simulating long-distance seed dispersal in a dynamic vegetation model
TRANSCRIPT
RESEARCHPAPER
Simulating long-distance seed dispersalin a dynamic vegetation modelRebecca S. Snell*†
Faculty of Forestry, University of Toronto,
Toronto, ON, Canada M5S 3B3
ABSTRACT
Aim Predicting the migration of vegetation in response to climate change is oftendone using a climate-driven vegetation model; however, the assumption of fullmigration (where seeds are not limited by distance or barriers) is a commoncriticism. Previous efforts to incorporate limitations on seed dispersal haveoccurred exclusively in bioclimatic envelope models. This paper describes howlimitations on seed dispersal were integrated into a physiologically based dynamicvegetation model, LPJ-GUESS.
Location An idealized landscape, representative of temperate and boreal forests inNorth America.
Methods LPJ-GUESS already simulates establishment, growth, reproduction andcompetition. I used a generic seed dispersal kernel to determine the probability ofdispersal between grid cells, and a logistic function to determine the spread betweenpatches within a grid cell. Plant functional types were parameterized to representthree temperate tree species, Acer, Pinus and Tsuga, by using published dispersalkernels and life-history measurements. Simulations were run with full and limitedmigration, and compared with past vegetation migration rates.
Results Using the old assumption of full migration, the entire landscape wascolonized at the same time (migration rates of 270–380 m year−1). With the newlimited dispersal, species colonized the landscape one row at a time, at rates whichcorresponded well with independent migration estimates based on genetic orpollen reconstructions (Acer, 141 m year−1; Pinus, 76 m year−1). Tsuga was the onlyspecies where simulated migration rates (85 m year−1) were quite a bit slower thanhistorical migration estimates.
Main conclusions The new model was able to simulate reasonable migrationrates, which is a substantial improvement over previous assumptions of full migra-tion. Migration estimates which include the effects of limitations on dispersal,demography,competition and plant physiology will also improve our understandingof how climate change and various other processes can influence plant range shifts.
KeywordsHemlock, Latin hypercube, long-distance seed dispersal, LPJ-DISP, maple,migration, pine, sensitivity analysis, simulation modelling.
*Correspondence: R. S. Snell, Faculty ofForestry, University of Toronto, Toronto, ON,Canada M5S 3B3.E-mail: [email protected]†Present address: Forest Ecology, Institute ofTerrestrial Ecosystems, Department ofEnvironmental Systems Science, ETH Zürich,8092 Zürich, Switzerland.
INTRODUCTION
As the amount of CO2 in the atmosphere continues to rise over
the next century, future climate change scenarios predict a rapid
shift in temperature and precipitation (IPCC, 2007). The north-
ern latitudes are likely to experience the most extreme change,
with a minimum warming of 5 °C and a 20% increase in pre-
cipitation (IPCC, 2007). One goal of climate change research is
to anticipate how these shifts in climate will affect the distribu-
tion and functioning of species (Thomas et al., 2004; Morin
et al., 2008; Doxford & Freckleton, 2012). Plants are of particu-
lar interest, not only because of the potential feedbacks between
bs_bs_banner
Global Ecology and Biogeography, (Global Ecol. Biogeogr.) (2013) ••, ••–••
© 2013 John Wiley & Sons Ltd DOI: 10.1111/geb.12106http://wileyonlinelibrary.com/journal/geb 1
climate and vegetation (Cramer et al., 2001; Purves & Pacala,
2008; Sitch et al., 2008) but also due to the unique challenges
plants face in being able to track climate change (Malcolm et al.,
2002; Midgley et al., 2006). Plants shift their distributions
through their offspring, relying on rare long-distance seed dis-
persal events and successful establishment in new habitats
(Pitelka et al., 1997). Predicting plant range shifts or plant
extinctions in response to climate change is an area where simu-
lation modelling can be a powerful tool.
Large-scale vegetation models traditionally assumed either no
migration or full migration, where plants have the ability to
migrate into any suitable habitat regardless of barriers or limi-
tations on seed dispersal (Cramer et al., 2001; Guisan & Thuiller,
2005). In reality, seed dispersal probably occurs somewhere in
between these two extremes. Efforts to correct this assumption
have occurred almost exclusively in species distribution models
(SDMs). SDMs correlate current species distribution with a
variety of climate and landscape variables (Guisan & Thuiller,
2005). Once this climatic niche has been established, future
climate scenarios can be used to test the potential shift in a
species’ distribution. Recent work has focused on incorporating
the effect of dispersal limitations and demography to improve
predictions of species ranges (e.g. Dullinger et al., 2012; Meier
et al., 2012; Pagel & Schurr, 2012). TreeMig, a process-based
forest landscape model, also includes competition between trees
(Lischke et al., 2006), which can influence the rate and success of
species migration. However, there has been very little progress in
adding dispersal limitations into the process-based, dynamic
global vegetation models (DGVMs).
DGVMs are physiologically based models which simulate
vegetation processes, hydrology and biogeochemical cycles in
response to climate change (Cramer et al., 2001). DGVMs are
often coupled with global climate models (GCMs) to simulate
the bi-directional feedback between biosphere and atmosphere;
climate-induced vegetation shifts can affect CO2 and water
exchange between land and air, which influence climate (Quillet
et al., 2010). However, DGVMs seldom impose any limitations
on migration (but see Sato & Ise, 2012). This assumption of full
migration could have significant impacts on predictions of
future climate change, and needs to be addressed within a
DGVM framework.
