1 Introduction

In boreal forests, stand composition and structure are influenced directly by fire events, insect outbreaks, windstorms, and industrial harvesting (James et al., 2011). Hence, the current state of the forest is affected by the spatial heterogeneity (patchiness and mosaics) of past forest disturbances; in turn, this spatial heterogeneity forms the template for future disturbance dynamics. Forest landscapes have a long memory of disturbances (Peterson, 2002), and their patterns can persist for several decades or even centuries (James et al., 2007). Consequently, disturbances are a crucial driver of forest landscape dynamics.

Forest disturbance regimes around the globe are changing rapidly because of climate change (Seidl et al., 2017). In boreal forest ecosystems, the coincidence of warmer and drier than average conditions consistently leads to increased disturbance activity (Seidl et al., 2020). As the climate system will likely continue to change in the coming decades and boreal regions warm more rapidly than other parts of the world (IPCC, 2013), forest disturbances in the boreal zone could increase in the future (Boulanger et al., 2014; Flannigan et al., 2009; Gauthier et al., 2015). Understanding and projecting future forest disturbance regimes is of paramount importance, as disturbances shape ecosystem structure and function and influence the ability of forests to provide important ecosystem services to society (Thom & Seidl, 2016). The main tools for making inferences on potential future disturbance trajectories and their impacts are models. This chapter reviews different approaches to modeling the most important natural disturbance agents in boreal forests: wildfire, wind and snow, and herbivory from pathogens, insects, and mammals. Given that the main platforms for modeling boreal forest disturbances are forest landscape models, we precede our discussion of modeling individual disturbance agents with a short introduction to forest landscape modeling.

2 Forest Landscape Modeling

Forest landscape models simulate forest dynamics beyond the stand scale in a spatially explicit manner and consider landscape-scale processes, such as the dispersal of seeds and the spread of fire across the landscape (Shifley et al., 2017). Because forest structures vary over time and space and landscape patterns create feedbacks affecting the frequency and severity of subsequent disturbances (James et al., 2007), simulating the spatiotemporal interactions between vegetation and disturbances has become a central purpose of forest landscape models. Several spatially explicit landscape models have been developed over the last decades, ranging from individual-based (SORTIE-NT, Beaudet et al., 2011; iLand, Seidl et al., 2012) to cohort-based models (LANDIS, He & Mladenoff, 1999). Many of these models aim to investigate the synergistic effects of the apparent stochasticity of natural disturbances (fire events, insect outbreaks) and scheduled human activities (forest management). Here, we refer the reader to excellent reviews that synthesize the wide range of landscape and disturbance models available to date (Keane et al., 2015; Perera et al., 2015; Seidl et al., 2011; Shifley et al., 2017, among others).

Most forest landscape models build on a common structure and add modules to incorporate more processes or features. For example, LANDIS (He & Mladenoff, 1999; He et al., 1999) models species by age cohort on a lattice where several processes must occur (seed dispersal, succession), while additional disturbances are optional, including fire, insect outbreaks, and harvesting). The PnET module of LANDIS-II (de Bruijn et al., 2014; Scheller et al., 2007) improved on the original LANDIS approach by adding ecophysiology and successional models to model biomass per age class and tree species. Similarly, LANDIS PRO (Wang et al., 2014; Xiao et al., 2017), which is derived from LANDIS, models biomass on the basis of tree density and size per cell; this makes it possible to interface the model directly with forest inventory data. Other forest landscape models have been developed using dedicated modeling languages, such as SELES (Spatially Explicit Landscape Event Simulator, Fall & Fall, 2001). Another example is the Vermillion Landscape Model (VLM; James et al., 2007, 2011), which simulates the effects of fire events, insect outbreaks, and harvesting. Other models have used a state-and-transition modeling approach in which vegetation dynamics are simulated as transitions between discrete vegetation states (e.g., ST-SIM). For instance, Daniel et al. (2017) incorporated both deterministic state transitions from forest management plans and stochastic effects of fire events in their simulation of a study area in the boreal forest of Ontario. They showed the importance of including stochasticity caused by disturbances within forest management planning and highlighted the potential of disturbances to create shortfalls in timber harvest. We also note that in the context of forest management, many stand-level models are applied (e.g., Díaz-Yáñez et al., 2019; Valinger & Fridman, 2011).

