Unraveling past impacts of climate change and land management on historic peatland development using proxy‐based reconstruction, monitoring data and process modeling

Abstract Peatlands represent globally significant soil carbon stores that have been accumulating for millennia under water‐logged conditions. However, deepening water‐table depths (WTD) from climate change or human‐induced drainage could stimulate decomposition resulting in peatlands turning from carbon sinks to carbon sources. Contemporary WTD ranges of testate amoebae (TA) are commonly used to predict past WTD in peatlands using quantitative transfer function models. Here we present, for the first time, a study comparing TA‐based WTD reconstructions to instrumentally monitored WTD and hydrological model predictions using the MILLENNIA peatland model to examine past peatland responses to climate change and land management. Although there was very good agreement between monitored and modeled WTD, TA‐reconstructed water table was consistently deeper. Predictions from a larger European TA transfer function data set were wetter, but the overall directional fit to observed WTD was better for a TA transfer function based on data from northern England. We applied a regression‐based offset correction to the reconstructed WTD for the validation period (1931–2010). We then predicted WTD using available climate records as MILLENNIA model input and compared the offset‐corrected TA reconstruction to MILLENNIA WTD predictions over an extended period (1750–1931) with available climate reconstructions. Although the comparison revealed striking similarities in predicted overall WTD patterns, particularly for a recent drier period (1965–1995), there were clear periods when TA‐based WTD predictions underestimated (i.e. drier during 1830–1930) and overestimated (i.e. wetter during 1760–1830) past WTD compared to MILLENNIA model predictions. Importantly, simulated grouse moor management scenarios may explain the drier TA WTD predictions, resulting in considerable model predicted carbon losses and reduced methane emissions, mainly due to drainage. This study demonstrates the value of a site‐specific and combined data‐model validation step toward using TA‐derived moisture conditions to understand past climate‐driven peatland development and carbon budgets alongside modeling likely management impacts.

They occur mainly in the Northern Hemisphere circumpolar region, where low temperatures, high soil moisture and slow decay rates of litter input via net primary production (NPP) allow peat to form (i.e. a long-term positive balance between NPP and litter decay), often under conditions of high water-table depth (WTD). Crucially, this slow decay preserves an archive of peatland development (i.e. animal and plant remains) that can be dated and used to reconstruct past drivers of peatland growth, such as WTD and vegetation composition, providing key information on how peatlands respond to changes in climate.
Blanket bogs are a globally rare peatland habitat with the United Kingdom containing about 15% of this habitat (Tallis, 1998) but mostly in a degraded state (Natural England, 2008), largely due to past environment (e.g. N deposition) and management (e.g. drainage) impacts. In the UK blanket bogs represent about 90% of all peatlands (Bain et al., 2011), which are often managed for grazing and grouse shooting, commonly supported by draining the peat and regular burning of vegetation. The consequence is dominance of heather (Calluna vulgaris), very low overall plant biodiversity, suppressed cover of peat-forming Sphagnum mosses (Lindsay, 2010), and often drier and eroding peat. In fact, only about 12% of protected blanket bog sites are classified as in favorable condition (Natural England, 2008). However, little is known about management impacts on the long-term SOC accumulation and soil C emissions.
UK blanket bogs have accumulated peat since the start of the Holocene about 11.6 k years ago under varying climate, but current bioclimatic models highlight the threat by climate change to their natural range , suggesting that they might start to degrade as the climate warms (Gallego-Sala & Prentice, 2013) resulting in water-table drawdown and increased decomposition. However, existing bioclimatic models do not consider extremely important autogenic negative feedbacks within peatlands that may actually act as a "buffer" to climate change (Swindles, Morris, Baird, Blaauw, & Plunkett, 2012). Such feedbacks include ecohydrological linkages between changes in WTD and shifts in vegetation communities (with different rooting depth and litter quality and thus affecting SOC inputs across depth and peat decomposition rates). A better process-level understanding of climate-peatland SOC feedbacks is clearly needed (Davidson & Janssens, 2006) since the mineralization of peatland soil organic matter (SOM) has the potential to release vast amounts of previously locked-up C into the atmosphere.
