Introduction
Forests play a particularly important role in the global carbon
cycle. Forests store almost 50 % of the terrestrial organic carbon and
90 % of vegetation biomass . Globally, 70 % of
the forest is managed and the importance of management is still increasing
both in relative and absolute terms. In densely populated regions, such as
Europe, almost all forest is intensively managed by humans. Recently, forest
management has become a top priority on the agenda of political negotiations
to mitigate climate change (Kyoto Protocol,
http://unfccc.int/resource/docs/convkp/kpeng.pdf). Because forest
plantations may remove CO2 from the atmosphere, if used for energy
production, harvested timber is a substitute for fossil fuel. Forest
management thus has great potential for mitigating climate change, which was
recognised in the United Nations Framework Convention on Climate Change and
the Kyoto Protocol.
Forests not only influence the global carbon cycle, but they also
dramatically affect the water vapour and energy fluxes exchanged with the
overlying atmosphere. It has been shown, for example, that the
evapotranspiration of young plantations can be so great that the streamflow
of neighbouring creeks is reduced by 50 % . Modelling
studies on the impact of forest plantations in regions that are snow-covered
in winter suggest that because of their reflectance (the so-called albedo),
forest could increase regional temperature by up to four degrees
. Management-related changes
in the albedo, energy balance and water cycle of forests are of the same magnitude as the differences between forests,
grasslands and croplands . Moreover, changes in the
water vapour and the energy exchange may offset the cooling effect obtained
by managing forests as stronger sinks for atmospheric CO2
. Despite the key implications of forest management on the
carbon–energy–water exchange, there have been no integrated studies on the
effects of forest management on the Earth's climate.
Earth system models are the most advanced tools for predicting future climate
. These models represent the interactions between the
atmosphere and the surface beneath, with the surface formalised as a
combination of open oceans, sea ice and land. For land, five classes are
distinguished: glacier, lake, wetland, urban and vegetated. Vegetation is
typically represented by different plant functional types. ORCHIDEE is the
land-surface component of the IPSL (Institut Pierre Simon Laplace) Earth
system model. Hence, by design, the ORCHIDEE model can be run coupled to the
LMDz global circulation model. In this coupled set-up, the atmospheric
conditions affect the land surface and the land surface, in turn, affects the
atmospheric conditions. Coupled land–atmosphere models thus offer the
possibility to quantify both the climatic effects of changes in the land
surface and the effects of climate change on the land surface. The most
advanced land-surface models used, for instance, in Earth system models to
predict climate changes (see the recent CMIP5 exercise), account for changes
in vegetation cover but consider forests to be mature and ageless, e.g.
JSBACH , CLM , MOSES ,
ORCHIDEE and LPJ-DVGM . At present, none
of the predictions of future climate thus accounts for the essential interactions
between forest management and climate. This gap in modelling capability
provides the motivation for further development of the ORCHIDEE land-surface
model to realistically simulate both the biophysical and biogeochemical
effects of forest management on the climate. The ORCHIDEE-CAN (short for
ORCHIDEE-CANOPY) branch of the land-surface model was specifically developed
to quantify the climatic effects of forest management.
The aim of this study is to describe the model developments and
parametrisation within ORCHIDEE-CAN and to evaluate its performance.
ORCHIDEE-CAN is validated against structural, biophysical and biogeochemical
data on the European scale. To allow comparison with the standard version of
ORCHIDEE, ORCHIDEE-CAN was run with a single-layer energy budget. A more
detailed description and evaluation of the new multi-layer energy budget and
multi-level radiative transfer scheme is given by ,
and . A new forest management
reconstruction, which is needed to drive forest management in ORCHIDEE-CAN,
is presented in , and the interactions between forest
management and the new albedo scheme have been discussed by .
Model overview
The starting point: ORCHIDEE SVN r2243
The land-surface model used for this study, ORCHIDEE, is based on two
different modules their Fig. 2. The first module
describes the fast processes such as the soil water budget and the exchanges
of energy, water and CO2 through photosynthesis between the atmosphere
and the biosphere . The second module
simulates the carbon dynamics of the terrestrial biosphere and essentially
represents processes such as maintenance and growth respiration, carbon
allocation, litter decomposition, soil carbon dynamics and phenology
. The trunk version of ORCHIDEE describes global vegetation
by 13 metaclasses (MTCs) with a specific parameter set (one for bare soil,
eight for forests, two for grasslands and two for croplands). Each MTC can be
divided into a user-defined number of plant functional types (PFTs) which can
be characterised by at least one parameter value that differs from the
parameter settings of the MTC. Parameters that are not given at the PFT level
are assigned the default value for the MTC to which the PFT belongs. By
default, none of the parameters is specified at the PFT level; hence, MTCs
and PFTs are the same for the standard ORCHIDEE-trunk version. A concise
description of the main processes in the ORCHIDEE-trunk version and a short
motivation to change these modules in ORCHIDEE-CAN is given in
Table .
Concise description of the modules in the standard ORCHIDEE version
with the motivation to change the modules in ORCHIDEE-CAN.
Module
Description
Motivation for change
Albedo
For each PFT the total albedo for the grid square is computed as a weighted average of the vegetation albedo, the soil albedo, and the snow albedo.
The scheme overlooks the effect of vegetation shading bare soil for sparse canopies and gives the ground in all PFTs the same reflectance properties as bare soil.
Soil hydrology
Vertical water flow in the soil is based on the Fokker–Planck equation that resolves water diffusion in non-saturated conditions from the Richards equation . The 2 m soil column consists of 11 moisture layers with an exponentially increasing depth .
No change
Soil temperature
The soil temperature is computed according to the Fourier equation using a finite difference implicit scheme with seven numerical nodes unevenly distributed between 0 and 5.5 m .
No change
Energy budget
The coupled energy balance scheme, and its exchange with the atmosphere, is based on that of . The surface is described as a single layer that includes both the soil surface and any vegetation.
A big leaf approach does not account for within canopy transport of carbon, water and energy. Further, it is inconsistent with the current multi-layer photosynthesis approach and the new multi-layer albedo approach.
Photosynthesis
C3 and C4 photosynthesis is calculated following and , respectively. Photosynthesis assigns artificial LAI levels to calculate the carbon assimilation of the canopy. These levels allow for a saturation of photosynthesis with LAI, but have no physical meaning.
The scheme uses a simple Beer law transmission of light to each level, which is inconsistent with the new albedo scheme.
Autotrophic respiration
Autotrophic respiration distinguishes maintenance and growth respiration. Maintenance respiration occurs in living plant compartments and is a function of temperature, biomass and, the prescribed carbon/nitrogen ratio of each tissue . A prescribed fraction of 28 % of the photosynthates allocated to growth is used in growth respiration . The remaining assimilates are distributed among the various plant organs using an allocation scheme based on resource limitations (see allocation).
No change
Carbon allocation
Carbon is allocated to the plant following resource limitations . Plants allocate carbon to their different tissues in response to external limitations of water, light and nitrogen availability. When the ratios of these limitations are out of bounds, prescribed allocation factors are used.
The resource limitation approach requires capping LAI at a predefined value. Due to this cap, the allocation rules are most often not applied, reducing the scheme to prescribing allocation.
Phenology
At the end of each day, the model checks whether the conditions for leaf onset are satisfied. The PFT-specific conditions are based on long- and short-term warmth and/or moisture conditions .
No change
Mortality and turnover
All biomass pools have a turnover time. Living biomass is transferred to the litter pool; litter is decomposed or transferred to the soil pool.
This approach is not capable of modelling stand dimensions.
Soil and litter carbon and heterotrophic respiration
Following , prescribed fractions of the different plant components go to the metabolic and structural litter pools following senescence, turnover or mortality. The decay of metabolic and structural litter is controlled by temperature and soil or litter humidity. For structural litter, its lignin content also influences the decay rate.
No change
Forest management
An explicit distribution of individual trees is the basis for a process-based simulation of mortality. The aboveground stand-scale wood increment is distributed on a yearly time step among individual trees according to the rule of : the basal area of each individual tree grows proportionally to its circumference.
The concept of the original implementation were retained, however, the implementation was adjusted for consistency with the new allocation scheme and to have a larger diversity of management strategies.
Before running simulations, it is necessary to bring the soil carbon pools
into equilibrium due to their slow fill rates, an approach known as model
spin-up . For a long time, spin-ups have been
performed by brute force, i.e. running the model iteratively over a
sufficiently long period which allows even the slowest carbon pool to reach
equilibrium. This naïve approach is reliable but slow (in the case of
ORCHIDEE it takes 3000 simulation years) and thus comes with a large
computational demand, often exceeding the computational cost of the
simulation itself. Alternative spin-up methods calling only parts of the
model, e.g. subsequent cycles of 10 years of photosynthesis only followed by
100 year cycles of soil processes only, have been used for ORCHIDEE to reduce
the computational cost in the past. These approaches, however, tend to lead
to instabilities in litter and carbon pools. In recent years, semi-analytical
methods have been proposed as a cost-effective solution to the spin-up issue
. A matrix-sequence method has been
implemented in ORCHIDEE following the approach used by the PaSim model
. The semi-analytical spin-up implemented in ORCHIDEE relies
on algebraic methods to solve a linear system of equations describing the
seven carbon pools separately for each PFT. Convergence of the method and
thus equilibrium of the carbon pools is assumed to be reached when the
variation of the passive carbon pool (which is the slowest) drops below a
predefined threshold. The net biome production (NBP) is used as a second
diagnostic criterion to confirm equilibrium of the carbon pools. In order to
optimise computing resources, the semi-analytical spin-up will stop before
the end of the run once the convergence criteria are met. ORCHIDEE's
implementation of the semi-analytical spin-up has been validated on regional
and global scales against a naïve spin-up, and has been found to converge
12 to 20 times faster. The largest gains were realised in the tropics and the
smallest gains in boreal climate (not shown).
Modifications between ORCHIDEE SVN r2243 and ORCHIDEE-CAN SVN r2290
Schematic overview of the changes in ORCHIDEE-CAN. For the trunk the
most important processes and connections are indicated in black, while the
processes and connections that were added or changed in ORCHIDEE-CAN are
indicated in red. Numbered arrows are discussed in Sect. 2.2.
One major overarching change in the ORCHIDEE-CAN branch is the increase in
internal consistency within the model by adding connections between the
different processes (Fig. , red arrows). A more specific novelty is
the introduction of circumference classes within forest PFTs, based on the
work of . For the temperate and boreal zone, tree height
and crown diameter are calculated from allometric relationships of tree
diameter that were parametrised based on the French, Spanish, Swedish and
German forest inventory data and the observational data from
. The circumference classes thus allow calculation of the
social position of trees within the canopy, which justifies applying an
intra-tree competition rule to account for the fact that
trees with a dominant position in the canopy are more likely to intercept
light than suppressed trees, and, therefore, contribute more to the stand
level photosynthesis and biomass growth. To respect the competition rule of
, a new allocation scheme was developed based on the pipe
model theory and its implementation by
. The scheme allocates carbon to different biomass pools
(leaves, fine roots, and sapwood) while respecting the differences in
longevity and hydraulic conductivity between the pools. In addition to the
biomass of the different pools, leaf area index (LAI), crown volume, crown density, stem diameter, stem height
and stand density are calculated and now depend on accumulated growth. The
new scheme allows for the removal of the parameter that caps the maximum LAI
(Table ).
The calculation of tree dimensions (e.g. sapwood area and tree height) that
respect the pipe theory supports making use of the hydraulic architecture of
plants to calculate the plant water supply (Fig. , arrow 1), which is
the amount of water a plant can transport from the soil to its stomata. The
representation of the plant hydraulic architecture is based on the scheme of
. The water supply is calculated as the ratio of the
pressure difference between soil and leaves, and the total hydraulic
resistance of the roots, leaves and sapwood, where the sapwood resistance is
increased when cavitation occurs. Species-specific parameter values were
compiled from the literature. As the scheme makes use of the soil water
potential, it requires the use of the 11-layer hydrology scheme of
(Table ). When transpiration based on
energy supply exceeds transpiration based on the water supply, the latter
restricts stomatal conductance directly, which is a physiologically more
realistic representation of drought stress than the reduction of the
carboxylation capacity done in the standard version of
ORCHIDEE (further also referred to as the “trunk” version). In line with
this approach, the drought stress factor used to trigger phenology and
senescence is now calculated as the ratio between the transpiration based on
water supply and transpiration based on atmospheric demand (Fig. ,
arrow 2).