DGVMs already include many of the processes important for
migration of vegetation, such as establishment, carbon assimila-
tion, vegetation growth, reproduction and competition (Cramer
et al., 2001). LPJ-GUESS is a hybrid model (i.e. it incorporates a
forest gap model within a DGVM framework; Smith et al., 2001)
and is particularly suitable for simulating seed dispersal due to
the way it represents vegetation within a grid cell. Most DGVMs
simulate one average individual for each plant functional type
(PFT) within a grid cell. This means that PFTs arrive and establish
as one large individual that immediately travels the length of the
grid cell. LPJ-GUESS simulates a number of replicate patches
within each grid cell, where each patch contains several individ-
uals for each PFT at different ages. New PFTs could arrive and
establish in just one patch. PFTs would then be forced to disperse
between patches to cross a grid cell.
To improve our predictions on how climate change will affect
vegetation distributions, this paper describes how long-distance
seed dispersal was incorporated into the dynamic vegetation
model LPJ-GUESS. Seed dispersal kernels were used to predict
the probability of dispersing between grid cells, and limitations
were placed on patch-to-patch movements within a grid cell.
The goal of this study was to simulate plant migration based
on dispersal limitations alone, so variations in atmospheric
carbon, precipitation, soil and landscape heterogeneity were
not included. The success of the new dispersal module was
determined with a sensitivity analysis and by comparing simu-
lated migration rates with migration rates for three temperate
tree species reconstructed from pollen and genetic data.
METHODS
The LPJ-GUESS model
LPJ-GUESS is a generalized ecosystem model that combines the
dynamic global vegetation model LPJ with a forest gap model
(Smith et al., 2001). Carbon assimilation is calculated using
a modified Farquhar photosynthesis scheme (Haxeltine &
Prentice, 1996), which is influenced by temperature, atmos-
pheric CO2 concentration, absorbed photosynthetically active
radiation and stomatal conductance. At the end of a year, the
amount of carbon available for tree height and growth is
reduced by maintenance respiration, growth respiration, leaf
and root turnover and a fixed allocation to reproduction (Smith
et al., 2001).
LPJ-GUESS was run in cohort mode, where each grid cell
contains a number of replicate patches (400 in the present study,
which was the minimum number of patches required to simu-
late dispersal between grid cells). Since each age cohort has
different properties (i.e. height, leaf area index, biomass), the
model can successfully simulate intra- and interspecific compe-
tition for light, space and resources. The herbaceous layer is
simulated as one individual for each patch. All patches within a
grid cell have the same climatic and environmental properties.
The variability between patches results from stochastic processes
such as age-related mortality and disturbance.
Additional details on LPJ-GUESS can be found in Smith et al.
(2001), Sitch et al. (2003) and Gerten et al. (2004). To distin-
guish between LPJ-GUESS with full migration or with limited
migration, LPJ-DISP refers to the new version with limited
migration.
Communication between cells
The first step was to fundamentally change the way the program
runs, from isolated single cells to a two-dimensional landscape.
LPJ-GUESS simulates one grid cell at a time, discarding the grid
cell object after it had reached the total number of simulation
years. This makes it impossible to transfer seeds between neigh-
bouring grid cells as they don’t exist in memory at the same
time.
R. S. Snell
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd2
I chose to distinguish between a spin-up period and a migra-
tion period. During the spin-up period, the grid cells are simu-
lated independently with full migration capabilities for all PFTs.
This allows the PFTs which are in equilibrium with the current
climate to establish and form stable communities. However,
after the spin-up period, each grid cell (and all patch and veg-
etation information contained within it) is retained in memory.
During the migration period, each year is simulated across all
grid cells before moving forward to the next year. Each grid cell
accesses the vegetation composition for its neighbours from the
previous year.
Dispersal between grid cells
In this study, each grid cell is 0.166° or approximately 18 km.
Although most dispersal events would occur within a grid cell,
long-distance dispersal between cells will determine the migra-
tion of vegetation across a landscape. The number of seeds
arriving from distance x, is a product of the number of seeds
produced in neighbouring grid cells, seedn, and the probability
of those seeds travelling that distance, k(x),
seed seedx k xn( ) = ∗ ( ). (1)
Seed production (seedn)
Seed production in LPJ-GUESS is represented by the variable,
cmass_repr. This variable represents the total carbon allocated to
reproduction for each PFT from all the patches in a grid cell.
However, only those patches located close to the edge of the grid
cell are likely to have seeds disperse into neighbouring grid
cells. So seed production for each PFT within a grid cell
(pft.cmass_repr) was scaled by the proportion of patches close to
the edge of the grid cell (pPFT),
seed PFTn pft cmass repr p= ∗. _ . (2)
Number of patches close to the edge of thecell (pPFT)
Patches in LPJ-GUESS do not have locations within the grid cell,
making it impossible to identify the actual patches located close
to the edge. To generate a formula to describe the relationship
between patch number, distance and probability of containing
the PFT in question, it is assumed that the patches are randomly
located throughout the grid cell.
Using the spatial statistical package spatstat within the R
program (http://www.r-project.org), 100 random points were
generated with a Poisson distribution within an 18-km square.
Each point represents one patch, and the square represents a
grid cell in LPJ-GUESS. The 100 random points were generated
1000 times to determine that on average 0.0028% of the patches
are within 500 m from one of the edges. Given that long-
distance dispersal can occur over distances greater than 500 m
(Cain et al., 2000), the proportion of patches within 1, 2, 3, 4 and
5 km from the edge were also calculated.
