Introduction
The Earth's land surface is a heterogeneous mixture of vegetation types, lakes, wetlands,
and bare soil. Correct representation of such small-scale heterogeneities in
climate system models is a challenge. How can models better account for the
small-scale features in the large-scale climate system? Proposing a new
parameterization to fill a scaling gap between local and larger scales is the
main focus of this paper. Many recent studies have focused on different
approaches to simulate local small-scale characteristics of the land surface,
with climate enforcing evolution of different soil surface heterogeneities
and small-scale vegetation patterns . In turn, small-scale heterogeneity could influence the
land–atmosphere fluxes on a larger scale. Several studies have addressed the
hydrological cycle in drylands, where water recycled by vegetation may play
an important role in the local water budget .
In particular, showed that the amount of water
transferred through transpiration may change up to 10 % if one considers
different vegetation patterns, even with the same biomass density and the
same spatial scale. Recent efforts have also been focused on downscaling
remote sensing information to simulate subgrid surface heterogeneities
e.g.,, and to scale up information across
scales using network techniques .
Effects of small-scale heterogeneities on land–atmosphere fluxes are of
especial interest in northern peatlands because of the great amount of carbon
stored in the soil . Recent studies have
shown that greenhouse gas fluxes, in particular of methane, strongly depend
on the micro-topographic features of such environments , and that local hydrology is regulated by
micro-relief
. In particular, a typical
feature of methane-emitting landscapes is the nonlinear relationship between
fluxes and emitting surface area. A small fraction of the total landscape can
therefore function as a “hotspot” for methane fluxes. Recent eddy covariance
measurements in northern peatlands showed how the saturated surface, with
water table near to the surface level, despite covering only 10 % of the
total landscape, is responsible for up to 45 % of the total methane
emissions .
This “hotspot” feature of methane emissions potentially constitutes a large
local and even regional feedback to the climate system, which is neglected in
the current global circulation models (GCMs), as shown by, e.g.,
. Because of the complexity of the small-scale
biogeochemical and hydrological interactions that regulate this “hotspot”
effect, it is computationally feasible to represent such nonlinear phenomena
only in local mechanistic models i.e.,, with a fine-grained resolution (10-2–100 m). The “hotspot” effect is due to the nonlinear relationships between
decomposition and its drivers (e.g., soil temperature and water level), and
therefore a spatially explicit model able to identify such “hotspots” is
likely to perform better in representing methane emissions
.
developed the Hummock–Hollow (HH) model, a model
for resolving micro-relief features in a typical boreal peatland (hummocks
and hollows) and coupled this hydrological model to a process-based model for
methane emissions developed by . They found that a
micro-topography representation is necessary to correctly capture
hydrological dynamics and methane fluxes, as the water table position
regulates the depth of the oxic zone, where part of the methane coming from
the anoxic zone is oxidized and emitted to atmosphere as CO2.
Global land surface models such as JSBACH , the
land component of the Max Planck Institute Earth System Model MPI-ESM
, operate at a spatial resolution analogous to the
atmospheric one, which is about 50 km × 50 km at the finest feasible
scale. To include a representation of the “hotspot effect” on this scale, new
subgrid-scale parameterizations are needed.
In the present paper we propose a novel method to fill the scaling gap from
local mechanistic models to large-scale mean field approximations, using the
output of the local fine-grained model to tune and modify the coarse-grained
bucket-like model, in order to upscale the local information (100–101 m) to the landscape scale (e.g., 103 m).
We present an application of this upscaling method to the HH model, where we
analyze the dynamics of the area which we assume to be a hotspot for methane
emissions. We then use this information to modify a version of the HH model
without representation of micro-topography, which originally failed to represent
the magnitude of methane fluxes. In this paper we present (i) results for the
average climatology of the past 30 years, for which we calibrated the
parameterization, and (ii) results for the next century, testing the robustness of
the parameterization under a different forcing.
