We used the University of Victoria Earth System Climate Model (UVic ESCM) version 2.9 to study the climatic effects of fire and fossil fuel CO2 emissions. The UVic ESCM computes at a resolution of 3.6∘×1.8∘ (longitude × latitude) the exchanges of carbon, energy, and water among the land, atmosphere, and ocean . The land module consists of a simplified version of the MOSES land surface scheme coupled to the TRIFFID DGVM . TRIFFID simulates the competition among five different plant functional types (PFTs): broadleaf tree, needleleaf tree, C3 grass, C4 grass, and shrub, accounting for the dynamics of different carbon pools for vegetation (leaves, stem, and roots) and soil–litter. The UVic ESCM computes the atmospheric energy and moisture balance with dynamical feedbacks, and its ocean module represents three-dimensional circulation, sea ice dynamics and thermodynamics, inorganic carbon, and ecosystem/biogeochemical exchanges .

The UVic ESCM can account for various types of prescribed forcings, including the emissions of CO2, other greenhouse gases, and sulphate aerosols, land-cover changes, volcanic aerosols, and land ice . In this study, we also used the UVic ESCM fire module developed by . In each grid cell, this module estimated the gross CO2 emissions coming from combustion as the product of prescribed burned area (see Sect. ), fuel density (simulated by the UVic ESCM), and PFT-specific combustion fractions for the different fuel types (Table ). The carbon contained in the vegetation killed by fire but not combusted was transferred to the soil–litter pool, where it decomposed and released additional CO2 at a rate that depended upon the simulated soil temperature and moisture. Since we were interested in non-deforestation fires, the different PFTs could compete and grow back in the recently burned area, giving rise to a regrowth CO2 flux influenced by the climate–carbon feedbacks simulated by the UVic ESCM (e.g., fire-induced CO2 fertilization and temperature changes). The model further accounted for the post-fire changes in land surface exchanges due to the modified vegetation cover, including the increase in land surface albedo (αL, unitless). In all simulations, we included only the CO2-related effects of fire and fossil fuel combustion and not the associated aerosols and non-CO2 greenhouse gases. We note that fire releases some carbon as carbon monoxide (CO) and methane (CH4); however, these species constitute less than 10 % of the fire-emitted carbon and get mostly oxidized to CO2 on a timescale shorter than the one of interest here . Similarly, we did not include here the short-term albedo decrease due to surface blackening.

We based the prescribed burned area on the January 2001 to December 2012 monthly data from version 4 of the Global Fire Emissions Database (GFED4), which was derived from satellite observations . We then simplified the GFED4 data set in order to retain its most essential features only. Each grid cell from the UVic ESCM was labelled as a “fire cell” if it had been affected by fire at least once over the 2001–2012 period according to GFED4 (Fig. ). The main simplification here was that the burned area fraction was set equal across all the UVic ESCM fire cells, with the specific burned area fraction value varying across fire simulations (see Sect. ). The use of this binary distribution of burned area fractions (i.e., the same value for all fire cells and 0 for all other cells) was necessary in order to reach the target fire CO2 emissions while ensuring that the burned area fractions were proportional for all fire cells across the different fire simulations. Given that the actual burned area fractions are already relatively close to 100% in various regions , upscaling the original GFED4 data would not have resulted in the same relative changes for all fire cells. Fire happened once per year in each of the UVic ESCM fire cells, during the month of highest burned area according to the mean 2001–2012 value from GFED4 data (Fig. ).

We started with an equilibrium run of the climate system for the year 1750, using the prescribed forcings from for solar radiation, atmospheric CO2 (fixed at 277 ppmv), non-CO2 greenhouse gases, land-cover changes, land ice, and volcanic aerosols. Five groups of transient simulations then branched off from this equilibrated climate, in addition to a control transient simulation; in all cases, the forcings from year 1750 were maintained, except that the climate and carbon cycle were free to respond to the effects of the fire and fossil fuel experiments.

