3. Forest modelling as tools to address key ecological
questions
Forest models from different communities have been following converging
trajectories of development, leading to a generation of models capable
of addressing similar topics and taking on an increasing role to address
novel ecological questions beyond their traditional focus. We detected
different ecological fields for which we expect forest modelling to make
important contributions in the next decade, by increasing our
understanding of forest ecosystems and helping generalize ecological
findings. To illustrate this, we now provide examples of recent model
applications to these topics, and suggest 10 important questions for
future studies (Table 2).
Carbon stocks and fluxes
Quantifying forest carbon stocks and fluxes is an important task, in
particular to inform climate change mitigation policies such as REDD
(Gibbs et al. 2007). However, substantial uncertainties remain in
estimated carbon and other element stocks and fluxes associated with
forest locally and worldwide (Bonan 2008; Pan et al. 2011; Réjou-Méchain
et al. 2019). Their quantification has motivated large efforts of data
collection (Table 1), including labor-intensive forest inventories
(Brienen et al. 2015), flux measurements (Falge et al. 2002), or
remote-sensing (Running et al. 2004; Saatchi et al. 2011). Forest models
provide a framework to connect empirical data of various nature, and
this connection is even more powerful as models adopt resolutions that
match with a broader range of empirical data, such as individual-based
modelling approaches, including individual-based DGVMs (Smith et al.
2001; Sakschewski et al. 2015; Rödig et al. 2017; Fisher et al. 2018).
Models have thus been used to upscale and infer dynamic estimates of
forest productivity and biomass (e.g. Fischer et al. 2015) using
allometries derived from field measurements (Chave et al. 2005, 2014).
Recently, assimilation of remote-sensing data within forest models has
allowed to account for the heterogeneity in forest structure and
land-use history in those estimates at stand to continental scales
(Joetzjer et al. 2017; Rödig et al. 2017, 2018). Beyond estimations of
carbon stocks and fluxes, forest models can be used to understand the
drivers of their spatial variation. For example, through simulation
experiments using an IBM, Fyllas et al. (2017) showed that solar
radiation and trait variation driven by spatial species turnover explain
the decline of forest productivity along a tropical elevation gradient.
Similarly, using a forest demographic model, Berzaghi et al. (2019)
showed that elephant disturbances enhance carbon stocks in central
African forests through their effects on forest structure and
composition. Models can also prove useful to create benchmarks against
which methods to estimate carbon stocks and fluxes from measurements can
be evaluated and improved (e.g. LiDAR, Knapp et al. 2018; eddy-flux
tower, Jung et al. 2009).
Tree mortality and carbon allocation are key drivers of forest
productivity and biomass (Bugmann & Bigler 2011; Johnson et al. 2016)
but remain poorly understood processes (Holzwarth et al. 2013; Malhi et
al. 2015; Hartmann et al. 2018b; Merganičová et al. 2019), and future
modelling studies should seek to foster our understanding of these
critical processes through model-data fusion approaches (Q9, Q10, Table
2).
Biodiversity and ecosystem functioning
Understanding the link between biodiversity and ecosystem functioning is
of high interest in the context of global biodiversity loss (Naeem et
al. 2009). Long-term effects remain underexplored, and underlying
mechanisms are still under debate (Loreau et al. 2001; Scherer-Lorenzen
2014). By virtually manipulating the composition of simulated forest
communities, forest IBMs have proven useful in exploring the effect of
species richness and functional composition on ecosystem properties
(e.g. Fischer et al. 2018). Simulations reproduced positive
relationships between (species or functional) diversity and productivity
or biomass, in agreement with the few observed patterns (Morin et al.
2011; Maréchaux & Chave 2017), further motivating a finer-grained
representation of diversity in DGVMs. These studies demonstrated how
competition for light can induce this positive effect in heterogeneous
forests. Going beyond the effect of bulk species richness, Bohn & Huth
(2017) showed that this positive effect is stronger if species are well
distributed across the forest canopy vertical gradient. García-Valdés et
al. (2018) showed that climate change-driven extinctions of tree species
may affect forest productivity or biomass more severely than random
extinctions. Schmitt et al. (2019) found that the mechanisms through
which biodiversity influences forest functioning depend on the ecosystem
state, shifting from a dominant complementary effect in recently
disturbed systems to a selection effect in anciently disturbed systems,
suggesting a way to reconcile contrasting results obtained with
snapshots of ecosystem state in empirical studies.
A more detailed modelling investigation of the effect of tree species
diversity and species loss on other forest ecosystem functions (e.g.
water and nutrient cycles) should follow in the near future (Q1, Q3,
Table 2). Another potential field of model exploration considers the
influence of species diversity on crown- and surface-fire intensity as
recently investigated empirically for the boreal zone (Rogers et al.
2015). Forest models, including flexible-trait DGVMs (Scheiter et al.