The new allocation scheme also drastically changed the way forests are
represented in the ORCHIDEE-CAN branch. Although the exact location of the
canopies in the stand is not known, individual tree canopies are now
spherical elements with their horizontal location following a Poisson
distribution across the stand. Each PFT contains a user-defined number of
model trees, each one corresponding to a circumference class. Model trees are
replicated to give realistic stand densities. Following tree growth, canopy
dimensions and stand density are updated (Fig. , arrow 3). This
formulation results in a dynamic canopy structure that is exploited in other
parts of the model, i.e. precipitation interception, transpiration, energy
budget calculations, a radiation scheme (Fig. , arrow 4) and absorbed
light for photosynthesis (Fig. , arrow 5). In the trunk version these
processes are driven by the big-leaf canopy assumption. The introduction of
an explicit canopy structure is thought to be a key development with respect
to the objectives of the ORCHIDEE-CAN branch, i.e. quantifying the
biogeochemical and biophysical effects of forest management on atmospheric
climate.
The radiation transfer scheme at the land surface benefits from the
introduction of canopy structure. The trunk version of ORCHIDEE prescribes
the vegetation albedo solely as a function of LAI. In the ORCHIDEE-CAN branch
each tree canopy is assumed to be composed of uniformly distributed single
scatterers. Following the assumption of a Poisson distribution of the trees
on the land surface, the model of calculates the
transmission probability of light to any given vertical point in the forest.
This transmission probability is then used to calculate an effective LAI,
which is a statistical description of the vertical distribution of leaf mass
that accounts for stand density and horizontal tree distribution. The
complexity and computational costs are largely reduced by using the effective
LAI in combination with the 1-D two-stream radiation transfer model of
rather than resolving a full 3-D canopy model. By using the
effective LAI, the 1-D model reproduces the radiative fluxes of the 3-D
model. The approach of the two-stream radiation transfer model was extended
for a multi-layer canopy to be consistent with the
multi-layer energy budget and to better account for non-linearities in the
photosynthesis model. The scattering parameters and the background albedo
(i.e. the albedo of the surface below the dominant tree canopy) for the
two-stream radiation transfer model were extracted from the Joint Research
Centre Two-stream Inversion Package (JRC-TIP) remote sensing product
(Sect. 4.7). This approach produces fluxes of the light absorbed,
transmitted, and reflected by the canopy at vertically discretised levels,
which are then used for the energy budget (Fig. , arrow 6) and
photosynthesis calculations (Fig. , arrow 5).
The canopy radiative transfer scheme of separates the
calculation of the fluxes resulting from downwelling direct and diffuse
light, with different scattering parameters available for near-infrared (NIR)
and visible (VIS) light sources. The snow albedo scheme in the trunk does not
distinguish between these two short-wave bands. Therefore, the snow scheme of
the Biosphere-Atmosphere Transfer Scheme (BATS) for the Community Climate
Model was incorporated into the ORCHIDEE-CAN branch,
since it distinguishes between the NIR and VIS radiation. The radiation
scheme of requires snow to be put on the soil below the
tree canopy instead of on the canopy itself. The calculation of the snow
coverage of a PFT therefore had to be revised according to the scheme of
, which allows for snow to completely cover the ground at
depths greater than 0.2 m. The parameter values of were
used in the ORCHIDEE-CAN branch.
The ORCHIDEE-CAN branch differs from any other land-surface model by the
inclusion of a newly developed multi-layer energy budget. There are now
subcanopy wind, temperature, humidity, long-wave radiation and aerodynamic
resistance profiles, in addition to a check of energy closure at all levels.
The energy budget represents an implementation of some of the characteristics
of detailed single-site, iterative canopy models
e.g. within a system that is coupled
implicitly to the atmosphere. As an enhancement to the trunk version of
ORCHIDEE (Table ), the new approach also generates a leaf
temperature, using a vegetation profile and a vertical short-wave and
long-wave radiation distribution scheme , which will be
fully available when parametrisation of the scheme has been completed across
test sites corresponding to the species within the model . As
with the trunk version, the new energy budget is calculated implicitly
. An implicit solution is a linear solution in
which the surface temperature and fluxes are calculated in terms of the
atmospheric input at the same time step, whereas an explicit solution uses
atmospheric input from the previous time step to calculate the surface
temperature and fluxes. Although it is less straightforward to derive, the
implicit solution is more computationally efficient and stable, which allows
the model to be run over a time step of 15 min when coupled to the LMDz
atmospheric model – much longer than would be the case for an explicit
model. Parameters were derived by optimising the model against the
observations from short-term field campaigns. The new scheme may also be
reduced to the existing single layer case, so as to provide a means of
comparison and compatibility with the ORCHIDEE-trunk version.
The combined use of the new energy budget and the hydraulic architecture of
plants required changes to the calculation of the stomatal conductance and
photosynthesis (Fig. , arrow 7). When water supply limits
transpiration, stomatal conductance is reduced and photosynthesis needs to be
recalculated. Given that photosynthesis is among the computational
bottlenecks of the model, the semi-analytical procedure as available in
previous trunk versions (r2031 and further) is replaced by an adjusted
implementation of the analytical photosynthesis scheme of ,
which is also implemented in the latest ORCHIDEE-trunk version. In addition
to an analytical solution for photosynthesis, the scheme includes a modified
Arrhenius function for the temperature dependence that accounts for a
decrease in carboxylation capacity (kVcmax) and electron transport
capacity (kJmax; see Table for variable
explanations) at high temperatures and a temperature-dependent
kJmax/Vcmax ratio . The temperature response of
kVcmax and kJmax was parametrised with values from
reanalysed data in the literature , whereas
kVcmax and kJmax at a reference temperature of
25 ∘C were derived from observed species-specific values in the TRY
database . As the amount of absorbed light varies with
height (or canopy depth), the absorbed light computed from the albedo
routines is now directly used in the photosynthesis scheme, resulting in full
consistency between the top of the canopy albedo and absorption. This new
approach replaces the old scheme which used multiple levels based on the leaf
area index, not the physical height.
Variable description. Variables were grouped as follows: F= flux,
f= fraction, M= pool, m= modulator, d= stand dimension,
T= temperature, p= pressure, R= resistance, q= humidity,
g= function.
Symbol in text
Unit
Symbol in ORCHIDEE-CAN
Description
Frm
gC m-2 s-1
resp_maint
Maintenance respiration
Frg
gC m-2 s-1
resp_growth
Growth respiration
FLW,i
W m2
r_lw
Long-wave radiation incident at vegetation level i
FSW,i
W m2
r_sw
Short-wave radiation incident at vegetation level i
FTrs
m s-1
Transpir_supply
Amount of water that a tree can get up from the soil to its leaves for transpiration
Ta,i
K
temp_atmos_pres, temp_atmos_next
Atmospheric temperature at the “present” and “next” time step, respectively, at level i
TL,i
K
temp_leaf_pres
Leaf temperature at level i
qa,i
kg kg-1
q_atmos_pres, q_atmos_next
Specific humidity at the “present” and “next” time step, respectively, at level i
qL,i
kg kg-1
q_leaf_pres
Leaf-specific humidity at level i
Ml
gC plant-1
Cl
Leaf mass of an individual plant
Ms
gC plant-1
Cs
Sapwood mass of an individual plant
Mh
gC plant-1
Ch
Heartwood mass of an individual plant
Mr
gC plant-1
Cr
Root mass of an individual plant
Mlinc
gC plant-1
Cl_inc
Increment in leaf mass of an individual plant
Msinc
gC plant-1
Cs_inc
Increment in sapwood mass of an individual plant
Mrinc
gC plant-1
Cr_inc
Increment in root mass of an individual plant
Mtotinc
gC
b_inc_tot
Total biomass increment
Minc
gC plant-1
b_inc
Increment in plant biomass of an individual plant
Mswc
m3 m-3
swc
Volumetric soil water content
mw
–
wstress_fac
Modulator for water stress as experienced by the plants
mψ
MPa
psi_soil_tune
Modulator to account for resistance in the soil-root interface
mNdeath
–
scale_factor
Normalisation factor for mortality
mLAIcorr
–
lai_correction_factor
Adjustable parameter in the calculation of gap probabilities of grasses and crops
dh
m
height
Plant height
dl
m-2
–
One-sided leaf area of an individual plant
ds
m-2
–
Sapwood area of an individual plant
dhinc
m
delta_height
Height increment
ddbh
m
dia
Plant diameter
dba
m2 plant-1
ba
Basal area
dbainc
m2 plant-1
delta_ba
Basal area increment
dcirc
m
circ
Stem circumference of an individual plant
dind
trees
n_circ_class
Number of trees in diameter class l
dc
m2
crown_shadow_h
Projected area of an opaque tree crown
dcsa
m2
csa_sap
Projected crown surface area
dLAI
mleaf2 mground-2
–
Leaf area index
dLAIeff
–
laieff
Effective leaf area index
dLAIabove
–
lai_sum
Sum of the LAI of all levels above the current level
dA,i
m2
–
Cross-sectional area of vegetation level i
dhl,i
m
delta_h
Vegetation height of level i
dV,i
m3
–
Volume of vegetation level i
drd
–
root_dens
Root density
dλ
ind m2
–
Inverse of the individual plant density
pdelta
MPa
delta_P
Pressure difference between leaves and soil
pψsr
MPa
psi_soilroot
Bulk soil water potential in the rooting zone
pψs
MPa
psi_soil
Soil water potential for each soil layer
Rr
MPa s m-3
R_root
Hydraulic resistance of roots
Rsap
MPa s m-3
R_sap
Hydraulic resistance of sapwood
Rl
MPa s m-3
R_leaf
Hydraulic resistance of leaves
Rtemp
MPa s m-3
–
Hydraulic resistance of roots, sapwood or leaves adjusted for temperature
Ra,i
s m-1
big_r
Aerodynamic resistance of vegetation at level i in the canopy
Rs,i
s m-1
big_r_prime
Sum of the stomatal and leaf boundary layer resistance terms for latent heat
Continued.
Symbol in text
Unit
Symbol in ORCHIDEE-CAN
Description
fPwc
–
Pwc_h
Porosity of a tree crown
fPgaptrees
–
PgapL
Gap probability for trees
fPgapgc
–
PgapL
Gap probability for grasses and crops
fPgapbs
–
PgapL
Gap probability for bare soil
fdeathicir
–
mortality
Mortality fraction per circumference class
fKF
–
KF
Leaf allocation factor
fLF
–
LF
Root allocation factor
fγ
–
gamma
Slope of the intra-specific competition
fs
m
s
Slope of linearised relationship between height and basal area
frl
–
leaf_reflectance
Reflectance of a single leaf
ftl
–
leaf_transmittance
Transmittance of a single leaf
fRbgd
–
bdg_reflectance
Reflectance of the ground beneath the canopy
fColl,vegfR
–
Collim_alb_BB, Isotrop_alb_BB
Reflected fraction of light to the atmosphere which has collided with canopy elements, separated for direct and diffuse sources, respectively
fUnColl, bgdfR
–
Collim_alb_BC, Isotrop_alb_BC
Reflected fraction of light to the atmosphere which has not collided with any canopy elements, separated for direct and diffuse sources, respectively
fUnColl, vegT
–
Collim_Tran_Uncoll
Transmitted fraction of light to the ground which has not collided with any canopy elements
fColl, bgd,1fR
–
–
Reflected fraction of light which has struck the background a single time and has collided with vegetation
fColl, bgd,nfR
–
–
Reflected fraction of light which has struck the background multiple times and has collided with vegetation
z
m
z_array
Height above the soil
θz
radians
solar_angle
Solar zenith angle
θμ
radians
–
Cosine of the solar zenith angle
gG
–
–
Leaf orientation function
gσ
–
sigmas
Cut-off circumference of the intra-specific competition, calculated as a function of kncirc
ORCHIDEE-CAN incorporates a systematic mass balance closure for carbon
cycling to ensure that carbon is not getting created or destroyed during the
simulation. Hence, budget closure is now consistently checked for water,
carbon and energy throughout the model.