Regardless of the selected modeling approach, all models must be parameterized, and an understanding of the direct and indirect relationships among the modeled processes is a prerequisite. Models developed for a specific region usually cannot be applied to another region without first calibrating the model with new data or evaluating it against independent data obtained from the new study area. A key difference exists between empirical models, which are fit to available data, and process-based models, which are built from a quantitative understanding of the processes underlying forest dynamics. Whereas empirical models use the available data for model building—and are thus often more precise in their projections—process-based models, which are more general and can also robustly capture the effects of future environmental conditions not represented in past data, require data from the study area to evaluate whether the model can reproduce observed patterns (Grimm et al., 2005).

3 Fire

A tough challenge in projecting future changes in boreal forests is the inclusion of disturbance processes, such as wildland fire, that are driven by extreme events, exhibit nonlinear dynamics, and involve spatial relationships. Simulating fire involves emulating dynamic and sometimes stochastic processes of fire ignition, spread, and extinguishment controlled by a host of climatic, geoenvironmental, and societal factors, such as wind speed, slope, aspect, and proximity to human development. These factors vary widely in their relative importance depending on the type of simulated fire, i.e., natural wildfires (lightning-caused), human-caused accidental fires, and prescription burning. Lightning, for example, is the cause of most fires in Alaska (Kasischke et al., 2010); however, the majority of fires in Siberia are human-caused (Achard et al., 2008). This causes differences in the spatial and temporal pattern of ignitions and affects the rate of spread and the potential for fire suppression. Across all types, fire is sensitive to vegetation composition and structure (Johnson, 1996), but fire also has a significant effect on the rate and successional sequence of vegetation and carbon cycling (Agee, 1996; DeBano et al., 1998). Creating models that simulate the timing, pattern, and severity of different fire types, while allowing for nonlinear changes in vegetation responses to climate change is not a simple task (Fig. 24.1).

Fig. 24.1
A set of 2 photos. A photo on the left depicts a grassland near a hill which have taken on a brown tint as a result of the fire. A photo on the right depicts a close-up shot of one of the plants.

Photo credits Melissa Lucash

The complex patterns created by wildfires in boreal forests (left) and their impacts on vegetation (right) as evidenced by the Hess Creek Fire, which burned 76,634 ha in central Alaska in 2019.

Full size image

One of the most effective tools for simulating wildfire is the landscape fire succession model, which runs the gamut from simple models of successional pathways and stochastic wildfire (e.g., SIMPLEE, Chew et al., 2004) to complex models simulating individual trees, biogeochemistry, and climate (e.g., Fire-BGC, Keane et al., 1996). These models vary in their ability to simulate different ignition types, i.e., natural or lightning-caused, suppression activities, the degree to which they rely on first principles, the level of stochasticity, and the appropriate scale to which they should be applied (Keane et al., 2004; McKenzie & Perera, 2015). Despite the wide range of approaches and applications, all landscape fire models share four essential components: (1) fire ignition, (2) fire spread, (3) vegetation, and (4) fire effects.

The ignition component of a fire model simulates the initiation of a fire event, which has both spatial and temporal aspects owing to variations in climate, vegetation, and topography, which affect the probability of a successful ignition. Wildfire ignitions are often simulated stochastically by applying a user-defined number of ignitions combined with a probability distribution function (e.g., Weibull, zero-inflated Poisson, and Pareto). These functions are then calibrated to match fire data from a fire-history database or perimeter atlas for the study region. However, this simulates the pattern but not necessarily the underlying mechanisms. Models differ in the factors that influence burn probability; some include weather (e.g., BFOLDS, Perera et al., 2002, 2008; FlamMap, Finney, 2006), a flammability coefficient or stand age (ALFRESCO, Rupp et al., 2000b), fire return interval (FireBGCv2, Keane et al., 2011), or fuel moisture and type (BFOLDS, Perera et al., 2002, 2008). Most models do not account for the different spatial and temporal ignition patterns between human and natural fires. Human-caused fires often occur in areas of high accessibility and on holidays and weekends (Beale & Jones, 2011; Maingi & Henry, 2007), whereas natural fires are driven more by fuel type, fuel moisture, and climatic conditions that favor lightning. High-quality data are needed to parameterize or calibrate approaches that explicitly account for factors controlling ignitions (Prestemon et al., 2013), and recent efforts have compiled large databases for public use in the United States and Canada (e.g., NRC, 2020; Short, 2017). These databases are not available for all circumboreal forests, notably Siberia, and have limitations; for example, accidental fires may be reported within minutes of ignition, but lightning fires may not always be detected because they can smolder for days before growing to a detectable size often in remote locations. A decision-tree analysis is often employed for simulating human-prescribed burning in boreal ecosystems, whereby a maximum allowable number of ignitions is user-prescribed, and weather-conditional statements are applied to determine whether the fire ignites (e.g., SCRPPLE, Scheller et al., 2019). Including all physical processes that affect wildfire ignitions for the various causes of fire is a complicated task (Prestemon et al., 2013) and has yet to be fully integrated into forest simulation models.

