Wildfire exacerbates high-latitude soil carbon losses from climate warming

2.1.1. Carbon uptake

2.1.2. Soil carbon decomposition

In this study, we initialized the model with equal seed carbon of five plant functional types (deciduous, evergreen, sedge, moss, lichen) across all grids. Soil attributes were obtained from large-scale gridded datasets to initialize the model. The clay and sand fraction, pH, cation exchange capacity, and bulk density in the model were initialized from the Unified North America Soil Map (Liu et al 2013) gridded to a 0.25° × 0.25° spatial resolution across vertical soil profiles. The SOC resolved across vertical soil profiles was initialized using data from the Northern Circumpolar Soil Carbon Database (Hugelius et al 2013). Temporally dynamic climate forcing and atmospheric CO₂ concentrations were used for historical and future simulations. Surface air temperature, precipitation, incoming shortwave radiation, relative humidity, and wind speed from 1980 to 1989 taken from the 3 h time-step North American Regional Reanalysis (NARR) (Wei et al 2014), linearly interpolated to 1 h, and cycled through 1900–1979 were used to spin-up the model. The earlier 10 years of NARR were selected to reduce the effects of amplified warming events during the later years on model spin-up. The full NARR time series was used to force the model from 1979 to 2018. Anomalies for future climate (through 2300) were derived from Representative Concentration Pathway 8.5 (RCP8.5) of the CCSM4 climate model. The CCSM4 was selected since it was shown to have a long-term (to year 2300) representation of the permafrost in the northern ecosystems and exhibited higher performance compared to current temperature and precipitation estimates (McGuire et al 2018).

Historical and projected stand-replacing fire time series were applied in ecosys (figure S2). The frequency of Alaska wildfire under past climate was modeled statistically using the Mean Fire Return Interval (MFRI) dataset of the LANDFIRE product (Rollins 2009), which estimates the average time between presumed past fire events (Mekonnen et al 2019). A temporal distribution of individual fire events was modeled in ecosys on the basis of stand-age-dependent fire-event-return intervals generated from a normal distribution of the base MFRI for each grid cell and the probability of fire occurrence in a grid cell was set to be dependent on the stand age as described in Mekonnen et al 2019. Increase in wildfire frequency under future climate was applied to the baseline to match projected changes in burned area. Projected changes in burned area over the 21st century were applied to the base MFRI using an estimated rate of increase of 350% by 2100 (Balshi et al 2009b). The burned area beyond the 21st century was set to increase with a rate normalized by changes in mean annual maximum temperature of CCSM4 (figure S1). All fire events were set as stand-replacing with prescribed fractions of combusted below- and above-ground biomass, SOC, nitrogen, and phosphorus. In addition to direct losses of nutrients from combustion, the model also simulates losses from leaching. We prescribed a depth of burn to 15.1 cm on the basis of observed mean depth of burn (median = 14.2 cm, and standard deviation = 6.2 cm) (Boby et al 2010, Turetsky et al 2010, Rogers et al 2014) from 235 burned sites across Alaska. Fire is projected to increase with warming and therefore in our simulation the depth of burn was set to linearly increase to a maximum of 26.5 cm (equivalent to the 95th percentile of data from the 235 burned sites) by 2100 and was maintained beyond 2100 through 2300. This increase in burned depth was prescribed to mimic expected increases in burn severity while maintaining a realistic range based on severe current fire conditions in the current Alaskan boreal forest. We conducted a sensitivity simulation in the absence of fire after the year 2000 while keeping all other model forcing and parameters the same, to partition the effects of fire under the future climate on SOC dynamics.

Here, we further evaluated modeled biomass (Mekonnen 2022) using a data-derived biomass product (Xu et al 2021) of Alaska (figure 2). The modeled biomass agreed well across the region with the data-derived biomass (geographically weighted regression, GWR, R² = 0.62). We also tested the modeled burned area, carbon emissions, and carbon combustion losses against three remote sensing and observation-derived products (Alaskan Fire Emissions Database (AKFED)) (Veraverbeke et al 2015), the Wildland Fire Emissions Information System (WFEIS) (French et al 2011), and the Global Fire Emissions Database version 4 with small fires (GFED4s) (Randerson et al 2012). The modeled annual burned area (5810 ± 185 km²), annual carbon emissions (14.4 ± 3.9 Tg C), and mean carbon combustion (2.47 ± 0.9 Kg C m⁻²) are within the uncertainty ranges of the datasets, except for the annual carbon consumption estimates from WFEIS, which is much higher compared to the model and the AKFED and GFED4s products (figure 2). We note that our modeled burned area was in the lower range of the observations, suggesting slightly conservative model estimates (figures 2(c)–(e)). The modeled carbon combustion values are also consistent with measurements at field sites (2.88 ± 0.23 kg C m⁻²) in interior Alaska (Rogers et al 2014).