Depending on the water-table level, C emissions from decomposition are either as CO 2 (oxic acrotelm) or CH 4 (anoxic catotelm), the latter particularly responsible for exacerbating climate change and affecting the overall greenhouse gas (GHG) emissions. However, a key limiting issue in more accurate predictions of future climate is still the inadequate model representation of climateterrestrial carbon (C) cycle feedbacks, particularly of peatland soil organic carbon (Limpens et al., 2008). Several peatland models of varying complexity and feedback mechanisms have been developed (Baird, Morris, & Belyea, 2012;Bauer, Gignac, & Vitt, 2003;Clymo, 1984;Frolking et al., 2010;Gignac, Vitt, & Bayley, 1991;Heinemeyer et al., 2010), which have often been compared to measured C stocks. However, there is a lack of validating the C cycle underpinning hydrological model predictions against past palaeo records.
Peat archives from peat cores are also important for testing peatland development models, which predict peat layer accumulation and their associated chemical (e.g. carbon content) and physical (e.g. bulk density) properties, enabling comparison between the two.
Data-model comparisons have revealed uncertainties in peat accumulation processes , particularly considering hydrology, vegetation and NPP. Furthermore, recent peat core studies  indicate increased C accumulation during warmer periods due to increased NPP outweighing higher decomposition, which contradicts most global earth system model carbon cycle simulations (Friedlingstein et al., 2006 Testate amoebae (TA) are commonly used to reconstruct peatland hydrological changes over the Holocene, as indicator species are aligned across a gradient of wet to dry conditions based on present-day associations (Amesbury et al., 2016;Turner, Swindles, & Roucoux, 2014). Peat cores provide a stratigraphic (i.e. temporal) archive of past TA species composition, thus allowing to predict past moisture (and likely water-table) conditions from understanding of contemporary ecology. Quantitative transfer functions are used to establish a relationship between present-day species composition and hydrological data and then applied to subfossil data from cores (Amesbury et al., 2016).
Model predictions of peat carbon stock and flux changes rely on capturing seasonal and interannual WTD changes. Although in the short-term site measurements can be used for model validations, validation over longer-time scales relies on comparing model predictions to WTD reconstructions, for example, based on TA taxa found within the peat core. In addition to other dating tools such as radiocarbon (Turner et al., 2014), the use of Spheroidal Carbonaceous Particles (SCPs; Swindles, 2010) allows the generation of temporally constrained records of palaeo-hydrological conditions in the recent past. Together with dating tools, palaeo-reconstructions thus offer crucial insights into peatland development, (i) understanding baseline trajectories in peatland development, and (ii) assessing long-term resilience and recovery rates of peatlands to climate or management impacts . TA based reconstructions of past WTDs can then be compared to process model predictions, providing an important hydrological model validation tool. A good fit between TA and model predicted WTDs supports applying model scenarios to explore past management impacts on peatland functioning and C storage. The MILLENNIA peatland model predicts peat hydrological conditions (i.e. water-table depth) based on climatic conditions and long-term peat column growth either for annual or monthly time steps. Here, we used the annual MILLENNIA version (Heinemeyer et al., 2010) for long-term peat accumulation during the Holocene until 1914, and then either the annual or monthly version (Carroll et al., 2015)  WTDs from different management model scenarios were then compared to the TA-based WTD, and related to C accumulation and C emissions affecting the GHG balance.

