Low impact of dry conditions on the CO2 exchange of a Northern-Norwegian blanket bog

Northern peatlands hold large amounts of organic carbon (C) in their soils and are as such important in a climate change context. Blanket bogs, i.e. nutrient-poor peatlands restricted to maritime climates, may be extra vulnerable to global warming since they require a positive water balance to sustain their moss dominated vegetation and C sink functioning. This study presents a 4.5 year record of land–atmosphere carbon dioxide (CO2) exchange from the Andøya blanket bog in northern Norway. Compared with other peatlands, the Andøya peatland exhibited low flux rates, related to the low productivity of the dominating moss and lichen communities and the maritime settings that attenuated seasonal temperature variations. It was observed that under periods of high vapour pressure deficit, net ecosystem exchange was reduced, which was mainly caused by a decrease in gross primary production. However, no persistent effects of dry conditions on the CO2 exchange dynamics were observed, indicating that under present conditions and within the range of observed meteorological conditions the Andøya blanket bog retained its C uptake function. Continued monitoring of these ecosystem types is essential in order to detect possible effects of a changing climate.


Introduction
Northern peatlands are important ecosystem types in a climate change context, as they hold large amounts of organic carbon (C) in their soils, amounting to about half of the current atmospheric C pool (Gorham 1991). Peatlands are wetlands that during the last millennia have converted atmospheric carbon dioxide (CO 2 ) into soil organic material, i.e. peat, because of reduced decomposition rates due to anoxic soil conditions. Changes in temperature and soil wetness can modify the C sink functioning of peatlands, with potential feedback effects on the climate system (Ise et al 2008, Dorrepaal et al 2009.
For peatlands as well as for most other ecosystem types, net ecosystem exchange (NEE) of CO 2 is the main component of the C budget. However, due to prevalent wet conditions, CH 4 emissions and C loss through runoff can also be of importance for the peatland C budget (Roulet et al 2007, Koehler et al 2011. Hydrological conditions exert a strong control on peatland NEE (Limpens et al 2008, Lafleur 2009, Lund et al 2012. Drier soils can lead to increased soil respiration as well as decreased plant photosynthesis (Lafleur 2009). Dependent on timing, severity and duration of a drought, the effects on NEE, gross primary production (GPP) and ecosystem respiration (R eco ) may differ (Lafleur 2009, Lund et al 2012. In addition, hydrological settings, primarily whether the peatland is connected to the groundwater system (fen) or not (bog), as well as vegetation composition, regulate peatland response to drought periods (Sulman et al 2010, Lund et al 2012. Blanket bog is a distinctive peatland type restricted to maritime climates, where a positive water balance allows ombrotrophic vegetation to develop over extensive areas (Charman 2002). Palaeoecological records indicate that climatic variability, affecting soil wetness, has regulated the development of blanket bogs (Ellis and Tallis 2000). A recent modelling study showed that blanket bogs are endangered by climate change, because of marked shrinkage of their present bioclimatic space as a consequence of global warming, which may lead to peat erosion and vegetation change (Gallego-Sala and Prentice 2012).
Few studies exist on the contemporary C exchange in unmanaged blanket bogs, with the exception of a temperate Atlantic blanket bog in Ireland (Glencar) from which a host of work has been published on CO 2 fluxes (e.g. Sottocornola and Kiely 2005, Sottocornola and Kiely 2010, McVeigh et al 2014 as well as the exchange of CH 4 and DOC (Koehler et al 2011). In addition, Beverland et al (1996) studied the exchange of CO 2 and CH 4 in a blanket bog area in Scotland. To our knowledge, there has been no extensive study on land-atmosphere C exchange from more northerly situated blanket bogs.
In this study, we present 4.5 years of eddy covariance (EC) measurements of the land-atmosphere exchange of CO 2 in the Saura blanket bog area on the island of Andøya in northern Norway. The purpose of the study was to describe the multi-year CO 2 exchange in the bog, and to investigate impacts of dry conditions (i.e. low soil water content (SWC) and high vapour pressure deficit (VPD)) on NEE, GPP and R eco .