Once a fire ignites, the spread components of the model determine the shape and extent of the fire, applying either a lattice approach or a vector strategy (Gardner et al., 1999). The lattice approach simulates fire spread from one raster pixel to another using cellular automata (EMBYR, Hargrove et al., 2000) or bond percolation (SpaDES, Marchal et al., 2020). These models allow the stochastic spread of fires between raster pixels on the basis of (1) probability distributions (e.g., Base Fire in LANDIS-II, He & Mladenoff, 1999; WMFire, Kennedy & McKenzie, 2010, McKenzie & Kennedy, 2012), (2) stochasticity combined with empirical relationships derived from laboratory experiments or field data (e.g., iLand, Seidl et al., 2012; SCRPPLE in LANDIS-II, ALFRESCO, Rupp et al., 2000b), or (3) physics-based combustion and spread models (e.g., the WFDS, Mell et al., 2007; Coupled Atmospheric Weather-Fire Experiment, Coen et al., 2013). Vector strategies use raster maps of ignitions, but they allow fire to spread using two-dimensional vertices that increase in number as the fire grows (Finney, 1998). The spread is driven stochastically, empirically with generalized linear modeling, or via algorithms of physical processes (e.g., FARSITE, Finney, 1998; FARSITE in Fire-BGC, Keane et al., 1996). Spread in both cellular automata and vector approaches is influenced by vegetation succession, which can be simulated using (1) a state-and-transition model of user-defined community types and pathways, (2) a cohort model of species and age, or (3) an individual plant model that simulates each tree or plant on the landscape. State-and-transition models, like ALFRESCO, have been widely used in boreal forests to characterize changes in vegetation type (Johnstone et al., 2011; Rupp et al., 2000a), tree-line expansion (Hewitt et al., 2016), and vegetation-climate feedbacks (Euskirchen et al., 2016) in response to climate change. Cohort models, e.g., LANDIS-II, have seldom been used in boreal forests; an exception is their application to characterizing the importance of timber harvesting in driving long-term succession in Siberia (Gustafson et al., 2011). Although studies of postfire boreal succession have, to date, relied primarily on simpler, more deterministic models, future studies will focus on modeling fire spread and vegetation development to capture the emerging nonlinear dynamics stemming from the increased fire frequency in these systems (Johnstone et al., 2010; Kasischke et al., 2010).

Fire effects are often simulated very simplistically using either rule-based methodology (e.g., SIMMPLE, Chew et al., 2004; LANDSUM, Keane et al., 2006, 2008; TELSA, Klenner et al., 2000, Kurz et al., 2000) or mechanistic mortality probabilities (Fire-BGC, Keane et al., 1996, SCRAPPLE in LANDIS-II, Scheller et al., 2019). In some individual models, all trees die if a fire burns in a cell (e.g., Base Fire in LANDIS-II, He & Mladenoff, 1999; He et al., 1999), whereas state-and-transition models use rules to determine the fate of a vegetation type (i.e., transition to a different state, Rupp et al., 2001). A more mechanistic approach relies on empirically derived logistic regression probabilities to model species or age-specific mortality (e.g., SCRAPPLE in LANDIS-II, Scheller et al., 2019); however, this has not been used in boreal ecosystems to date.

Future attempts to project how boreal forests will be affected by wildland fire and climate change could be improved by (1) capturing the different mechanisms and spatial patterns between human-caused fires and wildfires, (2) establishing direct linkages to smoke models to estimate the impacts of smoke on human health, (3) creating models that couple processes of fire, vegetation, permafrost, and hydrology, and (4) ensuring the models capture nonlinear, emergent fire and vegetation behavior under a changing climate. Improved projections of wildland fire and smoke in boreal forests will help identify urban and rural communities at risk and determine the most effective strategies for developing future land-use plans.

