Blanket peatlands constitute a rare ecosystem on a global
scale, but blanket peatland is the most important peatland type on the British Isles. Most
long-term peatland development models have focussed on peat bogs and
high-latitude regions. Here, we present a process-based 2-D hillslope model
to simulate long-term blanket peatland development along complex hillslope
topographies. To calibrate the model, the peatland architecture was assessed
along 56 hillslope transects in the headwaters of the river Dee (633 km
Peatlands occur across the globe and contain up to one third of the global soil carbon stock, despite covering approximately less than 3 % of the Earth's surface (Gorham, 1991; Xu et al., 2018). Especially at higher latitudes, peatlands are an important ecosystem type, and their dynamics have profoundly influenced the terrestrial carbon cycle throughout the Holocene (Yu et al., 2011). Unfortunately, little is known about long-term peatland dynamics and their response to climatic and land cover changes (Wu, 2012).
Blanket peatlands are spreads of peat of varying thickness, covering the underlying topography, thus “blanketing” the landscape (Lindsay, 1995). This peatland type occurs in hyperoceanic climates with cool and moist conditions throughout the year, and it is mostly confined to the maritime edges of the continents (Gallego-Sala and Prentice, 2013). Due to its location in the landscape, blanket peatland formation is more controlled by topography, compared to other peatland types (Parry et al., 2012). Although rare on a global scale, up to 6 % of the area of the United Kingdom is covered by blanket peatland (Jones et al., 2003). The large area of the Scottish blanket peatlands, covering 23 % of the country, compared to the international rarity of these environments, makes the Scottish peatlands a high-value target for conservation efforts (Fyfe et al., 2013; Tipping, 2008).
During the Holocene, large areas of blanket peatland have developed throughout the Scottish Highlands, and this shift from mineral to waterlogged and nutrient-poor organic soils is one of the most important Holocene landscape changes in Scotland. Different hypotheses have been raised regarding the cause of this peatland development (Tipping, 2008). The original hypothesis, as proposed by Moore, linked the blanket peatland initiation to human impact, where anthropogenic land use change and increased grazing during the Neolithic period led to a shift in the hillslope hydrology resulting in the paludification of the upland soils (Moore, 1973). While this hypothesis has been supported by local studies throughout the British Isles, other authors have suggested that, at least for Scotland, the initiation of blanket peatlands resulted from climatic changes during the Atlantic period (Ellis and Tallis, 2000; Huang, 2002; Simmons and Innes, 1988; Tipping, 2008). A recent study based on a database of basal radiocarbon dates shows regional differences in the timing of the blanket peatland development with an earlier timing for central and southern Scotland, compared to the other regions of the British Isles (Gallego-Sala et al., 2016). Most of the case studies studying the blanket peatland initiation are based on field data such as pollen cores and radiocarbon dating, but studying causalities based on timing alone is difficult (Gallego-Sala et al., 2016). Process-based modelling of this landscape transformation could prove to be a useful technique, complementary to the field data, to provide insight into the underlying processes and mechanisms.
In recent decades, several peatland models have been developed, varying in spatial and temporal scale and in model complexity (Frolking et al., 2010). A good overview of the models developed for simulating long-term peatland behaviour is given by Baird et al. (2012). Currently, several long-term peatland models such as DigiBog and the Holocene Peatland Model (HPM) allow for simulations of peatland processes and the feedbacks between ecology, hydrology and peat properties over Holocene timescales (Baird et al., 2012; Frolking et al., 2010). These models have been applied successfully within the context of peat bogs, but they are difficult to transfer to blanket peatlands for two reasons. Firstly, these models are developed as cohort models, where each year a new peat layer is added to the soil profile and included in the calculations for the remaining part of the simulations. As a result, these allow for the simulation of temporal changes in peat properties such as hydraulic conductivity within the peat profile, but as the simulated time period increases, these models become computationally expensive, especially when a spatial dimension is added. Secondly, these models have been developed for peat bogs, which have a different peatland architecture compared to blanket peatlands, and they are therefore not always adapted to simulating peatland processes along complex hillslope topographies. These issues have been partially resolved by the MILLENNIA model, which has been designed specifically for blanket peatlands (Heinemeyer et al., 2010). This model is also a cohort model but incorporates additional processes which are specific to blanket peat such as runoff-driven peat erosion. However, the high degree of detail in the model domain and the representation of the processes make it difficult to apply cohort models at the landscape scale.