Methods
The HH model
The Hummock–Hollow (HH) model simulates peatland
micro-topographic controls on land–atmosphere fluxes. It is suited to work
at 1 m × 1 m resolution, which is the typical spatial scale of
peatland micro-topography. Each grid cell of the HH model represents just one
micro-topographic feature, namely a hummock or a hollow. The model simulates
a 1 km × 1 km peatland and its parameters are tuned with values
for a typical peatland in northwestern Russia. In particular, we use the
model to simulate the Ust-Pojeg mire in the Komi Republic
(61∘56′ N, 50∘13′ E; 119 m a.s.l.). The site has been
extensively studied, and recent efforts described peat characteristics
, fluxes of water vapor , carbon dioxide
, and methane , as well as energy and
water balance and spatial distribution of dissolved
organic carbon (DOC) . The micro-topography
is initialized with micro-topographic data collected through surveying with a
theodolite. An elevation distribution is derived from the data, and it is
possible to randomly assign an elevation at each grid cell for more
information,. Depending on the elevation, the grid cell
is therefore either a hollow or a hummock (Fig. ).
For each grid cell (i.e., for each micro-topographic unit) we compute the
water balance as
dWi,jdt=Sn+P-ETi,j-Ri,jsi,j,
where Wi,j is the water table level in the grid cell at the position
(i,j) relative to the surface level, Sn is the snowmelt, P is the
precipitation input, ETi,j is the evapotranspiration, Ri,j is the
lateral runoff, si,j is the drainable porosity, and t is time. The
time step is δt=1 day. Terms without the indices (i,j) are applied
uniformly over the model domain. Water table is computed in respect to the
micro-topographic surface, and it is positive above the surface, and negative
below it. For a description of the parameterization of Sn and ETi,j, see
Appendix A. This version of the model with the explicit representation of
hummocks and hollows is called the Microtopography configuration.
Schematics of the HH model showing two grid cells: a hummock and a
hollow. The model represents a 1 km × 1 km peatland, and works at a 1 m × 1 m grid cell. It is therefore able to resolve the
micro-topographical features such as hummocks and hollows. The figure shows
two typical grid cells, a hummock and a hollow, and the variables needed for
the water table dynamics (Eq. in the text). Each grid
cell has an elevation which is randomly assigned from the distribution of
elevation data collected in situ. For each grid cell we simulate a dynamical
water table, which changes with snowmelt (Sn), precipitation (P),
evapotranspiration (ET), and lateral runoff among the different grid
cells (Rhummock/hollow). These quantities regulate the change in water
table depth (W).
The HH model can also run in the Single Bucket configuration, where all
quantities are averaged over the model domain. Equation ()
becomes therefore
dWdt=Sn+P-ET-Rs.
The lateral flux is implemented in the same way in the two versions, but in
the Microtopography version the water can flow from cell to cell, while in
the Single Bucket version the water simply flows out of the system.
showed that the Single Bucket configuration, despite
being computationally much faster, fails to represent the peatland hydrology,
constantly underestimating the water table position in comparison to
measurements. This is due to the strong runoff that washes away the water at
the beginning of the simulation. Because of the more rugged, hummocky surface
represented in the Microtopography version, the runoff is delayed. This
behavior better agrees with in situ measurements for water table position
, whereas the water table position simulated by the HH
model in the Single Bucket configuration is too low. Table
describes the main differences between the two configurations of the HH
model, and the Hotspot parameterization we present in this paper.
Coupling to a process-based methane emission model
The HH model is coupled to a process-based model for methane emissions, in
order to quantify the effect of surface heterogeneities on greenhouse gas
fluxes. The model developed by is a quite general model
for methane emissions, and it can be applied to peatlands in different
environments. It is the same model that is used and coupled with some
dynamical global vegetation models (DGVMs)
e.g.,. We tuned the
model to perform in a typical peatland at the latitude of the Ust-Pojeg mire
complex. In the Microtopography configuration, we computed methane fluxes
locally and we averaged over the model domain in order to upscale the local
fluxes at the landscape scale. The process-based model for methane emissions
provides an output of methane fluxes FCH4i,j as a
function of the water table (computed by the HH model), net primary
productivity (NPP), and soil temperature (T):
FCH4i,j(t)=f(Wi,j,(t),NPP(t),T(t)),
where Wi,j is the water table depth with respect to the surface computed
at each position (i,j). All variables are represented at the daily time
step. We force the model with time series of T and NPP taken from CMIP5
experiments performed by the MPI-ESM model. We then considered the model
output for the grid cell which corresponds to the Ust-Pojeg mire (see
Sect. 2.4). The amount of methane which is emitted by each kind of surface
class changes according to the relative position of water table and surface.