2013; Sakschewski et al. 2015), can further investigate how functional
diversity supports forest productivity and carbon storage under climate
change, from local- to biome-scale.
Resilience and stability
Forest responses to perturbations can be complex and non-linear (Ives &
Carpenter 2007; Felton & Smith 2017), but their understanding is
critical in an epoch of global change, including changes in intensity
and frequency of climate extremes (Field et al. 2012; Reichstein et al.
2013) and disturbances (Seidl et al. 2017). Forest models can help to
disentangle the different mechanisms shaping forest responses to
perturbations through virtual experiments unreachable by empirical
approaches. Simulations using a individual-based and trait-based DGVM
showed that a higher trait diversity increases the resilience of the
Amazon rainforest under future climate (Sakschewski et al. 2016). This
positive effect was attributable to ecological sorting, in agreement
with results from forest IBMs in temperate (Morin et al. 2018) or
tropical (Schmitt et al. 2019) forests. Higher temporal stability of
productivity for forests with higher diversity was also attributed to
the asynchrony of species responses to small disturbances (Morin et al.
2014). Using a multimodel analysis, Radchuk et al. (2019) showed that
the multiple properties of stability, such as resistance, recovery or
persistence (Donohue et al. 2013) can vary independently depending on
the disturbance type.
However, we still have an insufficient understanding of forest ecosystem
stability (Donohue et al. 2016), and future modelling studies should
help disentangling the multiple drivers of forest resilience while
paying attention to the elements leading to feedbacks (e.g. the adult –
regeneration feedback). This will foster our predictive ability of
potential critical transitions (Q5, Q6, Q10, Table 2).
Community assembly
Understanding the drivers of community assembly, i.e. the processes that
shape the number, identity and abundance of co-occurring species, has
been an important question in ecology since its inception (Clements
1916; Gleason 1926; MacArthur & Levins 1967; McGill et al. 2006).
Forest models allow to separate the effect of different drivers through
the use of null models and sequential simulation set-ups. For instance,
forest IBMs have been recently used to investigate the role of
trait-mediated trade-offs and their size dependency in shaping forest
community (Kunstler et al. 2009; Chauvet et al. 2017; Falster et al.
2017). In doing so, they used a more realistic modelling framework than
most theoretical investigations generally developed to address these
questions and typically restricted to systems with few species. This
approach may be further developed and applied to various forest
communities as trait data is being increasingly available. Modelling
also helps to disentangle the contribution of stochastic vs.
deterministic processes through the assessment of variability among
repeated runs (Savage et al. 2000).
Although many mechanisms have been empirically detected to contribute to
species coexistence in forest communities (Nakashizuka 2001; Wright
2002), their relative strengths in observed communities across
environmental gradients remain poorly known. Forest modelling could help
quantifying their relative contributions through a combination of simple
theoretical models and data-driven simulation experiments, and exploring
the debated role of intraspecific variability on species coexistence
(Lischke & Löffler 2006; Hart et al. 2016; Q2, Q4, Table 2). To do so,
models need to include key aspects of community assembly or known
coexistence mechanisms, such as regeneration processes (Vacchiano et al.
2018), negative density-dependence (Lischke & Löffler 2006; Maréchaux
& Chave 2017), or functional trade-offs (Sakschewski et al. 2015) in a
heterogeneous environment.
Biodiversity conservation
Conservation efforts have so far not been successful to alleviate
biodiversity loss across the globe (Butchart et al. 2010), calling for
renewed effort and biodiversity forecasts (Urban et al. 2016). As SDMs
could be calibrated for almost all species for which reliable
distribution data are available, these models have long been identified
as tools for conservation (Davis & Zabinski 1992; Guisan et al. 2013;
Araújo et al. 2019). Predictions of SDMs under climate change scenarios
could be used to help refine conservation areas (Ferrier 2002), or
predict invasion ranges of introduced species (Thuiller et al. 2005;
Broennimann et al. 2007). Although this claim is still put forward very
often (Fernandes et al. 2018), case-studies reporting applications
remain sparse (Mouquet et al. 2015), likely because of the uncertainty
of SDMs predictions (Barry & Elith 2006; Dawson et al. 2011; Journé et
al. 2019).
Mixed predictions carried out jointly with different model types
(process-based or hybrid distribution models, Morin & Thuiller 2009;
Evans et al. 2015; Box 2) could make more robust projections available
to conservation managers (e.g. Thom et al. 2017). Such an approach
appears especially feasible for tree species, as individual- and
process-based models are typically more available for forests than for
other ecosystems. Therefore, DGVMs and gap models should be encouraged
to address the challenges of biodiversity conservation planning (e.g.
Fischer et al. 2016), in complement of species-level process-based
models already available (e.g. Chuine & Beaubien 2001; Keenan et al.
2011; Serra‐Diaz et al. 2013; Q1, Q4, Q8, Table 2).