4 Wind and Snow

Wind is a major disturbance agent in coastal forests around the globe. The risk of wind disturbance generally decreases with distance from the coast. High gust speeds are the primary trigger of wind disturbances, with individual trees falling when gusts exceed approximately 30 m·s−1, and marked wind impacts occur when gusts exceed 40 m·s−1 (Gardiner et al., 2010). The main causes of strong winds are (1) cyclonal storms resulting from large-scale pressure differences in the atmosphere; these storms are generally responsible for the most extensive wind disturbances in forest ecosystems; (2) thunderstorms, often with very high wind speeds but only local impacts; (3) katabatic winds resulting from cold air pooling over ice masses; and (4) winds resulting from weather differences between the windward and leeward sides of mountain ranges (e.g., foehn, Chinook). Strong winds can generally cause a wide variety of disturbance patterns in forest ecosystems, ranging from small-scale canopy openings via the replacement of individual trees to large-scale, high-severity disturbance patches (Fig. 24.2). Wind impact is strongly modulated by forest structure, with tree height and species identity being the most prominent predictors (Díaz-Yáñez et al., 2017; Valinger & Fridman, 2011). The main impact of wind on trees is stem breakage and uprooting. As this fundamental impact is the same for snow disturbances, the two agents are often modeled similarly and are addressed jointly here. Snow-related disturbances require the presence of snow, which limits them to areas having frequent snowfall or long periods of snow cover, such as the boreal zone. However, critical for the occurrence of snow-related disturbances are individual heavy snowfall events or rain-on-snow events, which cause heavy snow loads in tree canopies. The risk of snow-related disturbance is generally considered high when the cumulative snow load exceeds 20 kg m2 (Kilpeläinen et al., 2010).

Fig. 24.2
A group of 2 photos. A photo on the left depicts a forest where few trees are fell on the ground and few are sliced in half, and a photo on the right depicts few plant stalks have fallen to the ground.

Photo credits Rupert Seidl

The impacts of wind and snow on forest ecosystems range from individual tree death or damage (left) to the killing of trees at the stand to landscape scale (right).

Full size image

Three crucial processes must be addressed to capture the dynamics of wind and snow disturbances in models: (1) the occurrence of strong winds or heavy snow loads, (2) the susceptibility of forests to wind and snow disturbance, and (3) the impacts of wind and snow on vegetation. Disturbances by wind and snow are triggered by climatic extremes, such as high winds and extreme snow loads. The occurrence probability of such events can be derived from statistical analyses of climate data, e.g., using extreme value theory (Bengtsson & Nilsson, 2007). However, good climate observations—a prerequisite for such analyses—are often not available for remote forested areas. The occurrence of extreme wind and snow conditions has thus frequently been modeled as dependent on topographic variables (Ruel et al., 1997; Suárez et al., 1999), describing the varying exposure to such disturbances in a landscape. Most existing dynamic forest landscape models trigger wind and snow disturbances stochastically. Increasingly, however, detailed local airflow models are used to derive critical windspeeds for forest landscapes. Approaches such as WAsP (Zeng et al., 2006) and MS-Micro/3 (Talkkari et al., 2000) have been applied to model the wind development over forest canopies, accounting for the effects of topography and forest structure at the landscape scale. Such models can also be applied to downscale projections from regional climate models to obtain detailed wind projections for forests (WINDA, Blennow & Olofsson, 2008).

Forest structure and composition strongly determine how vegetation responds to strong winds and high snow loads. In general, the susceptibility of forests to wind increases with tree height (Valinger & Fridman, 2011). Crown shape, stem taper, and species-specific wood properties also influence the sensitivity of forests to wind and snow. A common approach to modeling these susceptibility differences is fitting regression models to observational data (e.g., Díaz-Yáñez et al., 2017; Jalkanen & Mattila, 2000). Such models can then be implemented in simulation models that dynamically project forest structure and composition. Dose–response models are a more mechanistic approach to modeling vegetation susceptibility to wind and snow. These models typically quantify tree and stand attributes related to resisting the physical forces of wind and snow, such as tree height, modulus of rupture, and rooting strength. They subsequently determine the critical loads required for breaking or overturning a tree (GALES, Gardiner et al., 2000, 2008; HWIND, Peltola et al., 1999a; see also Fig. 24.3). Tree-pulling experiments provide an important empirical database for the parameterization of these models (Nicoll et al., 2006). Because frozen soil can considerably improve the anchoring of trees, soil frost has also been considered in modeling wind disturbances (Peltola et al., 1999b; Seidl et al., 2014). Furthermore, the spatial context of a stand is an important factor determining its wind risk, e.g., whether there is a large upwind gap or not, a situation considered explicitly in some simulation frameworks (HWIND, Zeng et al., 2009; iLand, Seidl et al., 2014).