Site description
The site is located in the middle boreal vegetation zone (Moen 1999) at the Saura peatlands on the island of Andøya, Nordland County, northern Norway (69°08' N, 16°01' E, 17 m.a.s.l.; figure 1). Despite the high latitude, the site does not have permafrost due to the maritime influence of the nearby Atlantic Ocean. Long-term (1961Long-term ( -1990 mean annual air temperature is 3.6°C, with February being the coldest month (−2.2°C) and July and August the warmest (both 11.0°C). Long-term mean precipitation is 1060 mm per year (data from station 87 110 operated by Norwegian Meteorological Institute, located approximately 17 km north of the Saura site).
The peatlands on northern Andøya are dominated by ombrotrophic bogs and poor fens (Buys 1992). The un-eroded concentric raised bogs of Andøya are the most northern within Europe (Vorren et al 2007), many assessed to be of national and international conservation value. Intermediate fens are scattered in areas influenced by former sea shore shell deposits. The Saura blanket bog is dominated by hummocks with a relatively dry surface. Peat depth is expected to be approximately 2-3 m, similar to the raised bog Sellevollmyra (Vorren et al 2007) located ca. 7 km southwest of Saura, underlain by late glacial and Holocene raised beaches. Hollows are present between the hummocks. Vegetation and microtopography surrounding the EC system was investigated in August 2009. Plots (1 m 2 ) were established in a cross centered close to the EC system, and sites were selected at a distance of 50, 100 and 200 m from the centre. At each site two plots were established, one at a hummock and one in a hollow. We recorded both the field (vascular plants) and bottom layer (mosses and lichens) at each plot.

Instrumentation
The EC system, consisting of a LI-7500 open-path gas analyzer (Li-Cor, USA) and a CSAT3 3D sonic anemometer (Campbell Sci., UK), was installed at a height of 2 m during the summer of 2008. Data from both sensors was collected at a frequency of 10 Hz on a CR3000 data logger (Campbell Sci., UK). Supporting half-hourly ancillary data includes air temperature (T a ) and relative humidity (RH; HMP45C, Vaisala, Finland), photosynthetic photon flux density (PPFD; LI-190, Li-Cor, USA), net radiation (R n ; Q*7, REBS, USA), soil temperature (T s ; TCAV-L, Campbell Sci., UK) and SWC (CS616, Campbell Sci., UK). The SWC probes were not calibrated to the local soil characteristics, but were considered to provide a good measure of relative differences.

Data processing
Raw data files were processed with the EdiRe software package (Robert Clement, University of Edinburgh) producing half-hourly fluxes and averages. Fluxes were calculated based on standard flux community methodology (see Aubinet et al 2000), including despiking (Højstrup 1993), 2D coordinate rotation, time lag removal by covariance optimization, block averaging, frequency response correction using model spectra and transfer functions (Moore 1986) and WPL correction (Webb et al 1980). It has recently been suggested that measurements using an open-path gas analyzer need an additional term in the WPL correction, to account for the local heat flux created by the instrument itself during cold conditions (Burba et al 2008). In this study, we have applied method 4 in Burba et al (2008) to correct measurements obtained during cold periods, here defined as days with mean T a < 5°C. Effects of the self-heating correction are taken into account in the uncertainty assessment (see below).