The trunk uses 13 PFTs to represent vegetation globally: one PFT for bare
soil, eight for forests, two for grasslands, and two for croplands. The
ORCHIDEE-CAN branch makes use of the externalisation of the PFT-dependent
parameters by adding 12 parameter sets that represent the main European tree
species. Species parameters were extracted from a wide range of sources
including original observations, large databases, primary research and remote
sensing products (Sect. 4). The use of age classes is introduced through
externalisation of the PFT parameters as well. Age classes are used during
land cover change and forest management to simulate the regrowth of a forest.
Following a land cover change, biomass and soil carbon pools (but not soil
water columns) are either merged or split to represent the various outcomes
of a land cover change. The number of age classes is user defined. Contrary
to typical age classes, the boundaries are determined by the tree diameter
rather than the age of the trees.
Finally, the forest management strategies in the ORCHIDEE-CAN branch were
refined from the original forest management (FM) branch
. Self-thinning was activated for all forests regardless
of human management, contrary to the original FM branch. The new default
management strategy thus has no human intervention but includes
self-thinning, which replaces the fixed 40 year turnover time for woody
biomass. Three management strategies with human intervention have been
implemented: (1) “high stands”, in which human intervention is restricted
to thinning operations based on stand density and diameter, with occasional
clear-cuts. Aboveground stems are harvested during operations, while branches and
belowground biomass are left to litter; (2) “coppices” involve two kinds of
cuts. The first coppice cut is based on stem diameter and the aboveground
woody biomass is harvested, whereas the belowground biomass is left living.
From this belowground biomass, new shoots sprout, which increases the number
of aboveground stems. In subsequent cuts the number of shoots is not
increased, although all aboveground wood biomass is still harvested; and (3)
“short rotation coppices”, where rotation periods are based on age and are
generally very short (3–6 years). The different management strategies can
occur with or without litter raking, which reduces the litter pools and has a
long-term effect on soil carbon . All management types are
parametrised based on forest inventory data, yield tables and guidelines for
forest management. The inclusion of forest management resulted in two
additional carbon pools, branches and coarse roots (i.e. aboveground and
belowground woody biomass) and therefore required an extension to the
semi-analytical spin-up method (Sect. 2.1). The semi-analytical spin-up is
now run for nine C pools.
Description of the developments
Allocation
Following bud burst, photosynthesis produces carbon that is added to the
labile carbon pool. Labile carbon is used to sustain the maintenance
respiration flux (Frm), which is the carbon cost to keep existing
tissue alive . Maintenance respiration for the whole plant
is calculated by summing maintenance respiration of the different plant
compartments, which is a function of the nitrogen concentration of the tissue
following the Beer–Lambert law and subtracted from the whole-plant labile
pool (up to a maximum of 80 % of the labile pool).
The remaining labile carbon pool is split into an active and a non-active
pool. The size of the active pool is calculated as a function of plant
phenology and temperature and was formalised following ,
and . The remaining non-active pool is
used to restore the labile and carbohydrate reserve pools according to the rules proposed
in . The labile pool is limited to 1 % of the plant
biomass or 10 times the actual daily photosynthesis. Any excess carbon is
transferred to the non-respiring carbohydrate reserve pool. The carbohydrate
reserve pool is capped to reflect limited starch accumulation in plants, but
carbon can move freely between the two reserve pools. After accounting for
growth respiration (Frg), i.e. the cost for producing new tissue
excluding the carbon required to build the tissue itself ,
the total allocatable C used for plant growth is obtained
(Mtotinc).
New biomass is allocated to leaves, roots, sapwood, heartwood, and fruits.
Allocation to leaves, roots and wood respects the pipe model theory
and thus assumes that producing one unit of leaf mass
requires a proportional amount of sapwood to transport water from the roots
to the leaves as well as a proportional fraction of roots to take up the
water from the soil. The different biomass pools have different turnover
times, and therefore at the end of the daily time step, the actual biomass
components may no longer respect the allometric relationships. Consequently,
at the start of the time step carbon is first allocated to restore the
allometric relationships before the remaining carbon is allocated in the
manner described below.The scaling parameter between leaf and sapwood mass is
derived from:
dl=kls×mw×ds
where dl is the one-sided leaf area of an individual plant,
ds is the sapwood cross-section area of an individual plant,
kls a parameter linking leaf area to sapwood cross-section area,
and mw is the water stress as defined in Sect. 3.2.
Alternatively, leaf area can be written as a function of leaf mass
(Ml) and the specific leaf area (ksla):
dl=Ml×ksla.
Sapwood mass Ms can be calculated from the sapwood cross-section
area ds as follows:
Ms=ds×dh×kρs,
where dh is the tree height and kρs is the
sapwood density. Following substitution of Eqs. (2) and (3) into Eq. (1),
leaf mass can be written as a function of sapwood mass:
Ml=Ms×fKF/dh,
where,
fKF=kls×mw/ksla×kρs,
where kls is calculated as a function of the gap fraction as
supported by site-level observations :
kls=klsmin+fPgap, trees×(klsmax-klsmin).
klsmin is the minimum observed leaf area to sapwood area ratio,
klsmax is the maximum observed leaf area to sapwood area ratio and
fPgap,trees is the actual gap fraction. By using the gap fraction
as a control of kls more carbon will be allocated to the leaves
until canopy closure is reached.
Following , sapwood mass and root mass (Mr)
are related as follows:
Ms=ksar×dh×Mr,
where the parameter ksar is calculated according to
(their Eq. 17):
ksar=(krcon/kscon)×(kτs/kτr)×kρs,
where krcon is the hydraulic conductivity of roots, kscon
is the hydraulic conductivity of sapwood, kτs is the
longevity of sapwood and kτr is the root longevity.
Following substitution of Eq. (4) into Eq. (7) and some rearrangement, leaf
mass can be written as a function of root mass:
Ml=fLF×Mr,
where,
fLF=ksar×fKF.
Parameter values used in Eqs. (1) to (9), i.e. klsmax,
klsmin, ksar, ksla, kρs,
krcon, kscon, kτs and kτr, are based on literature review (Tables S1, S2 and S3 in the
Supplement). The allometric relationships between the plant components and
the hydraulic architecture of the plant (Sect. ) are both
based on the pipe model theory; hence, both the allocation and the hydraulic
architecture module use the same parameter values for root and sapwood
conductivity.
In this version of ORCHIDEE, forests are modelled to have kncirc
circumference classes with dind identical trees in each one. Hence,
the allocatable biomass (Mtotinc) needs to be distributed across
l diameter classes:
Mtotinc=∑(l)[dind(l)×Minc(l)],
where Minc(l) is the biomass that can be allocated to diameter
class l. Mass conservation thus requires:
Minc(l)=Mlinc(l)+Mrinc(l)+Msinc(l),
where Mlinc(l), Mrinc(l) and, Msinc(l)
are the increase in leaf, root and wood biomass for a tree in diameter class
l, respectively. Equations () and () can be rewritten as
(Ml(l)+Mlinc(l))/(Ms(l)+Msinc(l))=fKF/(dh(l)+dhinc(l))(Ml(l)+Mlinc(l))=(Mr(l)+Mrinc(l))×fLF
An allometric relationship is used to describe the relationship between tree height and basal area :
dh(l)=kα1×(4/π×dba(l))(kβ1/2).
The change in height is then calculated as
dhinc(l)=[kα1×(4/π×(dba(l)+dbainc(l)))(kβ1/2)]-dh(l),
where dba(l) and dbainc(l) are the basal area and its
increment, respectively. kα1 and kβ1 are allometic
constants relating tree diameter and height. The distribution of C across the
l diameter classes depends on the basal area of the model tree within each
diameter class. Trees with a large basal area are assigned more carbon for
wood allocation than trees with a small basal area, according to the method
of .
dbainc(l)=fγ×dcirc(l)-km⋅gσ+(km×gσ+dcirc(l))2-(4×gσ×dcirc(l))/2,
where km is a parameter, fγ and gσ are calculated
from parameters and dcirc(l) is the circumference of the model
tree in diameter class l. gσ is a function of the diameter
distribution of the stand at a given time step.
Equations (10) to (16) need to be simultaneously solved. An iterative scheme
was avoided by linearising Eq. (15), which was found to be an acceptable
numerical approximation as allocation is calculated at a daily time step, and
hence the changes in height are small and the relationship is locally linear:
dhinc(l)=dbainc(l)/fs,
where fs is the slope of the locally linearised Eq. (15) and is
calculated as
fs=kstep/(kα1×(4/π⋅(dba+kstep))(kβ1/2)-kα1×(4/π×dba)(kβ1/2)).
Equations (10), (11), (12), (13), (14), (16), (17) and (18) are then solved
for fγ. fγ distributes photosynthates across the
different diameter classes and as such controls the intra-species competition
within a stand. fγ thus depends on the total allocatable carbon and
needs to be optimised at every time step. Once fγ has been
calculated, Mlinc(l), Mrinc(l) and
Msinc(l) can be calculated.
Hydraulic architecture
The representation of the impact of soil moisture stress on water, carbon and
energy fluxes has been identified as one of the major uncertainties in
land-surface models . Neither the empirical functions nor
the soil moisture stress functions, which are commonly used in land-surface
models, fully capture stomatal closure and limitation of C uptake during
drought stress . Therefore, we replaced the soil
moisture stress function which limits C assimilation through a constraint on
kVcmax in the ORCHIDEE trunk, with a constraint based on the amount
of water plants can transport from the soil to their leaves.
The model calculates plant water supply according to the implementation of
hydraulic architecture by . Plant water supply is the
amount of water the plant can transport from the soil to its stomata,
accounting for the resistances to water transport in the roots, sapwood and
leaves. If transpiration rate exceeds plant water supply, the stomatal
conductance is reduced until equilibrium is reached.
The water flow from the soil to the leaves is driven by a gradient of
decreasing water potential. Using Darcy's law , the supply of water for transpiration through stomata can be
described as
FTrs=pdelta/Rr+Rsap+Rl,
where pdelta is the pressure difference between the soil and the
leaves; and Rr, Rsap and Rl are the
hydraulic resistances of fine roots, sapwood and leaves, respectively.
pdelta is calculated following :
pdelta=pψsr-kψl-(dh×kρw×kg)
where kψl is a PFT-specific minimal leaf water potential,
which means that plants are assumed to maximise water uptake by lowering
their kψl to the minimum, if transpiration exceeds
FTrs . The product of dh, kρw and kg accounts for the loss in water potential by
lifting a mass of water from the soil to the place of transpiration at height
dh, kρw is the density of water, and
kg is the gravitational constant. The soil water potential in the
rooting zone (pψsr) was calculated by adding a modulator
(mψ) to the bulk soil water potential, which was calculated as the sum
of the soil water potential in each soil layer weighted by the relative share
of roots (drd) in the individual soil layer:
pψsr=∑(l)[pψs×drd]+mψ.
The soil water potential for each layer pψsl is calculated
from soil water content according to .
pψs(l)=1kavMswc-kswcrkswcs-kswcr-1/kmv-11/knv,
where Mswc is the volumetric soil water content, kswcr
and kswcs are respectively the residual and saturated soil water
content and kav, kmv and knv are parameters.
Root resistance is related to the root mass and thus can be expressed as :
Rr=1(krcon×Mr),
where krcon is the fine root hydraulic conductivity per unit
biomass. Sapwood resistance is calculated according to :
Rsap=dh(ds×kscon),
where kscon is the sapwood-specific conductivity, which is
decreased when cavitation occurs. The loss of conductance as a result of
cavitation is a function of pψsr and was implemented by using
an s-shaped vulnerability curve
kscon=kscon×e(-pψrs/kψ50)kc,
where kψ50 is the pψsr that causes 50 % loss of
conductance; and kc is a shape parameter.