In the process-based methane emission model developed by ,
the water table is a key variable in methane fluxes, because of the oxidation
processes simulated as the water table drops below the surface and as the
oxic zone deepens. The HH model in the Microtopography configuration
reasonably represents the hydrological interactions among hummocks and
hollows and the variability of emissions within the peatland. In the Single
Bucket configuration the water table drops quickly below the surface after
the snowmelt due to a strong runoff, and thus most of the methane transported
from below ground is oxidized. Parameters for the methane emission model are
described in Appendix B.
Description of the different configurations of the Hummock–Hollow
(HH) model used in the present paper.
Configuration
Properties
Resolution
Microtopography
Explicitly resolves micro-topography. Computationally expensive and requires fine-scale data for initialization.
1 m × 1 m
Single Bucket
Averages quantities over the domain. Does not consider micro-topography. Computationally fast and requires minimal information for initialization.
1 km × 1 km
Hotspot
Averages quantities over the domain. Considers micro-topographic information. Computationally fast and requires minimal information for initialization.
1 km × 1 km
The Hotspot parameterization
The HH model has a critical scale of about 0.01 km2 at which seasonal
results do not change for finer resolutions . Even at
this resolution it is unfeasible to include a micro-topography
parameterization in the current GCMs.
The general purpose of our Hotspot parameterization is to upscale information
from the local to the atmospheric scale. The HH model identifies different
surface types depending on the relative position of the water table W and
the surface:
W>ϵa⇒wet surface,-ϵb≤W≤ϵa⇒saturated surface,W<-ϵb⇒dry surface.
Here we assume, after , the following because of the importance of such thresholds for methane emissions:
ϵa=15 cm,ϵb=10 cm.
We assume the saturated surface to be the surface class which dominates the
methane emission dynamics, as a water table near to the surface prevents
oxidation.
After obtaining the seasonal behavior of the desired surface class, we aim to
parameterize of the area covered by the saturated surface class with a
fractional number q, which represents the fraction of the total surface
which is saturated at each time step. This information results in a different
water table behavior which in turns controls methane emissions. By knowing
the fraction q of saturated surface at each time step t, we implicitly
subdivide the domain of the HH model in the Single Bucket version A in
unsaturated surface Aunsat and saturated surface
Asat:
A=(1-q)Aunsat+qAsat.
The position of the water table in Asat stays between -ϵb≤Wts≤ϵa, which is given by the definition of the saturated
surface, and therefore we assume
Wts=-ϵb+(ϵa+ϵb)r,
where r is a random number between 0 and 1. The position of the water
table in Aunsat, instead, is the one computed by the HH model in
the Single Bucket configuration, i.e., W in Eq. (), which
responds to precipitation and evapotranspiration. Methane fluxes are
calculated as a function of the water table assuming a linear relationship
between emitting area and methane fluxes:
FCH4=(1-q)FCH4SB(W)+qFCH4sat(Wts)
where FCH4 is the methane flux from the whole domain,
FCH4SB the flux from the HH model in the Single
Bucket version, and FCH4sat the flux from the
saturated area Asat. The saturated area fraction q is defined
in Eq. (). The other forcing variables for
FCH4 stay unchanged, as in Eq. ().
The specific form of q as a function of time will be inferred by the
analysis of the saturated area dynamics, an output of the HH model in the
Microtopography configuration.
Forcing data
The HH model is forced with prescribed snowmelt, precipitation, and
evapotranspiration (Eq. ). The simulated Sn is a
stochastic input that functions as initialization parameter for the water
table. It is parameterized to gain the same magnitude of the observational
data . Evapotranspiration is simulated
according to observations of using an empirical
parameterization. All parameterizations are described in more detail in the
Appendices. In Eq. () we assumed Sn and P to be uniform
over the whole simulated domain and we did not apply any downscaling further.