In this study, a process-based peatland model is presented which is able to simulate the hillslope hydrology and peatland dynamics along topographically complex hillslopes on Holocene timescales. Additionally, the representation of the model domain is relatively simple using a diplotelmic peat profile, making it computationally feasible to study peatland development on a landscape scale by simulating a large number of hillslope cross sections. The model is applied to the Upper Dee area in the Cairngorms National Park in eastern Scotland. The goal of this study is twofold: firstly, to apply a relatively simple process-based peatland model to study the long-term blanket peatland development in the Scottish Highlands on a landscape scale and secondly, to identify the relative importance of climatic and land cover changes for long-term blanket peatland development.
The study area consists of the headwaters of the river Dee in eastern Scotland, with an elevation ranging from 322 to 1309 m a.s.l. The area lies in the centre of the Cairngorms National Park and is managed by the Mar Lodge, Invercauld and Mar estates (Fig. 1). The geology of the Dee catchment is characterized by metamorphic and igneous rocks, with schists and granulites in the southern part of the study area and granite batholith intrusions in the north (Maizels, 1985). The entire area was glaciated by the Scottish ice sheet during the last ice age, which retreated between approximately 16 kyr and 13.6 kyr BP. In contrast to the western Highlands, the Cairngorms massif was not subjected to widespread glacial expansion during the Younger Dryas (Loch Lomond Stadial). During this period, the glacial activity remained largely restricted to the cirques (Everest and Kubik, 2006). The development of the current landscape and soils in the Upper Dee area has been influenced by the deglaciation, forming a wide variety of glacial and fluvioglacial landforms (Ballantyne, 2008). In many parts of the study area, the bedrock is covered by glacial till of varying thicknesses (Maizels, 1985). The summits and ridges mostly carry skeletal soils and bedrock outcrops, while the slopes are covered by blanket peat and alpine podzols (Smith, 1985). The peat deposits are found both lying directly on bedrock and overlying a layer of mineral sediment. This mineral substrate consists of gravel-rich silt loam and sandy loam in the southern part of the study area and sandy loam to loamy sand in the northern part.
Currently, the area is dominated by semi-natural land cover, including
alpine and montane heath vegetation on the highest summits, heather moorland,
and small pockets of natural forest
(Tetzlaff and Soulsby, 2008). The total
annual precipitation ranges from 800 mm in the eastern part of the study
area to almost 2000 mm on the mountain tops, with a significant proportion
of the precipitation falling as snow during the winter months
(Dunn et al., 2001). The temperature regimes
can vary considerably within the study area. The town of Braemar (339 m a.s.l.) has a mean annual temperature of 6.8
Early Holocene traces of human presence have been found within the study area, with archaeological evidence indicating the presence of Mesolithic hunter–gatherer structures as early as 8.2 cal kyr BP in the western part of the study area (Warren et al., 2018). The first traces of permanent settlement in the Upper Dee are from the village of Braemar around 1000 CE (Paterson, 2011). In contrast to the western part of Scotland, the study area shows no traces of large scale peat extraction, which is probably due to the relatively thin peat profiles and difficult access to the area (Maurer, 2015).
Location of the Upper Dee area, with an indication of the hillslope transects and the pollen sites used for the land cover reconstruction.
