3.2. Probabilistic Fire
When the probabilistic fire is activated, the model appropriately simulates stand composition, with pine and larch co-dominating the landscape and only a few dark conifers and hardwoods. Conifers comprise 90% of vegetation. Simulated 110–200-year-old forest was compared to the average conditions along the Angara River described by Kukavskaya et al. [29
]. The simulated basal area (32.5 ± 0.5 m2
) is in line with the observed basal area for this region (30–50 m2
]). The 110–200-year-old simulated forests (from bare ground) contain pine stands with 195 ± 19 m3
biovolume, larch stands with 112 ± 11 m3
biovolume and spruce and fir mostly in the understory and representing 28 ± 1 m3
and 9 ± 1 m3
, respectively. This compares well with the reported forest structure from Severny leshoz, wherein over 50% of pines are in the mature and over-mature stage and have an average biovolume of 210 m3
; larch stands were at an average age of 113 years at the time of observation with a biovolume of 138 m3
]. Spruce and fir stands contribute less to overall forest composition and are concentrated along river banks and on the upper slopes, respectively. Spruce and fir are mostly in the understory in the simulation and fairly equally dispersed across the simulated landscape, which represents 81-ha of the more upland territory.
In Severnyi leshoz, pine comprises approximately 50% of the stands, larch 29.3%, spruce 5.2%, fir 2.6%, and there are birches and aspen regenerating in burned or clearcut areas (<20%) [68
]. Correspondingly, simulated stands are comprised predominantly of pine and larch (Table 5
). The simulated stand average biovolume of dark conifers and hardwoods is smaller than observed. The former is likely due to not capturing the niches occupied by firs and spruce in the selected DEM section for the simulation. The latter, however, is likely due to two reasons: (1) not accounting for the timber harvest and management that has occurred on leshoz territory since 1958; and (2) the nature of the probabilistic fire subroutine.
The region of central Siberia along the Angara River is one of the most heavily disturbed areas by timber harvest. Hardwoods (birch and aspen) regenerate in the clearcuts, which are not represented in the simulation because SIBBORK does not have a timber harvest subroutine at this time. Additionally, the burning of the plots in this probabilistic fire subroutine is not synchronized, so some plots in the 81-ha simulation burn each year, creating small 10 m × 10 m gaps in the canopy. This only slightly suppresses the stand aggregate biovolume, and the shift of composition toward a higher contribution from birch and aspen is not observed. The 10 m × 10 m gaps are likely not large enough to promote regeneration of these shade-intolerant species. However, the presence of small gaps in the canopy likely promotes regeneration by growth of branches of trees already dominant in the area, or the promotion of trees from the subcanopy into the main canopy, rather than sprouting and promotion of young saplings into the sub or main canopies. This may explain the higher contribution from subcanopy species (fir, spruce) in the simulation, compared to observations. Otherwise, the simulated species composition compares well to stand characteristics reported by Gustafson and colleagues [17
] (Table 5
The annual area burned using the probabilistic fire mode under a historical climate (1.87% ± 0.47%, range: 0%–3.4%) compares well to some of the estimates suggested in the literature for central Siberian mixed forests (1.89%, range: 0.4%–4.7% [5
]) and for western and eastern Siberia in general (1.76%, Wirth, 2005; see Table 3
for others). The simulated area burned is somewhat larger than suggested by Furyaev [12
], who estimate that with a 50 ± 20 year MFRI 0.6% of total Russian forest area burns annually. However, estimates of annual area burned, whether across all of Russian boreal forests or within central Siberia, vary widely across literature and official government reports from the USSR [60
]. Figure 2
shows the wide array of estimates across several decades. Using the largest and smallest forest area burned estimates and the official estimate of total forest cover of 797 Mha in Russia [78
], 0.025% to 4.8% of forest burn annually. This is comparable to the estimate for central Siberia of 0.4%–4.7%/year [5
] and is captured by the simulated range. Scots pine forests in central Siberia and near the Angara river burn more frequently [12
], but the reported area burned may be significantly underestimated, because the fire regime across Russian boreal forests is dominated by ground fires, which are notoriously difficult to detect in satellite-based observations [5
] and may not be adequately represented in the data presented in Figure 2
. Although the crown-to-ground fire ratio varies from year to year in the simulation, over the long-term (>decade) approximately 20% are crown fires, which result in complete mortality on the plot, and 80% are ground fires, which selectively remove species from the plot, significantly changing the light environment on the affected and nearby plots. In some years, no crown fires are observed within the 81-ha simulated area, and over the course of the 300-year simulation, a few years have no fires at all. The simulated crown-to-ground fire ratio is in alignment with observed and estimated crown fire fraction: 16%–24% [32
], 22% [14
], 20% [3
], 33.6% [77
], 25% crown [2
], 6.5% crown [5
] and a similarly low estimate of 7% crown with 22% intense ground (may consume much vegetation and exhibit similar intensity as crown fires) [72
]. In larch- and pine-dominated forests, crown fires can comprise as much as 42% of total fires [77
3.3. Drought-Triggered Fire
