Fire disturbance and climate change: implications for Russian forests

Boreal forests sit atop the largest reservoir of terrestrial carbon (C), two-thirds of which are in Russia (Hare and Ritchie 1972, Alexeyev and Birdsey 1998). Fire is a dominant disturbance in this region that alters C stores and forest composition (Sukhinin et al 2004). Warming has resulted in intensification in both fire frequency (i.e. the number of fire occurrences per unit time within an area (Pyne et al 1996)) and severity, as well as larger areas burned (Kasischke et al 2004, Kasischke and Turetsky 2006, Kharuk et al 2008). Boreal fire frequency and area burned are predicted to continue increasing (Overpeck et al 1990, Stocks et al 1998, Stocks et al 2000, Flannigan et al 2009), potentially leading to shifts in species composition and tree demography, and alteration of the annual C budget. The temporal effects of fire depend on the combined effects of fuel state, residence time of burn and fire intensity (Van Wagner 1973, Pyne et al 1996). The subsequent loss of above- and belowground organic material defines the fire severity (Keeley 2009). Fire severity and variations in duff consumption shape post-fire successional pathways (Johnstone and Kasischke 2005, Shorohova et al 2009, Johnstone et al 2010a, Barrett et al 2011, Shenoy et al 2011). Low-severity, frequent fires can prevent young trees from reaching maturity, and severe, stand-replacing fires reset the successional cycle (Johnstone et al 2010a, Johnstone et al 2010b). Some species tolerate and regenerate quickly after fire, and others are negatively impacted by frequent or severe fires, thus impacting composition (Johnstone and Kasischke 2005, Shorohova et al 2009, Johnstone et al 2010a, Shenoy et al 2011, Schulze et al 2012). Through its effects on stand structure and species composition, fire will be a strong driver of forest composition and C dynamics within Russia.

In addition to its impact on fire dynamics, climate change is also likely to bring about changes to aboveground boreal C dynamics through alteration of tree growth, vigor, and mortality (Allen et al 2010, Kharuk et al 2005, Schaphoff et al 2016, Soja et al 2007, Tchebakova et al 2009). It is unclear how these may interact, especially with the concurrent effects of changing fire disturbance. Schaphoff et al (2016) reported that the current fire-affected area is ∽20% larger than mean area found for the period of 1960–2007 (Soja et al 2007). Forest lost to stand-replacing fires has also increased (Schaphoff et al 2016). Within Siberia, more frequent or more severe fire alters stand density and diameter increment, resulting in increased aboveground C storage (Kharuk et al 2005, Furyaev et al 2001, Alexander et al 2012, Kharuk et al 2011). High severity fires and the combustion of duff and organic soil layers can create a shift in the dominant species through changes in soil depth and moisture, impacting species-specific recruitment and growth (Johnstone et al 2010a, Johnstone et al 2010b, Barrett et al 2011). Field studies have demonstrated differential species response to fire, with Pinus sylvestris forests in lower and central Siberia displaying a 50% and 83% loss of biomass, respectively, whereas lower Siberian larch (Larix spp.) forests lose between 47% and 23% of biomass to fires of variable severities (Ivanova et al 2011, Kukavskaya et al 2014). Evaluating the ultimate response of boreal forests and associated aboveground C storage thus requires consideration of the response of individual species.

Individual tree-based gap models, which simulate the establishment, growth, and death of trees on patches across a landscape, can explore the impact of changing disturbance and climate regimes on forest characteristics and aboveground biomass. Gap models established according to the approach of Botkin et al (1972) and Shugart and West (1977) are driven by successional dynamics and the concept of 'gap phase' replacement introduced by Watt (1947). As a simulated tree dies, new and existing trees establish and grow within this gap through greater access to light and resources (Shugart and West 1980). Unlike dynamic global vegetation models (DGVMs), which typically only simulate biomass at the level of plant functional types (PFTs) the strengths of gap models lie in their simulation of trees and species, thus allowing prediction of fine-scale changes in forest structure and composition. Field studies demonstrate that prior stand structure, regeneration, and within-stand disturbances act as important controls on landscape aboveground C storage (Kashian et al 2013). By tracking each stem, individual-based models capture changes to aboveground biomass and stand structure resulting from environmental factors such as shading and moisture stress, or size- and species-specific mortality events such as fires.