For the study area, the blanket peatland architecture was assessed along 56 hillslope transects across the study area during field campaigns in 2015 and 2017 using soil corings, laboratory analysis of peat samples and pollen cores (Fig. 1). The soil corings were taken along the hillslope transects with a spacing of approximately 50 m using a gauge auger. The hillslope topography was measured using post-processing real-time kinematic (RTK) GPS measurements. The transect locations were selected in order to include a wide variety of lithologies, elevation zones and topographic parameters such as aspect, slope and curvature. Additionally, the carbon content, dry bulk density and water content of the peat deposits were derived from 35 field samples, collected as core sections with a length of 5 cm. These samples were collected at random coring locations and at depths ranging from 5 to 165 cm below the surface. The regional vegetation evolution over the past 12 000 years was reconstructed based on seven pollen cores located within the study area (Fig. 1). These cores provide vegetation information from different elevation zones with varying distances to the low-lying valleys of the Dee and the Spey (Birks, 1969; Hunter, 2016; Huntley, 1994; Paterson, 2011). Using the REVEALS (Regional Estimates of Vegetation Abundance from Large Sites) model (Sugita, 2007), the pollen percentages were converted to regional vegetation fractions. REVEALS was developed to reconstruct regional vegetation composition using pollen data from large lakes, but previous studies has shown that a group of sites can also be used to estimate regional vegetation cover (Fyfe et al., 2013; Mazier et al., 2012; Trondman et al., 2016). Pollen type parameters (pollen productivity and fall speed) were based on the standardized set of Mazier et al. (2012). The regional vegetation fractions were grouped in five classes (coniferous trees, deciduous trees, shrubs, heather, grasses and herbs) and used as land cover input in the hillslope model. As the land cover reconstruction for Scotland based on REVEALS by Fyfe et al. does not include pollen data from high-elevation sites, a new land cover reconstruction was made by Hunter for this study using local pollen data (Fyfe et al., 2013; Hunter, 2016).
At all coring locations, the soil profiles were described based on visual inspection, analysing the colour, texture and possible presence of macroscopic remains (charcoal, wood, etc.). Based on the coring descriptions, the peat thickness could be derived. In this study, peat is defined as a dark organic-rich layer of at least 10 cm thick without or with a minimal presence of mineral material based on visual inspection. Organic-rich horizons with a clear presence of mineral material were not classified as peat.
In total, 34 peat samples were radiocarbon dated at 17 locations throughout the study area, encompassing a range of topographic situations and peat thicknesses (see Appendix A2 for dating details). All radiocarbon dates were performed by the Belgian Royal Institute for Cultural Heritage and calibrated using the software OxCal 4.3 and the IntCal13 calibration curve (Bronk Ramsey, 2009; Reimer et al., 2013).
The model presented here is based on the concept of impeded drainage, where environmental conditions such as bedrock topography can cause waterlogging and peat formation (Alexandrov et al., 2016; Clymo, 1984; Ingram, 1982). The basic structure of the hillslope peatland model consists of a hydrology module simulating the water table behaviour along the hillslope, which is coupled to a peat growth module simulating biomass production and decomposition (Fig. 2). The model domain consists of a two-dimensional hillslope cross section, which is discretized in a series of model grid points. The stratigraphy consists of impermeable bedrock, overlain by a layer of glacial till, which is assumed to be porous. Over time, a peat profile can develop on top of this till substrate when the right environmental conditions are met. The hillslope topography is based on the detailed GPS measurements for each coring location.
General model workflow. For a more detailed description of the model structure, the reader is referred to the text.
The water table dynamics are modelled using a variant of the Boussinesq
equation for a non-constant slope
(Hilberts et al., 2004):
The diplotelmic nature of the model is represented by the depth-integrated saturated hydraulic conductivity. Each stratigraphic unit (mineral substrate, catotelm and acrotelm) has a specific saturated hydraulic conductivity value. The bedrock is excluded from the water table depth calculations as it is assumed to be impermeable.
A simple snow module is included in the hydrological model, with
precipitation falling as snow during periods with sub-zero temperatures. The
amount of melt is based on a degree day factor model. Snow sublimation is
not explicitly represented in the model. For each time step, infiltration
and saturation excess overland flow is calculated. The produced runoff is
assumed to leave the hillslope before the next time step. For open peatland
vegetation types, all rainfall is assumed to be able to infiltrate for
intensities below 2 mm h
The potential plant transpiration and soil evaporation (mm d
The actual evapotranspiration rate (AET) (mm d
Since detailed local climate reconstructions are scarce, several peatland
modelling studies have used continental or at best regional climate
reconstructions (mostly pollen-based) which were fine-tuned using local
climate information
(Frolking
et al., 2010; Heinemeyer et al., 2010; Morris et al., 2015). Here, a similar
approach is used. Input data for temperature and precipitation values are
based on a European gridded dataset of mean annual temperature and mean
monthly precipitation anomalies for the last 12 000 years derived from
pollen data with a spatial resolution of
Reconstructed mean annual temperature (
The peat accumulation at each grid point is calculated as the balance between
biomass production and decomposition. In the literature, several relatively
simple equations can be found to calculate the net primary production (NPP)
based on climatic data (mean annual temperature, total annual rainfall or
potential evapotranspiration) such as the Miami model or the Thornthwaite
memorial model of which some have been implemented in peatland models
(Heinemeyer et al., 2010; Lieth,
1973; Lieth and Box, 1972). In some cases, such as the Miami model, the NPP
is calculated based on multiple climatic variables (rainfall and
temperature), using the minimum value of both equations as the final
NPP value. However, given the climatic conditions of the Scottish Highlands,
precipitation is not a limiting factor for biomass production. As a
consequence, the biomass production is simulated as a function of the mean
annual temperature using a power function regression equation based on field
data for the Moor House–Upper Teesdale National Nature Reserve in northern England (Garnett,
1998), corrected for the woodland cover. A possible disadvantage of this
simple approach is the dependence of the biomass production calculations on
the quality of the climate and land cover reconstructions.