| Site location and environmental conditions
The study site, Moor House National Nature Reserve (NNR), covers about 75 km 2 of a typical UK blanket bog with much known about its ecology (Garnett, Ineson, Stevenson, & Howard, 2001;Heal & Perkins, 1978) together with detailed meteorological (i.e. weather station) and hydrological (i.e. water-table) data collected by the Environmental Change Network (ECN). It is located in the northern Pennines across an elevation range of 290-850 m (a.s.l.) and is characterized by a subarctic-oceanic climate with an average longterm mean annual temperature (MAT) and precipitation (MAP) at 550 m altitude of about 5. 1°C (1931-1997) and around 2,000 mm (last 20 years), respectively (cf. Garnett, 1998). Peat formation at the study site started about 9,000 years ago (see Heinemeyer et al., 2010 for more site information). This study focuses on a square kilometre around the ECN meteorological station at 550 m (NY757328;

| Peatland model predictions
The MILLENNIA peatland model (Heinemeyer et al., 2010) predicts peat hydrological conditions (i.e. water-table) based on climatic conditions either for annual or monthly time steps. The hydrological conditions, in connection with internal factors (e.g. litter quality) and external modifiers (e.g. temperature and oxygen availability), then determine decomposition rates of old and new carbon fractions (as in Bauer, 2004) across the peat profile based on litter inputs via NPP as a function of potential evapotranspiration (PET) based on Lieth and Box (1972). Detailed model explanations are provided in Heinemeyer et al. (2010;annual model) and Carroll et al. (2015; monthly model). Here, we used the annual version for the long-term peat accumulation during the Holocene and for the period until 1914, and either the annual or monthly version until 2012 (reflecting model application and climate data availability). However, model outputs are only provided as annual averages in relation to average TA predictions. The main model driver is climate with either monthly or annual temperature and rainfall amounts as inputs. Both versions consider topography (e.g. slope affecting temperature, runoff and erosion), vegetation (e.g. affecting NPP, litter quality and transpiration losses) and proceeding hydrological conditions (e.g. high watertables leading to higher run-off) to predict a new water-table, vegetation (based on the preceding 5-year average water-table) and corresponding changes in soil carbon fluxes (i.e. CO 2 and CH 4 from decomposition), peat depth increments (i.e. accumulation) and soil C budgets (change in soil C stock in relation to input from NPP and losses from soil C fluxes from decomposition and erosion).
Further changes were implemented to improve hydrological and C flux process representation by calculating water filled pore space in the peat, bedrock drainage, plant-mediated transport and methane oxidation (oxidation): • The available pore space in relation to the height above the watertable depth (WTD) was based on data by Hayward and Clymo (1982; see Figure 4, but ignoring the short term hysteresis effect); an exponential relationship is assumed between the distance to the water-table and the available pore space (0.2*EXP (1.6*WTD 2 )), such that available space increases with distance from the water-table. Total space is then calculated by integrating the available pore space over the available unsaturated peat cohorts. By combining the water entering the system with the available space, a new WTD is calculated.
• To simulate drainage of the peat column into the bedrock, two drainage factors are included, specific yield (SY) and hydraulic conductivity (HC). These are set to default values of 0.02 (SY in %) and 0.1 (HC in cm/year) reflecting average values for clay reported by Johnson (1967) for SY (2%) and for unweathered clay based on Bear (1972) for HC (10-5 feet/day). However, SY and HC can be altered as a user input.
• The plant functional type (vegetation) composition of shrub, sedge, rush, grass, herb, Sphagnum, other moss) is based on a moving average of 5 years of previous water-tables, allowing representing a more stable/resilient vegetation in the case of only a few very dry or wet years.
• The anoxic ratio is set to 0.035, similar to values reported in previous literature (Bauer, 2004) ranging from 0.025 to 0.0625.
• Methane oxidation is set to 0.05 g C g À1 year À1 and reflects the range of the very scarce data available on methane oxidation in relation to dry peat and/or on a carbon (mass) basis, i.e. McDonald, Hall, Pickup, and Murrell (1996) provided incubation values at HEINEMEYER AND SWINDLES | 4133 20°C for UK peat of around 0.08 g C g À1 year À1 , Watson, Stephen, Nedwell, and Arah (1997) quoting 0.012 g C g À1 year À1 (i.e. 0.001021 mol C g À1 year À1 equal to 1.021e À3 mol C g À1 year À1 ), but Yrj€ al€ a et al. (2011) quoted only 0.0009 g C g À1 year À1 (0.2 lmol g À1 DW day À1 ) and Whalen and Reeburgh (2000) measured around 0.002 g C g À1 year À1 .
We used available reconstructed Holocene climate data (based on a combination of recent instrumental data and a variety of existing multiproxy climate reconstructions, see Morris, Baird, Young, & Swindles, 2015) to model long-term peat accumulation, Met Office 5 km gridded data (Perry & Hollis, 2015) for the recent past  and ECN data (ECN Data Centre: http://data.ecn.ac.uk) for recent present periods . Met Office data were adjusted for elevation for the Moor House site in order to achieve the same long-term average temperature and rainfall amounts as the Moor House ECN data (see Carroll et al., 2015). Model predictions of monthly WTD could be compared to averaged ECN hourly automated dipwell data (UK grid location: NY 75072 33425; missing data were gap-filled by interpolation of manual data) for the period of 1999-2012. Water-table depth standard deviation for a Moor House model evaluation (see supplementary data in Carroll et al., 2015) was predicted to within 0.32 cm (linear R 2 = 0.57).
