CLM5, a land component of the NCAR Community Earth System Model (version 2.0.1), is a grid-based computational model (Lawrence et al., 2019). Each grid cell is comprised of sub-grids that represent the land cover type (i.e., glacier, lake, wetland, urban, and vegetated). The 17 plant functional types (PFTs) are represented in the vegetated land cover. The model represents the instantaneous exchange of energy, and water and momentum were simulated between land and atmosphere across a variety of spatial and temporal scales at the sub-grid level. Furthermore, hydrological processes including evapotranspiration, surface runoff, sub-surface runoff, streamflow, aquifer recharge, and snow are simulated at the sub-grid level. When the BGC module is adopted (i.e., CLM5-BGC), the carbon and nitrogen cycles and seasonal vegetation phenology are simulated for the atmosphere, vegetation, and soil organic matter at the PFT level. These cycles, which are linked to climate, land cover and land use, fires, and atmosphere CO2 level, affect other cycles such as hydrological cycles and energy fluxes.

In CLM5-BGC, fire is simulated based on a process-based fire parameterization developed by Li et al. (2012). There are four types of fire in CLM5-BGC: non-peat fire, agriculture fire, deforestation fire, and peat fire. For non-peat fires, the number of fire ignitions is calculated as the sum of natural and anthropogenic ignitions. The estimation of natural ignition sources is based on the NASA Lightning Imaging Sensor (LIS)/Optical Transient Detector (OTD) lightning frequency datasets. The frequency of cloud-to-ground lightning that ignites fires is estimated with the latitudinally varying ratios of the total lightning frequency obtained from remotely sensed data (i.e., LIS/OTD), which include two different types of lightning, i.e., cloud-to-ground and the cloud-to-cloud lightning. Furthermore, the ignition source from human activity is calculated based on the human population density. The fire spread rate is then calculated by considering wind speed and vegetation condition (Arora and Boer, 2005). Socioeconomic influences are parameterized using gross domestic product (GDP) and population density, which means that higher populated and more developed regions will have a better fire suppression capacity.

In CLM, the burned area is calculated at the grid level, and the fire emissions are calculated at a PFT level. Once a grid-level burned area is calculated, the same fractional area burned is imposed on each PFT in the grid. The PFT-level carbon emission from the fire is calculated as follows (Li et al., 2012): 1CE=A⋅C⋅CC, where CE is the carbon emission; A is the fractional area burned; C is a vector with the carbon density of leaves, stems, and roots, carbon transfer, and carbon pools; and CC is the corresponding combustion completeness factor vector.

Leaves and roots may be damaged in burned areas, which reduces their carbon-capturing productivities (Reyer et al., 2017; Seo and Kim, 2019; Swezy and Agee, 1991). In CLM5-BGC, the amount of leaf carbon to litter (Ψ) caused by fire is calculated as follows (Li et al., 2012): 2Ψ=Abfi⋅Ag⋅Cleaf(1-CC)⋅M, where Ab is the calculated burned area, Ag is the area of the grid cell, fi is the fraction of coverage of each PFT, Cleaf is the amount of leaf carbon, and M is the mortality factor vector for each PFT. The leaf area index (LAI) is recalculated based on the adjusted amount of leaf carbon. In addition, the methods by which the amount of carbon in live stems, dead stems, and roots and the storage pool is adjusted due to fires are similar to those mentioned above.

Leaf area controls canopy evaporation and transpiration as well as carbon fluxes (gross primary production (GPP), net primary production (NPP), net ecosystem production (NEP), and net ecosystem exchange (NEE)). NEE, which represents the total carbon fluxes between an ecosystem and the atmosphere, is calculated by using the NEP and carbon emissions from wildfires. The equations for these carbon fluxes are as follows: 3NPP=GPP-Rp,4NEP=NPP-Rh,5NEE=-NEP+CE, where Rp is plant respiration and Rh is heterotrophic respiration.