Using 100 points assumes that the PFT of interest can be
found growing in 100% of the patches. Thus, random point
generation was repeated using 90 points (assuming that 90% of
the 100 patches contain the PFT), 80, 70, 60 . . . and so on. The
resulting formula is:
p x cell size x nPFT PFT( ) = ( )[ ]∗0 1. _ (3)
where x is distance, cell_size is the size of the grid cell in km and
nPFT is the number of patches within the grid cell that contain
the PFT in question. Combining equations 1–3 results in the
following formula:
seed PFTx pft cmass repr cell size x n k x( ) = ( )[ ]∗{ } ( ). _ . _ .0 1 (4)
Seed dispersal kernels (k(x))
The probability of a seed travelling a specific distance, x, is
known as the dispersal kernel. A generalized dispersal kernel has
been described by Clark (1998):
k xc
c
x c
( ) =( )
⎛⎝⎜
⎞⎠⎟ −⎛
⎝⎜⎞⎠⎟2 1α αΓ
exp (5)
where Γ() is the gamma function, c is a shape parameter and αis a distance parameter. By choosing different values for c and α,
the formula can be used to describe different kernels. For
example, c = 2.0 is the Gaussian kernel and c = 1.0 is the expo-
nential. Kernels with c < 1.0 are leptokurtic. A leptokurtic dis-
persal kernel describes a distribution which has a higher peak
around the mean (i.e. most seeds are clustered around the
parent tree) and a ‘fat tail’ which captures rare long-distance
seed dispersal.
Since different dispersal vectors operate on different spatial
scales, an increasing distance function was used, as opposed to a
predetermined value for distance. The minimum distance evalu-
ated was 500 m. Distance was increased in 500-m steps until the
dispersal probability became too small to consider.
Spread between patches within a grid cell
Just as there were no limitations to dispersal between cells, LPJ-
GUESS also has no restrictions on dispersal between patches
within a cell. Each grid cell has a common propagule pool which
every patch contributed seedlings to and took seedlings from.
Theoretically, a new PFT could travel all the way across the cell
(i.e. 18 km) in 1 year. Ideally, seeds coming in from neighbour-
ing cells should only arrive in the patches closest to the edge,
establish, spend a few years growing before reproducing, and
then disperse to nearby patches. It should take years for a new
PFT to travel all the way through the cell.
The first step was to eliminate the common propagule pool
within each grid cell. PFTs should only contribute seedlings to
Incorporating seed dispersal into a DVM
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd 3
their own patch and are unable to access seedlings from different
patches (except through dispersal).
The second step was to restrict where seeds coming in from
neighbouring cells land. Equation (3) was used to calculate the
number of patches located close enough to an edge to receive
seeds. These patches remain constant throughout the simulation
and are the only patches that can receive seeds from a neigh-
bouring grid cell (e.g. patch1, patch2 and patch3 will receive seeds
from the southern neighbour; patch4, patch5 and patch6 will
receive seeds from the eastern neighbour . . . and so on).
The third step was to add a new variable, age_repr. This
parameter prevents a PFT from allocating carbon to reproduc-
tion until it reaches a minimum age. Maturation age is a com-
monly measured parameter (Clark, 1998) and when used with
the already present variable reprfrac (the fraction of net primary
productivity allocated to reproduction), it can represent a spe-
cific life history for each PFT. For example, cherry birch trees
delay reproduction for many years but have a very high fecun-
dity once they do start producing seeds, compared with flower-
ing dogwood trees which start reproducing at a much younger
age but have lower fecundity (Clark, 1998). Adding a minimum
age for reproduction delays the rapid migration through the cell,
but only by the set number of years (i.e. the minimum age).
The final step was to limit patch-to-patch dispersal within a
cell. Again, this is more complicated since patches don’t have real
locations within the grid cell. However, knowing the proportion
of patches containing the PFT, we can calculate the probability
of having at least one neighbouring patch that also contains the
PFT. For example, if only one patch within the cell contains the
PFT then the probability of having a neighbouring patch
contain that PFT is very low. If more than 50% of the patches
contain the PFT, the probability of a patch having at least one
neighbour with the PFT is quite high. The logistic growth curve,
a relatively common function in ecology, can be used to describe
this relationship:
P tKP
K P
rt
rt( ) =
+ −( )0
0 1
e
e, (6)
where P is the population size at time t, r is the growth rate and
K is the carrying capacity or the largest size that the population
can reach given unlimited time. To suit my purpose, the formula
was modified as follows:
P pK
K
rp
rp( ) =
+ −( )e
e 1, (7)
where P is the population of patches available for receiving seeds
(i.e. those having at least one neighbouring patch that contains
the PFT) when there are p patches that contain reproducing
adults for that PFT. The carrying capacity, K, is the total number
of patches in one grid cell. The initial population size (P0) is
always set to 1 since this formula is only used if the PFT is
already present in the grid cell. The growth rate (r) was set to 0.1
and kept constant for all simulations (see Appendix S2 in Sup-
porting Information for a sensitivity analysis and discussion
about this parameter).
Using equation 7 in this way means that there is not neces-
sarily any growth each year, unlike in the traditional population
growth model. It is used to calculate the number of new patches
which have the potential to receive seeds from neighbouring
patches; however, seeds may not establish upon arrival due to
competition for space and resources. It may take several years for
a PFT to successfully establish in a new patch. The growth rate
(r) is not intended to represent any other processes which influ-
ence the success of seeds, such as dormancy, disease or seed
predation.
Simulation protocol
LPJ-DISP was tested using an imaginary landscape (eight rows
of ten grid cells across). The top five rows were assigned a boreal
climate (mean annual temperature 4.68 ± 0.45 °C) and the
bottom three rows were assigned a warmer, temperate climate
(mean annual temperature 15.19 ± 0.41 °C). The climate data
were extracted from actual boreal and temperate regions from
the Climatic Research Unit (CRU) global gridded data set (New
et al., 2002). The CRU data are composed of mean monthly
surface climate from 1961 to 1990, at a resolution of 0.166° or
18 km.
Three boreal PFTs (a shade-intolerant, intermediate shade-
tolerant and shade-tolerant tree), three temperate PFTs (a
shade-intolerant, intermediate shade-tolerant and shade-
tolerant tree) and one C3 grass were used (Appendix S1). Their
temperature ranges were modified to ensure complete separa-
tion between the temperate and boreal PFTs until after the
climate started to warm. This was done to make plant migration
easier to track.
