Effect of the 2018 European drought on methane and carbon dioxide exchange of northern mire ecosystems

We analysed the effect of the 2018 European drought on greenhouse gas (GHG) exchange of five North European mire ecosystems. The low precipitation and high summer temperatures in Fennoscandia led to a lowered water table in the majority of these mires. This lowered both carbon dioxide (CO2) uptake and methane (CH4) emission during 2018, turning three out of the five mires from CO2 sinks to sources. The calculated radiative forcing showed that the drought-induced changes in GHG fluxes first resulted in a cooling effect lasting 15–50 years, due to the lowered CH4 emission, which was followed by warming due to the lower CO2 uptake. This article is part of the theme issue ‘Impacts of the 2018 severe drought and heatwave in Europe: from site to continental scale’.


Introduction
During the summer of 2018, Northwestern Europe experienced an exceptional drought and heatwave, also affecting Fennoscandian mire ecosystems [1][2][3]. The drought and associated warm temperatures can alter the short-term hydrological status of mire ecosystems, leading to alterations in biogeochemical processes within these ecosystems. These changes can have a drastic effect on greenhouse gas (GHG) exchange between the mires and the atmosphere [4].
Northern mire ecosystems are characterized by two considerable GHG fluxes, viz. carbon dioxide (CO 2 ) uptake and methane (CH 4 ) emission, that generate opposite radiative forcing (RF) [5]. On longer timescales, e.g. over millennia, carbon uptake and storage as peat, i.e. sequestration of CO 2 from the atmosphere, results in a climate cooling effect. Methane emission, on the other hand, has an intense short-term warming effect on the atmospheric radiative balance [6].
The seasonal variation in the CO 2 and CH 4 fluxes between the atmosphere and mires has generally been observed to be related to temperature and water table position [7][8][9][10][11]. Dry conditions and lowered water tables hinder CO 2 uptake [7,12], but they also lead to a reduction in CH 4 emission [4,13,14]. Thus, the same environmental forcing of GHG exchange of mires can lead to counteracting climatic effects.
To assess the climatic impact of weather events through ecosystem GHG exchange, the differing radiative properties and atmospheric lifetimes of GHGs need to be accounted for. Global warming potential (GWP) is a commonly used metric that integrates the radiative forcing due to a GHG pulse emission over a prescribed time (typically 20 or 100 years) and is expressed as CO 2 equivalents, i.e. the cumulative RF relative to that of CO 2 [15]. A more dynamic approach to compare the effects of different GHG fluxes is to examine the development of instantaneous RF due to these fluxes [16,17].
In addition to effective metrics, reliable data on ecosystem GHG exchange are needed to assess the climatic impact of weather events. In Sweden and Finland, GHG fluxes are measured at several mire ecosystems using eddy covariance (EC), mostly within national networks of the Integrated Carbon Observation System (ICOS-Sweden and ICOS-Finland). Appropriate environmental parameters are also measured at each site. In this paper, we will use EC and auxiliary data from five sites to analyse the effect of the 2018 European drought on the CO 2 and CH 4 fluxes of mire ecosystems in relation to changes in key environmental drivers. Furthermore, we will analyse the climatic effect of the drought-induced changes in GHG fluxes by using both the GWP and dynamic RF approaches.