We use the individual-based gap model University of Virginia Forest Model Enhanced (UVAFME) to simulate forest dynamics across Russia in response to historical fire disturbance probability and fire disturbance altered by changing climate. The fire disturbance portion of the model is updated to include size- and species-level effects of fire on tree mortality and regeneration. The historical fire return interval is calculated for eco-regions across Russia, from which we derive the probability of fire. The associated percent of stand mortality across Russia is derived and applied to calculate the mean intensity of fire at model-simulated locations. Using four simulation scenarios, which include or exclude fire and use either historical or altered climate, we evaluate the impact of changing climate and fire disturbance regimes on species biomass, composition, and stand characteristics across Russia. These results are an important step towards understanding how C and species dynamics may change.

Model description

Fire within the model

Fire disturbance within the model has been updated to allow for variable fire intensity as well as species- and size-specific effects on tree survival and regeneration. A detailed description of this module and parameterization is included in Supplement S2. Fire occurs stochastically based on a site-specific fire probability, set according to the fire return interval (FRI) in years determined from historical data (figure 1). FRIs for this study are based on values derived across Russia for eco-regions using the Regionally-Adjusted MODIS Burned Area (RAMBA) method of mapping burned area for the period of 2001 to 2014 (Loboda et al 2007, Loboda et al 2011) with the methodology of Soja et al (2006). According to Soja et al 2006, the FRI is the average amount of time required to burn an area equivalent to an entire ecosystem. Using this method, a landscape with an average FRI of 100 years would be expected to experience 1/100 or 0.01 of that landscape to burn annually. The percentage of stand-replacing fires calculated according to Krylov et al (2014), which represents the percentage of total forest stand mortality (figure 1), is used to derive the site-specific mean fire intensity. Fire severity based on vegetation mortality has been shown to correlate with fire intensity in forests dominated by conifers (Wade 1993, McCaw et al 1997, Keeley 2009). Individual tree mortality is evaluated in the model based on size, percentage of crown scorch and cambial damage based on bark thickness. Therefore, for trees the same size and with the same crown scorch damage, those with thicker bark will have a lower probability of mortality. Removal of soil organic material by fire is not included in this version of the model. Fire affects seeds and seedlings based on species-specific characteristics with either an increase, decrease or no change to the seedling bank. Equations within this updated fire module have been successfully utilized within the western United States (Hood et al 2007, Keane et al 2011, Reinhardt and Crookston 2003, Ryan and Reinhardt 1998) and have also been successfully tested with UVAFME in the southern Rocky Mountains (Foster et al (2017)).

Zoom In Zoom Out Reset image size

Modelled fire probability increases with increasing site aridity based on the ratio of precipitation to potential evapotranspiration as in Feng and Fu (2013) (equation (5) Supplement S2). This modification allows the probability of fire to increase with increasing evaporative demand, either due to higher temperatures or lower precipitation.

Model simulations

The FRI and percent stand mortality across Russia show broad similarities, with the map of percent stand mortality displaying more heterogeneity than the FRI map (figure 1). The FRI shows, on average, long FRI in the north across much of Russia, and shorter FRI along the southern border and in the Amur region of the Russia Far East. Low stand mortality (i.e. less than 30%) occurs along the southern boundary of Russia and corresponds to areas of more frequent fire (figure 1). Small areas with high mortality are embedded within areas classified as having infrequent fires in far northeastern Russia.

Total mature forest biomass for historical climate simulations with (figure 2(a)) and without fire (figure S1) show areas of high biomass in the Amur region of the Russian Far East and European Russia. After 600 years of simulation for historical climate with fire (scenario 2) there is an increase in total biomass compared to simulation without fire (scenario 1) (table 1). Under scenario 3 (altered climate without fire) the increased mean total biomass (94.50 tC ha⁻¹) reflects increased biomass across northern and southern portions of Russia (figure 2(b)). With fire disturbance under altered climate conditions (scenario 4) biomass decreases in the southern portion of western Russia and increases within far northeastern Siberia and along the Amur River border between China and the Russia (figure 2(c)). Across all of Russia, scenario 4 results in increased mean total biomass (112 tC ha⁻¹) (table 1).

