Rapid loss of an ecosystem engineer: Sphagnum decline in an experimentally warmed bog

Abstract Sphagnum mosses are keystone components of peatland ecosystems. They facilitate the accumulation of carbon in peat deposits, but climate change is predicted to expose peatland ecosystem to sustained and unprecedented warming leading to a significant release of carbon to the atmosphere. Sphagnum responses to climate change, and their interaction with other components of the ecosystem, will determine the future trajectory of carbon fluxes in peatlands. We measured the growth and productivity of Sphagnum in an ombrotrophic bog in northern Minnesota, where ten 12.8‐m‐diameter plots were exposed to a range of whole‐ecosystem (air and soil) warming treatments (+0 to +9°C) in ambient or elevated (+500 ppm) CO2. The experiment is unique in its spatial and temporal scale, a focus on response surface analysis encompassing the range of elevated temperature predicted to occur this century, and consideration of an effect of co‐occurring CO2 altering the temperature response surface. In the second year of warming, dry matter increment of Sphagnum increased with modest warming to a maximum at 5°C above ambient and decreased with additional warming. Sphagnum cover declined from close to 100% of the ground area to <50% in the warmest enclosures. After three years of warming, annual Sphagnum productivity declined linearly with increasing temperature (13–29 g C/m2 per °C warming) due to widespread desiccation and loss of Sphagnum. Productivity was less in elevated CO2 enclosures, which we attribute to increased shading by shrubs. Sphagnum desiccation and growth responses were associated with the effects of warming on hydrology. The rapid decline of the Sphagnum community with sustained warming, which appears to be irreversible, can be expected to have many follow‐on consequences to the structure and function of this and similar ecosystems, with significant feedbacks to the global carbon cycle and climate change.


| INTRODUC TI ON
Boreal and subarctic peatlands contain large amounts of carbon (C), estimated to be as much as one-fifth to one-third of the world's soil C pool (Ciais et al., 2013;Gorham, 1991;Yu, 2012). The C accumulated in peat over centuries and millennia, because cold, acidic, and waterlogged conditions retard decomposition. Peatland C stocks are thought to be especially vulnerable to climate change because rising temperatures and associated hydrologic changes are expected to accelerate decomposition of surficial C stocks (He, He, & Hyvonen, 2016;Wilson et al., 2016), increase ecosystem respiration (Samson et al., 2018), and cause northern peatlands to become net sources of carbon to the atmosphere and exacerbating climatic warming (Gallego-Sala et al., 2018). Hence, peatlands may be one of the most important ecosystems providing feedbacks to global climate change (Bridgham, Pastor, Dewey, Weltzin, & Updegraff, 2008;Hilbert, Roulet, & Moore, 2000;Moore, Roulet, & Waddington, 1998). The source of much of the C accumulated in peatlands is the mosses of the genus Sphagnum (Clymo & Hayward, 1982; Figure 1). Hence, understanding and predicting the responses of Sphagnum to climatic change is essential for the assessment of the responses of the peatland ecosystem and its contribution to global C budgets (Moore et al., 1998).
Sphagnum moss, a keystone component of boreal peatlands, is considered to be an "ecosystem engineer" that creates its own favorable conditions while forming adverse conditions for vascular plants (van Breemen, 1995;Clymo & Hayward, 1982;Granath, Strengbom, & Rydin, 2012). Its unique characteristics, including adaptations for low nutrient availability, acidification of its surroundings, and a chemical composition that retards decomposition, drive local environmental conditions that influence the presence and performance of co-occurring plants. Sphagnum species are ecological specialists, sorting out in relation to pH, cation content of water, water level, and shade (Vitt & Slack, 1984).
Alteration of any of those conditions brought on by atmospheric and climatic change can be expected to alter growth, vitality, or composition of the Sphagnum community with feedbacks to the functioning of the bog ecosystem.
We are studying Sphagnum responses to experimental air and soil warming in the Spruce and Peatland Responses Under Changing Environments (SPRUCE) project  https ://mnspr uce.ornl.gov/). Located in an ombrotrophic bog in northern Minnesota, USA, at the southern edge of the boreal zone, this ecosystem is expected to be especially vulnerable to climate change. The intact ecosystem, comprising black spruce (Picea mariana (Mill.) B.S.P.) and tamarack (Larix laricina (Du Roi) K. Koch) trees, shrubs, and a nearly complete cover of Sphagnum mosses, is being exposed within large enclosures to a range of air and soil warming treatments in ambient or elevated atmospheric CO 2 . A comprehensive set of questions and hypotheses are being addressed on ecosystem productivity and C balance, hydrologic and nutrient cycling responses, microbial responses, and plant community ecology, all of which are closely integrated with model development. The range of attained warming of the whole ecosystem includes treatments spanning mild to very aggressive treatments (+1.6 to +10.0°C above ambient temperature), which is broader than that used in other warming experiments. The range was chosen to encompass the estimate of 7°C (with high uncertainty) for the threshold for boreal forest dieback (Lenton et al., 2008) and the projected July temperature increase for the site (~6.8°C; Peacock, 2012) under the IPCC high CO 2 emission scenario (RCP8.5) by the year 2,100.
We structured our measurements of Sphagnum responses to the imposed warming and CO 2 treatments to address the following objectives and hypotheses based on our understanding of the ecological attributes of the Sphagnum species at our site and observations from previous experiments.