Rl is related to the specific leaf conductivity per unit leaf
area (kl) and the leaf area index:
Rl=1(klcon×dLAI).
The response of water viscosity to low temperatures increases the resistance
. The relationship is described as
Rtemp=R(kα1v+kα2v×T),
where kα1v and kα2v are empirical
parameters , Rtemp is the temperature adjusted
Rl, Rsap or Rr, T is air temperature for
Rl and Rsap and T is soil temperature for
Rr.
If, for any time step, the transpiration calculated by the energy budget
exceeds the amount of water the plant can transport from the soil to its
stomata, transpiration is limited to the plant water supply. As the
transpiration is now reduced, the initial calculations of the energy budget
and photosynthesis, solely based on atmospheric information, are no longer
valid. As a result the energy budget and photosynthesis must be recalculated
for the time step in question. For this recalculation, stomatal conductance
at the canopy level is calculated such that transpiration equals the amount
of water the plant can transport. Owing to the feedback between stomatal
conductance, leaf surface temperature and transpiration, this calculation may
require up to 10 iterations to converge, using a stationary iterative method.
When the multi-layer energy budget is reduced to its single-layer
implementation, however, canopy level stomatal conductance is decomposed to
obtain the stomatal conductance at each canopy layer assuming that each layer
is equally restricted by drought stress. Finally, the restricted stomatal
conductance is used to calculate CO2 assimilation rate according to the
photosynthesis model by Farquhar, von Caemmerer and Berry (Sect. 3.6).
Canopy structure
Stand structure controls the amount of light that penetrates to a given depth
in the canopy. For example, the amount of light reaching the forest floor
will be higher for a stand with few mature trees compared to many young
trees, even if both stands have the same leaf area index. Where a big leaf
approach assumes a homogeneous block-shaped canopy (as in the trunk version
of ORCHIDEE) and can therefore rely on the law of Beer and Lambert, a
geometric approach is required to calculate light penetration through
structured canopies. Light penetration needs to be simulated to calculate
albedo (Sect. 3.4), photosynthesis (Sect. 3.6), partitioning of energy fluxes
(Sect. 3.5) and the amount of light reaching the forest floor (see for
example Sect. 3.1). The gap fraction, which is the basic information in
calculating light penetration at different depths in the canopy, is
calculated following the approach presented by and
formalised in their semi-analytical model. Rather than a spatially explicit
approach, follow a statistical approach which reduces the
memory requirements for the simulations and limits the space requirements for
storing the model output files.
The model of represents the canopy by a statistical height
distribution with varying crown sizes and stem diameters for each height
class. The crown canopies are treated as spheroids containing homogeneously
distributed single scatterers. Although this fPgap model can
explicitly include trunks, we made the decision to exclude them, as the
spectral parameters for our radiation model (Sect. 3.4) are extracted from
remote sensing data (Sect. 4.8) without distinguishing between leafy and
woody masses. This gives the gap probability for trees as a function of
height (z) and solar zenith angle (θz):
fPgaptrees(θz,z)=e-dλ×dc(θz,z)×(1-fPwc(θz,z))‾,
where dλ is the inverse of the tree density, dc is the
projected crown area (for an opaque canopy), and fPwc is the mean
crown porosity. The overbar depicts the mean over the tree distribution as a
function of tree height or, in our case, the mean over the l circumference
classes. Following minor adaptations, the implementation of Haverd and Lovell
was incorporated into ORCHIDEE-CAN. As there also exist
crops, grasses, and bare soil in the model, fPgap was adjusted for
these situations as well. For grasses and crops, the same formulation is
used:
fPgapgc(θz,z)=e-0.5×dLAIabove×mLAIcorr/cos(θz),
where dLAIabove is the total amount of LAI above height z, and
mLAIcorr is a correction factor to account for the fact that
grasses and crops are treated as homogeneous blocks of vegetation with no
internal structure, and is often referred to as a clumping factor. Here it is
treated as a tunable parameter and therefore the term “correction factor”
was used. For bare soil, there is no vegetation to intercept radiation, and
therefore fPgapbs(θz,z) is always unity.
Multi-layer two-way radiation scheme for tall canopies
Species-specific radiation absorbance, reflectance and transmittance by the
forest canopy were calculated from a radiation transfer model
which was parametrised by satellite-derived
species-specific scattering values (Sect. 4.8). Given the complexity of
radiation transfer, it remains challenging to accurately simulate radiation
transfer through structurally and optically complex vegetation canopies
without using explicit 3-D models. The applied 1-D model belongs to the
family of two-stream models and thus calculates
transmittance, absorbance and reflectance of both the incoming and outgoing
radiation. The calculation of the reflectance at the top of the canopy due to
a collimated source (i.e. the Sun) is divided into three components:
scattering of radiation between the vegetated elements with a black background
fColl, vegfR=f(θmu,frl,ftl,gG,dLAIeff)
scattering of radiation by the background with a black canopy
fUnColl, bgdfR=fRbgd×e(-dLAIeff/(2×θmu))×fUnColl,
vegT
multiple scattering of radiation between the canopy and the background
fColl, bgdfR=fRbgd×fColl,
bgd,1fR+fColl, bgd,nfR
Term (1) is widely used in cloud reflectance calculations, and depends on the
cosine of the solar zenith angle (θmu), the reflectance and
transmittance of the single leaves (frl and ftl,
respectively), the leaf orientation function (gG), and the
effective LAI (dLAIeff). The exact definition of this term is given
in Eq. (B2) in . In term (2), fRbgd is the
reflectance of the ground beneath the canopy and fUnColl,
vegT is the transmitted fraction of light to the ground which has
not collided with any canopy elements. In term (3), fColl,
bgd,1fR is the fraction of light which has struck vegetation
and collided with the background a single time, while fColl,
bgd,nfR is the fraction which has collided multiple times
(n) with the background.
The sum of the three components results in the canopy albedo
. Similar equations can be derived for light originating
from diffuse sources (e.g. clouds and other atmospheric scattering)
. Implementations of the calculations of the canopy fluxes
for a single level are available from the JRC, and these implementations were
used as the basis of the routines put into ORCHIDEE-CAN for both the single-
and multi-level cases . This implementation relies on the
use of the effective LAI, which is the LAI that needs to be used in a 1-D
process representation to obtain the same reflectance, absorbance and
transmittance as would be obtained by a 3-D-canopy representation
. In this study, the effective LAI was calculated by first
computing the canopy gap probability, i.e. the probability that light is
transmitted to a specified height in the canopy at a given solar angle. The
gap probability is then converted into the effective LAI by passing it as an
input to the inverted Beer–Lambert law (with an extinction coefficient of 0.5
to ensure compatibility with the two-steam inversion of
).
dLAIeff=-2.0×cos(θz)×log(fPgap),
where fPgap can be fPgaptrees,
fPgapgc, fPgapbs. Following the
introduction of multi-layer photosynthesis and energy budget submodels, the
approach proposed by had to be adjusted such that it could
be applied for every level for which absorbance needs to be known to
calculate photosynthesis (Sect. 3.6) and reflectance needs to be known to
calculate the net short-wave radiation (Sect. 3.5). The multi-layer approach
basically applies the 1-D two-stream canopy radiation transfer model by
to each canopy level where the light transmitted by the
overlaying level becomes the input for the lower level.
As the multi-level approach is built around the solution of the one-level
scheme for each canopy level, no new equations are introduced. The method can
be summarised by the following algorithm for which the details are given in
. First, three fluxes are calculated for each level
independently: the fraction of light transmitted through the layer without
striking vegetation, the fraction of light reflected after striking
vegetation, and the fraction of light transmitted through the layer after
striking vegetation. These three fluxes represent the only possible fate of
light (any light not taking one of these paths must be absorbed for energy
conservation). Next, an iterative approach is invoked which follows the path
of a single photon entering the top level. Based on the solutions for each
single level, probabilities can be calculated that the photon will be
transmitted to a lower level or reflected to a higher level. Any fraction
which is reflected upwards from the top level is added to the total canopy
albedo and is not considered further. The fraction which is transmitted
through the top level enters the next highest level, and again the
single-level solutions determine where this light goes. Any fraction
reflected upwards is considered in the next iteration as part of the light
entering the upper level. The steps continue until the bottom canopy level is
reached. Here, any fraction which is transmitted into the soil is removed
from consideration and added to the total transmittance through the canopy.
The algorithm then proceeds to the above canopy level. Now the transmitted
fluxes are moving in the upwards direction towards to the sky, while
reflected fluxes are moving towards the ground. The code continues towards
the top level, taking as input from below both the flux reflected by
downwelling light from the level below the current level and the flux
transmitted from the lower level by upwelling light. After each iteration
(moving from the top of the canopy to the bottom and back to the top), the
total amount of light considered active has been reduced by light escaping to
the sky or being absorbed by the canopy or ground. Eventually, this
“active” light falls below a pre-defined threshold, and the calculation is
considered to be converged.
Due to the iterative procedure, energy is not strictly conserved, although we
have attempted to choose a threshold which minimises this loss. The
multi-level albedo calculation is currently the most expensive part of the
model, due to the iterations and the fact that it must be performed over all
canopy levels (currently set to 10), grid points, and PFTs at every physical
time step. Levels with no LAI are no less expensive to compute, either,
although we have arranged our canopy levels to make sure no levels are empty
in most cases.
Multi-layer energy budget
The present generation of land-surface models have difficulties in
reproducing consistently the energy balances that are observed in field
studies . The
ORCHIDEE-CAN branch implemented an energy budget scheme that represents more
than one canopy layer to simulate the effects of scalar gradients within the
canopy for determining more accurately the net sensible and latent heat
fluxes that are passed to the atmosphere. As outlined in ,
the use of an implicit solution for coupling between the atmospheric model
and the surface layer model is the only way to keep profiles of temperature
and humidity synchronised across the two models when the coupled model is run
over large time steps (e.g. of 30 min). The difference between explicit and
implicit schemes is that an explicit scheme will calculate each value of the
variable (e.g. temperature and humidity) at the current time step entirely in
terms of values from the previous time step. An implicit scheme requires the
solution of equations written only in terms of those at the current time
step.
The modelling approach formalises three constraints that ensure energy
conservation. The three equations that describe the main interactions are the
following.
The energy balance at each layer is the sum of incoming and outgoing fluxes of latent and sensible heat and of short-wave and long-wave radiation:
klhc,ikρv,iδTL,iδt=-kshckρa(TL,i-Ta,i)Ra,i-kλ,LEρaqL,i-qa,iRs,i+FSW,i+FLW,i1Δdhl,i,
where FLW,i is the sum total of long-wave radiation, that is,
the net long-wave radiation absorbed into layer i,
and FSW,i is the net absorbed short-wave radiation as calculated
by the radiation scheme in Sect. 3.4. kshc is the specific heat
capacity of air. The source sensible heat flux from the leaf at level i is
the difference between the leaf temperature (TL,i) and the
atmospheric temperature at the same level (Ta,i), divided by
Ra,i, which is the leaf resistance to sensible heat flux (a
combination of stomatal and boundary layer resistance). Similarly, the source
latent heat flux from the leaf at level i is the difference between the
saturated humidity in the leaf (qL,i) and that in the
atmosphere at level i (qa,i), divided by Rs,i,
which is the leaf resistance to latent heat flux. Ra,i is
calculated based upon the leaf boundary layer resistance, and is described in
the present model according to . Rs,i is an abbreviation for the sum of the stomatal and leaf boundary layer resistance terms for latent heat.
The sensible heat flux between the vegetation (“the leaf”) and the surrounding
atmosphere at each level, and between adjacent atmospheric levels above and below, is provided by the following expression:
δTa,iδtΔdV,i=kk,iδ2Ta,iδz2ΔdA,i-TL,i-Ta,iRa,i1Δdhl,iΔdV,i,
where z denotes the height above the soil surface. We have re-written the
scalar conservation equation, as applied to canopies, in terms of the
sensible heat flux, temperature and source sensible heat from the vegetation
at each layer.