Fig. 24.3
A flowchart depicts framework of H W I N D dose-response model. It describes the critical wind speed and snow loading for uprooting and stem breakage formation outputs from various types of inputs with the help of derived values and relationships, and model calculation.

Modified with permission from Canadian Science Publishing, permission conveyed through Copyright Clearance Center, Inc., from Peltola et al. (1999a)

The conceptual design of the HWIND dose–response model applied to simulate wind and snow disturbance in Finland.

Full size image

The impacts of wind and snow on forests can be manifold, ranging from broken branches and roots to stem breakage and the uprooting of trees. Furthermore, the frequent exposure to wind and snow can result in acclimation processes within a tree, e.g., increased allocation of carbohydrates to roots, changed canopy structure. To date, these processes have rarely been explicitly considered in disturbance models. Most models of wind and snow disturbance impacts consider only stem breakage and uprooting and determine whether a tree survives a given event or not. Some models additionally consider that trees can be killed by falling neighboring trees in a windthrow (ForGEM-W, Schelhaas et al., 2007). In cohort-based approaches, wind disturbances reset the forest development of a cohort (LANDIS, He et al., 1999), whereas in structurally simple big leaf ecosystem models, wind impacts are simulated by removing biomass from the respective pools (BiomeBGC, Lindroth et al., 2009). In contrast to fire, the spatial extent of wind and snow disturbances is usually not determined through an active spread process. However, by updating the vegetation structure during a wind event and accounting for newly exposed trees, the landscape patterns created by windthrow can be mimicked closely in simulations (iLand, Seidl et al., 2014).

Climate change will profoundly influence the occurrence and severity of wind and snow disturbances. Changes in peak wind speeds remain difficult to project (Shaw et al., 2016); nonetheless, warmer temperatures and increased levels of atmospheric CO2 could lead to taller trees in the boreal forest, which are more susceptible to windthrow. Moreover, warming will reduce soil frost, with adverse effects on the anchoring of trees (Gregow et al., 2011). Snow disturbances are generally expected to decrease under climate change (Seidl et al., 2017), yet the prevalence of wet snow events could also increase locally with warming. To improve projections of future wind and snow disturbances, we need better information on future wind and snow conditions and improved process models.

5 Herbivory

Biotic disturbance agents, such as insects, pathogens, and mammals, consume plant biomass in the boreal forest, causing growth loss and tree mortality. The extent and severity of biotic disturbances range from small, low-severity, gap-type dynamics, typical in Fennoscandia (Kuuluvainen & Aakala, 2011), to large-scale high-severity outbreaks, such as the spruce budworm (Choristoneura sp.) in Canada (Navarro et al., 2018) or the Siberian silkmoth (Dendrolimus superans) in Siberia (Kharuk et al., 2007) (Fig. 24.4). Although a tiny bark beetle may seem very different from a gigantic moose (Alces alces), they—from a perspective of disturbance modeling—share common processes that can be harnessed for modeling. All herbivores (1) disperse, (2) establish, (3) reproduce and die, and (4) affect their host in various ways (Fig. 24.5). The details of each process vary between agents and systems; however, these processes form the basic building blocks of models that simulate the dynamics of biotic disturbance agents in the boreal forest.

Fig. 24.4
2 photographs depict the side and top view of a forest with a great height of mangrove trees on the green earth's surface.

Photo credits Juha Honkaniemi (left), Miguel Montoro Girona and Janie Lavoie (right)

Disturbances caused by biotic agents vary by extent and severity; (left) a low-severity, small-scale disturbance caused by the European spruce bark beetle (Ips typographus) in Finland, and (right) a high-severity, large-scale disturbance caused by eastern spruce budworm (Choristoneura fumiferana) in Quebec, Canada.

Full size image
Fig. 24.5
An illustration depicts recycle process of dispersal, establishment, population growth, and impact from disturbance dynamics such as host, and the environment by using a biotic agent.