| Objective: Establish response surfaces of Sphagnum productivity response to warming
Although growth rates of Sphagnum spp. are sometimes positively related to moderate temperature increases (Dorrepaal, Aerts, Cornelissen, Callaghan, & van Logtestijn, 2004;Gunnarsson, 2005), such positive effects are expected to be countered by associated drying of surficial peats (i.e., the acrotelm; Robroek, Limpens, F I G U R E 1 Sphagnum magellanicum, S. fallax, and S. angustifolium are dominant at the site of the SPRUCE experiment in northern Minnesota, USA. Photograph by Dave Weston Breeuwer, & Schouten, 2007) and may not persist as the degree of warming increases (Bragazza et al., 2016).
Hypothesis: Sphagnum productivity will increase with modest warming and decrease as warming increases.

| Objective: Determine whether elevated CO 2 alters Sphagnum response to warming
The primary responses of vascular plants to elevated CO 2 are an increased rate of photosynthesis and decreased stomatal aperture.
Sphagnum photosynthesis, however, is primarily controlled by tissue water content that regulates both hydration and CO 2 diffusion (Proctor, 2009;Schipperges & Rydin, 1998;Weston et al., 2015;Williams & Flanagan, 1996), and stomatal responses are precluded (Sphagnum has no stomata). While increased atmospheric CO 2 should increase dissolved CO 2 in the wet Sphagnum surface and thus CO 2 available for photosynthesis, free-air CO 2 enrichment experiments in four peat bogs across Europe reported no effect of elevated CO 2 on Sphagnum productivity (Hoosbeek et al., 2001).
Hypothesis: Elevated CO 2 will have no measurable effect on Sphagnum growth or its response to warming.

| Objective: Detect and quantify changes in Sphagnum community composition
Bryophyte species on hummocks (e.g., S. magellanicum Brid. and Polytrichum strictum Brid.), which are more distant from the water table, may be expected to be disproportionately affected by warming and the associated drying of the acrotelm (the surface layer of peat containing living plants) compared with species in the hollows (e.g., S. fallax (Klinggr.) Klinggr.). Alternatively, since hummock species are more adapted to drier conditions, the composition of the Sphagnum community may shift toward hummock species (Breeuwer, Heijmans, Robroek, & Berendse, 2008;Robroek, Limpens, Breeuwer, Crushell, & Schouten, 2007 Jensen fraction will increase. An alternative hypothesis is that S. magellanicum will be relatively favored because it is more adapted to drier conditions than the other species.
The research questions raised by these hypotheses have been tested in other experiments, but here we address them in situ at an ecosystem scale in an intact peatland system. We use a regression approach that permits us to develop response surfaces with temperature and explore whether elevated CO 2 alters the response of Sphagnum to warming. Our results provide a foundation for a complete assessment of carbon cycling responses in this peatland ecosystem when integrated with other ongoing investigations of tree, shrub, and belowground responses.