The model was run for 1000 years with a stable climate and no
restrictions on dispersal (spin-up period). Over the next 200
years, the boreal climate was warmed by an average of
9.51 ± 0.35 °C and the temperate climate was warmed by an
average of 1.74 ± 0.06 °C. The temperature increases were
chosen based on the climatic tolerances for the temperate PFTs
(i.e. the boreal climate needed to warm by c. 10 °C to reach the
minimum temperature requirements for temperate PFTs to
establish and grow; Appendix S1). Although the degree of
warming in the boreal cells is more extreme than in future
climate change predictions (IPCC, 2007), the goal of this study
was to test migration of plants based on dispersal limitations
alone. If a more moderate warming was applied, dispersal and
climate would have been limiting factors. The model continued
to run at the new warmer temperatures for 1000–2000 years,
until the species had migrated through all the grid cells.
Test species
To test how well the model simulates dispersal, the temperate
PFT was parameterized to represent three different species: Acer
rubrum, Tsuga canadensis and Pinus rigida (Table 1). These tree
species were selected since they had published dispersal kernels
(Clark, 1998) for equation 5 and reconstructed migration rates
following the retreat of the last glacier in North America (Davis,
R. S. Snell
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd4
1981; Delcourt & Delcourt, 1987; McLachlan et al., 2005). The
published dispersal kernels were used to parameterize the model
and the reconstructed migration rates were used as an inde-
pendent comparison with the simulated migration rates. Since
individual species are not usually distinguished in the pollen
record, simulated rates were compared with the appropriate
genus (Table 2). The three species also represent different life-
history strategies (Table 1) which should have an effect on how
quickly they migrate across a landscape.
Sensitivity analysis
Performance of a sensitivity analysis on LPJ-DISP is important
to reduce model uncertainty, improve confidence in model pre-
dictions and identify areas that require further research. The
analysis identifies which input parameters have a significant
effect on the simulated migration rates. If the most influential
parameters are those which can be measured and known exactly,
this increases confidence in the results. A full description of the
methodology is provided in Appendix S2.
Calculating migration
LPJ-DISP was run for each of the test species twice, once with
the assumption of full migration and once with the new limited
dispersal described above. Once the species had migrated
through all the grid cells, the average migration rate (time to
move through each grid cell, 8 km) and the overall migration
rate (time to move the entire distance, 90 km) were calculated. A
species was considered to have entered a grid cell once a repro-
ducing adult could be found in at least one patch. The establish-
ment of saplings was not considered since other events (e.g.
competition, disturbance) might prevent that sapling from
reaching reproductive maturity. This is also comparable to how
migration is calculated in the pollen record (i.e. immature trees
don’t produce pollen). A species was considered to have crossed
a grid cell after mature trees were found in more than 80% of the
patches. Due to random patch-destroying events, individual
species almost never occupied 100% of the patches.
Migration rates for the first row of grid cells were much
slower since both climatic and dispersal limitations were in
effect. Thus, an average migration rate was calculated for the
first row (n = 10) and another average migration rate was cal-
culated for the next four rows (n = 40).
RESULTS
Sensitivity analysis
The sensitivity analysis demonstrated the range of migration
rates produced by LPJ-DISP by varying the parameters within
their published ranges. For example, the overall migration rate
ranged from 15–108 m year−1, the average migration rates
ranged from 16–49 m year−1 for the first row and from 37–162 m
year−1 for all subsequent rows. In general, migration rates were
the most sensitive to maturation age, as this parameter was
Table 1 Values used to parameterize the three test species (additional parameter values are listed in Appendix S1).
Maturation age αdisp reprfrac kest_repr Longevity Other
Acer rubrum 8 30.8 0.1 5000 80 Temperate, broadleaf, deciduous
Pinus rigida 12 15.1 0.1 1097 100 Temperate, needleleaf, evergreen
Tsuga canadensis 15 22.8 0.1 2317 500 Temperate, needleleaf, evergreen
αdisp, a distance parameter used in the dispersal kernel; reprfrac, the fraction of carbon allocated to reproduction; kest_repr, a constant in the equation forseed production.
Table 2 Simulated migration rates for each of the test species. The first row is the most southerly row, subsequent rows calculate themigration rate from all rows excluding the first. Shown is the average ± standard deviation (minimum – maximum values). Thereconstructed migration rates are either from pollen records (Davis, 1981; Delcourt & Delcourt, 1987) or phylogenetic methods(McLachlan et al., 2005).
Simulated migration rates (m year−1)
Reconstructed migration rates (m year−1)Average (first row) Average (subsequent rows) Overall
Acer rubrum 48 ± 8 (34–64) 141 ± 9 (122–162) 111 Acer rubrum 80–90 McLachlan et al. (2005)
Acer 126 (80–172) Delcourt & Delcourt (1987)
Acer 200 Davis (1981)
Pinus rigida 30 ± 4 (24–37) 76 ± 12 (55–101) 60 Pinus (northern) 135 (104–170) Delcourt & Delcourt (1987)
Pinus (southern) 81 (22–174)
Tsuga canadensis 47 ± 5 (40–53) 85 ± 16 (25–100) 79 Tsuga 202 (113–278) Delcourt & Delcourt (1987)
Tsuga 200–250 Davis (1981)
Incorporating seed dispersal into a DVM
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd 5
consistently ranked first under all the various sensitivity meas-
ures (Appendix S2). The second most important parameter was
usually r (equation 7). Although reproduction had only a weak
effect on migration rates (Appendix S2), it was always a negative
relationship (i.e. a higher proportion of carbon allocated to
reproduction resulted in slower migration).