Material and methods
We selected natural mire ecosystems that have EC measurements of both CO 2 and CH 4 fluxes during 2018 and at least one additional reference year of data. The sites are listed in table 1 and their locations are shown in figure 1. Many of these stations are either ICOS stations or in the process of becoming such and thus the measurements follow the ICOS protocols of CO 2 and CH 4 fluxes, and those of auxiliary parameters [24][25][26][27]. The average temperature at the sites ranges from −1.4°C to +6.8°C. None of these sites contain permafrost. The vegetation at the mires is listed in table 2, with associations to different mire types.
The effect of drought on GHG fluxes was estimated as differences in the cumulative annual CO 2 and CH 4 fluxes between 2018 and a reference period (ΔF CO2 and ΔF CH4 ). The reference period was selected as a single year or several years with rainfall and temperature close to the 30-year average (tables 3 and 4).
However, flux data availability places a strong constraint on this. For some sites, only a few years of data exist on both CO 2 and CH 4 fluxes, and the maximum length of time series for any of the sites was 15 years. As a result of flux data availability, these reference years vary among different mire sites and the environmental conditions during these years may slightly deviate from the long-term average climatological conditions. We related the changes in annual cumulative fluxes to average changes in temperature and water table in summertime, as the drought and heatwave were most conspicuous during this period. The significances of these relations were estimated by non-parametric Spearman's rank correlation test (Matlab Matlab R2015b, corr function). We also compared the apparent temperature dependence of methane emission during the drought and reference years using bin-averaged daily mean methane fluxes. For this, we used daily mean peat temperatures and 2°C bins starting at 0°C.
For the long-term climate reference, we used the 1981-2010 monthly precipitation and monthly average air temperature  Table 2. Dominating vascular plant vegetation on the five mire sites (1 = presence of the species). Mire type indicates the species main distribution range according to the Northern vegetation classification by Påhlsson [28]; nutrient poor ombrotrophic bog (B) and minerotrophic fens in order of increasing nutrient availability: poor fen (PF), intermediate fen (IF) and moderate fen (MF). G indicates that the species can be found in all four mire types, and if present in several types the preferred mire type is indicated by *.   due to atmospheric conditions not fulfilling micrometeorological flux quality criteria or due to instrument malfunctions were filled in the time series. The CO 2 fluxes were gap filled as in Wutzler et al. [29]. For CH 4 fluxes, daily averages were calculated [10] and gap filling was conducted by linear interpolation. The uncertainty caused by linear interpolation was assessed by creating artificial gaps in data, representing the number and distribution of gaps in original data, and interpolating the resulting time series. Repeating this 100 times indicated uncertainty generally below 10%. As the automatic water table measurement at the Kaamanen site was not operational in 2017-2018, we used averages of manual measurements to calculate the monthly water table depths at this site. The statistical significance of difference in the daily CO 2 and CH 4 fluxes between 2018 and the reference years was tested using the non-parametric Mann-Whitney-Wilcoxon test (Matlab R2015b, ranksum function, 5% significance level). The test was conducted for July-August, which was the peak flux period and had a large 2018-to-reference difference in water table at most sites.
To compare the climatic effects of the drought-induced changes in CO 2 and CH 4 fluxes, we used the GWPs of CH 4 with two different time horizons, 20 and 100 years, referred to as GWP 20 (=84) and GWP 100 (=28). Multiplying the change in the annual CH 4 mass flux by these GWP values results in fluxes in which CO 2 and CH 4 are expressed in common units, i.e. CO 2 equivalents [15]. To characterize the dynamics of the radiative effect of the GHG flux changes in more detail, we calculated the radiative forcing due to these changes, i.e. again adopting a 'normal' year as a reference. We used the impulseresponse model described by Lohila et al. [16] and subsequently updated to include the indirect RF due to atmospheric CH 4 -to-CO 2 oxidation [30], revised radiative efficiencies [31] and  Table 4. Annual carbon dioxide and methane fluxes, and the corresponding GWP-based CO 2 equivalents of the difference between 2018 and the reference year (ΔCO 2 -eq). Global warming potentials of CH 4 [15]: GWP 20 = 84, GWP 100 = 28. future GHG concentration scenarios [32]. The use of this method allowed us to estimate the decline of an atmospheric GHG perturbation and the related instantaneous RF over time. We also positioned the mires to the instantaneous RF switchover time diagram based on the ratio of changes in annual CH 4 and CO 2 fluxes [6].   All the sites showed a typical annual cycle of both daily CO 2 and CH 4 fluxes, with CO 2 uptake in summer and release outside the growing season, and CH 4 emission peaking during summer months (figures 5 and 6). The effects of drought and heatwave on CO 2 exchange is conspicuous, with reduced summertime CO 2 uptake at Degerö, Mycklemossen and Siikaneva, and increased uptake at Lompolojänkkä. Summertime CH 4 emission is reduced at all sites except Lompolojänkkä. The difference in daily CH 4 fluxes during July-August between 2018 and the reference period was significant for all sites. For CO 2 fluxes, the difference was significant for all sites except Lompolojänkkä.
The cumulative CO 2 fluxes at Degerö, Lompolojänkkä and Siikaneva showed annual CO 2 uptake in the reference years, whereas at Mycklemossen and Kaamanen the cumulative net CO 2 uptake was close to zero (figure 5). In 2018, annual CO 2 uptake was reduced at all sites except for Lompolojänkkä and three sites acted as CO 2 sources at an annual timescale. The annual cumulative ecosystem CH 4 emission was reduced during 2018 as compared to the reference years, except at Lompolojänkkä (figure 6). Not accounting for this site, the change in the annual CO 2 flux (ΔF CO2 ) ranged from 3 g C m −2 to 100 g C m −2 , while the change in annual CH 4 flux (ΔF CH4 )

Degerö Degerö
Kaamanen Kaamanen Lompolojänkkä Lompolojänkkä  The relations of ΔF CH4 and ΔF CO2 to the 2018-to-reference difference in summertime water table position (ΔWT) or the difference in air temperature were not significant (ΔT) ( figure 7). However, at all sites with a lowered water table, the CH 4 emissions at a given temperature bin were generally lower during the drought year than in the reference years ( figure 8).
For all mires with a substantial water table lowering in 2018, the drought-induced changes in the annual CO 2 and CH 4 balances, estimated above, correspond to a cooling effect when the fluxes are expressed as GWP 20 -based CO 2 equivalents (negative CO 2 equivalents, table 4). This indicates the short-term dominance of reduced CH 4 emissions. However, the corresponding GWP 100 -based values were positive, indicating warming, at all sites except for Kaamanen.
The instantaneous RF due to GHG flux changes, caused by dry conditions, show an initial cooling effect resulting from the reduced CH 4 emission at all sites except for Lompolojänkkä ( figure 9a). Later, the effect of reduced CO 2 uptake will dominate, causing a warming effect at these sites. Lompolojänkkä, with opposite changes in GHG fluxes as compared to other mires, shows an initial warming and a subsequent cooling effect. The switchover of the instantaneous RF from cooling to warming takes place 15-50 years after the drought year for the mires which experienced a water table drawdown (figure 9b), while at Lompolojänkkä the transition was from warming to cooling (figure 9c).

