculated based on the single-point CO 2 concentration measurements and added to the flux. Data was screened for periods of low atmospheric mixing (defined as friction velocity < 0.1 m s −1 ). The system setup and data processing are described in more detail in Lund et al (2007).
Gaps in CO 2 flux data were filled using the following approaches. Firstly, small gaps (≤2 h) were filled using linear interpolation. Secondly, the Misterlich function (Falge et al 2001) was parameterized for daytime periods (PPFD > 10 µmol m −2 s −1 ) using an 8 day moving window (time step 1 day) with PPFD as independent variable; where F csat is CO 2 exchange at light saturation (mg CO 2 m −2 s −1 ), R d is dark respiration (mg CO 2 m −2 s −1 ) and α is initial slope of the light response curve (mg CO 2 µmol −1 PPFD). The parameterization of the light response curve was only considered significant when all parameters (F csat , R d , α) were significantly different from zero (p < 0.05). Gaps in PPFD were filled using data from STRÅNG (http://strang. smhi.se/), an operational mesoscale radiation model from SMHI (Swedish Meteorological and Hydrological Institute) that produces radiation data covering Scandinavia at a resolution of about 11 km. The obtained light response curve parameters and PPFD data were subsequently used to fill gaps in CO 2 flux data. Remaining CO 2 flux gaps, smaller than 7 days, were filled with mean diurnal variation using an 8 day window (Falge et al 2001). Remaining longer gaps (DOY 208-255 and 300-327 in 2006;DOY 97-108 in 2007;DOY 225-235 in 2008;DOY 161-169, 221-231 and 233-240 in 2009) were filled using artificial neural networks (Papale and Valentini 2003). The total uncertainty in annual CO 2 sums was estimated based on the procedure described in Elbers et al (2011) and Lund et al (2012). In addition, the effect of the long gaps (>7 days) was evaluated by using alternative gap-filling methods. The parameters of the light response curve (F csat , R d , α) were estimated based on available PPFD and NEE data 4 and 8 days prior to and following the gap, respectively. In case the parameterization of the light response curve was not significant, the gap was filled using average NEE 4 and 8 days prior to and following the gap, respectively. The uncertainty related to long data gap-filling was determined from the standard deviation of the three resulting NEE sums (calculated from artificial neural network and alternative gap-filling method based on 4 and 8 days, respectively).
Gross primary production (GPP) was subsequently modelled using the light response curve (equation (1)) by subtracting R d . Daytime ecosystem respiration (R eco ) was calculated as the difference between measured and gap-filled NEE and modelled GPP; while night-time R eco equalled measured and gap-filled NEE.
To extract seasonality and reduce random noise, time series of original MODIS NDVI were smoothed using local Savitzky-Golay curve fits in the TIMESAT software (Jönsson and Eklundh 2004). Based on the quality information of the MOD13Q1 data, higher weights were assigned to data points with higher quality and lower weights to those with lower quality (Schubert et al 2010).

Meteorological conditions
The annual temperature during the study period was on average 1.6 • C above the long-term (1961-90) average (figure 1). Most of the excess heating occurred during the winter months; the period DJF was on average 2.1 • C warmer while MAM, JJA and SON ranged from 1.0 to 1.5 • C warmer. The annual precipitation was on average slightly above the long-term average (+88 mm); however, there were large variations within and between years. For example, in 2006 the months of June and July only received 31% of normal (1961-90) precipitation, while in 2007 the same months received 272% of normal precipitation.