Using the drought-triggered fire mode and historical climate conditions, the simulated fire regime is very similar to the probabilistic MFRI-based fire, because this region has not been experiencing significant drought conditions during the 20th century, except in pine stands on sandy soils (Figure 3
). Historically, droughts were experienced 1–3-times per decade during the 19th and 20th centuries in this region [15
]. Over the course of the 190 years of simulation under a historical climate, only 2–3 severe drought events occur with >70% of the growing season at or below wilting point that trigger more fires than in the probabilistic fire mode. Although warming has been observed in central Siberia over the last several decades, no appreciable changes in precipitation have been noted (Table 6
]). However, if precipitation is decreased by as little as 5% while maintaining the historical temperature conditions, drought does occur, and the simulated fire regime changes, along with forest structure and composition. This middle taiga ecosystem is likely to be affected by increases in temperature, because the increase in PET demands is likely to outpace any marginal increases in precipitation. In the simulation, drought events, with as much as 70% of the growing season at or below wilting point, begin to occur in 2060. Furthermore, when the drought-triggered fire mode is activated under a warming scenario, the annual area burned and the crown:ground fraction increase to 2.69% ± 1.87% (range 0.40%–7.89%) and 32.8%, respectively, by late 21st century (time average of years 2050–2100). This shifts forest composition toward greater contribution of hardwoods (Betula
spp.), especially in areas where gaps are formed the size of multiple plots. Simulated post-fire regeneration is predominantly comprised of pine and hardwood saplings which compares well with observations from central Siberia [58
Using the probabilistic fire regime, the burn areas are randomly distributed and only small gaps are formed in the canopy. This facilitates the growth of canopy dominant larch and pine to close up the small gaps, and suppresses the dark conifers (fir, spruce) in the understory. This is shown by an increase in DBH of the former and a decrease in the latter over the course of 1–3 decades following a fire event on a plot. In the central Siberian Angara region, few droughts occur under a historical climate. Historical droughts affect multiple plots, and in the drought-triggered mode, larger fires are generated in the simulation during those years. Regeneration of hardwoods occurs following drought-triggered fires, but is negligible in the probabilistic fire mode, where only small, 10 m × 10 m gaps are generated in the forest.
Under a warming scenario not accompanied by changes in precipitation (Figure A1
c), droughts become more frequent and severe (affecting more plots) after the year 2050. Prior to the onset of more frequent droughts, hardwoods regenerate in gaps, whereas after 2050 pulses of fir and spruce saplings regenerate and release occur following drought-triggered fire in the simulation. The average (Lorey’s) height of canopy dominant larch and pine decreases between the years 2050 and 2100. The average DBH and leaf area of cedar (Pinus sibirica
) spruce, fir, and larch decrease significantly during the second part of the 21st century, and dark conifers almost disappear from the simulation by the year 2070. These simulated structural and compositional changes are similar to the observed suppression of canopy-dominant pine that facilitates a release of subcanopy and understory spruce and fir trees following an increase in area burned in recent decades [6
]. When severe droughts begin to occur more frequently (2–3-times per decade), all species become suppressed. The stand basal area decreases by 20% between 2050 and 2070. The change in forest structure (more gaps, leading to more light within and below the canopy) and decrease in competition releases pine (accelerated rate of DBH increase), as well as promotes new growth and its foliage to fill in the gaps (increase in stem density and leaf area). However, these pines experience stunted growth, with stand biovolume decreasing by almost 50% by the year 2100. The disappearance of dark conifers following intensification of the fire regime has been suggested by Furyaev [12
] and others (Gorbachev and Popova, 1996; Sannikov and Goldammer, 1996) [48
]. Furthermore, field observations show that under drier (soil) conditions and following an increase in fire activity (frequency and intensity), forested ecosystems do not regenerate. Even in central Siberia, some areas that are repeatedly burned are unable to regenerate as forest, and transition to grassland/meadow or steppe [28
3.6. Drought-Triggered Fire
When the drought-trigger is activated, average annual area burned under a historical climate increases to 0.30% ± 0.12% (range: 0%–0.77%), with 31% of those as crown fires. This range still encompasses the observed multidecadal average (0.25% [56
]). However, reports of area burned vary widely for this region [83
], and in severe fire years (e.g., 2002) it is estimated that as much as 9% of this region burns (steppe and forest included) [84
]. This value cannot be compared directly to simulated area burned, because the simulation only includes forested area, except that we expect the simulated area burned in severe fire years to be significantly less than 9%, because the predominant vegetation that burns in this region is the steppe, which is not included in the simulation. An increase in crown fire proportion is appropriate in drought conditions, because the vegetation is drier and facilitates ignition and higher severity burning. The slight increase in annual area burned in this mode is representative of the specific forested region near the southern boreal zone boundary, rather than the average conditions in this region, much of which is covered by steppe.