Zoom In Zoom Out Reset image size

Plant biomass and SOC dynamics under changing climate and wildfire are linked. We first describe modeled plant changes and then discuss SOC changes and relevant plant-soil interactions. We found that warmer climate and higher levels of atmospheric CO₂ (figure S1) resulted in NPP increases in both Alaskan Arctic tundra and boreal forests by 2100 (figure 3). Gains in NPP and the transition from black spruce to fast-growing deciduous trees (aspen) in the boreal forest resulted in greater biomass. Plant NPP increases continued after 2100 and resulted in further biomass accumulation (figure 3(c)). The spatial mean of biomass modeled during 2100 (3691 gC m⁻²) and 2300 (4459 gC m⁻²) are within the ranges of currently observed biomass values from warmer areas in the boreal forest (figure 2(a)).

Zoom In Zoom Out Reset image size

The modeled gains in plant carbon are attributed to warming-induced enhanced carbon fixation rates and nitrogen availability from more rapid microbial mineralization driven by warmer soils, deeper active layers, and more rapid symbiotic and non-symbiotic N₂ fixation. These carbon fixation responses drive more rapid autotrophic and heterotrophic respiration and greater root nitrogen uptake (figure S3). We showed that, although there is substantial spatial heterogeneity in NPP and biomass changes over the 21st century and beyond, sustained increases in productivity integrated over Alaska are expected (figures 3(b) and (d)), suggesting that warmer climates will likely result in plant carbon gains in cold tundra and boreal forest systems.

Post-fire ecosystem carbon trajectories varied under different future climates (figure 4). On average across Alaska, modeled NPP recovered about 30 years after fire under recent climate vs. a more rapid recovery (about 15 years) in the warmer climate around 2100 (figure 4(b)). NPP was also shown to rapidly increase in the first few years following fires at the end of the 20th and 21st centuries and continued steadily for 50 years post-fire (figure 4(b)). Despite a rapid NPP recovery under future warming (e.g. after fire events during 2240–2250), we modeled a decline in NPP after about 20 years post-fire.

Zoom In Zoom Out Reset image size

Similar to NPP, post-fire SOC stock trajectories were modeled to vary with climate (figures 4(a) and (c)). SOC dynamics following a fire are affected by factors that determine soil litter inputs, carbon removal through combustion, and surface and subsurface losses. Residue carbon (total carbon on the soil surface from litterfall and in the soil profile from root litter and exudation, and microbial carbon) increased in the first year following fire events and then rapidly declined in subsequent years as a result of accelerated decomposition from warmer post-fire soil temperatures (figure 4(a)). Post-fire residue carbon may take more than 50 years to recover to pre-fire levels, with slower recovery rates from lower residue carbon inputs under current vs. warmer climates (figure 4(a)). Residue carbon pool was the largest under future climate from the greater modeled net primary production that resulted in more residue carbon inputs following fire. Similarly, post-fire humus carbon substantially declined under past and future climates (figure 4(c)). The decline in post-fire humus carbon was higher during fire events under warmer climates, which we partly attribute to greater combustion intensity with climate warming (Methods). During fires in 2250, the overall decline in humus carbon stock post-fire was modeled to be high, from previous frequent fire events and rapid decomposition of the top soil layer that reduced soil carbon stocks. Under all past and future climatic conditions, post-fire humus carbon declined concurrently with rapid decomposition of residue carbon (figure 4(c)). Further, humus carbon stocks did not recover within 50 years postfire. The most rapid decline of humus carbon was modeled under warmer climate conditions that led to accelerated decomposition rates. This result suggests that climate warming beyond 2100 under the RCP8.5 climate scenario may result in severe effects on SOC stocks following wildfires. This result is consistent with other studies that show severe fire events leading to higher combustion losses of carbon and slower recovery in the tundra and boreal ecosystems (Jin et al 2012, Holden et al 2016). Under future climate, these losses may also be exacerbated by greater fuel availability (Gaboriau et al 2020, Walker et al 2020) and plant litter inputs with greater NPP driven by warmer climate and elevated atmospheric CO₂.

The overall effects of fire on SOC (residue + humus), net ecosystem productivity (NEP = NPP- RH), and carbon losses through combustion and surface and subsurface transport of dissolved organic and inorganic carbon resulted in a post-fire net biome productivity (NBP = NEP−disturbance losses) of −2500 gC m⁻² yr⁻¹ averaged during fire events in years 2000–2010 (figure 4(d)). During this period, NBP recovered in less than 10 years following fires. We modeled that the fire-induced carbon losses were larger under warming beyond 2100 (e.g. postfire carbon losses were >4000 gC m⁻² during fire events in years 2090–2100 and 2240–2250), suggesting that wildfires under future climate may accelerate net ecosystem carbon losses. After a rapid initial brief recovery from net ecosystem carbon losses following fire events in years 2240–2250, we modeled that carbon losses continued for 50 years after fires, implying an adverse long-term impact of fire with further warming (figure 4(d)).