We forced the process-based model for methane emissions developed by
(Eq. ) and the water balance (Eq. ) with prescribed time series of NPP and T, and of
precipitation P respectively. The time series are computed from simulations
performed for the CMIP5 experiments with the MPI-ESM model at T63 resolution
for the grid cell which corresponds to the Ust-Pojeg mire. The potential bias
introduced by using NPP of C3 grasses and not the one for mosses (not
included in the MPI-ESM model) is negligible as discussed by
.
We used the P, T, and NPP from the last 30 years of the IPCC historical
simulations and forced the model to infer a parameterization of the saturated
area (Eqs. and ) for the past 30-year
climatology. To assess the robustness of our parameterization for future
simulations we chose three Representative Concentration Pathways (RCP)
scenarios , and we therefore considered the identical set
of variables from the RCP2.6, RCP4.5, and RCP8.5 experiments from year 2006
to 2099 on daily resolution .
Area densities for dry (red line), wet (blue line), and saturated
(green line) grid cells. The solid lines represent the different surface
class dynamics averaged over 30 years, from 1976 to 2005. Shaded areas
represent standard deviations over the same period of time. The dynamics of
the saturated grid cells are mimicked by the empirical Hotspot
parameterization (black dotted line), Eq. () in the text.
Results and discussion
Hotspot area dynamics
By averaging the output of the model over 30 years of simulations, from 1976
to 2005 we calculated the average dynamics of the three surface classes: wet,
saturated, and dry. In particular, we are interested in the 30-year average
of the saturated area Asat dynamics (Eq. ).
After snowmelt, most of the simulated peatland surface is either saturated, or wet (Fig. ). As the
simulations continue, surface and subsurface runoff wash water out of the
peatland, changing the relative composition of the area densities. More and
more cells become dry by having a water table lower than 10 cm
below the surface. Grid cells belonging to the wet surface class,
with a high water table, become saturated and towards the beginning of August
virtually no grid cell displays a water table higher than 15 cm above the
surface level. At the end of the simulations, almost in all grid cells the
water table lies more than 10 cm below the surface level, and the peatland is
relatively dry by the end of October.
We used the output of the spatially explicit HH model to describe the
dynamics of methane emission hotspots, assuming that the saturated
grid cells are the ones where methane emissions are higher. We therefore
infer the dynamics of the saturated grid cells from Fig. and obtain the following parameterization for methane
emission hotspots:
q(t)=qin+qmax-qint1-t0(t-t0) if t≤t1qmax if t1<t≤t2qmax+qmin-qmaxt3-t2(t-t2) if t2<t≤t3qmin otherwise,
where t is the daily time step of the simulation, and the parameters ti
and qj are tuned quantities obtained according to the dynamics of
saturated grid cells in Fig. . Values for
the parameterization are described in Table . We slightly
overestimate the amount of saturated grid cells in order to take
into account the potential methane emission hotspots belonging to the
wet surface class.
Parameter values for Eq. (). We
infer the values from the dynamics of the grid cells belonging to the
saturated surface class as in Fig. . Days
are computed according to the Julian calendar.
Symbol
Meaning
Value
t0
Initial day of simulation
79
t1
Initial day of maximum saturation
110
t2
Final day of maximum saturation
170
t3
Initial day of minimum saturation
260
qin
Initial saturation area density
0.52
qmax
Maximum saturation area density
0.8
qmin
Minimum saturation area density
0.5
We illustrate the empirical parameterization of the area density computed by
Eq. () in Fig. (black dotted line).
This parameterization represents the average dynamics of methane emission
hotspots for the 30-year period 1976–2005.
Methane emissions for 1976–2005
We compared methane emissions from the Ust-Pojeg mire simulated over a
30-year period (1976–2005) in the three versions of the HH model
(Table ). We then averaged the 30 simulations and studied the
differences in dynamics among the different HH model versions. The
Microtopography configuration (black line in Fig. )
produces seasonal fluxes that more than double the cumulative methane fluxes
produced by the HH model in the Single Bucket configuration (red line in
Fig. ). In particular towards July and August, when
temperatures are higher and methane fluxes larger, the two versions of the HH
model diverge in flux estimation and the Single Bucket configuration largely
underestimates methane fluxes .