Full migration
Using the traditional assumption of full migration, all five rows
were colonized at the same time (Fig. 1), just after the 200-year
warming period. This is the equivalent of moving 383 m year−1
for Acer, 273 m year−1 for Pinus and 360 m year−1 for Tsuga.
Limited migration
Simulations with limited dispersal gave more moderate overall
migration rates. The simulated migration rates for Acer (110 m
year−1) were consistent with phylogenetic reconstructed migra-
tion rates (80–90 m year−1; Table 2). Simulated migration rates
for Pinus (60 m year−1) were also consistent with reconstructed
rates from the southern pine group. However, simulated migra-
tion for Tsuga (80 m year−1) was considerably slower than the
pollen-reconstructed rates (Table 2).
Using a logistic curve to describe the within-cell filling of
patches produced a consistent pattern. The initial stage, where
the test species was maintained in only a few patches for a long
period of time, is followed by a rapid spreading stage, once the
species had established in c. 30% of the patches (Fig. 1). Stochas-
tic processes, disturbances and cell-to-cell climate differences
caused considerable variability between grid cells even within
the same row (Fig. 1). For example, there were c. 300 years
between the first and last grid cell to be colonized in the same
row for the Pinus simulation (Fig. 1b).
DISCUSSION
DGVMs include many of the processes which are lacking in
some of the correlative SDMs, such as competition, demography
and disturbances (e.g. Higgins et al., 2012). However, the
assumption of full migration has limited their application for
questions of species range shifts. That fact that LPJ-DISP was
able to simulate the movement of vegetation across a theoretical
landscape at a rate that is consistent with historical vegetation
migrations (Table 2) represents a substantial step forward.
Simulated migration rates were two to three times slower in the
first row (Table 2) when plants were limited by dispersal and as
well as climate. Palaeo reconstructed migration rates also show a
range of values (Delcourt & Delcourt, 1987), and it is also
thought that slower migration rates were due to climatic or
other environmental constraints and the fastest migration rates
occurred when the only limiting factor was dispersal.
Acer had the fastest average and overall migration rates. This
in not unexpected considering it has the youngest maturation
age, highest fecundity and largest distance parameter (Table 1).
The overall migration rate was reasonably close to both the
Figure 1 Simulated migration in LPJ-DISP for the three testspecies in an idealized landscape (five rows of 10 grid cells). Eachline represents one grid cell (black in print, different coloursonline where each colour represents one column). The thin greylines represent the simulation with the traditional assumption offull migration. The vertical dotted line indicates the end of thewarming period (i.e. climate is now suitable for species in all gridcells). The first 1000 years were spin-up and are not shown. Eachgrid cell was c. 18 km (0.166°). For graphing purposes, distancewas calculated as the proportion of filled patches within a grid.For example, if 30% of the patches were occupied in year 2150then it had moved 5.4 km through that grid cell.
R. S. Snell
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd6
pollen-reconstructed rate from Delcourt & Delcourt (1987) and
the genetic-reconstructed rate from McLachlan et al. (2005).
That the model was able to simulate migration rates similar to
reconstructed rates is a very exciting outcome, considering the
values used to parameterize the model are completely independ-
ent of the estimates of migration rates from palaeo and genetic
data.
Pinus migration was parameterized using data for Pinus
rigida, a northern pine; however, the simulated migration rates
were more similar to the southern pine group (Table 2). This is
probably an effect of the maturation age (identified as being the
most significant parameter for simulated migration rates in LPJ-
DISP; Appendix S2). The maturation age for Pinus rigida is 12
years (Table 1), which is the same as the average maturation age
for the southern pine group. Species in the northern pine group
had a younger average maturation age (9 years) which would
lead to faster migration rates.
Tsuga was the only test species where there was a large differ-
ence between the simulated and pollen reconstructed rates
(Table 2). Both Davis (1981) and Delcourt & Delcourt (1987)
estimated Tsuga trees to have migrated more than 200 m year−1,
which is more than twice as fast as the simulated rate. As the
model already includes long-distance seed dispersal, perhaps
this is a case of unidentified refugial populations. Recent
phylogeographic studies have been able to identify refugial
populations that persisted close to the edge of a retreating
glacier (Anderson et al., 2006; Hu et al., 2009). Refugial popu-
lations are too small to appear in the pollen record, but factoring
in their existence drastically lowers our estimates of historical
migration rates (Anderson et al., 2006). Of the three test species,
migration rates based on genetic evidence were only available
for Acer rubrum (McLachlan et al., 2005), which matched the
simulated rates quite well (Table 2).
One way to achieve rapid migration rates similar to those
observed in the pollen record is by increasing the proportion of
long-distance seeds (Clark, 1998). For all three species, I used a
‘fat-tailed’ dispersal kernel (equation 5; c = 0.5) with different
values for α (Table 1). This ensured that there was some long-
distance dispersal, but didn’t specify how much. To reproduce
the rapid migration rates observed in the palaeo-record, Clark
(1998) allocated up to 10% of the seeds to the tail. In reality, such
large amounts of long-distance dispersal are highly unlikely.
Dispersal within a grid cell
The use of a logistic curve may be a novel approach for describ-
ing seed dispersal between patches, but such curves have long
been used in epidemiology to describe the spread of diseases
(Berger, 1981). In spite of an unconventional spread pattern
within an individual grid cell (Fig. 1), a logistic curve does
produce migration rates at a landscape scale that are consistent
with historical vegetation migration rates (Table 2). The size of
the grid cell does influence the size of the ‘jump’ (Appendix S3).
In order to maintain the relationship across scales, the growth
rate (r) needs to be adjusted. For example, reducing the grid cell
resolution to 9 km2 requires a doubling of r to maintain the same
migration rate (Appendix S3).