The latent heat flux between the vegetation and surrounding atmosphere at each level, and between adjacent
atmospheric levels above and below is described in a form that is analogous
to Eq. (36), above:
δqa,iδtΔdV,i=kk,iδ2qa,iδz2ΔdAi-qL,i-qa,iRs,i1Δdhl,iΔdVi)
In addition to these three basic equations, various terms had to be
parametrised. The
1-D second-order closure model of was used to simulate
the vertical transport coefficients kk,i within the canopy while
accounting for the vertical and horizontal distribution of LAI (Sect. 3.3).
This set of equations was then written in an implicit form and solved by
induction. More details on the implicit multi-layer energy budget and a
complete mathematical documentation are given in .
To complete the energy budget calculations, the multi-layer 1-D canopy
radiation transfer model (Sect. 3.4) was used to calculate the net short-wave
radiation at each canopy layer. Furthermore, the canopy radiation scheme
makes use of the Longwave Radiation Transfer Matrix (LRTM) . This approach separates the calculation of the radiation
distribution completely from the implicit expression. Instead, a single
source term for the long-wave radiation is added at each level. This means
that the distribution of LW radiation is now explicit (i.e. makes use of
information only from the “previous” and not the “current” time step),
but the changes within the time step were small enough not to affect the
overall stability of the model. However, an advantage of the approach is that
it accounts for a higher order of reflections from adjacent levels than the
single order assumed in the process above.
Analytical solution for photosynthesis
The photosynthesis model by Farquhar, von Caemmerer and Berry
predicts net photosynthesis of C3 plants as the minimum
of the Rubisco-limited rate of CO2 assimilation and the electron
transport-limited rate of CO2 assimilation . The
ORCHIDEE-CAN branch calculates net photosynthesis following an analytical
algorithm as described by . In addition, the C4 photosynthesis
is calculated by an equivalent version of the Farquhar, von Caemmerer and
Berry model that was extended to account for noncyclic electron transport
. A detailed derivation of the analytical solution of the
Farquhar, von Caemmerer and Berry model is given in . Although
the exclusion of mesophyll conductance from the photosynthesis model could
lead to an underestimation of the CO2 fertilization effect in Earth
system models , mesophyll conductance was not included in
ORCHIDEE-CAN, to maintain compatibility between the model formulation and its
parametrisation. Because values of kVcmax and kJmax
differ between different formulations of the photosynthesis model
and the parametrisation that was used in
ORCHIDEE-CAN did not include mesophyll conductance, it was also not accounted
for in the model formulation. The analytical photosynthesis model implemented
in ORCHIDEE-CAN could be easily extended to include mesophyll conductance,
but that would require reparameterising the photosynthesis model.
Owing to the canopy structure simulated in this model version and the
layering of the canopy, the amount of absorbed light now varies with canopy
depth. This new approach replaces the old scheme which uses multiple levels
based on the leaf area index, not the physical height within the canopy.
Photosynthesis is now calculated at each vertically resolved canopy level
independently, using the total amount of absorbed light calculated by the
radiation transfer scheme, which means that radiation transfer inside the
canopy and photosynthesis are now fully consistent. In the new photosynthesis
scheme, photosynthesis thus indirectly depends on canopy structure.
Forest management and natural mortality
Although forest management has developed a wide range of locally appropriate
and species-specific strategies , the nature of
large-scale land-surface models such as ORCHIDEE-CAN requires only a limited
number of contrasting strategies that are expected to be relevant on the
spatial scale (e.g. 50×50 km) of global and regional modelling
studies. Four management strategies were implemented based on their expected
impact on biogeochemical and biophysical processes.
In unmanaged stands self-thinning drives stand dynamics and continues until too few trees
are left on site. Subsequently, a stand replacing disturbance moves all standing biomass into
the appropriate litter pools and a new stand is established.
High stand management is characterised by regular thinning and a final harvest cut.
Thinning is decided on the basis of the deviation between the actual and potential stand density
for any given diameter. This approach relates to the so-called relative density index
, the land use disturbance index or hemeroby and naturalness
approaches . Exceeding a threshold diameter results in a clear cut and
the stand is replanted in the next year. For both thinning and harvest, leaves, roots and belowground
wood are transferred to the appropriate litter pools, whereas the aboveground woody biomass is removed
from the site and stored in a product pool. Trees with a diameter below a species-specific threshold
are stored in a short-lived product pool which mimics wood uses for fuel, paper and cardboard. Trees
with larger dimensions are moved to medium- and long-lived product pools which mimic, for example,
particle boards and timber usages, respectively.
Coppicing of the aboveground biomass is decided on stem diameter. At harvest, the root system is
left intact and, in between coppicing, no wood is harvested. Note that at present it is not possible
to simulate coppicing-with-standards in ORCHIDEE-CAN.
In ORCHIDEE-CAN, stands under short rotation management are limited to poplar (Populus spp.)
and willow (Salix spp.) forests. Stands are harvested at a prescribed age. Following a set number of harvest cycles, the root system is uprooted and the whole stand is replanted.
Different age classes are distinguished to better account for the structural
diversity and its possible effects on the element, energy and water fluxes. A
clear hierarchy was established for the mortality processes regarding the
actual killing of trees (i.e. move their biomass to the litter or harvest
pools). All of the processes determine first how much biomass they would
remove in the absence of all the other processes. Afterwards, the killing is
arranged in the most realistic way possible. A clear-cut event has the
highest priority, followed by human thinning and finally natural mortality
including self-thinning. If, for example, a forest is scheduled to be
clear-cut, the entire forest biomass is subjected to the rules of the
clear-cut and no other mortality occurs in that time step.
In addition to forest management and natural prescribed mortality, a variety
of changes have been made to processes involving vegetation mortality. A
whole PFT within a grid cell is now killed if, at the end of the day, the
labile pool is empty and there is no carbon available in the leaf or
carbohydrate reserve pool to refill it. In this situation, it will be
impossible for the plant to assimilate new carbon from the atmosphere as it
will not be able to grow new leaves and thus initiate plant recovery.
Furthermore, a forest can die if the density falls below a certain prescribed
value. In the next time step a new young forest will be prescribed.
If a forest is thinned, it is assumed that the weakest trees will be thinned,
and therefore human thinning reduces or even eliminates the natural mortality
for that time step. Natural mortality still happens on a daily time step,
while human-induced mortality happens only at the end of the year.
Self-thinning, as described below, takes priority over environmental
mortality, which is the mortality of individuals by insects, lightening,
wind, drought, frost and heart rot. Environmental mortality is calculated by
multiplying the stand biomass by an assumed mortality fraction of
1/ktresid. Where self-thinning is less than this assumed
environmental mortality, self-thinning is complemented by additional
mortality to reach the set environmental mortality. Where self-thinning
mortality exceeds the set environmental mortality, simulated self-thinning is
assumed to include environmental mortality. The fire module that is available
for the trunk but not for the ORCHIDEE-CAN branch simulates stand replacing
fires rather than individual-tree-based mortality due to lightening. The
approach implemented in the ORCHIDEE-CAN branch could therefore be extended
with models that simulate stand replacing mortality from fire, insects and
storms.
The use of circumference classes adds a good deal of realism and flexibility
to the ORCHIDEE-CAN simulations, but it also raises additional questions. For
example, which trees should be targeted by which mortality? Given that
self-thinning reflects the outcome of continuous resource competition, the
largest trees are expected to be most successful when competing for
resources, and therefore we assume that the smallest trees die first to
reduce the stand density. Conversely, larger trees are more likely to die
because of environmental stress factors, being more prone to cavitation, wind
damage, lightening, and, heart rot. Therefore, we select more older trees to
die from environmental mortality. While doing this also trees in the other
circumference classes were killed based on the following recursive definition
cf.:
fdeathicir=fdeathicir-1×kddf1-(kncirc-1)mNdeath,
where kddf is the death distribution factor, which is the factor by
which the smallest and largest circumference classes differ (e.g.
kddf=10 means that the largest circumference class will lose 10
times as much biomass as the smallest as a result of the mortality),
mNdeath is a normalisation factor so that the sum of
fdeathicir is unity, and fdeath1 is set equal to
unity before normalisation. As the stands are very close to even-aged, we set
the factor kddf to be equal to 1. This means the same number of
trees is killed in each circumference class. If, for some reason, there is
not enough biomass in a given class to satisfy this distribution, the extra
biomass is taken from the next smallest class (in case the smallest class
does not have enough, it is taken from the largest class).
Related to mortality is the question of the circumference class distribution.
As mentioned above, trees in different circumference classes are
preferentially killed by different processes. If the simulation is long
enough (or if the morality is aggressive enough), eventually the number of
trees in some circumference classes may become 0. This would reduce the
numerical resolution of the allocation scheme. When only one circumference
remains populated, the scheme effectively loses its meaning, as all the newly
produced biomass is now being allocated to the only remaining circumference
class. In order to maintain the same level of detail through the simulation,
the distribution of all the circumference classes is recalculated at the end
of each day. A normalised target distribution is specified as an input
parameter (an exponential distribution is currently used), and this
distribution is scaled to produce a target distribution for the current
number of individuals. All of the current individuals are placed in these new
classes until the target distribution is satisfied. The target distribution
now contains, however, trees of multiple sizes, so we need to average them to
find the new model tree for each class. By changing the size of the model
tree in each class, we are able to preserve the total biomass of the stand as
well as the total number of individuals. Note that the boundaries of each
diameter class are recalculated at each time step; this approach is a
numerically efficient alternative to fixing the boundaries of each diameter
class with a varying distribution.
Description of the parametrisation
The ORCHIDEE-CAN branch was specifically developed to quantify the climate
effects of forest management over Europe. Although the developments are
sufficiently general to be applied outside of Europe, the model was initially
parametrised for the boreal, temperate and Mediterranean climate zones and
validation focused on Europe. Parametrisation of the tropical zone is subject
of a follow-up study. The parametrisation of the model, including parameter
optimisation and tuning, consisted of five major steps:
Parameters related to carbon allocation (Sect. ), forest management and mortality (Sect. ),
hydraulic architecture (Sect. , canopy structure (Sect. )), photosynthesis (Sect. ),
and canopy radiation transfer (Sect. ), and for which observations exist at the species level
(Sect. ),
were extracted from a wide range of sources (Tables S1–S5). Using the extracted species-level parameter values in
ORCHIDEE without further processing avoids hidden model tuning and largely reduces the likelihood that simulation results will be biased by hidden calibration owing to a poor
taxonomic definition of PFTs .
The phenology-related parameters of the deciduous MTCs were optimised by , using MODIS-derived NDVI data normalised to model fAPAR over the 2000–2008 time period.
The modulator (mψ) which accounts for processes in the the soil-plant continuum that are currently not modelled, was manually tuned against species distribution maps (Sect. ).
The coefficient for maintenance respiration was optimised making use of Bayesian calibration (Sect. ) against a compilation of 100+ observations of biomass production efficiency.
The leaf to sapwood area ratio was manually tuned (Sect. ) to match 100+ site-level gross primary production (GPP) and LAI observations recorded over Europe.
Introducing 12 new PFTs
Similarly to the ORCHIDEE trunk, the ORCHIDEE-CAN branch distinguishes 13
metaclasses (MTC) for vegetation. Outside Europe the original MTC
classification of ORCHIDEE was kept, while inside Europe 12 new parameter
sets representing the main European tree species were added. The default
vegetation distribution map in ORCHIDEE, i.e. , was replaced
by an up-to-date global MTC map which has been produced using the ESA CCI ECV
Land Cover map (http://www.esa-landcover-cci.org/) .
The mapping from land cover to MTC basically followed ,
although Table 5 (the “cross-walking” table) has been updated following
discussions with the LC-CCI team at Universite Catholique de Louvain. For the
European domain, the global MTC distribution was overlaid by a tree species
distribution map .