Because hydrological processes are highly linked to vegetation dynamics, fire processes may affect not only water cycles but also ecosystem products (Jiao et al., 2017). For instance, the water cycles on land surfaces, such as partitioning of ET, are affected by fires because the fire changes leaf area in ecosystems (Netzer et al., 2009; Park et al., 2020; Seo and Kim, 2019; Wang et al., 2019). More details on CLM5-BGC processes, including the equations for leaf phenology, hydrology cycles, fires, and carbon cycles, are described in Lawrence et al. (2019).

In this study, we focused on Alaska (61–70∘ N, 200–218∘ E) and Eastern Siberia (61–70∘ N, 130–148∘ E), which are located at northern high latitudes (Fig. 1). Both domains have the same size and latitudes. The average temperature based on Climate Research Unit (CRU)–National Centers for Environmental Prediction (NCEP) reanalysis data (2001–2012) is -5.11 and -15.28 ∘C in Alaska and Eastern Siberia, respectively. The average annual snowfall and rainfall are 83 and 218 mm in Alaska and 92 and 208 mm in Eastern Siberia, respectively.

There are differences in vegetation types in these regions, based on MODIS (Lawrence and Chase, 2007) (evergreen trees: 26.4 %, deciduous trees: 1.6 %, shrub: 28.5 %, grass: 34.5 %, crop: 3.9 %, and bare ground: 5.1 % in Alaska; evergreen trees: 1.2 %, deciduous trees: 14.9 %, shrub: 45.8 %, grass: 29.7 %, crop: 1.4 %, and bare ground: 7.1 % in Eastern Siberia). In summary, the tree fraction is higher in Alaska (28 %) than that in Eastern Siberia (16.1 %), and the fraction of low vegetation (i.e., grasses and shrubs) is lower in Alaska (63 %) than that in Eastern Siberia (74.5 %). Notably, the largest areas were the natural vegetation and crop land units, and the lake, urban, and glacier land units occupied less than 1 % in both regions.

In this study, we designed two sets of experiments to investigate the impact of burned area using fire simulation based on the study by Li et al. (2012) (i.e., CLM-Default) and satellite observations from GFED4 (i.e., EXP-GFED4) over Alaska and Eastern Siberia. Figure 2 shows the experimental process of this study. Our simulations started with a pre-existing initial condition state for the year 2000 at a 1.9∘ × 2.5∘ spatial resolution provided by NCAR. Because starting a new simulation at a different spatial resolution could introduce model artifacts, we ran CLM5-BGC at a 0.5∘ × 0.5∘ spatial resolution from the initial state, including the land use, such as cropland, for 200 years for the equilibration with repeated use of the Climate Research Unit (CRU)–National Centers for Environmental Prediction (NCEP) reanalysis climate data for 1981–2000. Then, CLM-Default and EXP-GFED4 were simulated for 12 years (2001–2012) at the 0.5∘ × 0.5∘ spatial resolution using CRU–NCEP atmospheric forcing, which includes precipitation, temperature, wind speed, surface pressure, specific humidity, longwave radiation, and solar radiation. While burned areas were simulated based on Li et al. (2012) in CLM-Default, the GFED daily burned area over the arctic region was directly inserted into CLM5-BGC in EXP-GFED4, with the daily data being equally divided into a half-hourly model time step (Seo and Kim, 2022).

In this study, we compared the carbon and water fluxes in CLM-Default and EXP-GFED4. In particular, carbon emissions and the NEE were evaluated using GFED4, AKFED, and GEOS-Carb CASA-GFED. Additionally, we analyzed the impacts of fire on carbon fluxes according to the distribution of PFT. Furthermore, comparisons of water fluxes such as ground evaporation, canopy evaporation, canopy transpiration, and soil moisture at grid level were performed to reveal the impacts of fire on water cycles.

GFED4, which is based on satellite data such as MODIS and the Tropical Rainfall Measuring Mission Visible and Infrared Scanner, provides gridded data on the global burned area, fire persistence, land cover distribution, and fractional tree cover distribution of burned areas, among others (Giglio et al., 2013). The data are provided at a 0.25∘ × 0.25∘ resolution and daily and monthly temporal resolutions. Furthermore, details on fire impacts, such as carbon emissions, dry matter emissions, biosphere fluxes (NPP, heterotrophic respiration), and emission factors data are included. The carbon emission data are based on burned areas and the Carnegie–Ames–Stanford Approach (CASA) carbon-cycle terrestrial model for each month. In this study, daily burned-area data from GFED4 were incorporated into CLM5-BGC, and monthly scaled carbon emission data from GFED4 were used to evaluate the model performance (Table 1).