Forest responses to global
change
Ongoing climate change has already altered forest functioning globally
(Nemani et al. 2003; Allen et al. 2010). Models represent a key tool to
assess forest responses to the interacting factors of future climate
change (Sabaté et al. 2002; Medlyn et al. 2011; Bugmann 2014).
Simulating the dynamics of vegetation, including forests, under climate
change is the main objective of DGVMs, and has been the focus of a
sustained effort from this modelling community (Mohren et al. 1997;
Jarvis 1998; Cramer et al. 2001; Alo & Wang 2008; Keenan et al. 2008;
Friend et al. 2014). However, stand-scale models, such as
individual-based gap models, have also been used to explore forest
dynamics under climate change scenarios (Pastor & Post 1986; Bugmann &
Fischlin 1996; Fischer et al. 2014; Reyer 2015; Collalti et al. 2018;
Shugart et al. 2018). Such finer-scale models can further inform the
role of forest composition and structure in shaping forest responses to
environmental drivers (Fyllas et al. 2017; Bohn et al. 2018).
Additionally, SDMs have been used to project species distributions under
future climate change (Thuiller 2004; Noce et al. 2017), although, as
mentioned above, their correlative nature has raised some criticisms
regarding their application to forecasting under no-present analogues.
Overall, a variety of models are utilized to simulate forest responses
to climate change, allowing comparisons of different approaches and
assessment of model uncertainties (Cheaib et al. 2012; Box 1), and
usually showing that process-based forest models are more conservative
than correlative SDMs (Morin & Thuiller 2009).
Some recent model developments further aim at accounting for other
components of global change (Pütz et al. 2014; Pérez-Méndez et al.
2016), such as the impacts of defaunation or fragmentation on forest
dynamics (Pütz et al. 2011; Dantas de Paula et al. 2015, 2018; Box 2).
Calls for a better integration of plant-animal (Berzaghi et al. 2018)
and plant-plant interactions, such as the effect of the increasing liana
abundance on tree growth and survival (Verbeeck & Kearsley 2016),
should further foster such developments (Pachzelt et al. 2013; di Porcia
e Brugnera et al. 2019). Another challenge is the representation of tree
species dispersal and migration of tree species at large scales (Neilson
et al. 2005; Snell et al. 2014; Lehsten et al. 2019; see Box 2, Q8,
Table 2), in combination with evolutionary processes to account for
species adaptive evolution and trait displacement under environmental
changes and fragmentation (DeAngelis & Mooij 2005; McMahon et al. 2011;
Scheiter et al. 2013). Moreover, accounting for the adaptive capacity of
tree individuals within their lifetime via acclimation and phenotypic
plasticity (Richter et al. 2012; Duputié et al. 2015) remains a
challenge, as knowledge about these processes remains incomplete. To
seek additional insights in estimating future forest responses, a number
of studies have used forest models to estimate past forest dynamics
(e.g. Heiri et al. 2006; Schwörer et al. 2014). Overall, these
developments should further help to understand the long-term effects of
multiple interacting factors of global changes on forests (Q4, Q5, Q10,
Seidl et al. 2017).
Forest management
Forests provide important ecosystem services, such as timber production,
carbon sequestration, recreation and protection against natural hazards,
whose persistence or improvement is of high societal relevance (De Groot
et al. 2002, MEA 2005). This is the focus of forest management (e.g.
Nabuurs et al. 2017; Yousefpour et al. 2018). Forest IBMs have a long
history in helping management planning (e.g. Mäkelä et al. 2000;
Courbaud et al. 2001; Huth & Ditzer 2001; Porté & Bartelink 2002; Huth
et al. 2005; Keenan et al. 2008; Pretzsch et al. 2008; Hiltner et al.
2018). As global change challenges current and future management
strategies (Seidl et al. 2014b), forest model developments have aimed to
help design adaptive forest management practices and mitigation
strategies under multiple disturbances (Fontes et al. 2010; Rasche et
al. 2011; Elkin et al. 2013; Kunstler et al. 2013; Lafond et al. 2014;
Maroschek et al. 2015; Reyer et al. 2015; Mina et al. 2017; Seidl et al.
2018). DGVMs have long disregarded the effect of forest management, as
their aggregated representation of vegetation structure typically
prevents a realistic representation of tree size distribution and
density relevant to simulate silvicultural practices. However, some
DGVMs used a simplified representation of wood extraction to simulate
its effect on forest carbon stocks (Zaehle et al. 2006), and recent
efforts have led to the development of more explicit forest management
modules, inspired by finer-scale forest gap models, forest growth and
yield models (Bellassen et al. 2010; Collalti et al. 2018).
The integration of societal and economic dynamics generate new
challenges (Q7 Table 2, Box 2), while future applications and
communications with forest stakeholders will benefit from developments
regarding visualisation of results from forest models (Fig. 1, section
2).
Table 2