This study focusses on tree species with a coverage of more than 2 % in
Europe, yielding seven species groups covering in total 78.8 % of the
European forest area: Betula sp., Fagus sylvatica,
Pinus sylvestris, Picea sp., Pinus pinaster,
Quercus ilex and a group combining Quercus robur and
Quercus petraea. For Pinus sylvestris, Picea sp.
and Betula sp. An additional distinction between boreal and
temperate forest was made for the species map and parametrisation: trees
located in Norway, Sweden and Finland were considered boreal, while trees
growing at lower latitudes were categorised as temperate. Given the potential
role of tree species of the Salicacea genus in short rotation coppice
management, a separate PFT was parametrised for Populus sp.
Furthermore, to improve the parametrisation of the MTC of boreal needleaved
deciduous forest, observations from Larix sp. were included when
possible.
For these 12 forest species, 12 new PFTs were created, with each PFT
belonging to a single MTC (Tables S2, S3 and S4). Almost 79 % of the
European forest was parametrised at the species level. The remaining 21 %
was reclassified into four residual groups, i.e. a temperate and boreal
needleleaf evergreen and a temperate and boreal broadleaved residual group.
For use outside Europe, the original MTC classification of ORCHIDEE was kept.
The parameters of the residual groups and MTCs are the mean of the parameters
of the species-level PFTs that are in the MTC, with the exception of albedo
parameters that could be extracted from remote-sensing products. Finally,
separate PFTs were introduced for boreal grasses and croplands, which allowed
for a boreal parametrisation of phenology, senescence and growth. This
approach, which distinguishes a total of 28 PFTs, allows a higher taxonomic
resolution over Europe, better defines forest types compared to the more
general MTC approach and facilitates the use of observations to derive
parameters.
Allocation
The allocation scheme relies on the leaf to sapwood area ratio
(Sect. ) and the relationship between diameter and height.
Following a logarithmic transformation of the more than 150 000 data points
from the national forest inventory data of Spain, France, Germany and Sweden,
the two parameters (i.e. kα1 and kβ1) describing the
relationship between diameter and height (Eq. ) were fitted at
the species level making use of a least square regression. Parameter values
for MTCs were derived by grouping the species into MTCs and fitting the
parameters. Data sources and parameter estimates are presented in Tables S2
and S3.
Forest management and mortality
Forest management and tree mortality are controlled by (Sect. 3.7):
(1) maximum tree diameter (no symbolic notation; called largest_tree_diam
in ORCHIDEE-CAN), (2) minimum stand density (no symbolic notation; called
ntrees_dia_profit in ORCHIDEE-CAN), (3) environmental mortality (no
symbolic notation; called residence_time in ORCHIDEE-CAN), (4) self-thinning
(kα2 and kβ2) and, (5) anthropogenic thinning (no
symbolic notation; called alpha_RDI_upper, alpha_RDI_lower,
beta_RDI_upper and beta_RDI _lower in ORCHIDEE-CAN) where the parameters
depend on the management strategy.
Maximum tree diameter was extracted from the French, Swedish, German and
Spanish forest inventories as the observed 50 % quantile for diameter at
breast height. The 50 % quantile rather than the observed maximum was used
to account for the fact that large-scale land-surface models are expected to
reproduce large-scale patterns rather than local extremes. Minimum stand
density was estimated as the expected stand density for the maximum tree
diameter for a stand under self-thinning. Although both criteria are related
to each other through the observed self-thinning relationship (see below),
the minimum number of trees is used to decide when unmanaged forests should
be replaced, whereas both the maximum diameter and the minimum number are
used for managed sites as criteria to initiate a clear cut. Parameters for
anthropogenic thinning are based on the national forest inventory data and
checked against the JRC database of species-specific yield tables. Parameter
values are presented in Table S5. Resource competition between trees in the
same stand has been reported to result in the so-called self-thinning
relationship that relates the number of individuals within a stand to the
stand biomass :
(Ms+Mh)×kρs=kα×(dind)-kβ,
where kα and kβ are the constants of the self-thinning
relationship. Furthermore, stem volume can be written as a function of tree
diameter (ddbh), tree height and stem form factor (kα′)
to account for the fact that the stem shape is not a perfect cylinder:
(Ms+Mh)⋅kρs=kα′×(ddbh)2×dh.
Following the allometric relationship given in Eq. (), tree
height can be written as a function of tree diameter. Hence, the
self-thinning relationship can be re-written to relate stand diameter to
stand density:
ddbh=kα2×(dind)-kβ2,
where, kβ2 relates to kβ1 (as in Eq. ) as
follows:
kβ2=-3/2×(2+kβ1)
kα1 and kβ1 were estimated by fitting Eq. ()
to observed diameter and height of individual trees from NFI of Sweden,
Germany, France and Spain. kβ2 was calculated from
Eq. () and kα2 was estimated by fitting
Eq. () to observations of the quadratic mean stand diameter
and stand density from NFI data.
Hydraulic architecture
Initial choices of parameters for this scheme were based on the values and
parameter sources listed by . All data sources were
revisited and the search was extended to obtain values at the PFT rather than
MTC level. Given that plant hydrology is rather well studied, observed
parameters were available for most of the species. Data sources are listed in
Table S1, whereas the parameter values are shown in Table S3. Our
implementation of hydraulic architecture required the introduction of a
tuning parameter (mψ) to account for processes that are currently
absent in the scheme, e.g. plant water storage and soil–root resistance. A
process-based description of these processes i.e. is being tested and should reduce the effect of the tuning
parameter and eventually allow its removal from the model.
For the time being, the modulator mψ was tuned manually against the
species distribution map to obtain a match between the simulated and observed
species distributions. When the modulator is set to zero, all PFTs experience
excessive water stress resulting in large-scale plant mortality. The
modulator was increased until the prescribed vegetation distribution which
was based on remote-sensing observations (Sect. ), survived
where it was prescribed. To this aim, the model was run for 50 years, forced
with v5.2 of the CRU-NCEP climatology for Europe (Climatic Research Unit,
University of East Anglia). Note that the values of the modulator depend on
the climate data that are used to force the model. Similarly the modulators
may need to be re-tuned when ORCHIDEE-CAN is coupled to an atmospheric model.
Canopy structure
The relationship between diameter and projected crown
surface area follows the model proposed by :
dcsa=kap×ddbhkbp
with parameters estimated using the data set presented in
. This data set contains diameter and projected crown
surface areas observations for over 37 000 individual trees in Europe
covering almost 30 species. Following logarithmic transformation of the
observations a linear least square regression was used to fit
species-specific parameter values. Parameter values are shown in Table S2.
Parameter values for MTCs were derived by grouping the species into MTCs and
fitting the parameters. No observations were available for the boreal zone
and temperate evergreen deciduous species. For the boreal species, a subset
of the temperate observations (Pinus sylvestris, Picea abies and Betula pendula) was used, i.e. the relationship between
dcsa and ddbh was fitted to all available data for
Pinus sylvestris. Next, all observations with a dcsa that
falls below the predicted dcsa were selected as considered to
represent a boreal subset. Given the importance of snow pressure on crown
structure, selecting observations with sub average dcsa is
justifiable as a first approximation. Subsequently, the parameters were
fitted to this subset of data. For Quercus ilex no data were
available and parameters were tuned such that the crown diameter was 0.85 m
less than the tree height.
Analytical solution for photosynthesis
Three originally MTC-specific photosynthetic parameters (kVcmax,
kJmax and ksla) were derived at the species level by
obtaining weighted site means for each species from the TRY global leaf trait
database and additionally from . Only
kVcmax and kJmax standardised to a common formulation and
parametrisation of the photosynthesis model by were
used. Most kVcmax and kJmax values in the TRY database
had already been standardised to a reference temperature of 25 ∘C
. Subsequently, a species-specific
kJmax,opt/kVcmax,opt ratio was calculated from the records
which included both kVcmax,opt and kJmax,opt
measurements. From this ratio, which was within a range of 1.91–2.47 for
each species, kJmax,opt was calculated for records which originally
only included kVcmax. Only geo-referenced observations within
Europe were used and the distinction between boreal and temperate forest was
made similar to the species map. Depending on the species this resulted in 5
to 183 observations for ksla and 11 to 173 observations for
kVcmax,opt and kJmax,opt. From these observations
species-specific means were calculated, weighted for differences in the
number of observations per site. The parameter values are shown in Table S3.
Multi-layer two-way radiation scheme for tall canopies
The radiation transfer scheme makes use of parameters describing leaf and
background properties, i.e. leaf single scattering and preferred scattering
direction (for both visible (VIS) and near-infrared (NIR) wavelengths) and
the so-called background albedo or the albedo of the surface below the
dominant tree canopy (VIS and NIR). All parameters were taken from the Joint
Research Centre Two-stream Inversion Package (JRC-TIP) . This is a software package which inverts a
two-stream model to best fit the MODIS broadband visible
and near-infrared white sky surface albedo from 2001 to 2010 at 1 km
resolution . The inverse procedure implemented in the
JRC-TIP is shown to be robust, reliable, and compliant with large-scale
processing requirements . Furthermore, this package ensures
the physical consistency between sets of observations, the two-stream model
parameters, and radiation fluxes.
Only parameter values for which the posterior standard deviation of the
probability density functions were significantly smaller than the prior
standard deviation were selected from the JRC-TIP optimisation
, since this condition ensures statistically significant
values. Species- and MTC-specific values were derived from JRC-TIP by
performing a multiple regression. This methods determines, in an objective
way, how the fractions of each MTC or species explain the JRC-TIP parameter.
The multiple regression was performed separately for the six parameters: the
single scattering of leaves (for both VIS and NIR), the scattering direction
of leaves (VIS and NIR) and the background albedo (VIS and NIR). Each JRC-TIP
parameter was used as the dependent variable and the independent variables
consisted of the fractions of each MTC or species
. These fractions were used to find a linear function that
best predicted each JRC-TIP parameter. The corresponding slope of a
regression of each MTC or species fraction gives the MTC or species dependent
JRC-TIP value. The multiple regression was performed without an intercept. To
avoid pollution by the seasonal cycle, the multiple regression was applied
only for the pixels of the Northern Hemisphere. Only pixels that were less
than 10 % covered by non-vegetative fractions where selected for the
analysis and only significant results following an F test and positive
r2 values were selected. The derived parameter values are shown in
Table S4.
Maintenance respiration
Both the trunk and ORCHIDEE-CAN branch reduce the definition of net primary
production to biomass production; hence, carbon leaching from the roots,
volatile organic emissions from the leaves, dissolved and particulate carbon
losses through water fluxes and carbon subsidies to mycorryhzae are not
accounted for in the model. These fluxes are (incorrectly) accounted for in
the modelled autotrophic respiration. Modelled autotrophic respiration should
therefore be considered an effective rather than a true value. For this
reason, the basal rate of autotrophic respiration was optimised against 126
site observations of the biomass production efficiency (kcmaint)
calculated as the ratio between annual biomass production and annual
photosynthesis , using a Bayesian optimisation
scheme. The scheme, for which more details are given in ,
uses a standard variational method based on the iterative minimisation of a
cost function that measures both the model data misfit and the parameter
deviations from prior knowledge .
The simulations that were used in the Bayesian optimisation prescribed a
20 m tall vegetation for temperate tree species, a 15 m tall vegetation for
boreal tree species and a 10 m tall vegetation for Mediterranean tree
species as its initial condition. This approach reduced the need for several
decades of simulations to a single year to grow a mature forests. In total,
the simulations were run for 10 years and covered the European domain. The
first year was discarded and the ratio between modelled GPP and NPP was
averaged over the remaining 9 years. Prior to the optimisation, the
observations were averaged for agricultural PFTs (0.57), and deciduous (0.44)
and evergreen (0.53) forest PFTs; the observed uncertainty was 0.03. The
parameter values were set to range between 0.0032 and 0.160. The optimisation
converged within 11 iterations and the optimised parameter values are shown
in Table S2.
It remains untested how well the simulated effective autotrophic respiration
represents the (rarely) observed autotrophic respiration. Note that in the
cases of both the trunk and the ORCHIDEE-CAN branch of ORCHIDEE, a match
between effective and observed autotrophic respiration should not be
interpreted as evidence of desired model behaviour because several components
of net primary production are not modelled yet.