It is not surprising that simulated migration rates were sen-
sitive to r, the parameter which describes the rate of within-cell
filling (equation 7). Although it would be ideal to choose a value
for r that reflects different dispersal vectors (i.e. wind, mammal,
bird), I suggest keeping this value constant until additional
research has been done. This ensures that the simulated migra-
tion rates are based on known life-history parameters which
reduces model uncertainty.
Wind versus animal dispersal
The three species in this study all use wind as their primary
method for seed dispersal (Clark et al., 1999). This is interesting
since the assumption of full migration has been argued to be a
reasonable representation for wind-dispersed trees (Higgins &
Richardson, 1999). However, this study clearly shows that full
migration is not the same as long-distance dispersal, even for
wind dispersal (Fig. 1).
To date, most of the effort in developing a generic model for
seed dispersal has focused heavily on wind-dispersed seeds (e.g.
Govindarajan et al., 2007; Nathan et al., 2011) as wind dispersal
is a common strategy in boreal and temperate trees. Alterna-
tively, animal-dispersed seed shadows are complex and rely
heavily on the behaviour of the dispersers (Gomez, 2003). A
mechanistic model for animal dispersal needs to account for
many additional factors such as animal movement, seed detach-
ment, gut throughput, landscape heterogeneity, interactions
with predators or competing species and the spatial distribution
and abundance of food sources (Gomez, 2003; Levey et al., 2008;
Cousens et al., 2010). Modelling animal dispersal at a large scale
may work really well, for several reasons. First, animals have the
potential for frequent and significant long-distance dispersal
(Levey et al., 2008; Campos-Arceiz & Blake, 2011). Second, most
of the complexity when simulating local or daily movements can
effectively be ignored since many of these processes would occur
within a grid cell. Using the framework outlined in this paper,
the only information that would be needed to simulated disper-
sal between grid cells is the amount of long-distance dispersal.
Perhaps studies that track animal movement and seed dispersal
between patches (Gomez, 2003; Levey et al., 2008) could be used
to describe patch-to-patch dispersal within a grid cell. Being able
to simulate range shifts in response to climate change for plants
with all modes of dispersal would be bring an enormous
improvement to our future climate change scenarios.
Life-history strategies and climate change
The sensitivity analysis identified age at maturation as the most
influential parameter for simulating migration rates in LPJ-
DISP. Maturation age is an important component in the life-
history strategies of trees (Loehle, 1988; Clark, 1991) and can
affect biodiversity, community composition and population
spread (Clark & Ji, 1995; Nathan et al., 2011). Delaying repro-
duction to invest in growth is a common strategy for trees in
Incorporating seed dispersal into a DVM
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd 7
stable environments. However, if trees delay reproduction for
too long they may be unable to respond quickly to climate
change and migrate to new locations.
The sensitivity analysis also illustrated another potential
trade-off for trees, namely the cost of reproduction (Obeso,
2002). A higher reproductive effort (represented in the model by
reprfrac) resulted in slower migration rates. One might have
expected that producing more seeds would result in faster
migration rates. However, allocating more carbon to reproduc-
tion means there is less carbon available for growth which puts
trees at a disadvantage when competing for light and resources.
In the light of future climate change and the importance of
being able to predict vegetation migration, it is interesting that
many life-history parameters can be influenced by environmen-
tal stress and elevated CO2 levels. Red mangrove trees started
reproducing at a much younger age when exposed to elevated
CO2, a full 2 years before seedlings typically start reproducing
(Farnsworth et al., 1996). Pinus taeda trees growing under
elevated CO2 levels also reached reproductive maturity at
younger ages and smaller sizes and produced more cones
(Ladeau & Clark, 2006). It is important to recognize that climate
change may cause modifications to life-history parameters
which can influence migration rates.
Competition between vegetation types
A strength of LPJ-DISP is that plants participate in intra- and
interspecific competition, a common deficiency in SDMs
(Hampe, 2004). Although the climate was suitable, the progres-
sion of the warm PFTs was slower since they were competing
with the existing cold-tolerant PFTs for space. Even at the end of
the Acer simulation (year 2199), 12 of the 50 grid cells still
contained at least one patch with the cold-tolerant PFT (results
not shown).
However, a limitation of this model is that incoming vegeta-
tion does not compete with other migrating species (i.e. other
warm PFTs). The warm shade-intolerant PFTs were unable to
migrate since they could never establish in the first row. First,
they were out-competed by the existing cool shade-tolerant
PFTs. Then, Acer/Pinus/Tsuga replaced the cool forest and con-
tinued to block out the shade-intolerant PFTs. The pollen record
has multiple examples of shade-intolerant species, such as larch
and jack pine, migrating after glacial retreat (Davis, 1981;
Delcourt & Delcourt, 1987). To simulate the movement of these
species, we would need to modify the landscape to include
natural open spaces, or increase the frequency and size of dis-
turbance events, creating more gaps in the forest and more
opportunities for shade-intolerant species to have a competitive
edge.
CONCLUSIONS
The lack of seed dispersal is a major criticism in all vegetation–
climate models. This paper presents the first time seed dispersal
has been included in a physiological dynamic vegetation–
climate model, at spatial and temporal scales that are suitable for
simulating vegetation migration. Although it can be difficult to
know for certain what vegetation migration rates were histori-
cally, the simulated migration rates can be considered reason-
able estimates and represent a significant improvement over
previous assumptions of full migration. The sensitivity analysis
also demonstrated the importance of a few key parameters,
including age at maturation and within-cell spread rates. As age
at maturation is a parameter which can easily be measured,
future work should focus on improving the parameterization of
within-cell spread (a methodological approach for choosing a
value for r). One suggestion would be to choose a value such that
the simulated migration rates correspond to independent
migration estimates for each species, if we have reason to believe
the independent migration estimates are correct. The source of
these independent estimates could be from pollen records,
phylogenetic data or from an independent, high-resolution dis-
persal model which operates over a smaller spatial extent. Ulti-
mately, adding migration limitations to dynamic vegetation–
climate models will help develop more realistic expectations for
how vegetation will respond to future climate change.