In order to reproduce the significantly larger crown:ground fire ratio during severe fire years, the average fraction of crown fires is increased to 35%. This fraction was determined iteratively and results in 20%–53% crown fires during severe fire years, and thus in the simulation, as in observations, the crown:ground fire ratio can be as high as 50% during severe fire years. It is possible that the crown ratio is overestimated, because “drought” is defined as a year during which at least one tree on a plot is stressed from lack of soil moisture. Based on this definition, under a historical climate, drought occurs in most years in the simulation of the southern taiga, but only during a few years in the simulation of middle taiga.
Interestingly, SIBBORK appropriately reproduces the distribution of vegetation (species and biovolume) along both the north- and the south-facing slopes (Figure 4
]. The model appropriately represents drought, and simulates differential changes in soil moisture with increase in temperatures not accompanied by appreciable changes in precipitation. Furthermore, drought-triggered fires ignite in areas of drought on the landscape and in the appropriate forest types: larch- and pine-dominated stands, but not usually in more humid dark conifer forests. The occurrence of higher-intensity crown and high-intensity ground fires is also appropriately larger in pine and larch stands [83
] and on south-facing slopes. The return of fires is more frequent to the south-facing slopes. The synchronization of the environmental conditions and the fire regime is not explicitly coded and is an emergent property of the SIBBORK model coupled to the DISTURB-F subroutine in drought-triggered mode.
Application of the drought-triggered fire regime to simulation of potential future climate based on linear extrapolation of observed warming (Figure A1
d) in the region along the southern boundary of the boreal zone results in greater incidence of drought, both on the south-facing slope and at lower elevations along north-facing slopes. Under this warming scenario, vegetation shifts upslope on both aspects, and the ribbon forests on the south-facing slopes decrease in vigor and biovolume, which is in accordance with observations [38
] and previous modeling studies [47
]. Warming ambient temperatures increase the PET demands on vegetation, but if no concurrent increase in precipitation occurs, vegetation becomes stressed and either dies from temperature-based drought stress or more easily succumbs to mortality from insects, fire, pathogens, or windthrow. In a warmer climate, both the annual area burned and fire intensity increase in southern taiga. Using the model, the annual area burned in 2100 is forecast to be 0.36% ± 0.16% (range: 0.2%–0.9%), with 35% as crown fires averaged across the mountain. This corresponds to 45% of forest fires occurring on the south-facing slope and 55% on the north-facing slopes, with 42% of the fires on the south-facing slopes as crown fires compared to 23% on the north-facing slope. Note that fire is only simulated on plots with trees, and fewer plots on the south-facing slope are populated by arboreal vegetation. Most of the plots on the south-facing slope do not have trees and correspond to steppe at lower elevations (below 1400–1500 m a.m.s.l.) and alpine tundra at higher elevations (above 2100–2200 m a.m.s.l.). Therefore, although only 45% of the simulated fires occur on the south-facing slope, the density of fire occurrence within the ribbon of forest on the south-facing slope is greater than on the north-facing slope.
Increases in annual area burned have been forecast and observed during recent severe fire years [18
]. Figure 5
shows the changes in distribution of fire along the idealized mountain from the historical climate fire regime to what the fire regime may resemble in 2100 if the warming trends observed during the last 25 years continue without appreciable changes in precipitation. Spatial patterns begin to appear in the fire distribution when the drought trigger is activated, with the forests on south-facing slopes more heavily damaged by fire. Crown fires tend to cluster on the south-facing slopes and lower north-facing slopes. Although Groisman et al. [21
] point out that no appreciable changes in precipitation have been observed in southern taiga over the last 50 years, Tchebakova and Parfenova [65
] argue that 4%–8% less precipitation is currently received in the boreal forests along the southern boreal boundary (Sayan-Altaj mountain complex and Tuva Republic) compared to the historical average. Decrease in precipitation will further drought-stress the vegetation, likely increasing the frequency of fire and the crown-to-ground fire ratio, and potentially facilitating a biome shift from forest to steppe on the south-facing slopes mid-21st century.