Data post processing in Matlab R2012a (The Mathworks, USA) included quality control, storage term calculation and gap-filling. When applicable for a given test, the growing season was defined as the period from the first three consecutive days with daily mean T a > 5°C until the first three consecutive days with daily mean T a < 5°C. Half-hourly flux values were excluded when (a) wind components and scalar concentrations were beyond preset ranges, (b) RH > 98%, (c) the difference between measured H 2 O concentration and modelled H 2 O concentration (based on T a and RH) deviated by more than 1 mmol mol −1 from a two-week running median of the difference, (d) growing season fluxes of latent (LE) and sensible heath (H) were more than 3 standard deviations (SD) off from half-monthly quadratic fits with R n , (e) friction velocity u * < 0.1 m s −1 and (f) daytime growing season fluxes of CO 2 were more than 2 SD off from half-monthly light response curve fits (equation (1)) and nighttime and cold season fluxes were more than 2 μmol m −2 s −1 off from a two-week running median (table 1).
Gap-filling was performed using a look-up table methodology based on Reichstein et al (2005) with slight modifications: a missing value was replaced with the mean of at least four values obtained during similar meteorological conditions (PPFD ±20 μmol m −2 s −1 , T a ±2.5°C, VPD ±0.5 kPa) within periods of ±5 days or ±10 days. Long gaps (>7 days) during non-growing season (15 December 2008-16 February 2009, 12 March-15 April 2010 were filled with the median flux of the week before and after the gap. Long gaps during the growing season (4 June-2 July 2009, 26 June-3 July 2012) were filled with a light response curve approach (Misterlich function (Falge et al 2001)), parameterized on data one week before and after the gap with PPFD as independent variable: where F csat is CO 2 uptake rate at light saturation, R d is dark respiration, and α is the initial slope of the light response curve. The light response curve (equation (1)) was parameterized for daytime periods using an eight day moving window (time step one day). The parameterization was only considered successful when based on more than 50 observations (half-hours) and when all parameters (F csat , R d , α) were significantly different from zero (p < 0.05).  44  38  36  30  29  24  23  2009  77  63  59  47  45  37  35  2010  85  69  63  53  51  43  40  2011  94  80  73  62  59  51  48  2012  88  74  Growing season GPP was modelled by using equation (1) and subtracting R d (Lindroth et al 2007).
Daytime R eco was calculated as the difference between gap-filled NEE and modelled GPP, while nighttime R eco corresponded to gap-filled NEE. The estimated uncertainty in annual NEE sums was based on Elbers et al (2011). Random error (E rand ) and frequency response correction uncertainty (E freq ) were assessed according to Aurela et al (2002) and u * threshold selection uncertainty (E ustar ) according to Elbers et al (2011). Gap-filling uncertainty (E gap ) was assessed by varying the length of the period during which similar meteorological conditions was sought (3-6 days and 7-14 days). E gap was calculated from the SD of the three NEE sums (gap-filling periods 3/6, 5/ 10 and 7/14 days, respectively). In addition, we assessed the self-heating correction (Burba et al 2008) uncertainty by using a deviation of ±5°C around the default definition for cold periods (thus days with mean T a < 0, 5 and 10°C, respectively), for which the self-heating correction was applied. The uncertainty, E burba , was determined as the SD of the NEE sums in these three periods.
The flux footprint of the EC system was estimated using the parameterization by Kljun et al (2004), to assess whether other landscape elements surrounding the peatland would have any influence on the measured flux. The streamwise dimension of the footprint x R , was calculated as where z m is measurement height (2 m), σ w is the SD of vertical wind speed and u * is friction velocity. Parameters c and d were calculated from equations (15) and (16) in Kljun et al (2004), where roughness length (z 0 ) is used as parameter. Roughness length was calculated as where d h is displacement height (2/3 of the mean height of obstacles, 0.1 m) and U is horizontal wind speed. Mean annual T a during the study period (2008-2012) was above long-term average (1961-1990: 3.6°C) for all years except for 2010, which had a mean annual T a of 3.5°C ( figure 2(a)). The extra warming during the study period was not equally distributed throughout the year. The winter months (December-February) were on average 2.2°C warmer compared with the long-term average (−1.8°C), while there was no significant difference (p > 0.05) for the summer months. Growing season onset occurred on average (±1 SD) at DOY 129 ± 6 (table 2), whereas the growing season ended on average at DOY 286 ± 12. Precipitation sum was below average (1060 mm) in all years, except for 2010 (1075 mm). The seasonal patterns in precipitation during the study period did not show as pronounced differences compared with long-term mean as was the case for T a . However, the spring period (March-May) was on average wetter than the long-term mean, whereas other seasons were generally drier ( figure 2(b)).