| Study site
The SPRUCE experiment is in an ombrotrophic peat bog in the Marcell Experimental Forest in northern Minnesota, USA (47.50283 degrees latitude, −93.48283 degrees longitude), at the southern edge of the boreal zone. Mean annual temperature  was 3.4°C, and average July temperature was 18.9°C, increasing 0.3°C per decade during summer months; average annual precipitation was 780 mm (Sebestyen et al., 2011).
The soil is a Typic Haplohemist, with average peat depths of 2-3 m (Parsekian et al., 2012). The bog has a developing hummock and hollow microtopography (Figure 2a), with shrubs located primarily on the hummocks. The perched water table, which has little regional groundwater influence, is typically 10-20 cm above the hollows after snowmelt, receding deeper later in the growing season (Iversen et al., 2018).
As described by Hanson et al. (2017) and Griffiths et al. (2017), Typical of bogs in northern Minnesota with an open tree cover (Vitt & Slack, 1984), there is a nearly continuous cover of mosses, primarily Sphagnum angustifolium, S. fallax, and S. magellanicum. [S. magellanicum has recently been separated into three species (Hassel et al., 2018), but the new taxonomic identity of the species in our bog has not yet been determined.] As in other similar bogs, S. fallax is found primarily in hollows, whereas S. angustifolium and S. magellanicum are found primarily in somewhat drier microhabitats, including lawns, low hummocks, and the flanks of high hummocks. Vitt and Slack (1984) determined S. angustifolium and S. magellanicum to have a high value of niche overlap with respect to water table height, but the niche breadth was much wider for S. magellanicum than for S. angustifolium or S. fallax. They note that species of the Sphagnum subsection Recurvum, including S. angustifolium and S. fallax, have long been a subject of controversy. S. angustifolium and S. fallax are very closely related phylogenetically, the microhabitat tolerances of the two species are superficially quite similar, and they can be difficult to distinguish in the field. In European peatlands, the two species group together in a cluster that has a high similarity in response to environmental variables (Robroek et al., 2017). At our site, S. angustifolium tends to dominate on hummocks and S. fallax in hollows, but they frequently intermix, and we have not attempted to separate them in our analyses; hence, we refer to them here as S. angustifolium/fallax.
An initial survey of the bog prior to establishment of experimental plots indicated the following composition of the bryophyte community: 68% S. angustifolium/fallax, 20% S. magellanicum, 1% S. fuscum, 8% Polytrichum strictum, 2% Pleurozium schreberi (Brid.) Mitt., and scattered other species.   Sebestyen & Griffiths, 2016), thereby mimicking the hydrologic condition that would obtain if the entire bog was warmed to the same degree as the plot. Air warming is achieved with propane-fired heat exchangers and a system of blowers and conduits . The air warming treatments were initiated in August 2015 and were maintained 365 days per year. Target values are +0, +2.25, +4.5, +6.75, and +9°C; the +0°C enclosures are generally 1-2°C warmer than outside ambient air. Soil (peat) warming was achieved with a belowground heating array of 3-m vertical low-wattage heating elements installed within plastic-coated iron pipe . The combination of air and soil warming compensates for heat losses to the surrounding bog and creates a more realistic temperature profile (Hanson et al., 2011). For our analyses, warming is characterized across the active season (Richardson et al., 2018) for Sphagnum, here defined as the period from 15 April to 15 October. One of the two enclosures at each temperature began to receive elevated CO 2 on 15 June 2016 to approach concentrations of 500 ppm greater than ambient, or about 900 ppm .

| Experimental treatments
Environmental monitoring (air and soil temperature, humidity, CO 2 concentration) is described in Hook (2016, Hanson et al., 2017). Beginning in June 2018, peat water content was measured by a new method using frequency domain capacitance probes (model 10HS,Meter Group,Inc.) preinstalled into open mesh cylinders filled with 0.1 g/cm 3 commercial peat, which were then placed laterally into the midhummock, 20 cm above the hollow position (n = 3). This standardized technique provided a stable relative measurement of volumetric water content (VWC) within the hummocks, whose internal structure is filled with voids and characterized by high spatial heterogeneity in bulk density. Water period in late spring when water levels had drained to hollow heights (day of year 147, 158, and 150 in 2016, 2017, and 2018, respectively).
Pretreatment surveys revealed no environmental gradients in peat depth (Tfaily et al., 2014), shrub cover, or foliar nitrogen or phosphorus concentrations across the potential enclosure locations that would require repeating treatments in separate blocks. Hence, treatments were assigned to enclosures randomly. The enclosure is the experimental unit, and in all cases where multiple samples were collected from within an enclosure, the data from individual samples were pooled. Pretreatment analyses  showed comparable environmental conditions within and across the constructed treatment plots. We are employing a regression-based design to focus on the pattern of response and potential response thresholds across all warming levels and whether elevated CO 2 alters the temperature response pattern. Regression is generally a more powerful approach than ANOVA, providing quantitative output that is more effectively incorporated into ecological models (Cottingham, Lennon, & Brown, 2005). The regression approach does not require replicating levels of warming, and indeed, if more plots had been available, they would have been assigned to intermediate levels of warming rather than replications of existing levels.
Although we cannot ask questions about any differences in response between individual enclosures, such questions are not relevant to our overall objectives.