We also used data on Alaskan carbon emissions from AKFED (Veraverbeke et al., 2015a) to evaluate the model performance for carbon emissions in Alaska (Table 1). Veraverbeke et al. (2015a) developed a statistical model to calculate the carbon consumption in Alaska between 2001 and 2012. They employed environmental variables such as elevation, slope, and day of burning to calculate ground-level carbon consumption. In addition, pre-fire tree cover and differenced normalized burn ratio are used to predict aboveground carbon emission. Veraverbeke et al. (2015a) estimated that the highest carbon emission was 69 Tg C in 2004, and the annual carbon emission was 15 Tg C.

We used monthly NEE products from GEOS-Carb CASA-GFED for 2003 to 2012 to evaluate the performance of EXP-GFED4 and CLM-Default. However, the definitions of NEE according to CLM5-BGC and GEOS-Carb CASA-GFED are quite different. In CLM5-BGC, the NEE is the final carbon flux between an ecosystem and the atmosphere. Thus, the carbon flux of burning was included when calculating the NEE (Eq. 5), but it was excluded in GEOS-Carb CASA-GFED. To unify the definition of NEE, we redefined the NEE in GEOS-Carb CASA-GFED as follows. 6NEE=NEEge+FireE+FuelE, where NEE is the total carbon flux between land and atmosphere including emission due to fires, NEEge is the value of NEE according to GEOS-Carb CASA-GFED, FireE is the wildfire carbon emissions, and FuelE is the carbon emissions from wood-fuel burning in GEOS-Carb CASA-GFED.

We first evaluated the performance of estimating burned areas in Alaska and Eastern Siberia from CLM5-BGC (i.e., CLM-Default) and compared it to that of GFED4 (Fig. 3). While an average of 0.42 Mha of burned area from 2001 to 2012 was observed in Alaska, the average annual burned area was estimated at 0.24 Mha in CLM-Default (Fig. 3a). In Alaska, there were large discrepancies in burned areas for 2004, 2005, and 2008 between GFED4 and simulation results. More than 1 Mha of burned area existed for 3 years (2004, 2005, and 2009), which is remarkably different from that of the other years. Studies suggested that these large burned areas were associated with a high lightning frequency and drought (Littell et al., 2016; Veraverbeke et al., 2017; Xiao and Zhuang, 2007). However, this phenomenon was not captured in CLM5-BGC, which predicts relatively constant annual burned areas. In contrast, the burned area was dramatically overestimated in Eastern Siberia (Fig. 3b). While an average of 0.29 Mha of burned area was observed, the average annual burned area was estimated at 2.14 Mha with CLM5-BGC. Although the GFED4 burned area in Eastern Siberia did not vary significantly over time, the simulated burned area increased from 2001 to 2012 at a rate of 0.33 Mha yr-1.

Figure 4 shows the spatial distribution of the burned areas of GFED4 and CLM-Default in 2004 over Alaska. The number of grid cells (0.5∘ × 0.5∘) in GFED4 where the burned areas exceeded 0.01 Mha in 2004 was more than 50. In contrast, there were two grid cells with more than 0.01 Mha of burned areas simulated using CLM5-BGC in Alaska (Table 2). Table 2 shows that CLM5-BGC has a limitation in simulating large burned areas in Alaska. Small fires were simulated in more grid cells, and the simulated burned areas were more widely distributed in CLM-Default than those in the GFED4 products.

We compared the carbon fluxes of CLM-Default and EXP-GFED4 to understand the impacts of fire on high-latitude regions (Figs. 5 and 6 and Table 3). The average carbon emissions were 11.87 and 21.11 Tg yr-1 in CLM-Default and EXP-GFED4 in Alaska, respectively, and 20.48 and 3.24 Tg yr-1 in CLM-Default and EXP-GFED4 in Eastern Siberia, respectively (Table 3). As expected, there were large differences in carbon emissions in CLM-Default and EXP-GFED4 in both regions because the simulated carbon emission was directly linked to burned areas. In the model, carbon emissions had a strong correlation with burned areas in both regions (Alaska: 0.99, Eastern Siberia: 0.89).