Discussion
The 2018 drought caused a widespread water table drawdown across North European mire ecosystems. This is a reflection of the dry and warm conditions during summer of 2018. In Sweden, precipitation deficit was observed in nearly the whole country from May to July, with Southern Sweden experiencing the highest deficit [1]. Furthermore, May and July were much warmer than the long-term average for the whole of Sweden, while June was somewhat warmer in the south and colder in the north [1]. In Finland, the early summer precipitation deficit was more pronounced in the southern part of the country [2], with warmer than average summer for the whole county [3]. However, local hydrological features related to e.g. topographical position can cause some mires to be less sensitive to climatic variations, as seen at the Lompolojänkkä mire.
We observed similarities in the change of annual GHG fluxes at all the mires with a water table drawdown, with a reduction of both CO 2 uptake and CH 4 emission. Three out of the four mires with lowered water table turned from CO 2 sinks to sources during 2018. The reduction of CH 4 emission was more moderate, the change being mostly less than 20% of the emission during the reference year, with the exception of Mycklemossen (42%). Mycklemossen is the southernmost  (table 2). As bogs typically have a lower water table as compared to minerogenic fens and have a lesser coverage of aerenchymatous plants, their CH 4 emission may be more sensitive to dry conditions. We could not establish statistically significant correlations between the changes of annual CO 2 exchange and annual CH 4 emission with summertime temperature or water table level. However, the apparent dependence of CH 4 emission on peat temperature shows a clear 2018-to-reference difference in all mires with a lowered water table. Similar differences in the apparent temperature dependence of CH 4 emission have also been observed previously [33,34]. At Lompolojänkkä, where the water table was not drawn down, the temperature response of CH 4 emission was similar in 2018 and the reference year. The high peat temperature at Lompolojänkkä in 2018 can explain the very high CH 4 emission in that year. On the other hand, at Degerö, which also had relatively high peat temperatures in 2018, the CH 4 emission was clearly lowered due to the lower water table. Thus, it seems obvious that both the water table level and peat temperature play a role in this variation. The use of such dependencies e.g. for upscaling the climatic effects of droughts would additionally require establishing a relationship between water table and precipitation, and peat temperature and air temperature, as water table position and peat temperatures are not parameters commonly measured by weather observation networks.
As the CH 4 emissions were reduced, this change first dominated the radiative forcing effects over the reduction in CO 2 uptake and resulted in a temporary cooling effect. According to our RF analysis, this cooling was in most cases limited to the first 15-50 years after the drought year. The length of this period depends on the ratio of the changes in the two GHG fluxes, while the strength of the cooling and warming effects depend on the magnitude of these changes. At Siikaneva and Degerö, with a small reduction in CH 4 emission as compared to a reduction in CO 2 uptake, the cooling period is short, whereas for Kaamanen, with a small change in CO 2 uptake, cooling lasts longer. Mires with a large change in CH 4 fluxes showed a large initial change in the instantaneous RF. Siikaneva, with the largest reduction in CO 2 uptake, showed the largest warming after the switchover from cooling to warming. The GWP 20 -and GWP 100 -based metrics, which essentially represent RF integrals, reflect the RF-based analysis.
The short-term climatic effect as shown by both the GWP and RF approaches is very sensitive to the changes in CH 4 fluxes. As the variation in annual CH 4 emissions from northern mires can be ca. 2 g C m −2 [10], the selection of reference years can have a large effect on the estimated short-term climatic forcing, which is affected more by CH 4 than CO 2 . Ideally, we should compare the CH 4 emissions during a drought year to a long-term average. Currently, however, very few CH 4 emission time series exceed 10 years. Thus, the development of long-term flux measurement networks, such as ICOS, is expected to lead to more representative datasets and improved understanding of these climate feedbacks.

Conclusion
The dry conditions in Northwestern Europe in 2018 led to a lowering of water table position at most, but not all, flux measurement sites on mire ecosystems. The lowered water table led to a reduction of both summertime CO 2 uptake and CH 4 emission, and annual exchange of these GHGs. The apparent temperature dependence of CH 4 fluxes was clearly affected by the lowered water table, but also temperature effects were obvious. Due to the different atmospheric residence times of these GHGs, the cooling effect due to the reduction of CH 4