Environmental characteristics
Daily values of T s and SWC at the Andøya peatland

CO 2 fluxes
The mean 90% footprint length during the entire measurement period was 89.5 ± 12.1 m (figure 4). Since the blanket bog extends >200 m in all directions surrounding the EC system (figure 1), the whole data set is considered reliable in terms of flux footprint. It should be noted that for the non-growing season south-westerly winds dominate, while north-easterly winds dominate during the growing season.
The temporal variation in the period 2008-2012 of mean daily NEE, GPP and R eco at the Andøya peatland is shown in figure 5. Mean July daily NEE means across all years was −1.40 ± 0.19 μmol m −2 s −1 , with highest uptake in 2009 (−1.65 μmol m −2 s −1 ) and lowest in 2010 (−1.16 μmol m −2 s −1 ). Mean July GPP and R eco were −2.55 and 1.17 μmol m −2 s −1 , respectively. Mean July fluxes in the Andøya bog are compared with other wetland sites in table 3. Average wintertime flux (December-February) during the whole study period was 0.32 ± 0.10 μmol m −2 s −1 .
The mixed peatland Stordalen in northern Sweden  and the fen Kaamanen in northern Finland (Aurela et al 2004) are situated at similar latitudes as the Andøya peatland. These three sites are located along a gradient ranging from maritime (Andøya) towards more continental (Kaamanen) climates. Therefore, we have paid particular attention to a comparison among these sites, as well as the extensively studied temperate Atlantic blanket bog Glencar (see McVeigh et al 2014) located in Ireland, due to its presumed functional similarity to Andøya. Parameters of the light response curve (equation (1)) derived from each site are shown in figure 6. In general, all parameters are lower for the two blanket bogs, except for the latter part of the season. Also, the increase in early growing season and the decrease in late growing season for F csat and R d for both blanket bogs occur at a lower rate compared with Stordalen and Kaamanen.
The mean annual CO 2 budget of the Andøya blanket bog across all complete measurement years (2009)(2010)(2011)(2012) amounted to −19.5 ± 18.3 g C m −2 (table 4). However, these estimates should be interpreted with caution as the total uncertainty was estimated to be on average 75.1 ± 4.9 g C m −2 . Of the separate components in the uncertainty analysis, the uncertainty relating to the choice of temperature threshold for applying the self-heating correction (Burba et al 2008) was overriding all other components (table 4). For the period May-September, the CO 2 budget was −111.8 ± 10.3 g C m −2 , with an associated uncertainty of 51.9 ± 5.5 g C m −2 (table 4).
Reduced SWC in the top-soil during summertime, as in 2008 and 2009 (figure 3), did not have an apparent effect on NEE, GPP and R eco . Instead, the wettest year, 2010, had the lowest summertime values of net CO 2 uptake (thus lowest NEE), GPP and R eco ( figure 5). This year was characterized by low T s (figure 3) and low PPFD during June-July (table 5), which may have slowed down vegetation growth. Light response curves based on data from July each year indicate that 2008 and 2010 had the lowest CO 2 uptake rates at PPFD > 1000 μmol m −2 s −1 (figure 7).