The peat column at each grid point is divided in an oxic and anoxic zone
based on the calculated mean annual water table height. The total
decomposition can thus be written as
Simulations start with a hillslope consisting of an impermeable bedrock covered by glacial till. As the thickness of the till is not known at each location, it is assumed to have a constant thickness of 50 cm. Over time, the organic matter accumulates within the upper 30 cm of the mineral soil forming an organic-rich horizon based on the balance between biomass production and decomposition. When a threshold is exceeded, additional organic matter which is produced starts to accumulate as peat at that location, with the properties of an acrotelm. In this study, the threshold is set at an amount of organic matter equivalent to a peat layer with a thickness of 10 cm using the median dry bulk density and organic-carbon percentage of the 35 peat samples collected in the field. This ensures that a similar definition for peat is used both for the hillslope corings as for the model simulations. Once the peat thickness exceeds the thickness of the acrotelm layer, the peat layer becomes diplotelmic, with the peat below the acrotelm having the properties of the catotelm. Once the biomass within the simulated peat profile decreases below the biomass threshold, the grid point is no longer considered to be covered by a peat layer and only mineral soil properties are taken into account.
The impermeable bedrock below the glacial till is used as a zero-flux boundary condition at the bottom of the model domain. At several locations throughout the study area, rivers have eroded the stream bank, exposing the peat. At the lower end of the hillslope, the water storage is thus set to a fixed value, representing the depth of the river. For the grid point at the top of the hillslope transect, a lateral zero-flux boundary is assumed.
Model calibration is based on the measured mean peat thickness per
topographic class. In total, nine topographic classes were defined by
dividing both the measured slope and curvature at each coring location into
three classes, resulting in nine possible combinations. The calibration
procedure resulted in topographic class limits of 0.098 and 0.135 m m
As an additional validation of the model behaviour, the simulated peat
growth initiation dates for all model grid points can be evaluated against a
dataset of basal radiocarbon dates for blanket peat deposits in the upland
regions of Scotland with an elevation above 300 m a.s.l. (
In total, soil coring descriptions were made at 866 locations throughout the
study area (detailed descriptions and location data can be found in the
data availability section, Swinnen, 2019a, b). Based on the definition of peat used in this study,
57 % of the coring locations contained a surface peat layer, with a
mean measured peat thickness over all coring locations of 36 cm and a maximal value of 3 m. The mean measured peat thickness per
hillslope transect varies between 0 and 96 cm (Fig. 4). Overall,
the transects with a high mean peat thickness can be found in the upstream
parts of tributaries of the river Dee. Strong spatial variability occurs,
even at small distances, making the peat cover throughout the area highly
variable in both occurrence and mean thickness. The mean measured peat
thickness per topographic class ranges from
Mean measured peat thickness at the hillslope transects (
As the model is based on the principle of impeded drainage, there is an assumed relationship between the bedrock slope and the peat thickness at a certain location. The peat thickness data indicate that this relationship is present, showing a clear decrease in the maximum thickness with an increasing bedrock slope. However, the variability, especially at lower slope angles, indicates that shallow peat layers or even the absence of peat is observed for every slope value (Fig. 5a). In the Boussinesq equation, the bedrock slope is used instead of the surface slope. One could argue that the bedrock slope might not relate directly to the surface slope in, for example, local depressions filled with peat. However, the comparison of the bedrock slope and surface slope for all coring locations indicate a clear and strong relationship between both, with the observed range in slope values strongly exceeding the differences between the bedrock and surface slope at a single location (Fig. 5b). As a result, the use of the bedrock slope is thus unlikely to introduce a bias in the modelled peat thickness values.