Cumulative emissions from different model configurations. The Single
Bucket configuration produces less than the half of the cumulative methane
emissions with respect to the model with micro-topography representation. By
inserting a simple parameterization of the saturated surface dynamics, we
improve significantly the seasonal methane emissions.
Symbol
Meaning
Value
Units
CH4SB
Cumulative emissions from the Single Bucket configuration
1.70±0.11×104
mg m-2
CH4Mic
Cumulative emissions from the Microtopography configuration
3.82±0.30×104
mg m-2
CH4HS
Cumulative emissions from the Single Bucket configuration with the Hotspot parameterization
3.47±0.25×104
mg m-2
Combining Eqs. () and (), and the empirical
parameterization of the hotspot area density q(t) (Eq. ), we
obtain a new flux dynamics (blue line in Fig. ). The new
parameterized fluxes display similar magnitude and dynamics as the fluxes
simulated by the Microtopography configuration, but at a much lower
computational cost. The main difference between the emissions from the Single
Bucket and Microtopography configurations is the large underestimation in the
central part of the summer season, i.e., in July and August. The Hotspot
parameterization, by changing the saturated area, improved this feature. The
visual improvement is confirmed by the large differences in the seasonally
cumulated methane emissions. The differences in cumulative emissions from the
three model configurations are summarized in Table 3.
Methane emissions from the HH model coupled with the
model. Solid lines are averages over 30 years (1976–2005)
and shaded areas represent standard deviations. Emissions are computed using
the HH model in the Microtopography configuration (black line), in the Single
Bucket configuration (red line), and in the Single Bucket configuration with
the Hotspot parameterization (blue line).
The Hotspot parameterization mimics the general magnitude and dynamics of the
emissions from the Microtopography configuration but fails to capture the
whole amplitude of methane emissions at the beginning and at the end of the
simulations. Such discrepancies might be caused by other variables which,
differently from the water table, remain averaged over the domain. In
particular, peat depth is uniform and the model does not have a heterogeneous
peat profile as in the Microtopography configuration. This difference may
influence the carbon available for methane emissions.
The Hotspot parameterization doubles the cumulative fluxes over the season
with respect to the Single Bucket configuration, despite its low
computational costs. From an ecological perspective, modeling CH4 fluxes
more accurately will improve our estimates of carbon stocks, which may help
constrain dynamic vegetation models, bacterial C consumption models, and
potential feedbacks with the atmosphere. Also, modeling hydroecological
effects of “slower” runoff from a peatland can potentially influence
vegetation dynamics of mosses in models including moss dynamics, e.g.,
. The HH model is novel in the physical representation of
lateral fluxes of water among hummocks and hollows, but other models
representing surface heterogeneity controls on water table
e.g., and methane fluxes e.g., display
similar effects. Therefore, and because of the process-based nature of the HH
model, we are confident in hypothesizing similar results if a Hotspot-like
parameterization was to be applied to other models.
Performances of the three configurations of the HH model for future
projections in different scenarios. Panels (a), (c), and
(e) represent seasonally cumulated methane emissions computed by the
HH model forced with CMIP5 data for the time period 2006–2099 from the
RCP8.5, 4.5, and 2.6 experiments respectively. Panels (b),
(d), and (f) represent the seasonal effectiveness of the
Hotspot parameterization for future projections, forced with CMIP5 data for
the time period 2006–2099 from the RCP8.5, 4.5, and 2.6 experiments
respectively. We illustrate the ratio of methane emissions with respect to
the Microtopography configuration for the Hotspot parameterization (in blue)
and the Single Bucket configuration (in red). We averaged each day of
simulation over the 2006–2099 period.