| Community composition
To monitor changes in the species composition of the bryophyte community, we established three transects within each plot, embedded within 1 m × 2 m plant community assessment areas where no destructive sampling or instrumentation was permitted. There are 5-10 permanently located 5 cm × 5 cm sample points in each transect, 25 total in each experimental plot. Although the total area surveyed is small, the linear extent of the transects is approximately 28% of the diameter of the measurable plot area. Permanent stakes mark the endpoints of the transects; we attach a removable, calibrated plastic tube to the stakes to ensure that we can locate the same sample points each year (Figure 2d,e; additional photographs and maps of these transects can be seen in the data archive  We also noted the percentage of the area that was bare or covered with dead moss. Each spot was coded as whether the location was hummock or hollow after measuring the distance to a level reference marker and the vertical distance from the reference line to the bottom of a nearby hollow, with sample points at heights 10 cm or more above the deepest nearby hollow considered on a hummock. All of the observations over the four years were made by the same observer. Evaluations over larger areas or from overhead photographs were precluded because of the shrub cover; it was necessary for the observer's eyes to be quite close to the surface for accurate assessments. To test the accuracy of our fractional cover estimates, we surveyed a similar 1-m transect outside of an enclosure and then harvested the ten 5 cm × 5 cm sample points and measured the area of S. angustifolium/fallax, S. magellanicum, and Polytrichum in the laboratory. Correlation coefficients were 0.66-0.70, and the averages of the field estimates of the 10 samples points were 6.5% low for S. angustifolium/fallax, 4.6% high for S. magellanicum, and 1.9% high for Polytrichum.

| Sphagnum growth
In preliminary measurements (Griffiths et al., 2017), we realized that common methods for measuring Sphagnum growth were not effective at this site. Crank or brush wires (Clymo, 1970 2013), which measure increases in stem length along a vertical wire, could not capture growth of Sphagnum in hollows, which were only rarely vertically oriented, and they could not be used over winter to capture growth immediately after snowmelt in spring. Bundles of 5cm-long Sphagnum stems secured with string and reinserted into the bog (Grosvernier, Matthey, & Buttler, 1997) could be used over winter, but retrieval in the fall was often compromised, especially when there was substantial growth. Furthermore, both of these methods have multiple sources of error when attempting to scale production to grams per square meter, which requires data on increase in length, mass per unit length, and number of stems per unit area.
In light of these difficulties, we developed a more direct approach. We measured Sphagnum growth (or dry matter increment) in mesh columns, constructed from rigid polypropylene mesh tubes (Industrial Netting, RN1500) (Figure 2g; additional photographs in the data archive ). The tubes have an inside diameter of 38 mm, with a wall thickness of 1.5 mm and 46% open area. We cut them to 22 cm length and attached a disk of mesh material to the bottom using small plastic cable ties. We initiated the annual growth assessment period in October after growth had ended for the year while ensuring the columns were in place to capture spring growth in the next year. We harvested clumps of Sphagnum from each plot, removed debris, and cut the stems to 7 cm length ( Figure 2f). We set up measurements for three types of Sphagnum: S. angustifolium/fallax from hummocks, S. angustifolium/fallax from hollows, and S. magellanicum from hummocks. We prepared two sets of the three types of Sphagnum by inserting the Sphagnum into the columns at a stem density similar to that of the bog (Figure 2g) and Sphagnum loss, so these measurements represent the growth potential. The fraction of growth that occurred in spring was determined as the ratio of length increment from October to May (with the assumption that all this growth occurred in spring) to total annual length increment.