ACKNOWLEDGEMENTS
I thank my PhD supervisor S. A. Cowling for her continuous
support. B. Smith kindly allowed me to use LPJ-GUESS. J. Nabel,
I. C. Prentice and one anonymous referee provided valuable
feedback on the manuscript. Funding for R.S.S. was provided by
OGS and NSERC graduate scholarships.
REFERENCES
Anderson, L.L., Hu, F.S., Nelson, D.M., Petit, R.J. & Paige, K.N.
(2006) Ice-age endurance: DNA evidence of a white spruce
refugium in Alaska. Proceedings of the National Academy of
Sciences USA, 103, 12447–12450.
Berger, R.D. (1981) Comparison of the Gompertz and logistic
equations to describe plant-disease progress. Phytopathology,
71, 716–719.
Cain, M.L., Milligan, B.G. & Strand, A.E. (2000) Long-distance
seed dispersal in plant populations. American Journal of
Botany, 87, 1217–1227.
Campos-Arceiz, A. & Blake, S. (2011) Megagardeners of the
forest – the role of elephants in seed dispersal. Acta Oecologica,
37, 542–553.
Clark, J.S. (1991) Disturbance and tree life-history on the shift-
ing mosaic landscape. Ecology, 72, 1102–1118.
Clark, J.S. (1998) Why trees migrate so fast: confronting theory
with dispersal biology and the paleorecord. The American
Naturalist, 152, 204–224.
Clark, J.S. & Ji, Y. (1995) Fecundity and dispersal in plant-
populations – implications for structure and diversity. The
American Naturalist, 146, 72–111.
Clark, J.S., Silman, M., Kern, R., Macklin, E. & HilleRisLambers,
J. (1999) Seed dispersal near and far: patterns across temper-
ate and tropical forests. Ecology, 80, 1475–1494.
R. S. Snell
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd8
Cousens, R.D., Hill, J., French, K. & Bishop, I.D. (2010) Towards
better prediction of seed dispersal by animals. Functional
Ecology, 24, 1163–1170.
Cramer, W., Bondeau, A., Woodward, F.I., Prentice, I.C., Betts,
R.A., Brovkin, V., Cox, P.M., Fisher, V., Foley, J.A., Friend,
A.D., Kucharik, C., Lomas, M.R., Ramankutty, N., Sitch, S.,
Smith, B., White, A. & Young-Molling, C. (2001) Global
response of terrestrial ecosystem structure and function to
CO2 and climate change: results from six dynamic global veg-
etation models. Global Change Biology, 7, 357–373.
Davis, M.B. (1981) Quaternary history and the stability of forest
communities. Forest succession: concepts and application (ed.
by D.C. West, H.H. Shugart and D.B. Botkin), pp. 132–153.
Springer-Verlag, New York.
Delcourt, P.A. & Delcourt, H.R. (1987) Long-term forest dynam-
ics of the temperate zone. Springer-Verlag, New York.
Doxford, S.W. & Freckleton, R.P. (2012) Changes in the large-
scale distribution of plants: extinction, colonisation and the
effects of climate. Journal of Ecology, 100, 519–529.
Dullinger, S., Gattringer, A., Thuiller, W. et al. (2012) Extinction
debt of high-mountain plants under twenty-first-century
climate change. Nature Climate Change, 2, 619–622.
Farnsworth, E.J., Ellison, A.M. & Gong, W.K. (1996) Elevated
CO2 alters anatomy, physiology, growth, and reproduction of
red mangrove (Rhizophora mangle L). Oecologia, 108, 599–
609.
Gerten, D., Schaphoff, S., Haberlandt, U., Lucht, W. & Sitch, S.
(2004) Terrestrial vegetation and water balance – hydrological
evaluation of a dynamic global vegetation model. Journal of
Hydrology, 286, 249–270.
Gomez, J.M. (2003) Spatial patterns in long-distance dispersal of
Quercus ilex acorns by jays in a heterogeneous landscape.
Ecography, 26, 573–584.
Govindarajan, S., Dietze, M.C., Agarwal, P.K. & Clark, J.S. (2007)
A scalable algorithm for dispersing population. Journal of
Intelligent Information Systems, 29, 39–61.
Guisan, A. & Thuiller, W. (2005) Predicting species distribution:
offering more than simple habitat models. Ecology Letters, 8,
993–1009.
Hampe, A. (2004) Bioclimatic envelope models: what they
detect and what they hide. Global Ecology and Biogeography,
13, 469–471.
Haxeltine, A. & Prentice, I.C. (1996) BIOME3: an equilibrium
terrestrial biosphere model based on ecophysiological
constraints, resource availability, and competition among
plant functional types. Global Biogeochemical Cycles, 10, 693–
709.
Higgins, S.I. & Richardson, D.M. (1999) Predicting plant migra-
tion rates in a changing world: the role of long-distance dis-
persal. The American Naturalist, 153, 464–475.
Higgins, S.I., O’Hara, R.B. & Romermann, C. (2012) A niche for
biology in species distribution models. Journal of Biogeogra-
phy, 39, 2091–2095.
Hu, F.S., Hampe, A. & Petit, R.J. (2009) Paleoecology meets
genetics: deciphering past vegetational dynamics. Frontiers in
Ecology and the Environment, 7, 371–379.
IPCC (2007) Climate change 2007: synthesis report. Contribution
of Working Groups I, II and III to the Fourth Assessment Report
of the Intergovernmental Panel on Climate Change (ed. by
R.K. Pachauri and A. Reisinger), p. 104. IPCC, Geneva,
Switzerland.