The pollen-based reconstructed land cover shows an Early Holocene woodland increase until the period 8.4–7.2 cal kyr BP (Fig. 6). This period is followed by a general woodland decline, with the woodland cover dropping below 5 % from 3.6 cal kyr BP onwards. The reconstructions for the individual pollen cores show an important east–west gradient in terms of maximal forest cover, with higher woodland percentages for the eastern and lower-lying part of the study area (Paterson, 2011). The woodland is a mixed type containing both coniferous species (Scots pine) and deciduous species (birch, rowan and aspen). A study by Fyfe et al. reconstructed the Holocene vegetation over Scotland using the REVEALS model for seven sites across the Scottish mainland, resulting in a maximal forest extent by 6.7 cal kyr BP (Fyfe et al., 2013). The data presented here show an earlier woodland cover decline around 7.2 cal kyr BP and a larger proportion of coniferous species in the forest composition. In comparison to the sites of Fyfe et al., the woodland cover appears to be relatively low, with an open heather landscape prevailing during the period under study, which can be attributed to the relatively high elevation of the study area.
Reconstructed vegetation proportions for the study area using the REVEALS model, based on seven pollen cores.
Point-by-point calibration resulted in poor correspondence between the
modelled and observed peat thickness. As a consequence, the model parameters
were calibrated based on the mean peat thickness per topographic class. In
total, nine topographic classes were constructed by classifying all coring
locations based on the slope and the hillslope curvature. The best fitting
parameter combination results in an acrotelm thickness of 10 cm, an
oxic decomposition rate at 10
Modelled and measured mean peat thickness per topographic class for both the calibration and validation transects.
The calibrated model was run to simulate the long-term blanket peatland
development since 12 kyr BP for the 56 hillslope transects using the
calibrated parameter values. The reconstructed land cover history (Fig. 6)
is used for vegetation evolution throughout the simulations. Overall, the
model simulations indicate that mean peat accumulation rates were low until
9.5 kyr BP, with small variations between the different grid points (Fig. 8).
Later, the accumulation rates increased and were high during two phases in
the Early Holocene: 9.5–8.5 and 8–6.5 kyr BP. From 6 to 2 kyr BP, the rates were relatively stable and slightly positive on average. A
long-term decrease in accumulation rates occurred between 2 and 1 kyr BP, which increased again to positive values around 0.5 kyr BP. The mean peat
and carbon accumulation rate over all grid points and for the entire studied
period is
Simulated mean peat accumulation rate and standard deviation for all grid points.
Figure 9 indicates the evolution of the mean peat thickness and
corresponding organic-carbon mass over all hillslope transects. The mean
peat thickness reaches a maximal value of 0.36 m around 2 kyr BP and
declines slightly afterwards to the current value of 0.33 m or 22.04 kg C m
Simulated mean peat thickness/carbon mass and standard deviation
for all grid points
Although the model calibration is solely based on the current peat
thickness, the timing of peatland development can be evaluated based on the
radiocarbon dating of peat profiles. Overall, the model simulations indicate
that peat growth initiated at most locations between 9.75 and 6.75 kyr BP. The basal radiocarbon dates (
Probability density function of the peat growth initiation dates
based on the model simulation and the radiocarbon dating database for
Scottish upland areas (above 300 m a.s.l., see Appendix A2) using a bin
width of 500 years.