| Net primary production (NPP)
We calculated NPP at the scale of the whole plot as the sum of dry matter increment of the three Sphagnum type times its respec-

| Data analysis
We evaluated the effects of temperature on Sphagnum DMI, NPP, and community composition data by stepwise multiple linear re-

| Sphagnum community composition
In October 2015, prior to initiation of the warming and CO 2 treatments, 70.1 ± 2.9% of the ground area within the experimental enclosures was covered by S. angustifolium or S. fallax and 20.3 ± 3.6% by S. magellanicum; Polytrichum strictum, Pleurozium schreberi, and other mosses or herbaceous vascular plants each covered less than 5% ( Table 2)

| Growth
Averaged across all enclosures, Sphagnum growth (annual DMI) was greater in hollows than in hummocks (Figure 4). Our results (after multiplying by 10 to convert to g/m 2 ) are consistent with reported mean growth rates of these species: S. fallax (primarily in hollows), 400 g m −2 year −1 ; S. angustifolium (primarily on hummocks), 180 g m −2 year −1 ; and S. magellanicum, 250 g m −2 year −1 (Gunnarsson, 2005). Approximately 30% of annual growth occurred in early spring of 2017 and 12% in 2018, but there were wide variation and no pattern with respect to species or treatment (data presented in . There were no effects of temperature or CO 2 on DMI in 2016 (Figure 5a,b) (Figure 5c), which was reflected in the cover-weighted average DMI (Figure 5d). Maximum growth occurred at 19.5°C, or 5.0°C above growing season ambient temperature. However, increased growth with modest warming did not persist in 2018.
Although there were no clear trends in any of the three Sphagnum groups (Figure 5e), cover-weighted average DMI declined linearly with increasing temperature in 2018, and DMI was less in elevated CO 2 (Figure 5f).

| Plot-level production
There was no effect of temperature on NPP (DMI × fractional cover) in 2016 (Figure 6a), but warming had a large effect on Sphagnum NPP in the second and third years after the onset of treatments (Figure 6b,c) due primarily to the loss of cover. In 2018, we also observed a negative effect of CO 2 enrichment on NPP, and there was a significant CO 2 × temperature interaction, indicating that the slopes of NPP versus temperature were different in the two CO 2 levels. The slope of the linear regressions indicates a loss of 8 g C/m 2 of NPP per degree temperature increase in 2017. The loss in NPP was greater in 2018:29 and 13 g C/m 2 of NPP per degree temperature in ambient and elevated CO 2 , respectively. The effect of temperature on NPP was related to the water status of the bog. The maximum depth to the water table increased, and VWC of hummocks decreased with warming (Table 1).
NPP in 2018 declined in relation to both VWC and water table depth, with no effect of CO 2 (Figure 7a,b). NPP also declined with increasing soil temperature (Figure 7c).

| D ISCUSS I ON
An important attribute of the regression-based design we employed in this experiment has been our ability to describe the response surface of Sphagnum in an intact ecosystem to increased air and soil F I G U R E 6 Plot-level net primary productivity (NPP) relative to average air temperature (T) measured at 0.5 m above hollow surface from 15 April to 15 October, plotted as in Figure 5. F I G U R E 7 Plot-level net primary productivity (NPP) in 2018 relative to secondary factors that changed with warming. (a) NPP in relation to the average volumetric water content (VWC) of hummocks from 3 July to 7 October 2018. Note that the x-axis is displayed in increasing dryness (decreasing VWC) since VWC declined with increasing temperature (Table 1). NPP = 1,404 × VWC − 657.8, r 2 = .71, p = .002. (b) NPP in relation to maximum depth to the water table WT. NPP = −5.26 × WT + 237.1, r 2 = .52, p = .02. (c) NPP relative to average soil temperature at 20 cm depth from 15 April to 15 October. NPP = −26.66 × Tsoil + 449.5, r 2 = .77, p = .001. [CO 2 ] and the interaction with [CO 2 ] were not significant in any of these relationships temperatures that encompass the range of warming that is predicted to occur over this century. We anticipated that warming of this bog ecosystem would cause a loss of Sphagnum productivity and that the loss in productivity would be driven primarily by the effects of temperature on water status. That expectation was realized, and the effects of warming were large, driven by a wide-scale desiccation and loss of the Sphagnum community with increasing temperature and with effects increasing over time as a cumulative response.
Reduction in growth potential also contributed to the loss in pro-