Recent field-based observations have found that the forests in southern Siberia are losing their ability to regenerate post fire and other landscape disturbances under a warming climate [29
]. The changes in the climate and disturbance regimes are not just shifting the southern taiga ecotone northward into what was historically middle taiga [56
], but the forest-covered area, both in middle and southern taiga, is decreasing [59
]. Suppressed forest regeneration was also apparent in our simulation studies.
Post-fire vegetation dynamics depend on fire size and intensity. Stand structure and, therefore, light geometry, differ significantly following a ground fire versus a stand-replacing crown fire, both in the field and in the simulation. Resistor trees (see Table 1
) survive the low-intensity ground fires, and the regeneration that follows facilitates the establishment of shade-tolerant coniferous saplings. After a stand-replacing, high-intensity crown fire, however, sufficient light environment facilitates sprouting of shade-intolerant broadleaf or deciduous conifer (larch) species, which matches observations from the boreal forest [8
]. High-intensity crown fires are often followed by regeneration of even-aged cohorts of shade-intolerant species (birch, aspen, larch), whereas low-intensity ground fires allow regeneration of cohorts of shade-tolerant (dark conifers) and somewhat-shade-tolerant species (pine) in the understory [17
]. Stand productivities along these two trajectories are vastly different, with shade-tolerant conifers exhibiting lower productivity.
Overall, the model appears to simulate fire in the appropriate locations on the landscape, and synchronicity is observed between environmental conditions and the fire regime. There is coupling between increased drought and increased occurrence of fire, as well as increased proportion of crown fires in the simulation alongside warming of ambient temperatures without appreciable changes in precipitation. We did not modify the MFRI or the crown-to-ground fire fraction for the future climate over the course of each simulation. Instead, we let the changes in the environmental conditions (temperature, precipitation, PET, soil moisture, growing season duration, fraction of growing season in drought) to affect the probability of ignition through the drought-trigger, allowing any changes in the fire regime to arise as emergent properties of the simulation. The growing season is determined based on simulated temperatures and, therefore, increases in duration under a warming scenario. Correspondingly, any drought that occurs within the new, longer growing season, can serve as a drought trigger for fire events, consequently lengthening the fire season, as has been observed in central and southern Siberia [28
]. As most of the precipitation is received during the summer months, increased drought and, therefore, fire, are observed during the spring and fall months within the growing season, which aligns with observed changes in the fire regime in southern and middle taiga [5
]. Although tree mortality in SIBBORK is assessed at an annual time step, drought conditions are computed for each month, capturing an increase in spring and autumn drought conditions. This is another example of emergent behavior of the coupled SIBBORK-DISTURB-F model.
Modeling facilitates the study of forest-fire dynamics and interpolates where the satellite record does not provide coverage due to cloud cover, smoke, insufficient resolution, or lack of fire activity during satellite overpass, which can account for as much as a third of satellite imagery over Eurasian boreal forests [5
]. Most importantly, however, we have no observations of the future, and the simulation of changes in forested ecosystems due to shifts in climate and fire regimes can facilitate insight into climate-fire-vegetation feedbacks and how the landscape-scale balance of carbon may change in the near future. Modeling results can be used to prioritize allocation of efforts toward conservation, mitigation, and forest management. For example, the effect of fires on the carbon balance can be estimated by using the simulated accumulated above- and below-ground biomass to estimate the amount of carbon sequestered from the atmosphere, and published estimates for emissions from fires [6
]. Conard and her colleagues [6
] estimated that 22.5 tC are emitted for each hectare burned in Eurasia via a high-intensity crown fire, 8.6 tC·ha−1
emitted in moderate-intensity ground fires, and 2.3 tC·ha−1
are emitted in low-intensity surface fires. These emission estimates can be used to convert area annually burned in the simulation to the amount of carbon released to the atmosphere during the associated simulation year. Using the historical climate coupled with drought-triggered fire mode, the amount of carbon released into the atmosphere would be approximately 12 tC·ha−1
in middle taiga and 4 tC·ha−1
in southern taiga during an average fire year; 53 tC·ha−1
and 6 tC·ha−1
, respectively, in more severe fire years, based on the above estimates. Alternatively, using the simulated biomass that is removed from the simulation following a fire, we estimate that in southern taiga carbon emissions during the second half of the 21st century will be double the historical amount. This is an underestimate, because the carbon-rich organic floor is not considered in this calculation.