Furthermore, the simulated Alaskan annual carbon emissions for CLM-Default and EXP-GFED4 were evaluated with AKFED carbon emission datasets and GFED4 (Fig. 5a and Table 4). The correlations of annual carbon emission between simulated carbon emissions (CLM-Default and EXP-GFED4) and GFED4 were 0.3 and 0.99, respectively. Moreover, the correlations between the simulated carbon emissions and AKFED carbon emissions were determined (CLM-Default: 0.31; EXP-GFED4: 0.96). While the root mean square error (RMSE) between the simulated carbon emissions and the AKFED carbon emissions decreased after applying the GFED4 burned area (CLM-Default: 20.48 Tg yr-1; EXP-GFED4: 10.98 Tg yr-1), the RMSE between the simulated carbon emissions and the GFED4 carbon emissions increased (CLM-Default: 11.02 Tg yr-1; EXP-GFED4: 20.93 Tg yr-1). This is because average carbon emissions for GFED4 were 8.36 Tg yr-1 and are relatively lower than carbon emissions in EXP-GFED4 and AKFED. The combustion completeness factor for leaves is 0.8, and that for stems ranges from 0.27–0.8, depending on the PFTs in CLM5-BGC. According to van der Werf et al. (2010), the combustion completeness factor of aboveground live biomass, which ranges from 0.3–0.4 in the boreal region, is lower than that in other regions. Therefore, the combustion completeness factors for boreal trees may be lower than the current default value in CLM5-BGC.

The carbon emission simulation was highly improved after replacing the fire simulation with GFED4 in Eastern Siberia (Fig. 5b); the correlation was improved from 0.41 in CLM-Default to 0.88 in EXP-GFED4, and the RMSE was reduced from 19.74 Tg yr-1 in CLM-Default to 4.2 Tg yr-1 in EXP-GFED4, compared with the GFED4 products. In Eastern Siberia, grasses are dominant, suggesting that the value of the combustion completeness factors for grass in CLM5-BGC is more similar to those of GFED4 products than to those of boreal trees.

Unlike carbon emissions, the regionally averaged GPP, NPP, and NEP (Fig. 6c–h) did not significantly change in EXP-GFED4. The differences in GPP, NPP, and NEP are less than 3 %, indicating that fires rarely impacted carbon fluxes related to vegetation and decomposition. This is because the ratio of the burned area to the total area was relatively small. For example, the highest annual burned area of all simulations was 6 Mha, which accounted for 6.87 % of our study domain. The simulated LAIs in Alaska and Eastern Siberia are presented in Fig. 6a and b, respectively. In Alaska (Fig. 6a), the difference in LAI between CLM-Default and EXP-GFED4 was the largest in 2005 (0.03 m2 m-2). Although the difference in burned area between CLM-Default and GFED4 (Fig. 3a) was the largest in 2004, the largest difference in LAI was in 2005 since vegetation damage caused by fire in 2004 had not fully recovered, and the difference in burned area in 2005 was also quite large. In Eastern Siberia (Fig. 6b), the difference in the simulated LAI between CLM-Default and EXP-GFED4 has been large since 2009, when the difference in the size of burned areas was amplified (Fig. 3b). Although the LAI, which affects primary GPP and other carbon fluxes, was reduced by fires, the LAI after fires was not substantially different owing to the small burned area compared to the total area.

However, NEE, which represents the net carbon fluxes between land and ecosystem (Eq. 5), was largely affected by fires, unlike other fluxes such as GPP, NEE, and NEP (Fig. 5c and d). NEE changed significantly with forcing of GFED4 into the model when the discrepancy in the burned area between CLM-Default and EXP-GFED4 was remarkable. Moreover, the NEE results for EXP-GFED4 and GEOS-Carb CASA-GFED had similar tendencies. For instance, we found that the net carbon in Alaska was emitted from land ecosystems to the atmosphere (i.e., positive NEE) in 2004, 2005, and 2009 in EXP-GFED4 and GEOS-Carb CASA-GFED, but it was absorbed (i.e., negative NEE) in CLM-Default. Although there was a change in NEE due to burned areas in Siberia, it was not as pronounced as that in Alaska.