The disturbance dynamics involving biotic agents are the result of close interactions between the agent, its host, and the surrounding environment. The common processes shared by biotic agents, together with the interactive variables of host vegetation and environment, form the basic building blocks for model development

Full size image

Dispersal is one of the relevant processes for all biotic disturbance agents. Many agents move autonomously (using their feet or wings) for dispersal, whereas others rely on external aid, such as wind or water. Dispersal can be modeled simply by the distance an individual moves in each time step. Many landscape models use probability density functions to describe the probability of an individual moving from point A to point B in time t (e.g., Pukkala et al., 2014). However, these models neglect the direct effects of landscape structure, the size of the individual agent, and the prevailing weather conditions. Consequently, more detailed approaches have been developed to consider these factors (e.g., Norros et al., 2014; Sturtevant et al., 2013).

Once the agent has moved into a new area, whether it can establish itself in that location is constrained by two factors: (1) habitat quality, i.e., host availability and climatic suitability, and (2) population density, i.e., the number of individuals required to maintain a viable population. These are most often included in models as Boolean filters to indicate the success or failure of establishment or are represented in indices of varying levels of complexity (e.g., Lustig et al., 2017; Sturtevant et al., 2004).

Population growth over time is crucial for an established population to thrive, and reproduction and mortality are central processes determining population growth. One of the simplest ways to model population dynamics is through logistic growth equations, where the population growth rate is defined by the birth and death rates of the agent, and the population size is constrained by external factors setting the carrying capacity of an area. However, population dynamics of well-studied biotic disturbance agents can be modeled more explicitly, e.g., using detailed phenological models (Baier et al., 2007; Bentz et al., 1991).

By definition, biotic disturbance agents affect their host species; these impacts can vary from decaying the root system to consuming foliage. Depending on the intensity of the biomass consumption and the compartment that is affected, these impacts can lead to growth loss or tree mortality. Modeling tree mortality probabilistically at different scales is one of the most common approaches to simulate biotic disturbance impacts (e.g., iLand, Seidl & Rammer, 2017; LANDIS BDA, Sturtevant et al., 2004). Some models incorporate more detailed approaches, such as the herbivory of foliage (Régnière & You, 1991) or the decay of root systems (WINDROT, Honkaniemi et al., 2017), and are thus capable of quantifying more detailed effects and various impact pathways in models.

The changing climate affects biotic disturbance agents in various ways. The ongoing changes in environmental conditions are causing the poleward migration of many species, introducing pests to new environments (Bebber et al., 2013; Økland et al., 2019). Many insect species also respond to increased temperatures with accelerated reproduction and increased winter survival (Bale et al., 2002). In general, insects and pathogens will adapt faster to a changing climate than their hosts, increasing the risk for large-scale, high-severity outbreaks in the future. Mammalian herbivores, such as moose and voles, do not generally respond to climate change by rapid range shifts. They may, however, change their behavior with a warming climate, which may affect where and when these species cause disturbance (Korpela et al., 2013; Melin et al., 2014). In addition to climate change, global trade has accelerated the introduction of nonnative pests and pathogens to new ecosystems, creating novel threats to forests globally (Chapman et al., 2017; Santini et al., 2013; Seebens et al., 2017). Both climate change and invasive alien species are serious challenges for boreal forests, for which simulation models can make a significant contribution.

The structure and type of models used for simulating biotic disturbances depend on the questions that must be answered. Some models are aimed to simulate the potential distribution, occurrence, and population dynamics of a pest, whereas others focus on quantifying pest impacts on forest ecosystems. Statistical models, such as logistic regression models that quantify short-term disturbance risks (e.g., Jalkanen, 2001; Magnussen et al., 2004) or climate-envelope models that predict potential species distributions (e.g., Vanhanen et al., 2007) are widely used. In a changing world, however, statistical models are not able to robustly capture the changes in agent dynamics. Moreover, the interaction effects between different disturbance agents are often impossible to capture with statistical models because of data limitations (e.g., Honkaniemi et al., 2018; James et al., 2011). Process-based models have been developed to simulate potential future scenarios more robustly, capturing the underlying processes and their cross-scale interactions (Malmström & Raffa, 2000; Seidl et al., 2011). Climate conditions drive many of the common biological processes shared by biotic agents, most often weather-related variables such as temperature or precipitation. In process-based models, these variables drive the physiological processes of agent dynamics, also making these models applicable for simulations under future climate conditions.