Although modules that simulate the spread of fire across the landscape exist (e.g., LANDIS [90
]), they either require atmospheric data at the sub-daily time step for parameterization or couple to models that simulate vegetation as cohorts or plant functional types (not individual trees). SIBBORK-DISTURB-F presents a platform for investigation of the non-linear effects of climate on the fire disturbance regime and post-fire regeneration dynamics in the boreal ecosystem. Changes in stand structure or species composition will affect the carbon storage capacity of the ecosystem, as well as the way the ecosystem interacts with the atmosphere (through carbon exchange, albedo, radiation budget, etc.). Simulated forest structure and composition can be used to estimate carbon emissions due to fire disturbance and the role the boreal forests may play in the global carbon cycle in the 21st century and beyond.
The SIBBORK vegetation model allows us to examine changes in vegetation structure and composition due to two concurrently changing regimes: climate and fire. Previous simulations estimated the disappearance of expositional larch forests on south-facing slopes in southern taiga by the year 2080 [47
] and the replacement of the southern half of the Siberian boreal forest by steppe by the year 2080 [91
], but those investigations only included the effect of climate on vegetation and did not consider disturbance. When concurrent effects of climate and fire are considered, our simulation studies show that significant portions of the forested landscape in mountainous southern taiga will transition to non-treed vegetation (steppe) by the year 2060. Upslope colonization is significantly slower than mortality from stress and fire within the historical range of these expositional larch forests. When only arboreal vegetation is considered and carbon emissions are computed based on above-ground biomass killed by the fires, our simulation shows that in the second half of the 21st century, the carbon emissions will double compared to the historical (1955–1990) emissions. Moreover, the steppe ecotone has a different fire regime with an annual fire return interval. The shift from the forest biome to a steppe changes the capacity of the landscape to sequester and store carbon, and the more frequent fire return interval changes its carbon emissions profile. These post-fire changes in vegetation can have further feedbacks to the regional climate, potentially increasing the drought risk due to decreased evapotranspiration from steppe compared to forest.
Furthermore, we applied SIBBORK in tandem with the new fire module DISTURB-F toward the investigation of how the middle taiga ecotone in central Siberia may be affected by the climate-driven changes in the fire regime. In our simulation, temperature-induced drought stress and repeat fires reduced the stand basal area by 20% during the 2050–2070 time frame and the biovolume by 50% by the year 2100. These are conservative estimates for changes in forest structure, because other disturbances, such as insect outbreaks, timber harvest and pathogens, are not considered in the simulation. Additionally, the linear extrapolation of observed warming trends may underestimate the climate changes this region is likely to experience in the near future.
These results are particularly timely due to the recent COP 21 meeting in Paris and the agreement to “cap” the global warming at +2 °C. Russia has been experiencing warming on the order of 2.5-times faster than the global average [22
], and the continental locations in Siberia have not experienced concurrent increases in growing season precipitation. Extrapolation of the observed warming rates from the 1990–2014 time frame forward shows that central Siberia is likely to experience +2 °C warming by the year 2040, while southern Siberia is likely to reach this state by the year 2050 (and on south-facing slopes, even sooner), compared to the historical (1955–1990) baseline [22
]. Increases in temperature not accompanied by concurrent increases in precipitation result in greater Potential Evapotranspiration (PET) demands on vegetation, resulting in temperature-induced drought stress. Stressed or dry vegetation is easier to ignite and facilitates the spread of wildfires. In this manner, climatological changes alter vegetation health, structure and composition, which can result in the modification and potential intensification of the fire regime (return interval, intensity, extent). Furthermore, repeat burns often result in suppression of tree sapling regeneration, and what used to be a forested landscape shifts to grassland/meadow or steppe [28
]. This biome shift, facilitated by changes in the fire regime, significantly alters the role of the landscape in the regional carbon cycle. Because Siberian forests cover millions of hectares and are experiencing accelerated change, alterations in the forest structure and composition and any biome shifts are likely to affect the global carbon cycle.
Large regions of southern taiga are likely to experience forest collapse and shift to a steppe ecotone within the next half century [91
]. Increased tree mortality often results in standing dry snags, which contribute to the fuel load and may lead to more extensive spread of wildfires in the area. The carbon dynamics within the southern tree line ecotone are changing and are likely to shift more toward that region acting as a source of atmospheric carbon. The SIBBORK-DISTURB-F simulation model can be used to inform the mitigation and possible shifts between economy sectors that may become necessary to assure adaptation of the local populations and those that depend on the timber resources of southern taiga.