Future projections with the Hotspot parameterization
The Hotspot parameterization mimics the simulated methane emissions of the
Microtopography configuration for the 1976–2005 period for which it has been
tuned. We now force the model for the 2006–2099 period with data from the
CMIP5 experiments. The HH model does not simulate an increasing trend for
methane emissions for the next 100 years, despite the generally higher
temperatures (Fig. a, c, and e). Even in the RCP8.5 scenario,
despite an increase of 4 K in average temperature in year 2099 in respect to
the RCP4.5 and the RCP2.6 simulations, we can not find any significant trend.
This result is in agreement with the findings from the Wetland and Wetland
CH4 Inter-comparison of Models Project (WETCHIMP) experiments
, which did not find a large significant trend in methane
emissions simulated by the models participating in the inter-comparison
project because of increased temperature or of precipitation trends. We use
these two variables to force the HH model coupled with the methane emission
model. Such an increase is suggested to reduce stomatal conductance, with the
same amount of evapotranspiration, thus increasing waterlogged surface area.
In particular, did not find a large significant trend in
methane emissions simulated by the models participating in the
inter-comparison project because of increased temperature or precipitation
trends, which are the two variables we use to force the HH model coupled with
the methane emission model.
Moreover, future changes in precipitation could potentially affect the water
table position and therefore the saturated area fraction which may not
correspond to the one described in the Hotspot parameterization. In the RCP
simulations, even if precipitation changes in respect to present day and
among the scenarios, the differences are not so large to cause significant
effects on methane emissions. In Fig. a, c, and e, the outputs
of the HH model in the Microtopography and Single Bucket configurations
(i.e., the black and red lines, respectively) have water table explicitly
depending on precipitation simulated in the RCP scenarios. The Hotspot
parameterization (i.e., blue lines), despite using the saturated area
dynamics for the years 1976–2005, is quite close to the methane emissions
from the Microtopography configuration. We then conclude that the potential
bias introduced by using a fixed saturated area dynamics (the one for the
period 1976–2005) and not a dynamic one is negligible.
The Single Bucket configuration estimates 42.8–50.8 % of the methane
emissions cumulated over the season simulated by the Microtopography
configuration with the RCP8.5 scenario forcing. These estimates are very
similar with forcing from the RCP4.5 scenario (44.3–50.4 %) and from the
RCP2.6 scenario (43.0–50.6 %). If we include the Hotspot
parameterization, the simulated annual methane emissions range from
[2.831–4.321] ×104 mg m-2 with the RCP8.5 forcing. This
is 83.9–101.5 % of the emissions simulated by the Microtopography
configuration. As for the Single Bucket configuration, the numbers are
similar for the other forcing scenarios. The simulated emissions range from
[2.771–4.056] ×104 mg m-2 (88.4–100.1 % of the
emissions in the Microtopography configuration) for the RCP4.5 scenario, and
[2.648–4.102] ×104 mg m-2 (87.7–104.3 % of the
emissions in the Microtopography configuration) for the RCP2.6 scenario. The
amplitude and timing of year-to-year variability of cumulative methane
emissions with the Hotspot parameterization are also comparable to the ones
simulated by the Microtopography configuration in all simulated scenarios.
These results increase the applicability of the Hotspot parameterization.
Despite being tuned for the 1976–2005 climatology, it works for the next
century of simulations under very different forcing scenarios. This is due to
the large differences in hydrological representations between the
Microtopography and Single Bucket configurations. Such differences are almost
totally overcome with the use of the Hotspot parameterization. These
improvements make the parameterization applicable also for future time
slices, despite the differences in temperature, precipitation, and NPP
forcing between the time period used for the parameterization tuning and the
scenario projections.
We also tested the effectiveness of the Hotspot parameterization over the
seasonal cycle. We averaged for each simulated day the methane emissions over
the 2005–2099 period for all model configurations and for all scenarios. We
then divided the daily emissions from the Single Bucket configuration and
from the Hotspot parameterization by the emissions from the Microtopography
configuration to investigate the impact of the new parameterization on the
seasonal cycle. In all simulated scenarios, the Hotspot parameterization
works very well during the mid-season. From mid-May till the beginning of
October, when methane emissions are higher, the ratio between the Hotspot
parameterization and the Microtopography parameterization is near one
(Fig. b, d, and f). The ratio between emissions from the
Single Bucket configuration and the Microtopography configuration reaches its
maximum only towards the end of the simulations, therefore missing the larger
methane emissions peaks in June, July, and August.