The processes of biotic disturbances are often complex and must be simplified in models. The specific processes needing to be simplified and to what degree they can be simplified is best decided on the basis of the model’s aims. Following the principles of pattern-oriented modeling (Grimm et al., 2005), a suitable trade-off between model complexity and payoff must be determined. For most biotic disturbance agents, however, we lack sufficient information on the species biology and ecology to freely choose a level of complexity in model development. Model complexity is dictated by the limited information available for parametrizing key processes foremost disturbance agents.

Agent-based modeling, in which the behavior of individuals or groups in an environment is simulated (Grimm & Railsback, 2006), is one of the most common types of process-based models of biotic disturbance agents. In particular, well-studied biotic disturbance agents, such as the European spruce bark beetle in Europe and the mountain pine beetle in North America, are simulated in highly detailed agent-based models. These models can answer a broad range of questions from agent dynamics to disturbance impacts on ecosystem services (e.g., Bone & Altaweel, 2014; Jönsson et al., 2012; Powell & Bentz, 2014; Seidl & Rammer, 2017).

However, the focus of model development on a small number of well-known agents is problematic, as it can overstate the susceptibility of certain host species over other hosts having less-known biotic disturbance agents. Therefore, the goal in developing process-based models for biotic disturbance agents in the future should be an inclusive and broad consideration of a wide variety of disturbance agents (see Honkaniemi et al., 2021; Lustig et al., 2017; Sturtevant et al., 2004 as examples of such models). General approaches to simulate the dynamics of different biotic disturbance agents will also help estimate the potential impacts of invasive alien pests in novel environments.

Projections of the future disturbance regimes in boreal forests suggest a marked increase of pests and pathogens (e.g., Seidl et al., 2017; Weed et al., 2013). The harsh climate of the boreal region is becoming more favorable to many species, and the poleward migration of species will increase pest introductions and their successful establishment (Hof & Svahlin, 2016; Vanhanen et al., 2007). As boreal forests usually have low tree species diversity, they are particularly vulnerable to changing biotic disturbances. Considering a broader spectrum of biotic disturbance agents and their potential interactions is essential for future improvements of disturbance modeling of boreal forests.

6 Outlook

An important challenge for modeling boreal disturbance regimes is capturing the interactions between disturbances. Frequently, disturbance agents do not act in isolation but are influenced by other disturbances (Buma, 2015). Many of these relationships are amplifying interactions, but dampening interactions can also occur, e.g., when disturbance agents compete for the same resource. As disturbances increase under climate change, interactions between disturbances are also likely to increase (Seidl et al., 2017). It is thus essential to address disturbance interactions in modeling. Dynamic landscape simulation models, such as LANDIS and iLand, provide a robust framework for addressing interactive disturbances because they can simulate multiple disturbance agents and their impacts on vegetation as an emergent property at the landscape scale (e.g., Lucash et al., 2018; Seidl & Rammer, 2017).

A second significant challenge for disturbance modeling is scaling. As the impact of forest disturbances for the provisioning of ecosystem services is increasingly recognized (Thom & Seidl, 2016), disturbance effects must be considered in assessments at policy-relevant scales (i.e., national to global scales). Disturbance modules are, for instance, currently being developed for many Dynamic Global Vegetation models (e.g., Huang et al., 2020; Kautz et al., 2018), which are used inter alia to inform climate policy in regard to the strength of the global vegetation carbon sink. Nonetheless, given the complex interplay between vegetation, climate, and disturbance processes, the scaling of disturbance dynamics is not trivial. An important tool for scaling could be metamodeling (Urban, 2005), i.e., deriving (more broadly applicable) models from existing models. Utilizing emerging machine-learning techniques can further contribute to scalable vegetation and disturbance models (Rammer & Seidl, 2019).

Finally, robust and powerful models are highly dependent on the data and information available for modeling. To improve process-based models of forest disturbance regimes, we need a better understanding of critical processes, such as the dispersal of biotic disturbances and the interactions between disturbance agents. The elimination of such knowledge gaps and the improvement of process-based modeling of forest disturbance regimes requires more experimental research. Notwithstanding limitations in process understanding, the development of forest models is progressing at an accelerating pace, fueled by an increasing computational capacity and the growing availability of data. Remote sensing is increasingly important, as forest disturbances across the globe can now be continuously detected and measured from space (Hansen et al., 2013; Senf & Seidl, 2021). Overall, the modeling of natural disturbances can make an important contribution to an improved understanding of boreal forest dynamics in a changing world and can inform decision-makers in forest management and forest policy regarding the potential consequences of and responses to changing forest disturbance regimes.