Zoom In Zoom Out Reset image size

The increased total biomass for scenario 4 is broadly associated with changes in larch, which is dominant across much of Russia (figures 3 and 4). Under scenario 3, evergreen conifers show increased biomass across central Russia, whereas in these same areas the biomass of larch decreases (table 1, figure 3 and S2). In contrast with fire, the biomass of larch increases for scenario 4 for areas in the Siberian transect spanning north from Lake Baikal, displaying increases of 100 tC ha⁻¹ compared to biomass for scenario 2 (historical conditions with fire) (figures 3 and 4 and S3).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Under altered climate, many species across interior Russia increase in biomass compared to historical climate conditions (figure S3). However, with the inclusion of fire in scenario 4 evergreen trees do not increase biomass compared to scenario 3 (figure 4). Warming climate and fire create a more complex pattern of height and stand age distribution across Russia. With fire and altered climate (scenario 4), UVAFME predicts, on average, taller trees in areas of increased biomass along the southern Siberia border, from the Lake Baikal region into northern Siberia, and in far northeastern Russia (figure S4). In scenario 2 the Lorey's height (i.e. mean height weighted by basal area) may reach 22 m in northern Siberia and 26 m in central and southern Siberia (figure 5). Fire disturbance with warming climate in scenario 4 results in patches of increased height in areas of newly increased biomass, i.e. far northeastern Russia (figure 5). The scenarios also display variable patterns in the frequency of mean stand age (figure 6). With historical climate and fire in scenario 2, there is a peak in frequency for the youngest 20 year age class, and a peak for the 280 year age class (figure 6 top). Under conditions of altered climate without fire in scenario 3, there is no true peak, but a broadly older forest, indicated by a cluster of age classes from 260 to 360 years (figure 6 middle). There is clear bimodal distribution of age classes for scenario 4, with peaks at the 100 year age class and again at 280 years (figure 6 bottom).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Individual-based models relate changes in external drivers (e.g. climate and fire regimes) with internal responses, and produce changes in ecosystem composition and age structure that further affect ecosystem function. Feedbacks among change in drivers and change in ecosystem function are seen in this investigation. The increase in biomass for both altered climate scenarios suggests that climate acts as a strong modulator on biomass and competition dynamics. Species with higher growth rates are more competitive in more favorable climate conditions (i.e. once released from temperature or water limitation). This increase in biomass is not seen with the inclusion of fire under historical climate. GDD increases under warming climate, resulting in increased biomass as species are released from temperature limitation (Schaphoff et al 2016). With increased AMI, drought tolerance is important, and larch effectively competes for resources, as seen in the increases in biomass (figures 3, 4 and S3). These results suggest that water limitation and disturbance will be determining factors for species composition in southern boreal forests. This highlights the importance of limiting species distributions based on available resources, rather than predefined limits, as in the use of a parabolic temperature response or a fixed climate envelope approach.

Peaks in age structure represent long-lived transients from stand replacement processes (i.e. succession) following stand initiation at year zero as well as the transient dynamics from changes in climate and fire in year 500 (figure 6). These results do not suggest a wholescale reduction in biomass in response to altered fire and climate, but rather an alteration of stand structure and thus C dynamics. Altered climate conditions without fire act to increase biomass, resulting in higher aboveground C stores as stand structure becomes heavily weighted towards older stands (figure 6). The inclusion of fire disturbance increases tree turnover rate, and for altered climate with fire shifts the age distribution to include young productive stands as well as mature older stands (figure 6). Associated changes in height occur alongside this change in stand age. Increases in stand height with warmer conditions (figure 5) agree with those of Tchebakova et al (2016), which demonstrated a similar increase in height in far northern Siberia using a simplified forest height stand model validated against the lidar-based height map of Simard et al (2011). The differences between the altered climate scenarios with and without fire suggest that the demographic and structural responses to fire are an important component driving biomass and species dynamics.

The increase in aboveground carbon storage from this study agrees with previous boreal forest modeling work by Kasischke et al (1995), which demonstrated an increase in aboveground storage with warming and increased fire frequency. In a field study of a transect in central Siberia which focused on fire frequency and its importance in determining species composition and performance, Schulze et al (2012) found that larch demonstrated a constant increase in biomass over 350 years of stand ages. The changes in biomass seen in this current study are largely due to increases in biomass for larch in the same region studied by Schulze et al (2012). These results demonstrate the competitive advantage that larch gains with fire disturbance due to their tolerance of fire and high growth rate compared to other species (Nikolov and Helmissari 1992, Hytteborn et al 2005).