Moor House was a formal shooting estate during  and grouse moor management scenarios reflected available site information (Rob Rose, CEH; personal communication), which indicated a 20 year burn rotation. The associated drainage was assumed to last from 1831, before intensification of grouse shooting (to enable enhanced heather growth and drier access conditions for gamekeepers), until the late 1950s. Burning was anticipated to have started in 1851 and to reduce NPP to 1% in the burnt year (and charcoal adding about 5% to an inert carbon pool), subsequently recovering in a sigmoidal shape to 100% by either 5 or 10 years after burning (based on field observations by A. Heinemeyer et al., unpublished data). Drainage was expected to reduce WTD on average by 5 cm, based on the field evidence of Wilson et al. (2010).
Reduced WTD were linked to enhanced decomposition and increased the associated CO 2 but decreased CH 4 emissions similarly to modeled impacts of natural WTD changes (Heinemeyer et al., 2010). Drainage (grip) effectiveness was assumed to be at optimum for 25 years and declining to 60% over the subsequent 15 years (renewed once in 1871 and then maintained at optimum until 1905, reflecting intense grouse shooting), finally declining to 0% by 1955.
Grazing pressure was anticipated to be insignificant above 450 m (i.e. no reduction in NPP at the modeled altitude of 550 m a.s.l.).
Further model scenarios considered a no management option (no shoot).

| Water-table reconstructions
A Russian peat core was taken from the top 50 cm of peat beside the Moor House ECN station (Lat. 54.695500; Lon. À2.387400) following De Vleeschouwer, Chambers, and Swindles (2010). The core was returned to the laboratory and stored at 4°C before analysis.
The top 20 cm of the core was sampled at 0.5 cm resolution and TA were prepared using the standard method of Booth, Lamentowicz, and Charman (2010). We applied the Northern England and pan-European transfer functions to the subfossil TA data to reconstruct WTD (Amesbury et al., 2016;Turner, Swindles, Charman, & Blundell, 2013). In both cases weighted-averaging tolerance downweighted regression with inverse deshrinking was used as it yielded the best performance statistics. The water-table reconstructions were standardized following Swindles et al. (2015). There is some variation in predicted water-tables between the two transfer functions which is caused by intermodel differences in WTD optima of key taxa. Some

| Peat core age-depth profile
SCPs have been deemed to provide reliable age information from peatlands in Northern Britain and Ireland for the last~150 years (Swindles, 2010;Swindles, Blundell, Roe, & Hall, 2010;Turner et al., 2014). Three unambiguous features can be used as age-equivalent stratigraphic markers: (i) the start of the record (c. 1850), (ii) the rapid increase in SCPs (c. 1950) and (iii) the peak (c. 1978). These represent (i) the start of high temperature combustion and power generation; (ii) the post-WW2 increase in power generation and fossil fuel burning and (iii) the peak of SCP production before reliance on other methods of power generation and the advent of clean fuel technologies (Rose, Harlock, Appleby, & Battarbee, 1995). We used the method of Swindles (2010) to prepare SCPs from the peat. Linear interpolation was used to generate a simple age-depth model between the three SCP age-equivalent stratigraphic markers and the date of core sampling (2011) as the uppermost time point; dates before the start of the SCPs are an extrapolation based on accumulation rate.

| Water table reconstructions
The testate amoebae data from Moor House (Figure 1

| Comparison of predicted water-tables for management scenarios and site records
Use of the available long-term climate data for a nearby Northern England peatland site (Morris et al., 2015), adjusted to the longterm mean temperature and total rainfall for Moor House, together  À0.04 AE 0.63 cm mean annual peat depth increment (with a mean WTD of 4.8 AE 1.9 cm). The burn but no drain 5-year NPP regrowth scenario reduced this further (i.e. 80% lower soil C and peat increment losses with a mean annual soil C budget of À10.9 AE 92.6 g C/ m 2 and a mean annual peat depth increment of À0.01 AE 0.64 cm.