ductivity. Recognizing that some studies have shown increases in
Sphagnum growth with warming (Dorrepaal et al., 2004;, our working hypothesis was that there would be a curvilinear response. This hypothesis was  (Dorrepaal et al., 2004). Biomass production of S. magellanicum and S. rubellum from a bog in Ireland was greater at 20°C than at the average site temperature of 15°C but was reduced when the water table was lowered (Robroek, Limpens, Breeuwer, Crushell, et al., 2007). Lichen and bryophyte cover decreased in response to 1-3°C warming at 11 tundra sites, but there was no significant effect on Sphagnum (Walker et al., 2006).
Increased soil temperature (1.7-4.5°C at 15 cm depth) did not affect bryophyte production in monoliths from a Minnesota bog, but production was driven strongly by water table depth (Weltzin, Harth, Bridgham, Pastor, & Vonderharr, 2001). Increased air temperature (+3.6°C) and associated increased evapotranspiration significantly reduced Sphagnum growth independent of water table in a poor fen in Sweden (Gunnarsson, Granberg, & Nilsson, 2004). S. fallax productivity decreased by 60% in mesocosms transplanted to a warmer (+5°C) location (Bragazza et al., 2016). Maximum photosynthesis of several Sphagnum species occurred at 30-35°C, but these observations were made on fully water-saturated Sphagnum stems in the laboratory (Haraguchi & Yamada, 2011); most temperate species have photosynthetic optima between 15 and 25°C (He et al., 2016).