Bark thickness is a key component determining tree mortality due to fire. Bark thickness increases with increasing tree diameter, thereby increasing fire resistance (Harmon 1984, Peterson and Ryan 1986, Ryan et al 1988, Ryan and Reinhardt 1988). Larch and P. sylvestris have the highest bark thickness value (0.063 cm bark per cm DBH), making them more tolerant of fires compared to other species, whose bark thickness ranges from 0.046 to 0.014 cm bark per cm DBH (table 2). Larch and P. sylvestris coexist across central Russia in mixed forests, with larch more dominant further north (Nikolov and Helmissari 1992, Schulze et al 1995, Kharuk et al 2011, Schulze et al 2012). Larch have a higher growth rate than does P. sylvestris, affording them an advantage in biomass accumulation following disturbance where both species survive. In our simulations, larch are dominant in biomass for conditions of warming climate and fire across the broad Siberian region. With fire, Pinus spp. have a similar mean biomass under historical and warming climate conditions (figure S6). This similarity indicates that while P. sylvestris may be fire tolerant, its growth rate is not fast enough to result in increased biomass following fire when in competition with larch. These results match findings of Schulze et al (2012) for studies in central Siberia, of Kharuk et al (2011) for northern Siberia, and of Schulze et al (1995) for eastern Siberia within the Lena basin, all of which found that fire cycle is essential in maintaining larch dominance. Field and remote sensing studies in larch forests of far northeastern Russia found a range of post-fire tree survival with biomass accumulation varying according to the age of surviving stands (Berner et al 2012). These results demonstrate that the ecosystem response to fire derives from individual-tree attributes, as both species and individual tree size are vital for determining fire mortality.

Though there have been improvements in modeling fire at the global scale, DGVMs do not consider complex age or size structure, nor species. The SPITFIRE model (Thonicke et al 2010) that captures vegetation mortality based on crown scorch and cambial damage has been used to improve fire simulation within multiple DGVMs (e.g. LPJ-LMfire (Pfeiffer et al 2013), ORCHIDEE (Yue et al 2014, Yue et al 2015), and JSBACH, the land portion of MPI (Lasslop et al 2014)). These DGVMs have shown fidelity in their ability to simulate historical global burned area and fire regimes with particular improvements within boreal regions where fire is strongly driven by climate variation. They all use PFTs to represent a group of similar species, which in the boreal region reduces the vegetation to a total of 2 or 3 PFTs depending on the model. DGVMs provide valuable tools for exploring fire impacts, but do not have the ability to track the complex interactions among individual trees and species. Size structure is essential for determining accurate biomass estimation and tracking mortality in forest systems globally, which demonstrate different behavior based on transitions following disturbance (e.g. dominance of different species in the boreal zone in response to fire as presented in this manuscript, and transitions between forest and grassland in tropical systems (Hoffman et al 2012, Lehmann et al 2014)). With their ability to provide detail size structure and demography, individual species-based models, such as UVAFME, could be used alongside DGVMs to provide a complete picture of vegetation structure and composition in response to changing disturbance and climate.

Along with forest composition and structure, FRI is an essential driver of forest dynamics. There is strong variation in FRI in association with forest type and latitude, with the northern forests having periods as long as 350 years between fires compared to the FRI of 20 to 60 years in the larch-taiga ecotone and southern Siberian mountains (Kharuk et al 2008, Ivanova et al 2010, Kharuk et al 2011, Soja et al 2006, Furyaev et al 2001, Ponomarev et al 2016). The map of burned area created by Schaphoff et al (2016), which utilized data from Sukhinin (2011) and Bartalev et al (2015), for the period from 1996 through 2010 agrees with the FRI data used in this study in identifying the southern Russian border as an area of high fire frequency. With classification based on burned area in large eco-regions the FRI data used for this study likely under-represent areas which locally have more frequent fires. This under-representation arises out of the temporal limitation of using the available 13–14 years of data in a region known to have extreme inter-annual variability in burned area and FRI (Soja et al 2006, Ivanova and Ivanov 2005, Kharuk et al 2011, Schaphoff et al 2016). By using the percent stand mortality as a determinant of mean fire intensity, UVAFME can simulate infrequent stand-replacing fires, as well as frequent low-level fires. The combined use of the detailed percentage of stand mortality of Krylov et al (2014) in conjunction with the broad eco-region FRI dataset allows for simulated fire intensity, and thus tree mortality, to be informed by the more detailed dataset, helping to reduce uncertainty associated with the FRI. Future analysis, however, will calculate FRI across smaller areas, as this may improve representation of FRI by capturing local areas with more frequent fires.

To determine how forest dynamics across the region might respond to a shorter FRI (i.e. a higher fire frequency), scenarios 2 and 4 are run with the longest FRI reduced to 350 years based on studies by Soja et al (2006), Kharuk et al (2008), Ivanova et al (2010), Kharuk et al (2011) and Ponomarev et al (2016). The results from these reduced FRI scenarios show that total mean biomass increased for both the reduced FRI fire scenario, to 86.11 tC ha⁻¹ compared to the previous mean of 82.06 tC ha⁻¹ with the eco-region-determined FRI, and for the altered climate with reduced FRI fire disturbance to 119.17 tC ha⁻¹ from the previous mean of 112 tC ha⁻¹ with the eco-region-determined FRI. Similar to the scenarios which used the eco-region FRI, the increase in biomass is from an increase in larch (figure S7). The similarity in results using a reduced FRI in comparison to those of the eco-region derived FRI suggests that competitive dynamics for resources (i.e. water resources), growth rate, and fire tolerance are key drivers determining biomass accumulation and species dominance.