Ladeau, S.L. & Clark, J.S. (2006) Elevated CO2 and tree fecun-
dity: the role of tree size, interannual variability, and popula-
tion heterogeneity. Global Change Biology, 12, 822–833.
Levey, D.J., Tewksbury, J.J. & Bolker, B.M. (2008) Modelling
long-distance seed dispersal in heterogeneous landscapes.
Journal of Ecology, 96, 599–608.
Lischke, H., Zimmermann, N.E., Bolliger, J., Rickebusch, S. &
Loffler, T.J. (2006) TreeMig: A forest-landscape model for
simulating spatio-temporal patterns from stand to landscape
scale. Ecological Modelling, 199, 409–420.
Loehle, C. (1988) Tree life-history strategies – the role of
defenses. Canadian Journal of Forest Research–Revue
Canadienne de Recherche Forestiere, 18, 209–222.
McLachlan, J.S., Clark, J.S. & Manos, P.S. (2005) Molecular indi-
cators of tree migration capacity under rapid climate change.
Ecology, 86, 2088–2098.
Malcolm, J.R., Markham, A., Neilson, R.P. & Garaci, M. (2002)
Estimated migration rates under scenarios of global climate
change. Journal of Biogeography, 29, 835–849.
Meier, E.S., Lischke, H., Schmatz, D.R. & Zimmermann, N.E.
(2012) Climate, competition and connectivity affect future
migration and ranges of European trees. Global Ecology and
Biogeography, 21, 164–178.
Midgley, G.F., Hughes, G.O., Thuiller, W. & Rebelo, A.G. (2006)
Migration rate limitations on climate change-induced range
shifts in Cape Proteaceae. Diversity and Distributions, 12, 555–
562.
Morin, X., Viner, D. & Chuine, I. (2008) Tree species range shifts
at a continental scale: new predictive insights from a process-
based model. Journal of Ecology, 96, 784–794.
Nathan, R., Horvitz, N., He, Y.P., Kuparinen, A., Schurr, F.M. &
Katul, G.G. (2011) Spread of North American wind-dispersed
trees in future environments. Ecology Letters, 14, 211–219.
New, M., Lister, D., Hulme, M. & Makin, I. (2002) A high-
resolution data set of surface climate over global land areas.
Climate Research, 21, 1–25.
Obeso, J.R. (2002) The costs of reproduction in plants. New
Phytologist, 155, 321–348.
Pagel, J. & Schurr, F.M. (2012) Forecasting species ranges by
statistical estimation of ecological niches and spatial popula-
tion dynamics. Global Ecology and Biogeography, 21, 293–304.
Pitelka, L.F., Gardner, R.H., Ash, J. et al. (1997) Plant migration
and climate change. American Scientist, 85, 464–473.
Purves, D. & Pacala, S. (2008) Predictive models of forest
dynamics. Science, 320, 1452–1453.
Quillet, A., Peng, C.H. & Garneau, M. (2010) Toward dynamic
global vegetation models for simulating vegetation–climate
interactions and feedbacks: recent developments, limitations,
and future challenges. Environmental Reviews, 18, 333–353.
Sato, H. & Ise, T. (2012) Effect of plant dynamic processes on
African vegetation responses to climate change: analysis using
Incorporating seed dispersal into a DVM
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd 9
the spatially explicit individual-based dynamic global vegeta-
tion model (SEIB-DGVM). Journal of Geophysical Research–
Biogeosciences, 117, G03017. doi:10.1029/2012JG002056
Sitch, S., Smith, B., Prentice, I.C., Arneth, A., Bondeau, A.,
Cramer, W., Kaplans, J.O., Levis, S., Lucht, W., Sykes, M.T.,
Thonicke, K. & Venevsky, S. (2003) Evaluation of ecosystem
dynamics, plant geography and terrestrial carbon cycling in
the LPJ dynamic global vegetation model. Global Change
Biology, 9, 161–185.
Sitch, S., Huntingford, C., Gedney, N., Levy, P.E., Lomas, M.,
Piao, S.L., Betts, R., Ciais, P., Cox, P., Friedlingstein, P., Jones,
C.D., Prentice, I.C. & Woodward, F.I. (2008) Evaluation of the
terrestrial carbon cycle, future plant geography and climate–
carbon cycle feedbacks using five dynamic global vegetation
models (DGVMs). Global Change Biology, 14, 2015–2039.
Smith, B., Prentice, I.C. & Sykes, M.T. (2001) Representation of
vegetation dynamics in the modelling of terrestrial ecosys-
tems: comparing two contrasting approaches within Euro-
pean climate space. Global Ecology and Biogeography, 10, 621–
637.
Thomas, C.D., Cameron, A., Green, R.E., Bakkenes, M.,
Beaumont, L.J., Collingham, Y.C., Erasmus, B.F.N., de
Siqueira, M.F., Grainger, A., Hannah, L., Hughes, L., Huntley,
B., van Jaarsveld, A.S., Midgley, G.F., Miles, L., Ortega-Huerta,
M.A., Peterson, A.T., Phillips, O. & Williams, S.E. (2004)
Extinction risk from climate change. Nature, 427, 145–148.
SUPPORTING INFORMATION
Additional supporting information may be found in the online
version of this article at the publisher’s web-site.
Appendix S1 Parameter definitions and values for each PFT.
Appendix S2 Sensitivity analysis for new dispersal parameters.
Appendix S3 Simulations with a smaller grid cell size.
BIOSKETCH
Rebecca Snell is a post-doctoral researcher in the
Forest Ecology group at ETH Zurich. This work was
part of her PhD dissertation at the University of
Toronto. She is interested in understanding
landscape-level distribution patterns and the processes
that influence species composition, structure and
dynamics.
Editor: Bill Shipley
R. S. Snell
Global Ecology and Biogeography, ••, ••–••, © 2013 John Wiley & Sons Ltd10