To study the model sensitivity to variations in parameter values, a sensitivity analysis is carried out. In total, seven parameters are varied over the range as mentioned in the literature. The parameters broadly cluster in two groups: peat properties and environmental parameters (Table 1). Each parameter is varied stepwise, while all other parameters are kept at the standard value. The sensitivity is evaluated as the current mean peat thickness over all grid points after a simulation over a 12 000-year period using the pollen-based climate and land cover variations as environmental boundary conditions (Fig. 11). Overall, the model appears to be most sensitive to the peat decomposition rate, mean annual temperature and woodland cover. The peat thickness shows no sensitivity towards changes in catotelm conductivity. This is probably because the low conductivity values of the catotelm result in slow drainage compared to other components of the hillslope hydrology and thus in quasi-permanent water saturation of the catotelm. The acrotelm conductivity shows the same behaviour except for the lowest value. The acrotelm is under oxic conditions for most of the simulated values. Only for the lowest conductivity value does the water table rise above the catotelm–acrotelm boundary, resulting in lower decomposition rates and a higher mean peat thickness.
Overview of the parameters used in the parameter sensitivity test, listing the standard value and the range over which the parameter is changed.
Mean simulated peat thickness for all variables used in the parameter sensitivity test.
The climate sensitivity of a model grid point appears to be dependent on the presence of a peat layer. Overall, the percentage of model grid points covered by peat appears to be more sensitive to precipitation changes than the mean peat thickness (Figs. 12, 13). This might be a result of the diplotelmic representation of the peat profile. The strong difference in saturated hydraulic conductivity for the two peat horizons results in minimal water table changes when the precipitation amount is varied. In other words, the use of a diplotelmic model results in a water table which fluctuates only slightly around the acrotelm–catotelm boundary. As a result, the peat accumulation rates and resulting peat thickness are not very sensitive to precipitation changes. However, this does not mean that the water table is located at the same depth for all grid points. Depending on the local hillslope topography, some locations will be fully saturated for most of the time, while at other locations, the water table will be located much lower in the peat profile. For the substrate, only a single saturated hydraulic conductivity value is used, which results in a more sensitive response to precipitation changes for the grid points which do not have a peat cover. Overall, the highest peat thickness and percentage peat cover can be found for the scenarios with a high temperature and precipitation amount.
Mean peat thickness for all combinations of temperature and precipitation changes.
Percentage of all model grid points with a peat cover of at least 10 cm for all combinations of precipitation and temperature variations.
Point-by-point calibration of the hillslope model resulted in a poor
correspondence between modelled and observed peat thickness. Using the mean
peat thickness per topographic class, however allowed for a calibration of the model
with an RMSE below 10 cm (Fig. 7). This indicates that a spatial
peatland model with a simplified representation of the peat profile is
unable to capture the local variability, but it can replicate the general
peatland evolution on the landscape scale. Similar model behaviour has been
found in sediment erosion modelling, where point-by-point comparison yields
poor correspondence but where the mean value per topographic class performs
sufficiently well (Peeters et al., 2006). The
calibrated acrotelm thickness of 10 cm fits well within the range
of 5–50 cm mentioned in the literature
(Ballard
et al., 2011; Belyea and Clymo, 2001; Clymo, 1984). The same holds true for
the calibrated oxic decomposition rate of 2.15 % yr
Overall, the peatland model is not able to simulate the high peat thickness values (larger than 1.5 m), observed at some locations in the landscape. This can be partially attributed to the relatively high calibrated decomposition rates (Fig. 14). A second reason might the fact that local depressions within the hillslope topography were filtered out to enable the use of the Boussinesq equation for the simulation of the subsurface flow. Since local depressions showed to contain thick peat deposits in the field, the filtering procedure on the hillslope topography reduces the potential of modelling high peat thickness values at these locations.
Frequency distribution of the measured and modelled peat thickness for all grid points.
The range of simulated peat accumulation rates appears to be realistic, with
periods of high mean accumulation rates coinciding with periods of
temperature increase (Figs. 3, 8). More specifically, the mean peat
accumulation rates were high during the periods 10–8.5 kyr BP, where the
mean annual temperature increased 3.74
In contrast to existing cohort models, which have shown to be capable of capturing local variations in dynamics within the peat profile, the relatively simple diplotelmic model presented here cannot reproduce the local dynamics with the same degree of detail (Frolking et al., 2010; Heinemeyer et al., 2010; Morris et al., 2012). However, the simple representation of the model domain leads to a decrease in computation time which allows for the application of the model over large spatial and temporal domains. In combination with the pollen-based climate and land cover reconstructions, it allows for studying peatland development on the landscape scale, rather than at the scale of a single peat bog or peat profile as is often the case for the cohort models, making it possible to answer different research questions.