The integrated effects of temperature and precipitation on the Fäjemyr peatland hydrology can be studied through measurements of water table depth (WTD; figure 2). In 2006, the water table was close to the surface during early growing season while it fell below 10 cm below ground on DOY 164 and reached −24.9 cm on DOY 208. Heavy rainfall in early August 2006 rapidly raised the water table above 10 cm on

Net ecosystem exchange of CO 2
The inter-annual variation in net ecosystem exchange (NEE) was large with a range of −108 to 87 g CO 2 m −2 yr −1 (figure 3). Most of this variability was caused by differences in NEE during the growing season between late May (DOY 150) and early September (DOY 250). In 2008, the summer-time CO 2 net uptake rate was lower compared with other years, which coincided with the drought period (figure 2). The summer-time net CO 2 uptake during 2008 barely balanced the emissions during previous winter and spring, and after taking the net CO 2 emissions during late part of the year into account the peatland acted as an annual source of 86.4 g CO 2 m −2 , with an associated uncertainty of ±29.1 g CO 2 m −2 . In 2006, the net uptake rates during early summer were similar to the wetter years (2007 and 2009); however, around DOY 201 the peatland no longer acted as a sink of CO 2 on a daily basis. Again, this event corresponded to drought conditions with low water table (figure 2). In 2006, the peatland was an annual source of 52.4 ± 26.0 g CO 2 m −2 while for 2007 and 2009, when water tables were close to the surface throughout the growing season, the annual CO 2 balances were −107.7 ± 28.1 and −106.1 ± 16.3 g CO 2 m −2 , respectively.
The four year CO 2 balance of this temperate bog (−18.8 ± 102.7 g CO 2 m −2 or −5.1 ± 28.0 g C m −2 ) was close to neutral, i.e., zero net CO 2 exchange. Other studies on comparable peatland sites have generally reported stronger CO 2 sinks: in a boreal poor fen in Sweden, the annual CO 2 -C balance ranged between −48 and −61 g C m −2 (Sagerfors et al 2008) based on three years of measurements. Roulet et al (2007) reported six years of annual CO 2 -C balances for a temperate Canadian bog ranging −2 to −112 g C m −2 . The annual CO 2 -C balance of an Irish blanket bog based on six years of measurements ranged −13 to −84 g C m −2 (Koehler et al 2011). Although being multi-year averages, these budgets may still not reflect the true contemporary CO 2 exchange in northern peatlands, as extreme events such as droughts may occur on timescales longer than 3-6 yr. Similarly, the CO 2 balance at Fäjemyr may be biased by the fact that summer-time droughts occurred during two out of the four years in this study.

CO 2 flux components
In 2008, ecosystem respiration (R eco ) rates were similar compared with other years (figure 4), while gross primary production (GPP) rates were generally lower (i.e. less negative GPP rates) during the drought period (DOY 126-217). In  2006, when the drought period began DOY 164, GPP rates were comparable to or even higher than the wet years of 2007 and 2009; however, R eco rates during the drought period of 2006 were also higher compared with other years.
Daytime NEE and GPP are largely regulated by the amount of incoming photosynthetic photon flux density (PPFD). To be able to assess how the drought periods affected the Fäjemyr peatland ecosystem independent from variations in PPFD, daily GPP rates at PPFD > 1000 µmol m −2 s −1 were calculated, which represent GPP at or close to light saturation (figure 5). Such GPP rates were not lower (i.e. less negative) during the drought period in 2006, while in 2008 GPP rates were consistently lower during DOY 140-220.
3.4. Effects of drought conditions on peatland vegetation and CO 2 exchange 3.4.1. Short-term effects (days to years).
The effects of drought and low water tables at Fäjemyr on the land-atmosphere exchange of CO 2 and its components were regulated by the timing of the drought event. In 2008, when water tables were low during early growing season (figure 2), the vascular vegetation of the peatland was likely affected by water stress during its developing phase. It was also observed in the field that Sphagnum mosses suffered from desiccation during this period. This resulted in decreased biomass build-up (figure 6) and lowered photosynthetic rates throughout large parts of the growing season (figure 5), and the peatland acted as a source of CO 2 on an annual basis. In 2006, the drought period occurred later when vascular plants were already established. Therefore, GPP at light saturation was not affected (figure 5); however, the unusually low water table exposed a thicker layer of peat for aerobic decomposition, and the ecosystem turned into a source of C at an earlier moment in time compared with the wet years. This argument is further supported by TIMESAT-smoothed NDVI data ( Previous studies have suggested that droughts in peatlands primarily affect GPP (Shurpali et al 1995, Arneth et al 2002, Lafleur et al 2003, while others have argued that R eco is more affected (Alm et al 1999, Aurela et al 2007. The results in this study indicates that the timing, severity and duration of the drought periods regulate the effects on GPP, R eco and NEE in a temperate bog. A short but severe drought period during the peak growing season may not significantly affect GPP, but can increase R eco due to increased heterotrophic respiration. On the other hand, a long drought period during the vegetation developing phase may significantly decrease GPP and to a lesser extent also R eco . Furthermore, it is likely that various peatland types will show dissimilar responses to dry conditions (Lafleur 2009, Sulman et al 2010. Aurela et al (2004) argue that peatlands characterized by hydrological buffers, such as fens (minerotrophic peatlands) with a dynamic connection to the catchment scale water system, will increase their C uptake in a warmer climate. Bogs (ombrotrophic peatlands) on the other hand, which are decoupled from the groundwater of surrounding watershed, are dependent on precipitation inputs to maintain their water balance. A possible generalization may thus be that bogs are more sensitive to dry conditions in terms of their CO 2 balance as compared with fens, since the former are rainwater-fed while the latter is fed by rain and groundwater. The relative contribution of vascular plants and mosses will act to regulate peatland ecosystem resilience to drought, due to differences in water uptake and evapotranspiration processes (Sulman et al 2010).