Summary and conclusions
We developed a new parameterization to bridge the scaling gap between a
process-based, small-scale hydrological model for peatlands, and a mean field
approximation, analogous to a large-scale parameterization in a DGVM. The
Hotspot parameterization uses the output of the HH (Hummock–Hollow) model
, which simulates a 1 km × 1 km peatland.
The HH model can work in both configurations, a spatially explicit one
working at 1 m × 1 m scale, simulating explicitly hummocks and
hollows (the Microtopography configuration), and a mean field approximation
of it, where all quantities are averaged over the domain (the Single Bucket
configuration). If coupled to a process-based methane emission model
the Microtopography configuration simulates more realistic
methane fluxes because of the better representation of hydrology due to the
explicit description of processes at 1 m scale, but at a much higher
computational cost. We assumed that the lack of representation of saturated
areas in the Single Bucket configuration, which are methane emission
hotspots, diminishes the cumulative emissions over the season by half.
We inferred a parameterization of this hotspot area for emissions for the
period 1976–2005, which are the last 30 years of the historical simulations
from the CMIP5 experiments. We analyzed the spatial pattern of the HH model
output in the Microtopography configuration averaged over the 30 simulated
years. We introduced this information in the Single Bucket configuration,
modifying the hydrology of the mean field approximation, obtaining the
Hotspot parameterization. This novel approach that takes into account the
information from the spatially explicit simulations bridges the gaps between
the simulated methane emissions. The Hotspot parameterization, due to its
higher modified water table, is able to mimic the general magnitude and
dynamics of the emissions from the model with micro-topography
representation.
By forcing the model with time series of temperature, NPP, and precipitation
for the next century from CMIP5 experiments in the RCP8.5, RCP4.5, and RCP2.6
scenarios, we assessed the robustness of the Hotspot parameterization under
forcing for which it was not originally calibrated. The parameterization
holds for years 2006–2099 for all three scenarios. Overall, the ratio
between the seasonally cumulated emissions from the HH model in the
Microtopography configuration and the ones simulated by the Hotspot
parameterization ranges between 0.84 and 1.04. This is a substantial
improvement in comparison to the methane emissions simulated by the Single
Bucket configuration, which only produces between 43 and 51 % of the
seasonally cumulated methane emissions. The Hotspot parameterization at
almost no computational cost therefore qualitatively changes and improves the
simulated system response for methane emissions.
We only applied this method to the HH model simulating a single peatland in
western Russia. This method, though, uses the information of a mechanistic,
spatially explicit model and it is a significant first step towards a full
parameterization of the micro-topographic impacts on complex ecosystems at
the DGVM scale. In order to develop such a parameterization we would need a
comprehensive and statistical analysis on the response of the mechanistic
local-scale model to different climatic forcing: we would need HH-like models
working at the micro-topographic scale applied at different peatlands in
other climatic zones. Another limitation of the applicability of this study
is its dependency on the availability of data to calibrate the original HH
model in its Microtopography configuration, as accurate measurements of
peatland micro-relief are needed to initialize surface height. While it is
not realistic to have theodolite micro-topographic measurements globally,
other methods and products could help provide similar information. Aerial
photographs provide some information on micro-topography, but generally at an
overly coarse scale. Statistical downscaling methods as the ones used, e.g.,
by and are therefore needed to infer
information on surface heterogeneities, but they are
not necessarily useful in identifying micro-topography distribution. Airborne
measurements could aid in giving qualitative and stochastic information also
on structural peatland patterns, such as the ones described by
. This information could be used by the HH model to
generate nonrandom configurations, potentially investigating the influence of
structured patterns on hydrology and methane emissions.
Introducing the analysis of spatial patterns produced by different
mechanistic models in multiple ecosystems is a powerful method to infer
landscape-scale dynamics and characteristics of patterns.