The results of the carbon fluxes at the grid level in Alaska and Eastern Siberia are investigated in Fig. 7, which shows the difference in carbon fluxes and burned areas between CLM-Default and GFED4 in Alaska for 2004 and in Eastern Siberia for 2012. As expected, the response of GPP, NPP, and NEP to fires were nonsignificant. However, fires significantly altered carbon emissions and the NEE in both regions, which can further alter the atmospheric carbon dioxide concentration and even climate. This suggests that high-latitude fires may influence the carbon sink or uptake markedly. Phillips et al. (2022) reported that boreal forest fires, which are largely distributed at high latitudes, make a significant contribution to releasing greenhouse gases. With an ESM combined with CLM5-BGC, the prediction of atmospheric carbon may become uncertain due to the limited performance of fire prediction models.

To investigate the fire impacts on water fluxes, we compared the results of ET and ET components, such as canopy evaporation, canopy transpiration, and ground evaporation, in six grid cells where the differences in burned area between CLM-Default and EXP-GFED4 are the largest in Alaska and Eastern Siberia (Fig. 8). Because the LAI decreases owing to wildfires, canopy evaporation and canopy transpiration decrease in the burned areas.

We observed that more rainfall reaches the ground, which would make the ground evaporation rate higher in regions with more burned areas, especially in 2004 and 2005 in Alaska. The differences in annual canopy evaporation, canopy transpiration, and ground evaporation between the two simulations were 5.41 and 13.37 mm, 2.3 and 6.26 mm, and -1.39 and -7.4 mm in 2004 and 2005, respectively. Canopy transpiration decreased by 3 %, canopy evaporation decreased by 12 %, and ground evaporation increased by 10 % in 2004 and 2005 after applying the GFED4 burned area to CLM. This is consistent with the findings of Li et al. (2017) and Seo and Kim (2019) showing that canopy evaporation and canopy transpiration decreased and ground evaporation increased when comparing the simulation with and without fire. Furthermore, the total ET in the presence of fire decreased by 6.32 and 12.08 mm in 2004 and 2005, respectively, indicating that canopy evaporation is more strongly influenced by fires over Alaska in CLM.

In Eastern Siberia, the patterns of canopy evaporation and ground evaporation were the same as those of Alaska. Canopy evaporation increased and ground evaporation decreased in EXP-GFED4 because the simulated burned area decreased, which was noticeable from 2009 to 2012 (Fig. 9f and h). However, the canopy transpiration of EXP-GFED4 was similar to that of CLM-Default. In other words, there was no significant change in canopy transpiration due to a change in burned area. Furthermore, the ET with the burned area applied changed slightly in Eastern Siberia. Differences in the average canopy evaporation and ground evaporation were -9.19 mm (-28 %) and 6.97 mm (10 %) from 2009 to 2012, respectively. The reasons for the smaller change in canopy transpiration are related to soil moisture and leaf area.

Figure 9 shows differences in the simulated soil moisture for CLM-Default and EXP-GFED4 at 0–20 cm (hereafter top soil) and 70–150 cm (hereafter bottom soil) in both regions. In Eastern Siberia, the top soil moisture and bottom soil moisture decreased after applying the observed burned areas. Although the leaf area increased with fewer burned areas applied, transpiration did not change significantly due to the decreased soil moisture. On the contrary, there was no considerable difference in the top and bottom soil moisture between CLM-Default and EXP-GFED4. Therefore, transpiration was positively correlated with leaf area. According to the McVicar et al. (2012) and Nemani et al. (2003), the Alaska region is drier and more water-limited than Eastern Siberia. Energy is sufficient to evaporate the increased stored water from the ground, which explains why soil moisture did not change considerably in Alaska.