To further investigate the role of dry conditions, measured July fluxes of NEE at light saturation (PPFD > 1000 μmol m −2 s −1 ) were arranged into VPD bins (table 6). Significant differences (p < 0.05) in NEE across bins were observed in all years except for 2012. In 2012, there were not enough observations in the 0.4-0.5 kPa VPD bin to calculate statistics, thus indicating a less dry summer from a meteorological perspective. In general, the net CO 2 uptake was lower (i.e. less negative NEE) at high VPD than at low VPD. This was especially true for 2009 and 2011; years that were characterized by below average precipitation through Table 2. Growing season periods start (GS start ), ending (GS end ) and length (GS length ) 2008-2012. GS start was defined as first of three consecutive days with daily average T a > 5°C; GS end was defined as first of three consecutive days with daily average T a < 5°C. June and July and low SWC. Since GPP and, subsequently, R eco were modelled using an eight day moving window (equation (1)), the instantaneous effect of high VPD on those flux components cannot be assessed. Instead, we used a separate approach to model R eco and GPP, hereafter denoted R eco,2 and GPP 2 : daily means of measured nighttime (PPFD < 20 μmol m −2 s −1 ) NEE were plotted against T s , and an exponential model was fitted to the data ( figure 8). The obtained model for each year was then fed with the mean T s from each VPD bin, providing estimates of R eco,2 (GPP 2 = NEE°−°R eco,2 ; table 6).
These estimates indicate that the difference in NEE across VPD bins can primarily be explained by variations in GPP 2 , whereas variations in R eco,2 have a lower influence.

Discussion
The estimated annual CO 2 budget 2009-2012 of the Andøya blanket bog (−19.5 ± 18.3 g C m −2 ) is higher (i.e. weaker CO 2 sink) than a 3-year mean from the Stordalen subarctic mixed peatland (−90.0 ± 5.6 g C m −2 ; Christensen et al (2012)), a 12year mean from the Degerö boreal fen (−58.0 ± 21.0 g C m −2 ; Peichl et al (2014)) and a 9-year mean from the Glencar Atlantic (i.e. maritime) blanket bog (−55.7 ± 18.9 g C m −2 ; McVeigh et al (2014)); but similar to a 6-year mean from the Kaamanen subarctic fen (−21.5 ± 19.8 g C m −2 ; Aurela et al (2004)). However, as noted previously, annual budget estimates derived from EC measurements with an open-path sensor should be interpreted with  caution, due to uncertainties regarding the application of the self-heating correction (Burba et al 2008). This correction especially applies to measurements during wintertime in cold areas, and, therefore, several previous studies on northern peatlands using a similar sensor have not applied the self-heating correction to growing season data (see Kwon et al 2006, Humphreys and Lafleur 2011, Parmentier et al 2011, McVeigh et al 2014. As such, our growing season fluxes are directly comparable with those studies. The seasonal (May-September) CO 2 sink at Andøya (−111.8 ± 10.3 g C m −2 ) was slightly stronger compared with Glencar, where the corresponding budgets varied from −75 to −100 g C m −2 (McVeigh et al 2014), likely due to higher mid-summer radiation and higher plant cover.
In terms of mean fluxes during July, the Andøya blanket bog generally had low fluxes of GPP and R eco compared with other northern wetland ecosystems (table 3). We attribute this to the low density of vascular plants at Andøya and the relatively high cover of mosses and lichens; these plant functional types are less productive than vascular plants. However, flux   Table 3. Mean (±SD in case of more than one year of data) July fluxes of NEE, GPP and R eco (μmol m −2 s −1 ) in various northern wetland ecosystems. The temporal evolution of the light response curve (equation (1)) parameters (F csat , R d , α) for the peatlands Andøya, Stordalen, Kaamanen and Glencar (figure 6), illustrates the influence of site specific settings on the CO 2 exchange dynamics. The generally lower parameter values for Andøya can be attributed to the more open and less mesotrophic vegetation type compared with Stordalen and Kaamanen. Also, the slower start and ending of peak activity period, illustrated by the rate of change especially for F csat and R d , can be related to more maritime conditions on Andøya attenuating seasonal temperature variations. Despite being located in the temperate zone, the onset of peak activity period does not occur earlier in Glencar, which may be explained by lower cryptogam (lichen and bryophyte) cover in Glencar (25%; Sottocornola et al 2009) compared with Andøya (76%). Cryptogams may start photosynthesizing as soon as there is sufficient light and mild subfreezing temperatures (Larsen et al 2007). Relatively high values of F csat in the late season (DOY 240 an onwards) for Andøya and Glencar can be attributed to on-going photosynthetic activity by evergreen shrubs, mosses and lichens during non-freezing conditions.