Based on the model simulations, the peat growth initiation dates cluster mostly within the period 9.75–6.25 kyr BP (Fig. 10), which corresponds largely to the Atlantic period, which is mentioned by other studies based on field data (Ellis and Tallis, 2000; Tipping, 2008). The database of basal radiocarbon dates shows a wider spread of initiation dates for upland Scotland between 12.25 and 3.25 kyr BP. However, the majority of the available basal dates within the dataset are based on a radiocarbon dating result higher in the peat profile, which is extrapolated to the bottom of the core using an age–depth model. One could question whether this extrapolation is justified. The deeper parts of the peat profile are older, and, as a result, a smaller fraction of the originally deposited biomass will remain. This leads to lower reconstructed accumulation rates further back in time. The reconstructed rates based on the available radiocarbon dates for the study area show a clear decreasing trend when going further back in time (Fig. 15, see Appendix A2 for dating details). When extrapolating to the bottom of a peat profile in order to obtain an initiation age, the accumulation rate of the above-lying part of the peat profile is used to extrapolate over the lowest part of the profile, which is older and has a lower fraction of remaining peat mass. As a result, these extrapolated ages will be biased towards younger basal ages. This effect increases with an increasing distance between the bottom of the peat profile and the deepest radiocarbon date. When excluding the extrapolated basal ages from the analysis, the probability density function shifts towards older ages, clustering between 11.25 and 4.25 kyr BP and corresponding much better with the probability density function of the model simulations (Fig. 10). Here, both the radiocarbon dates as well as the model simulations show peaks in peat growth initiation in the periods 10.25–8.25 and 7.75–6.25 kyr BP, which coincide with the periods of temperature increase. Overall, the simulations broadly show the same pattern as the radiocarbon dates, but there is a more pronounced difference between periods. The spikes in the simulated initiation dates might be attributed to the use of a single set of parameter values for the entire study area resulting in a relatively similar response of many model grid points to the changing environmental conditions during the Early Holocene. The more diffuse probability density function for the radiocarbon dates might thus, at least partially, be ascribed to local heterogeneity. Additionally, as the available basal radiocarbon dates come from areas all over Scotland, the probability density curve is likely to include regional differences in peatland initiation ages as well.
Reconstructed peat accumulation rates based on all available radiocarbon dates within the study area and the mean peat accumulation rate.
A study for the British Isles based on an envelope climate model for blanket peatlands finds a contraction in the area suitable for blanket peatland development in eastern Scotland since 6 kyr BP, and other studies find accumulation rates after 6 kyr BP to be relatively low (Gallego-Sala et al., 2016; Simmons and Innes, 1988). In this study, accumulation rates decrease from 8 to 6 kyr BP onwards (Fig. 7). Overall, the mean accumulation rate remains positive until approximately 2 kyr BP, but it never reaches the high values which occurred during the Early Holocene. This results in a slowdown of peatland development and carbon storage after 6 kyr BP (Fig. 9). The asymptotic behaviour of the peat thickness evolution at the landscape scale, stabilizing after 6 kyr BP, is also found in other studies modelling long-term peatland development at the local scale. While the modelled peat thickness trajectory is dependent on the specific conditions (climate, land cover, topography, etc.), it is clear that the carbon sequestering potential of a peatland has it limits at millennial timescales, as the balance between biomass production and decomposition comes in equilibrium with the environmental conditions (Frolking et al., 2010; Heinemeyer et al., 2010).
The conclusion that the blanket peatland development in the Upper Dee area
can be attributed to climate warming, independent of an increase in
precipitation, as demonstrated by the sensitivity analysis, is in line with
a study by a Morris et al. (2018), who compared a large dataset of peatland
initiation dates across the globe with GCM (general circulation model) paleoclimate simulations,
concluding that peatland initiation in formerly glaciated areas can be
attributed to rising growing season temperatures.