Relevant insights into Sphagnum responses under natural condi-
tions emerged from the record-breaking 2003 heat wave across Europe, which was associated with widespread die-off of Sphagnum (Bragazza, 2008); hummocks in 20 bogs Italian Alps had irreversibly desiccated Sphagnum with no sign of recovery 4 years later.
Sphagnum forming lawns and carpets also desiccated but recovered in subsequent years. Similarly, in our experiment there has been no indication of recovery of the Sphagnum community. Although we have not relaxed the treatments, the exposure temperatures follow a seasonal course, and recovery was not observed during the cooler, wetter conditions in spring in either hummocks or hollows.
Ecosystem warming can have multiple direct and indirect effects on ecosystem processes, and it can be difficult to tease the different effects apart. The decline in Sphagnum growth and NPP in this experiment might have been a direct result of effects on physiological processes in Sphagnum (Schipperges & Rydin, 1998;Van Gaalen, Flanagan, & Peddle, 2007;Walker et al., 2017) or indirect effects from changes in competition with shrubs for light or nutrients (He et al., 2016;Malmer, Svensson, & Wallen, 1994;Turetsky et al., 2012).
Increased mineralization of peat, as would likely occur with warming and lowering of the water table, has been shown to increase growth of vascular plants in peatland ecosystems, leading to increased shading and reduced growth of intolerant Sphagnum (Malmer et al., 1994). The most likely or most dominant mechanism of response, however, was probably through the effect of warming on depth to the water table and water content of the acrotelm (Grosvernier et al., 1997;Rydin, 1985;Weltzin et al., 2001), both of which responded to increasing temperature (Table 1). NPP in 2018 was negatively correlated with the greater depth to the water table and reduced VWC measured in the hummocks (Figure 7a,b). A low water table can reduce Sphagnum growth by reducing the capillary rise of water to the capitula (Goetz & Price, 2016). Desiccation of capitula due to increased evaporation associated with higher temperatures and vapor pressure deficits can reduce Sphagnum growth independent of the water table depth (Gunnarsson et al., 2004).
We saw no growth stimulation of Sphagnum in elevated CO 2 , consistent with previous observations and in support of our hypothesis for this experiment. Mini-FACE experiments in four predominantly ombrotrophic peat bogs in Finland, Sweden, the Netherlands, and Switzerland showed no effect of elevated CO 2 on Sphagnum biomass over 3 years (Hoosbeek et al., 2001). In a greenhouse experiment, Sphagnum growth was initially stimulated by elevated CO 2 , but the response did not persist in the second year of exposure, and in the field, Sphagnum responded more to spatial variation in hydrology than to atmospheric CO 2 concentrations (Toet et al., 2006). In contrast with vascular plants for which effects of elevated CO 2 on photosynthesis and stomatal conductance are well documented (Ainsworth & Long, 2005), Sphagnum photosynthesis is controlled more by water content (Schipperges & Rydin, 1998;Williams & Flanagan, 1996), and stomatal responses are precluded. Furthermore, methane can be a significant C source for submerged Sphagnum (Raghoebarsing et al., 2005); refixation of CO 2 derived from decomposition processes also is an important source of C for Sphagnum (Rydin & Clymo, 1989;Turetsky & Wieder, 1999). The reliance on substrate-derived CO 2 , especially in wet conditions when there is increased resistance to CO 2 diffusion from the atmosphere, can explain the lack of response of Sphagnum to atmospheric CO 2 enrichment (Smolders, Tomassen, Pijnappel, Lamers, & Roelofs, 2001). There were, however, significant negative effects of elevated CO 2 on DMI and NPP in 2018, which we assume was an indirect effect of CO 2 stimulation of shrub or tree growth, increasing shading of Sphagnum or altering water balance. S. magellanicum in particular has a narrow niche breadth with respect to shade (Vitt & Slack, 1984 on Sphagnum growth also were reported in a greenhouse study with peat monoliths (Heijmans, Klees, de Visser, & Berendse, 2002). They attributed the response to an interaction between CO 2 and the higher greenhouse temperature compared with outdoors, and possibly related to an accumulation of allelopathic substances, but not associated with changes in vascular plant cover.
The negative effect of elevated CO 2 on Sphagnum we observed in this experiment is surprising and emphasizes the importance of a whole-ecosystem analysis. Ongoing studies will focus on revealing the interactions between Sphagnum and other ecosystem components, particularly the interactions with shrubs and with nitrogen availability.
Our third hypothesis posited that Sphagnum species on the hummocks, which are further from the water  (Jassey et al., 2013). Subsequent measurements at this site showed variable responses of the two species to warming in wet and dry habitats (Buttler et al., 2015). Peat cores of lawns dominated by Sphagnum fallax that were transplanted to a warmer (+5°C) and drier location exhibited a 50% decline in S. fallax occurrence and the appearance of S. magellanicum after 3 years (Bragazza et al., 2016). These contrasting results suggest that the hummock-hollow microtopography has a larger influence on Sphagnum responses to warming than species-specific traits.
Decline in Sphagnum growth (due to fertilization) at Mer Bleue bog in Ontario, Canada, favored Polytrichum, which is better adapted to dry conditions than Sphagnum species (Bubier, Moore, & Bledzki, 2007), but we saw no evidence of an increase in the presence of Polytrichum in our experiment. Warming experiments represent responses to "changed climate" rather than to "climate change" (Frolking et al., 2010) and do not necessarily measure changes in vegetation resulting from altered competition or succession (Turetsky et al., 2012). It may well be that a more gradual warming that allowed more time for replacement of one species by another would elicit a different result than what we observed after 3 years. However, losses in productivity in this experiment occurred across the entire range of warming treatments, losses were especially severe for S. magellanicum, a species characteristic of low hummocks that is more tolerant of desiccation than other Sphagnum species (Hájek & Beckett, 2008), and we saw no evidence of adjustments that would lead to a reversal of the loss.  (Jassey et al., 2013), with potential effects on ecosystem C balance. In our study, warming caused a loss of 13-29 g C/m 2 of NPP per °C temperature increase in elevated and ambient CO 2 , respectively. Some of this loss is likely to be compensated for by an increase in vascular plant NPP with warming and CO 2 enrichment, especially if warming reduces the negative effect of mosses on vascular plant success (Jassey et al., 2013). Shrubs had the dominant influence on C sink strength in an ombrotrophic blanket bog in northern England (Ward et al., 2013), although in a transplant study (Bragazza et al., 2016), decreased Sphagnum productivity in elevated temperature was not compensated by enhanced vascular plant productivity.
Although compensatory C fluxes in response must be considered, the loss of Sphagnum productivity is nevertheless a substantial fraction of ecosystem NPP. Pretreatment NPP of trees and shrubs at the site was 187 g C/m 2 (Griffiths et al., 2017), so Sphagnum NPP of 133 g C/m 2 in +0°C enclosure in 2016 comprised 41% of total ecosystem NPP. Ongoing research at the site is determining all components of the C budget and will identify the effect of Sphagnum loss on overall NPP. Regardless of compensatory responses in NPP, however, peat accumulation generally declines as the fraction of productivity from moss declines (Frolking et al., 2010), so loss of Sphagnum could accelerate the reversal from carbon sink to source. Peatlands have been accumulating C for millennia because annual productivity exceeds annual decomposition, but modeling studies suggest that climatic warming will reverse this balance and become a positive

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are freely available from the SPRUCE project data archive: https ://doi.org/10.25581/ spruce.049/1426474