Within most years we found that the rate of CO 2 uptake decreased (i.e. NEE less negative) at high VPD (table 6)  Sphagnum mosses and lichens dry out and may even be damaged by long-term desiccation (Schipperges and Rydin 1998). This finding, that lowered GPP explains a majority of the reduction in CO 2 uptake during dry conditions, is in line with previous studies (Shurpali et al 1995, Arneth et  It is interesting to note that the warm and dry years from a meteorological perspective (2008,2009,2011), compared with long-term mean, had lower NEE and higher GPP (figure 5) and were stronger annual CO 2 sinks (table 4), compared with 2010 when both T a and precipitation were close to normal. This effect was observed despite higher VPD in July that lead to weaker NEE on a half-hourly basis (table 5), which suggests that drier conditions did not have persistent effects on the CO 2 exchange dynamics on longer time scales (seasonal-annual). It can be argued that the maritime conditions at the site reduced the frequency, duration and intensity of dry conditions (i.e. high VPD), and that other environmental characteristics were more important for the inter-annual variation in CO 2 exchange. Low T s in early growing season of 2010  as well as low PPFD levels during June-July (table 5) likely resulted in low biomass build-up compared with other years, which can be illustrated by the low CO 2 uptake capacity at high PPFD levels in 2010 (figure 7). The reduced CO 2 uptake capacity in combination with low levels of incoming light during summer (table 5) likely explains lower fluxes in 2010. Based on mean summer-time fluxes and annual budgets, there was no apparent long-lasting effect of dry conditions on the CO 2 exchange, indicating an inherent resistance of the Andøya peatland to dry conditions. However, for extended drought periods, increased heterotrophic respiration may become increasingly important for the R eco signal (Ise et al 2008). As the summer months during our study period were not significantly warmer than the longterm average, although slightly drier, we may not yet have captured an extreme drought event in our measurement record. Ground surface wetness has been found to have a significant influence on NEE interannual variation in Glencar Kiely 2010, McVeigh et al 2014), with highest summer-time CO 2 uptake observed for years with intermediate (not too cold, not too dry) rather than extreme meteorological conditions (Sottocornola and Kiely 2010).
As stated previously by several authors (e.g. Limpens et al 2008, Lafleur 2009, Lund et al 2012, the effect of a changing climate on peatland C exchange is dependent on site specific characteristics, most importantly hydrological settings and vegetation composition. As such, it is not feasible to draw general conclusions valid for all types of peatlands. However, since a positive water balance is a prerequisite for blanket bogs, future higher temperature must be followed by an increase in precipitation to maintain the water balance for such peatland types. If not, it is likely that vegetation change will occur in blanket bogs (Gallego- Sala and Prentice 2012), with uncertain consequences for the C budget.

Conclusions
We have used 4.5 years of EC measurements from the Andøya blanket bog in Norway to describe the multiyear CO 2 exchange and assess the impacts of dry conditions. Our main conclusions include; • The bog acted as a small sink for atmospheric CO 2 (−19.5 ± 18.3 g C m −2 ); however, uncertainties regarding self-heating correction of the open path analyzer were large.
• On a half-hourly scale, we observed reduced CO 2 uptake (i.e. higher NEE) during periods with high VPD, mainly caused by a decrease in GPP.
• On longer time scales, seasonal to annual, no persistent effects of dry conditions on the CO 2 exchange were observed. Instead, other variables such as growing season onset and amount of incoming light were important regulators for the between-year variation.