| Predicted impacts on soil C emissions, peat C budgets and peat accumulation rates
During 1831-1850, the period of drainage only (i.e. no burning), the managed scenario reduced the mean annual soil C budget by 27.8 g C/m 2 to À7.2 AE 15.8 g C/m 2 , which reflected an average reduction in mean annual WTD by 3.9 cm to 8.3 cm (Figure 8a).
These changes in soil C budget under drainage only corresponded to an annual peat depth increment reduction by 0.06 cm to À0.02 AE 0.63 cm compared to the unmanaged scenario (Figure 8b).
The grouse moor management (i.e. 1851-1950) not only impacted C dynamics via reduced water-tables from drainage (Figure 8a), it also altered C inputs and thus C dynamics via reduced NPP following burning. Overall, drained and 10-year NPP recovery scenarios reduced both mean annual C losses from soil CO 2 fluxes (308 AE 106 g C) and annual CH 4 emissions (4.9 AE 8.3 g C) compared to the unmanaged scenario for which mean annual values were 417 AE 60 g C for CO 2 and 13.2 AE 20.4 g C for CH 4 net emissions (i.e. including methane oxidation, ebullition and plant-mediated transfer (PMT) processes via sedge leaves and stems). However, whereas the no drain 10-year NPP burn scenario decreased CO 2 (292 AE 110 g C) and increased CH 4 (9.8 AE 18.0 g C) emissions (Figure 8a), the 5-year NPP scenario (data not shown) increased both CO 2 fluxes (348 AE 92 g C) and net CH 4 emissions (11.2 AE 18.8 g C) emissions, reflecting quicker vegetation regrowth and thus NPP and PMT recovery.

| DISCUSSION
This study provided novel insights into ecological applications of using TA-derived WTD reconstructions in a site-specific model validation context. The findings highlight the value of combining palaeoecological records with process level modeling to allow better understanding of the effects of climate and management on peat development and C cycling. This combination shows promising potential in allowing to understand the contributions of past environmental (e.g.   including active drainage ; drainage efficiency 25 years at maximum plus 15 years reduction to 60% and 0% by 1955) and burning (20 year cycles; with an exponential 10-year regrow period to full net primary productivity) during 1850 till 1950 climate) and human-induced (e.g. grouse management) changes in peatland development over time, C stocks and corresponding C fluxes.
Although the EU transfer function predictions (as used in Amesbury et al., 2016) showed wetter conditions overall (Figure 2), the model fit to the MILLENNIA predictions was not improved, indicating a better overall fit using the geographically closer NE transfer function, possibly applying across the wider UK context, which has yet to be tested. Previous work comparing TA to recent short-term monitored WTD by Amesbury et al. (2016)  and Archerella flavum (8 AE 7 cm mean WTD). This coincides with documented wet bog conditions across Ireland  and Northern Britain (Mauquoy, van Geel, Blaauw, & van der Plicht, 2002) during this period (i.e. the Little Ice Age) with potentially standing water for most of the year or water pools in hollows.
Such conditions of standing water are not specifically captured by either the model or TA WTD predictions, which are relatively insensitive to water-tables above the surface. Moreover, the climate data used in the model become less reliable before the 1850s, and this period might well have been wetter than the by Morris et al. (2015) reconstructed climate record suggests. Therefore, MILLENNIA predicted WTD during 1730-1830 might be under predicting the WTD in relation to uncertain climate input data.
