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