Additionally, a recent study on buried peat layers indicates that in
northern latitudes (> 40
The model simulations do not support the original hypothesis on the origin of the blanket peatlands, linking the peatland development to a deforestation-driven change in hillslope hydrology (Moore, 1973). Firstly, both the available basal radiocarbon dates and the simulated initiation dates indicate a shift towards peat soils during a period of increasing or stable woodland cover (Figs. 6, 10). Secondly, the parameter sensitivity analysis indicates that a decrease in tree cover, either by natural or anthropogenic causes, decreases the peat growth potential because the decrease in evapotranspiration due to a loss of tree cover is outweighed by the reduction in biomass production under the environmental conditions present in the study area. Tipping (2008) studied the Holocene blanket peatland development in five upland and northern sites in Scotland using a combination of geomorphic, archaeological and radiocarbon data, resulting in the hypothesis that blanket peatlands were common over large parts of the Scottish Highlands within the first few millennia of the Holocene either due to rapid soil development or climatic changes (Tipping, 2008). This study supports Tipping's hypothesis, as shown by the model simulations and peatland initiation dates, and it provides evidence that a changing climate (increasing mean annual temperature) was the main driver of blanket peatland development.
Although the simulated mean peat accumulation rates remain at low levels after 6 kyr BP, this does not mean peat profiles are unable to develop but rather that the peatlands at a landscape scale are in dynamic equilibrium with the stabilizing Holocene climate. This can be demonstrated using a forced model simulation where all peat soils are removed at 6 kyr BP. The resultant simulated peat thickness evolution indicates that peat starts to develop immediately after the peat removal (Fig. 16). After approximately 1500 years, the mean peat thickness over the study area again reaches the values of the standard model run. This indicates that the model can simulate peatland regeneration at locations which are impacted by the removal of peat cover either by natural processes (e.g. by shallow mass movements or gullies) or following anthropogenic peat cutting.
Simulated mean peat thickness/carbon mass and standard deviation
for all grid points
A new process-based model was presented to study long-term blanket peatland development along hillslopes. The simulations for the past 12 000 years indicate that a relatively simple diplotelmic model is able to capture long-term peatland dynamics on the landscape scale. However, point-by-point comparison still shows poor results, which can be attributed to the use of a single set of calibrated parameters and the idealized representation of the model domain. Overall, both the field data and model simulations indicate that the blanket peatlands in the Upper Dee area developed mostly during the Atlantic period, with a peak in peat growth initiation dates around 9 kyr BP. The timing of peatland initiation together with the results of the sensitivity analysis support the hypothesis of a climate-driven origin of the blanket peatlands in the Scottish highlands, where the peatland development shows to be driven by a long-term regional warming trend during the Early Holocene. A higher woodland cover leads to an increase in peat growth potential, contradicting the original hypothesis of Moore (1973), which identified deforestation as a potential driver of blanket peatland development. In more recent periods, the relatively stable climate and land cover within the study area since 6 kyr BP result in a stabilization of peatland development, indicating that the study area served as a terrestrial carbon sink mainly during the Atlantic period and has stabilized during the Late Holocene.
The supplementary data to this article consist of two datasets. A list of
all soil corings, including the coring location, elevation and measured peat
depth, is available online at
MIDAS (Met Office Integrated Data Archive System) weather stations used for the construction of the regression equations for orographic temperature and precipitation corrections.
Radiocarbon dating results. Calibrated ages were calculated using the software OxCal 4.3 and the IntCal13 calibration curve (Bronk Ramsey, 2009; Reimer et al., 2013).
The conceptualization and methodology development of this project was carried out by WS, NB and GV. The field work was performed by WS, NB and GV. WS carried out the lab work, developed the model code and performed the model simulations. GV and NB supervised the research. The writing of the paper was carried out by WS, NB and GV.
The authors declare that they have no conflict of interest.
The authors thank the Mar Estate, Mar Lodge Estate and
Invercauld Estate for permission to access the area. The authors thank
Teun Daniëls, Sofie De Geeter, Yasmine Hunter, Ellen Jennen, Vincent Lenaerts and Remi Swinnen for their assistance during the field campaigns.
Danny Paterson and Richard Tipping are thanked for sharing their pollen data
from the Upper Dee area. The pollen data of Birks (1969) and Huntley (1994)
were extracted from the European Pollen Database (EPD;
This research has been supported by the Fonds Wetenschappelijk Onderzoek (grant nos. 1167019N and G0A6317N).
This paper was edited by Alexey V. Eliseev and reviewed by Andreas Heinemeyer and one anonymous referee.