3.4.2. Long-term effects (years to centuries). If dry summer conditions become more abundant in the future, the C sink functioning of peatlands that are sensitive to variations in water balance may cease. Summer-time (JJA) temperature and precipitation in northern Europe are predicted to increase with on average 2.7 • C and 2%, respectively, for the period 2080-99 compared with 1980-99, according to 21 global climate models for the A1B scenario . According to our data the summer period (JJA) in 2006 was 2.6 • C warmer and received 7% more precipitation compared with long-term average 1961-90. The precipitation was, however, not equally distributed with 80% of the rainfall occurring in August. Still, the mid-summer drought period in 2006 was associated with the lowest water table and highest R eco rates during the study period. Modelling studies assessing the impact of water level variations on peatland CO 2 exchange have shown that heterotrophic respiration is most sensitive to changes in WTD (Yurova et al 2007, St-Hilaire et al 2010, due to increases in soil temperature and increased potential for aerobic decomposition. But, as heterotrophic respiration may contribute as little as 10% to peatland R eco , the relative impact of decreased GPP and autotrophic respiration through effects on vascular plant conductance and moss water content may be more important for the contemporary net CO 2 exchange (St-Hilaire et al 2010). However, on a longer timescale, an increased frequency and extent of drought conditions are likely to have important effects on peat soil decomposition, due to the feedback effect between water table and decomposition rates, known as the paludification (pond-making) process (Clymo 1984, Ise et al 2008. Lowered WTD reduces peat soil water retention capacity and the summer-time temperature insulation by peat, increasing the sensitivity of peat decomposition to temperature. Ise et al (2008) found that a long-term simulation with experimental warming of 4 • C caused a loss of 40 and 86% of shallow and deep peat soil organic C, respectively. However, their study did not include vegetation dynamics. Peatland plants will respond to changes in temperature, hydrology and nutrient status, and changes in vegetation structure and composition will act to regulate CO 2 dynamics. Drier and warmer conditions will stimulate vascular plant growth at the cost of Sphagnum mosses , Limpens et al 2008, resulting in increased litter decomposability. Trees may invade peatlands with sufficiently lowered water tables, however, nutrient-poor bogs can support only limited tree growth (Alm et al 1999). An increased amount of vascular plants, especially shrubs, will increase peatland above-ground biomass, and given increased soil decomposition rates there may thus be a transfer of C from the soil pool to standing biomass, although such C reallocation should be transient in time. By taking these factors into account, it is likely that if increased evapotranspiration rates are not balanced by increased precipitation rates in a warmer future, peatlands, especially bogs, will weaken their sink strength or even become persistent sources of atmospheric CO 2 .

Conclusions
This study underlines the importance of drought periods as regulators of inter-annual variation in CO 2 exchange in peatlands. Based on four years of eddy covariance CO 2 flux measurements, we found that timing, severity and duration of drought regulated the effect on the CO 2 dynamics. A short but severe drought period during mid-summer 2006 increased R eco rates, while GPP was not affected; while an early and prolonged drought period in 2008 decreased GPP rates. Data presented in this study can be useful to study the effects of droughts on the biophysical processes responsible for the net CO 2 exchange in more detail, using process-based peatland models with high temporal resolution such as McGill wetland model (St-Hilaire et al 2010) and the modified GUESS-ROMUL model (Yurova et al 2007).