SOC dynamics across Alaska varied with soil depth (figure 5(a)). Enhanced carbon uptake under warmer climates resulted in an increase in biomass and thus litter inputs (figure 5(b)) and litter layer carbon stocks through 2300. Litter carbon stocks increased under warming until about 2200 and then remained relatively stable (figure 5(a)). Although a sustained increase in litter inputs may result in SOC accumulation in the litter layer, our modeling analysis shows that combustion losses and accelerated SOC decomposition offset increases in soil carbon (figure 5(a)). Rapid decomposition from warmer soil and permafrost thawing caused decreases in deeper SOC; the largest declines occurred between 0 and 100 cm, followed by those between 100 and 200 cm depth. The deepest soil layer (200–300 cm) had a very slight decline in organic carbon. Overall, the spatially averaged SOC losses were modeled to be the highest (about 88%) in the top 1 m of soil and lower in deeper soil layers (10% and 2% in the top 1–2 m and 2–3 m intervals, respectively). Over the entire soil column, the spatially averaged SOC losses were about 2800 gC m⁻², equivalent to 4.4 Pg C across Alaska by year 2300 vs. 2000.

Zoom In Zoom Out Reset image size

Accelerated decomposition from warming was modeled to increase heterotrophic respiration (figure 6(a)). Rapid decomposition is primed by increased root carbon exudation (figure S4) driven by increasing NPP. On the other hand, increased N₂ fixation and N mineralization lead to increased plant N uptake (figure S3), increased CO₂ fixation, and thereby plant growth and litter inputs to the soil. Changes in SOC stocks are spatially heterogeneous across Alaska's Arctic and boreal forest regions (figure 6(c)). Spatially aggregated as a function of the distance from the treeline, modeled changes in SOC stocks show a decline in most regions north and south of the treeline, with interior Alaska experiencing the largest decline (about 5500 gC m⁻²) in 2300 vs. 2000.

Zoom In Zoom Out Reset image size

High-latitude region carbon budgets are primarily limited by temperature and nutrients, thus climate warming deepens the active layer, relaxes temperature limitations on SOC decomposition, enhances nutrient mineralization, enhances N₂ fixation, and thereby increases CO₂ fixation rates. We projected sustained increases in ecosystem NPP that peaks around the year 2200 and then slightly decreases under further warming (figure 6(a)). Increases in RH were sustained through the year 2300, resulting in a decreasing NEP. Consistent with other models (Qian et al 2010), our modeled spatial average NBP shows that the Alaskan ecosystem remains a strong carbon sink during the 21st century. However, wildfire and climate warming beyond 2100 led to a sustained decline in spatial average NBP resulting in a net ecosystem carbon source after 2200 (figure 6(b)). We modeled that, over the 21st century, SOC carbon losses were offset by increases in plant biomass gains (figure 5(b)). This result is consistent with those from Mack et al (2021), who showed that forest stands dominated by deciduous trees offset soil carbon losses. However, we found that warming and wildfire beyond the year 2200 result in sustained declines in SOC and plant biomass accumulation (figure 5(b)).

We found from historical and future simulations that wildfire substantially decreases high-latitude SOC stocks. In a sensitivity simulation with no fire after year 2000, we modeled a decline in the spatial average SOC stock by 740 gC m⁻² in year 2300 vs. 2000 (i.e. a 74% lower SOC loss than when considering fire (2800 gC m⁻²); figures 5(b) and S5). This result suggests that climate warming in high-latitude regions may drive increased soil organic decomposition and thus soil carbon losses, even in the absence of fire, but that fire will exacerbate the losses and be a dominant controller of Alaskan SOC dynamics over the next several centuries. This result is consistent with other studies that have reported that high-latitude wildfires control annual and interannual carbon balances (Bond-Lamberty et al 2007, Balshi et al 2009a, Walker et al 2019, Chen et al 2021, Mack et al 2021).

Although climate warming may drive plant carbon gains and thus greater litter inputs to soil, increases in combustion losses from projected frequent and intense wildfire and accelerated soil carbon decomposition substantially offsets plant carbon gains leading to high-latitude ecosystems becoming a carbon source (figure 5(a)). Our results are consistent with other studies which showed that under warmer climate, projected increases in wildfire will lead to net losses of carbon to the atmosphere (Kasischke et al 1995, Walker et al 2019). We note that the relative magnitude of SOC losses depends on other factors that may affect high-latitude soil carbon dynamics which we were unable to consider in this study: e.g. uncertainties associated with our burned area estimates, mechanistic processes that lead to ignition, changes in flammability with vegetation type, topography, and other disturbances such as grazing, insects, and three-dimensional changes in landscape-scale hydrological dynamics that may lead to thermokarst development and ground collapse. Accounting for abrupt thaw and ground collapse may increase permafrost carbon losses (Turetsky et al 2020) offset potential carbon sinks (Turetsky et al 2020), and wildfire may exacerbate permafrost degradation and ground subsidence (Brown et al 2015, Zhang et al 2015, Holloway et al 2020). Our simulation did not account for these carbon losses from ground collapse following fire, implying that our soil carbon loss projections may be conservative.