This version of UVAFME does not contain a feedback between vegetation and fire, nor does it track the impact of fire on the soil layers. Many studies have suggested a relationship between regional fire dynamics, species composition, and fuels (Rogers et al 2015, Kukavskaya et al 2014, De Groot et al 2013, Berner et al 2012, Schulze et al 2012, Johnstone et al 2010a, Furyaev et al 2001). The implications of this connection suggest that if there is a shift in species across Russia, there may be a corresponding shift in fire regime. For such a shift in fire regime to occur there would first be a conversion of vegetation in response to altered climate or fire conditions. Shifts in previously predictable forest states in response to altered fire and climate regimes are already being observed in the boreal forests of North America (Johnstone et al 2010b) and Eurasia (Cai and Yang 2016). Feedbacks between the soil state, existing species, and fire severity determine the resilience and stability of the boreal forest (Johnstone et al 2010a, Johnstone et al 2010b). Future modeling work which adds a feedback between vegetation and fire by tracking changes in size and moisture of fuels and soil layers would help to better resolve this possibility while preserving the ability to simulate novel species combinations. It is expected that low to mid-severity fires, which create a thin organic layer promoting warmer soil temperatures and eliminating moss, would favor larch, as moss and thick duff are a barrier to regeneration (Alexander et al 2012, Sofronov and Volokitina 2010). In contrast, high severity fires which expose mineral soil are expected to favor the regeneration of deciduous species (Johnstone et al 2010b, Cai and Yang 2016). Given the results of this study we expect these additional interactions to maintain larch regeneration, except in cases of high severity fire, which may favor deciduous broadleaf species.

The shifts in species composition and stand structure predicted in this study have the capacity to impact both the local and regional energy budgets through changes in surface albedo, soil moisture, and thus heating/cooling through sensible and latent heat fluxes (Betts 2000, Chapin et al 2000, Liu et al 2005, Liu and Randerson 2008, Jin et al 2012, Rogers et al 2013), as well as the permafrost-C-climate feedback (Loranty et al 2016). Boreal forest composition and stand age have been demonstrated to have varying effects on net radiation and albedo, with magnitudes that have the capacity to affect climate locally and regionally (Amiro et al 2006, Liu et al 2005, O'Halloran et al 2012). Because larch is deciduous, the seasonal component and presence of snowfall will be important in evaluating the net effect of vegetation change on the energy budget, as this genus has a particularly elevated albedo in snowy conditions (Betts and Ball 1997, Hollinger et al 2010). These local- to global-scale radiative effects of pre- and post-fire forests within Russia emphasize the importance of characterizing the post-disturbance forest as well as the importance of using individual tree-based models to make predictions about future forest conditions.

Fire has long been recognized as a critically important driver of compositional and structural dynamics within the boreal region, and exploration of this disturbance driver becomes more important as fire is expected to become more frequent and more severe with warmer and drier climate conditions. Because UVAFME explicitly tracks individual trees of multiple species as they compete for resources through time, it can be used as a tool to evaluate the response of the boreal forest to fire under conditions of historical and altered climate. Results suggest that altered climate conditions allow for increased accumulation of biomass, but that fire disturbance creates a competitive situation in which the deciduous conifer larch maintain dominance across the region due to their high growth rate and tolerance to fire. Fire disturbance with altered climate conditions increases stand turnover creating a bimodal distribution of age cohorts across the region with a peak of mature older stands with a mean stand age of 280 years and productive maturing stands with a mean stand age of 100 years. The changes in biomass and age distribution are associated with a more complex distribution of tree height associated with increased turnover, growth, and stand density. These results suggest that species and tree-size level interactions are important in capturing the response of forests to changing climate and fire.

This work was funded by the NASA grants Carbon Cycle project NNX11AE39G and Land Cover and Land Use Change project NNX12AD34G and the UVA Global Programs of Distinction. The authors acknowledge Advanced Research Computing Services, specifically Katherine Holcomb, at the University of Virginia for providing computational resources and technical support that have contributed to the results reported within this paper.