After the optimisation of the maintenance respiration coefficient
(kcmaint), the model simulates reasonable biomass production
efficiency for a unit of photosynthesis. Hence, the final step of the
parametrisation focussed on optimising the leaf area, as this is one of the
main drivers of photosynthesis.
Sapwood to leaf area ratio
The vegetation structure simulated by the ORCHIDEE-CAN branch is sensitive to
the value of kls which describes the ratio between the leaf and
sapwood area of an individual tree. The available observations show a wide
range within and across forest species. Dependencies of kls on tree
height , tree diameter following stand
thinning and CO2 have been
reported. Most observations, however, come from experiments where time was
substituted by space which hampers teasing apart the sources of variability.
Given the variation and uncertainty in the observations and the model
sensitivity to this parameter, we manually tuned its value within the
observed range, to match European-wide observations of leaf area index as
recorded in the Database of Global Forest Ecosystem Structure and Function
.
This database was used to calculate a mean and maximum observed leaf area
index at the species level for the temperate and boreal region. Initially
20 year long European-wide simulations were used to simulate leaf area index
of a species, when the large-scale leaf area index approached the mean target
value and did not exceed the maximum value, the simulations were extended to
reach 100 years for checking the temporal evolution of leaf area index. We
deliberately optimised the sapwood to leaf area ratio (kls) by
making use of stand-level data to reduce circularity with the model
validation (see below).
Limited tests over a period of 100 years in a Scots pine forest at
51–52∘ N, 13–14∘ E (Fig. S1 in the Supplement) suggested
that optimising kcmaint and kls had the largest effect on
the maximum LAI, which decreased by almost 17 % after optimisation compared
to a simulation with prior parameter values. Mean annual GPP, mean annual
transpiration and basal area decreased by, respectively, 6, 6 and 7 %
compared to a simulation with prior parameter values (Fig. S1).
Validation
ORCHIDEE-CAN is designed as the land-surface model to be coupled to the LMDz
atmospheric model. As such, future applications of ORCHIDEE-CAN are expected
to be regional to global in the spatial domain and to span several years in
the temporal domain. Given its anticipated uses, the ability of the model to
reproduce large-scale spatial patterns as well as their inter-annual
variability is essential. The first applications of the model, both offline
and coupled to the atmosphere, will focus on Europe. The validation,
therefore, reports performance indices both over Europe as over eight
separate regions within Europe . These eight regions,
which partially overlap, are defined after . Furthermore,
the performance indices are calculated for winter, spring, summer and autumn,
and thus allow one to evaluate the capacity of the model to reproduce
observed annual cycles.
In addition to the root mean square error, a land performance index (LPI)
based on the principles laid out for the Climate Performance Index
their SI was also calculated. LPI normalises the root
of the squared differences between the simulations and observations by the
observed spatial and temporal variance. The LPI was used to estimate the
likelihood that the simulated variable belongs to the same population as the
observed variable, defined as exp(-0.5LPI2). An LPI equal to 1
indicates that the model correctly reproduces the mean observed value and
implies a likelihood of 61 % that the simulations and
observations come from the same population. Similarly, an LPI of 2 reduces
this likelihood to 13 %. An LPI of less than 0.32 has a likelihood of more
than 95 % and therefore indicates a statistically significant result.
While developing ORCHIDEE-CAN, the numerical approaches that added
functionality to the code were selected on the basis of their performance at
the site level (see below). Rather than running the same site-level tests for
our implementation, we performed a complementary large-scale validation. The
strength of our approach lies not in the details, as is the case for
site-level validation, but in its width by simultaneously testing model
performance for structural variables such as basal area ,
canopy structure and canopy height ,
biogeochemical fluxes such as GPP , biophysical fluxes such
as albedo and fluxes at the interface of biogeochemistry
and biophysics such as evapotranspiration . The selection of
variables was limited by the availability of spatially explicit data-derived
products for Europe.
For the validation, both the trunk and ORCHIDEE-CAN branch were run from 1850
to 1900 using CRU-NCEP climate forcing from 1901 to 1950 at 0.5 degree
resolution. From 1901 until 2012, the corresponding CRU-NCEP forcing data for
each year were used. Both versions used the 11 layer soil hydrology, the
single-layer energy budget and the same land cover map .
Given that no European-wide, spatially explicit and data-derived products
were found for the validation of the net carbon flux, there was no need for a
carbon spin-up. For the ORCHIDEE-CAN branch, the observed tree height and
basal area were compared against the simulation values at the end of 2010
(the trunk does not simulate these variables). For both the trunk and the
ORCHIDEE-CAN branch, the observed GPP, evapotranspiration, effective LAI and
VIS and NIR albedos were compared against monthly means between 2001 and
2010.
Species versus PFTs
In ORCHIDEE-CAN the PFT concept was refined by parametrising the main
European tree species groups (Sect. ). To evaluate the
effect of the species parametrisation, we performed a companion simulation
for the configuration described above, but at the MTC level. Model
performance was barely affected by the use of the MTC parameters, compared to
the simulation with the species parameters (see Fig. S2 for RMSE scores).
Allocation
In ORCHIDEE-CAN, functional relationships which vary by species and light
stress are used to allocate carbon among the fine roots, foliage and sapwood.
The allocation scheme largely follows , who in turn was
inspired by . Approaches simulating allocation based on
functional relationships were found to out-compete allocation schemes based
on constant fractions or resource limitation . The ability
of these schemes to reproduce foliage, fine root and sapwood reported in
large observational data sets for example,
demonstrates that these schemes capture the main observed features
. In addition, allocation schemes making use of functional
relationships were also capable of simulating the observed effect of elevated
CO2 on two mature forest ecosystems . Despite these
successes, the schemes were reported to be sensitive to their
parametrisation. Differences in parameters were reported to result in
substantial differences in the simulated allocation. The parameters for the
functional relationships used in ORCHIDEE-CAN are given in Table S2. The main
conceptual difference between the allocation scheme by and
ORCHIDEE-CAN is that the latter was designed to simulate one or more diameter
classes.
Given that photosynthesis is still calculated at the stand level (and thus
not at the tree level) the allocation rule of was
integrated in the functional allocation scheme to account for light and
resource competition within a stand. Where the functional relationships are
used to simulate carbon allocation within an individual tree of a given
diameter, the rule of allocates carbon across the
different diameter classes. The allocation rule which models the radial
increment for individual trees in pure even-aged stands was successfully
tested for Norway spruce and Douglas fir stands in France
. A similar approach for modelling radial increment has
already been implemented in a version close to the trunk of ORCHIDEE
and was able to successfully simulate stand
characteristics such as height, basal area and stand diameter
. This previous implementation differs from the current
implementation in its time resolution (which is now daily instead of yearly),
its analytical solution and the underlying allocation scheme (which is now
based on functional relationships instead of resource limitation).
The aforementioned studies performed a detailed validation of the two
approaches dealing with carbon allocation, which were combined in
ORCHIDEE-CAN. Complementary to these studies, we performed a European-wide
validation of our implementation and parametrisation of these well-tested
schemes against a remote-sensing-based map of tree height ,
upscaled eddy-covariance observations for GPP and a map of
basal area based on national forest inventory data . The
model's ability to reproduce GPP is thought to reflect its capacity to
simulate the foliage biomass, a correct simulation of height reflects the
model's capacity to simulate aboveground woody biomass, and its capacity to
reproduce observed basal areas suggests that the interaction of stand density
and individual tree diameter are well captured.
Likelihood that the simulated variable comes from the same
population as the data. The ORCHIDEE-trunk version does not include effective
LAI, basal area and height. Note that the likelihood of Europe cannot be
derived from the values of the other regions due to the overlap between
regions.
ORCHIDEE-CAN
ORCHIDEE-TRUNK
GPP
EVAPO
ALB_NIR
ALB_VIS
EFFLAI
BA
HEIGHT
GPP
EVAPO
ALBEDO
EFFLAI
BA
HEIGHT
British Isles
0.91
0.87
0.78
0.45
0.55
0.47
0.13
0.91
0.49
0.74
0.04
-
-
-
Iberian Peninsula
0.80
0.80
0.73
0.65
0.60
0.09
0.66
0.65
0.37
0.25
0.04
-
-
-
France
0.86
0.90
0.92
0.46
0.60
0.66
0.60
0.69
0.46
0.75
0.02
-
-
-
Mid-Europe
0.92
0.93
0.88
0.86
0.68
0.80
0.76
0.81
0.48
0.64
0.46
-
-
-
Scandinavia
0.92
0.83
0.47
0.91
0.59
0.62
0.24
0.81
0.31
0.55
0.65
-
-
-
Alps
0.92
0.86
0.46
0.83
0.68
0.80
0.47
0.77
0.52
0.25
0.52
-
-
-
Mediterranean
0.84
0.77
0.77
0.80
0.65
0.51
0.72
0.54
0.45
0.43
0.45
-
-
-
Eastern Europe
0.93
0.94
0.70
0.93
0.73
0.71
0.76
0.84
0.52
0.51
0.75
-
-
-
Europe
0.91
0.87
0.71
0.92
0.67
0.72
0.68
0.79
0.45
0.61
0.69
–
–
–
The new implementation and parametrisation of the within-tree and
within-stand allocation schemes were found to have a 91, 68 and 72 % chance
that the simulations will reproduce the observations for GPP, tree height and
basal area for Europe, respectively (Table ). Given that basal
area and height are not available from the trunk version of ORCHIDEE, we
could not compare the performance of model versions in this respect. With
respect to GPP, the ORCHIDEE-CAN branch was found to outperform the trunk by
12 % and thus increased the likelihood that ORCHIDEE-CAN is an unbiased
simulator of the spatial and temporal variability of GPP from 79 to 91 %.
Improved performance of the ORCHIDEE-CAN branch compared to the trunk is
observed for all regions in summer where the RMSE of GPP was halved from
2.5–5 to 1–2 gC m-2 day-1 (Figs. , and
).
Root mean square error of ORCHIDEE-CAN for gross primary production,
evapotranspiration, visible and near-infra-red albedo, effective leaf area
index, basal area and height for different regions and periods (DJF:
December–February, MAM: March–May, JJA: June–August, SON:
September–November). The gray-scale of the symbols indicates the number of
pixels included in the calculation. The transition from green to white
indicates an RMSE of 100 %.
Root mean square error of ORCHIDEE trunk for gross
primary production, evapotranspiration and visible and
near-infrared albedo for different regions and periods (DJF: December–February;
MAM: March–May; JJA: June–August; SON: September–November). The
grey scale of the symbols indicates the
number of pixels included in the calculation. The transition from green to
white indicates an RMSE of 100 %.
Comparison between observations and simulations of
ORCHIDEE-CAN for gross primary production and basal area over Europe. Gross
primary production represents the mean for June–August between 2001–2010
and basal area is the value at the end of 2010.
Although part of the high likelihood could be due to the fact that the
observed GPP was upscaled making use of similar climatologies being used as
the forcings of the models, this circularity could neither have contributed
to the improved performance between the trunk and the ORCHIDEE-CAN branch nor
to the decrease in RMSE. The improvements are thought to be due to structural
changes to the model such as allocation, hydraulic architecture and canopy
structure as well as to the use of more consistent parametrisation.
Plant water supply
Our implementation of plant hydraulic architecture was largely based on the
scheme of , which was tested globally and at site level.
Global simulation results for actual evapotranspiration were found to
reproduce available data . At the site
level, the model agreed well with the magnitude and seasonality of
eddy-covariance measurements of actual evapotranspiration for 15 European
forest sites (EUROFLUX), with a tendency to slightly overestimate actual
evapotranspiration for 6 sites .
The maximum amount of water that can be transported by a tree relies on the
hydraulic architecture of the tree and therefore on the capacity of the model
to simulate tree and stand dimensions as well as on the model's capacity to
simulate soil water content. As an additional test, our implementation of the
model was compared against the upscaled eddy-covariance measurements for GPP
and actual evapotranspiration . The capacity to jointly
reproduce GPP and actual evapotranspiration is an indicator that the model
successfully reproduces the coupling between CO2 and water exchange. Model
validation showed 91 and 87 % chance (compared to 79 and 45 % for the
trunk) that ORCHIDEE-CAN reproduces the upscaled GPP and actual
evapotranspiration data (Table , Fig. ). The RMSE
for actual evapotranspiration during summer dropped well below
1 mm day-1 for most regions (Fig. ), whereas it never dropped
below 1 mm day-1 for the trunk (Fig. ).