Most interesting was the period between 1840s and 1940s, an unexpectedly drier period based on TA predicted hydrological conditions compared to the wetter conditions predicted by the MILLEN-NIA model based on climate only ( Figure 6). Moor House was a formal shooting estate throughout exactly this period (i.e. 1842-1951), based on grouse bags and predator control information (ECN data provided by Rob Rose from the Centre for Ecology and Hydrology (CEH) Lancaster, personal communication). Grazing intensity on true blanket bogs such as Moor House is low (0.5 sheep/ha according to Rawes & Heal, 1978), except perhaps for recently burnt areas; historically grazing was probably low overall because farmers were unable to grow sufficient feed to maintain their stock over the winter, particularly at altitudes above 500 m. Therefore, burning in connection with drainage seems to be the most likely factor explaining the water-table lowering. The available anecdotal evidence (R. Rose (CEH), personal communication) from some of the game keepers in the latter part of the period suggest that an average of 250 acres were burnt each year out of a total area of about 5,000 acres (i.e. averaging a 20 year burn cycle) and drainage ditches were maintained to aid the heather management. Such intense grouse moor management would have most likely resulted in a lowering of the water-table by around 5 cm, as has been observed in a grip blocking study by Wilson et al. (2010). Interestingly, the MILLENNIA scenarios simulating such grouse moor management resulted in the modeled WTD capturing this drier period as inferred from the TA record, which in the model scenario was mainly a result of drainage.
Moreover, the managed model scenario indicated considerable losses in the soil C budget and decreased peat accumulation rates.
Up-scaled to an intense grouse moor management period of 100 years these accumulated C losses (i.e. 1850-1950) relate to around 5 kg C/m 2 less soil carbon or 10 cm lower peat depth accumulation than predicted for the unmanaged scenario. Notably, burning and drainage contributed equally to the carbon loss, via reduced NPP (i.e. less litter input) and enhanced decomposition (i.e. lower WTD). Garnett et al. (2001) indicated a carbon loss and lower peat accumulation on burn management at Moor House (based on experimental burn plots) of around 73 g C m À2 year À1 , slightly higher but very similar to our model predictions of 51 g C m À2 year À1 . However, the Garnett et al. study contained some methodological uncertainties in relation to calculating C stocks (Clay, Worrall, & Rose, 2010) and the MILLENNIA model did not account for any potential charcoal impacts on hydrology and decomposition. In particular, impacts on peat bulk density (i.e. charcoal pieces entering peat pores) with possible changes in water holding capacity and charcoal inputs representing an inert C pool (i.e. biomass partly by-passing decomposition; see Clay et al., 2010) with likely additional effects on decomposition rates (e.g. negative priming; see Lu et al., 2014) might need to be considered in future model developments. Notably, the Garnett et al. (2001) study has been highlighted by Evans et al. (2014) as the only substantial study assessing burning impacts on UK blanket bogs. Therefore, burn management implications on the long-term peat C stock remain uncertain.
Although the estimated mean annual soil CO 2 fluxes of around 300-400 g C m À2 year À1 (i.e. range of managed vs. unmanaged scenario, respectively) are within the same order of magnitude as reported for a study at Moor House (Clay et al., 2010), the CH 4 emissions of around 5-10 g C m À2 year À1 were slightly higher compared to published values for the Flow Country (4.3 g C m À2 year À1 ; HEINEMEYER AND SWINDLES | 4139 Levey & Gray, 2015) and Moor House (3.9 g C m À2 year À1 ; Worrall, Armstrong, & Adamson, 2007;and 6.4 g C m À2 year À1 ; Worrall, Burt, Rowson, Warburton, & Adamson, 2009). However, the existing data on net CH 4 emissions from peatlands are very uncertain, with many older studies using inadequate methodologies (i.e. long chamber incubation periods and inaccurate analyser techniques such as gas chromatographs). Moreover, recent data from Moor House during two wet and warm years (2015-2016) measured much higher than previous average annual CH 4 emissions from peat decomposition of 2 to 400 g C m À2 year À1 (unpublished; pers. comm. Rob Rose at CEH). However, the largest uncertainty for field measurements is likely related to manual chamber measurements often not capturing plant-mediated transfer rates and ebullition, bypassing methane oxidation and thus leading to an overall underestimation of "true" methane fluxes. As a result of both model and measurement uncertainties, the presented CH 4 emissions should thus not be taken as real flux predictions but rather be seen as indicators of possible differences to be assessed by future monitoring.
In conclusion, peat cores provide a valuable archive for reconstructing peat development in relation to past climate, particularly TA-based water-table reconstructions as a driver of peat and carbon accumulation. We have shown here that combining this peatland archive with actual site measurements and model scenarios of past water-tables can provide additional information on likely impacts of land management, which are not easily detectable otherwise, but are crucial for explaining observed peat accumulation not explained by climate alone.