Canopy structure
The canopy structure model by was previously validated
against ground-based LIDAR data for several test sites with varying density,
structural complexity, layering and clumping .
Model-derived canopy gap probabilities compared with observations using a
one-sample t test were significant for 11 out of 12 test sites. We
considered this result to be a sufficient proof to use this canopy structure
model in the ORCHIDEE-CAN branch and added to its validation by comparing the
simulated canopy structure model over Europe against a remote-sensing-based
map of tree height and the JRC-TIP effective LAI product
. The effective LAI value expresses the capability of the
canopy to intercept direct radiation, and is thus associated with the
probability distribution function of the canopy gaps . Thus
the effective LAI contains information about the forest structure and leaf
distribution of the canopy. In the ORCHIDEE-CAN branch, canopy structure is
used to calculate the albedo, roughness length, absorbed light for
photosynthesis and leaf area that is coupled to the atmosphere for e.g.
transpiration and interception of precipitation.
The ORCHIDEE-CAN branch is the first branch of ORCHIDEE that makes use of an
effective LAI to calculate the interaction between the canopy and the
atmosphere. The LPI and RMSE of the branch, therefore, cannot be compared
against the trunk. Overall, the combined implementation of the allocation
scheme and the canopy structure model shows a 67 % chance to reproduce the
satellite-based estimates for effective LAI. Surprisingly, effective LAI is
better simulated in spring and autumn when dynamics within the canopy are
substantial due to leaf on-set and senescence. For the periods when the
effective LAI is expected to be most stable, i.e. summer and winter, LPI
approached and frequently exceeded 1 (data not shown). Part of this
shortcoming may be due to the lack of shrubs in the land cover
classification. In the model, shrublands are replaced by forest and/or
grasslands, likely resulting in differences between the observed and
simulated canopy structure. This lapse also appears in the RMSE of effective
LAI (RMSE higher than 0.8, Fig. )
Top of the canopy albedo
The radiation transfer model has been validated extensively
against realistic complex three-dimensional canopy scenarios
and as part of the RAdiation transfer Model Intercomparison
(RAMI) project. The 1-D canopy radiation transfer model by
was demonstrated to accurately simulate both the amplitude and the angular
variations of all radiant fluxes with respect to the solar zenith angle
. In addition, the radiation transfer model and its
effective values extracted from the JRC-TIP data set were successfully
applied to a single forest site .
Previously we reported on the capacity of the radiation transfer model to
simulate the effects of forest management on albedo . For the
latter, forest properties were prescribed and the radiation transfer model
was validated against top-of-the-canopy albedo data from five observational
sites. Differences in the spatial scales between the observed and simulated
albedo values were accounted for by presenting the mean June albedo during
2001–2010 . The simulated summertime canopy albedo falls
within the range of observation. However, there occurs a slight
overestimation in the near-infrared wavelength band compared to the single
site measurement. Overly high near-infrared single scattering albedo values
for pine, as obtained from the JRC-TIP product, are the most likely cause.
The observed deviation is not due to a shortcoming in the model itself, but
reflects the difficulties the JRC-TIP has with optimising parameter values in
the absence of field observations in the specific case of sparse canopies
.
For the spatial validation we use the white-sky albedo (VIS and NIR) from
Moderate Resolution Imaging Spectroradiometer MODIS, at
0.5∘ resolution (distributed in netCDF format by the Integrated
Climate Data Center (ICDC, http://icdc.zmaw.de) University of Hamburg,
Hamburg, Germany). Over large spatial and temporal domains the ORCHIDEE-CAN
branch reproduces the observed VIS and NIR albedo and its variability; LPI
for the albedo in the visible light is especially satisfying with a
likelihood of 92 % for the simulations to come from the same population as
the observations (Table ). This high overall performance index,
however, hides performance issues over Scandinavia and the Alps during the
snow season. The RMSE for VIS and NIR albedo without snow lies around 0.05,
whereas during the snow season the RMSE increases to 0.20 (VIS) and 0.18
(NIR) over these regions (Fig. ). When the ORCHIDEE-CAN branch is
coupled to an atmospheric model, however, these deviations will only have a
minor effect on the climate, owing to low incoming radiation during most of
the snow season, especially in Scandinavia.
Previous validation of the radiation transfer model showed that the largest
discrepancies were occurring in the near-infrared domain with a snow-covered
background . With the exception of the snow-covered season,
the new albedo scheme, which relies on the simulated canopy structure,
resulted in a substantial improvement of 0.05–0.15 compared to the trunk for
the RMSE in both the VIS and NIR range in Scandinavia and the Alps
(Figs. and ). The European LPI-based likelihood that our
model simulations come from the same populations as the MODIS albedo
increased by a remarkable 11 and 23 % for, respectively, NIR and VIS albedo
(from 61 and 69 % for the trunk to 72 and 92 % for the ORCHIDEE-CAN,
Table ).
Given that the parametrisation of the canopy radiation transfer model used in
ORCHIDEE-CAN relies on MODIS, the high likelihood may not come as a surprise.
However, our implementation of the radiation transfer model also relies on
the simulated absorbed light, simulated GPP, simulated allocation and
simulated canopy structure (which depends on mortality and forest
management). In the absence of all these processes our canopy radiation
transfer model is expected to reproduce the MODIS data with a probability of
100 %. Hence, the likelihood of 72 and 92 % (for NIR and VIS,
respectively) could also be interpreted as a verification of the
aforementioned calculations; all calculations that determine the canopy
structure reduce the reproducibility of the data by only 8–28 % (100 to 72
or 92 %).
Energy fluxes
The multi-layer scheme is in the process of a detailed evaluation across a
range of test conditions , and further validation across a
range of sites is ongoing. The scheme is able to produce within-canopy
temperature and humidity profiles, and successfully simulates the in-canopy
radiation distribution, as well as the separation of the canopy from the soil
surface. However, in order to preserve a measure of continuity with previous
evaluations of the model, the multi-layer solution is here set to
single-layer operation mode, which includes the effects of hydraulic
limitation (Sect. 3.2) and canopy structure (Sect. 3.3) on the energy budget.
The single-layer set-up of the multi-layer solution makes use of an improved
albedo estimation and is therefore expected to better simulate the net
radiation that needs to be redistributed in the canopy. This has been
confirmed at a single site with a sparse canopy .
Furthermore, the improvements in actual evapotranspiration in addition to the
low RMSE (Fig. ) are expected to be propagated in the performance of
the energy budget.
Forest management strategies
Model comparison has previously demonstrated that explicitly treating
thinning processes is essential to reproduce local and large-scale biomass
observations . This finding justifies the implementation of
generic approaches to forest management despite the difficulties associated
with defining and quantifying forest management and its intensity
. Although the use of so-called naturalness indices, in
which the current state of the forest in referenced against the potential
state of the forest, has been criticised because of difficulties in defining
the potential state of the forest , such approaches were
demonstrated to correctly rank different management strategies according to
their intensity .
Naturalness indices making use of only diameter and stand density or the
so-called relative density index (RDI) have been previously implemented at
the stand level as well as in large-scale models
. This approach was shown to successfully reproduce the
biomass changes during the life cycle of a forest . The implementation of a forestry model based on the relative
density index was reported to perform better than simple statistical models
for stand-level variables such as stand density, basal area, standing volume
and height . Although the performance of the model was
reported as less satisfying for tree-level variables, the approach is
nevertheless considered reliable for modelling the effects of forest
management on biomass stocks of forests across a range of scales from plot to
country .
Impact of the different forest management strategies
on an oak forest for unmanged (green), high stand (orange) and coppice (blue)
compared to a Poplar short rotation coppicing (red) at 48∘ N,
2∘ E. The simulation was run without spin-up to better visualise
carbon build-up in the coarse woody debris (C.W.D.) pool. Simulation cycled
of a single year (1990) of climate data to minimise the inter-annual
variability due to climatic year-to-year variability
In the absence of forest management, ORCHIDEE-CAN simulates that the stands
develop into tall canopy (Fig. a), with a high biomass
(Fig. b), a substantial dead wood and litter pool
(Fig. c) and no harvest (Fig. d). High stand
management reduces the height, standing biomass and litter pools
(Fig. a–c) but produces biomass for harvest
(Fig. d). Under coppicing, the reduction in forest age is
reflected in a shorter canopy and lower biomass and litter pools
(Fig. a–c) compared to high stand management. The harvest is
more evenly spread in time but falls below the harvest generated by high
stand management (Fig. d). Given the shorter rotations, canopy
height, standing biomass and litter pools are lower for short rotation
coppicing with poplar and willow compared to all other management strategies
applied on oak forest (Fig. a–c). Short rotation coppice was
harvested every 3 years resulting in a quasi-continuous supply of woody
biomass (Fig. d).
The forestry model implemented in ORCHIDEE-CAN is based on the RDI approach
by . We complemented earlier validation of such an
approach over France by a new European-wide validation
for basal area. On the European scale we verified the simulated basal area
and height against observed basal area from national forest inventories
and height from remote sensing . With an
RMSE of 3–7 for height and 7–15 for BA, and a chance of, respectively, 68
and 72 % to reproduce the data on the European scale
(Table ), our model is capable of correctly simulating the mean
height and basal area but fails to capture much of the spatial variability
(Fig. ; temporal variability was not considered because the
data products were only available for one time period).
Root mean square error (RMSE) of tree diameter for
different species (shown as different markers) for different regions over
France (shown as A to K). Open triangle, Pinus sylvestris; open
circle, Pinus pinaster; open square, Picea Sp.; filled
diamond, Quercus ilex/suber; filled triangle, Betula Sp.;
filled circle, Fagus sylvatica; filled square, Quercus robur/petraea.
Furthermore, we evaluated basal area and tree diameter at the species level
for 11 regions over France, which represents a finer spatial scale than
targeted by the model developments and their parametrisation. The data were
extracted from the French forest inventory between 2005 and 2010 and we used
the same simulations as for the European validation in the previous
paragraph. We selected pixels included in the French inventory data and for
both simulations and observations we calculated a moving average for the
diameter and basal area per age class to then calculated the RMSE
(Fig. ). To account for intrinsic species differences in
diameter and basal area, we normalised the RMSE. The normalised RMSE was
lower than 30 % of the mean tree diameter or mean basal area for each
region for Betula sp., Pinus pinaster and Quercus ilex. For Fagus sylvatica, Pinus sylvestris,
Picea sp. and Quercus robur/petraea the normalised RMSE of
diameter and basal area exceeded 50 % for one to four regions for tree
diameter and basal area (not shown).
The inability to fully capture the observed spatial variability in the
simulation could be due to the simulation protocol that started in 1850 with
2 to 3 m tall trees all over Europe. A longer simulation accounting for the
major historical changes in forest management such as the reforestation in
the 1700s following an all time low in the European forest cover, the start
of high stand management at the expense of coppicing in the early 1800s, and
the reforestation programs following World War II is
expected to improve the spatial variability in tree height and basal area.
Regional deviations such as those observed on the Iberian Peninsula or over
the entire Mediterranean (thus including part of the Iberian Peninsula) may
be due to the lack of shrubs in the land cover map and parametrisation of the
ORCHIDEE-CAN branch. Therefore the models simulates a higher stand density
and higher basal area for regions where in reality shrubs occur
(Fig. ).
The parametrisation of the forestry module strongly depends on the national
forest inventories from Spain, France, Germany and Sweden. Therefore
verification against the same data contains little information about the
model quality. Nevertheless, no time-dependent relationships were used in the
ORCHIDEE-CAN branch; thus the model's capacity to reproduce the relationship
between basal area and stand age, diameter and stand age or wood volume and
stand age could be considered a largely independent test of the model
quality. These tests were performed over eight bioclimatic regions of France
and the ORCHIDEE-CAN branch was found to largely capture the time
dependencies of basal area, diameter and wood volume (not shown).