First, we performed three simulations that each consisted of a single year of fire activity, followed by a return towards the pre-fire equilibrium conditions. The resulting fire pulses had sizes of 20, 100, and 200 PgC, based on their gross emissions (i.e., the carbon released from combustion only). We obtained these fire CO2 pulses by adjusting the single-year burned area fraction across all fire cells and designate these simulations as Fire20P, Fire100P, and Fire200P.

Second, we performed another set of fire experiments similar to the previous ones, except that the same burned area fractions were maintained year after year. We designate these stable fire regimes as Fire20S, Fire100S, and Fire200S, corresponding to the previous fire pulse experiments of 20, 100, and 200 PgC, respectively.

Third, we injected fossil fuel CO2 pulses of 20, 100, and 200 PgC into the atmosphere over a single year. The purpose of this set of three simulations was to compare the effects from fossil fuel CO2 emissions vs. the same amount (and timing) of gross fire emissions. We designate these simulations as FF20P-G, FF100P-G, and FF200P-G.

Fourth, we wanted to compare the effects from fossil fuel CO2 emissions vs. the same amount (and timing) of net fire emissions following each fire pulse. Each year, we computed the net fire emissions (land to atmosphere) as the annual change in total land carbon for the control simulation minus the annual change in total land carbon following the fire pulse (Fire20P, Fire100P, or Fire200P). We then injected into the atmosphere yearly fossil fuel CO2 emissions that were equal to these net fire emissions, including when they were negative (implying atmospheric carbon was sequestered back into geological reservoirs). We designate these simulations as FF20P-N, FF100P-N, and FF200P-N.

Fifth, we performed a set of three fossil fuel experiments in which the yearly fossil fuel CO2 emissions were this time equal to the net emissions from the Fire20S, Fire100S, and Fire200S stable fire regimes. We designate this last set of simulations as FF20S-N, FF100S-N, and FF200S-N.

The burned area fractions (unitless) in the fire cells for the 20, 100, and 200 PgC pulses were approximately equal to 0.09, 0.45, and 0.88, respectively. Since the 200 PgC pulse led to the burning of almost all the area within the fire cells, we used the results of this simulation to assess the post-fire simulated responses for changes in PFT cover, total biomass, and αL in different ecosystem types (Fig. ). In northern forests, the succession among the different PFTs (Fig. a) was qualitatively similar to, but noticeably slower than, observation-based trajectories . Simulated fire-caused changes also appeared reasonable when compared with field observations for biomass (Fig. c) and αL (Fig. e) . As expected , the return to pre-fire conditions was much faster in savannas (Fig. b, d, and f). Note that the very small increase in total biomass soon after the fire pulse (Fig. d) and the associated marginal decrease in αL (Fig. f; not visible) likely came from the CO2 fertilization effect caused by the long-lasting atmospheric CO2 anomaly (see Sect. ).

Additional simulations performed by further established the realism of results from the UVic ESCM fire module. First, they obtained gross fire CO2 emissions of 2.2 PgCyr-1 for the current fire regime, comparable to previous studies . The splitting of these gross emissions between vegetation (0.7 PgCyr-1) and soil–litter (1.5 PgCyr-1) also agreed with GFED-based estimates . Second, the differences in αL between the present-day fire regime and a no-fire world simulated by led to a global radiative forcing of -0.11 Wm-2 without the effect of surface blackening and -0.07 Wm-2 with surface blackening, in agreement with observation-based estimates (note that we did not include surface blackening in the current study).

The atmosphere, ocean, and land carbon pools responded as previously reported to the fossil fuel CO2 pulses (Fig. a). Part of the CO2 injected into the atmosphere progressively became absorbed by the land and ocean, so that 1000 years after the pulses, 60% of the additional CO2 was taken up by the ocean and the remaining 40% was divided almost equally between the land and atmosphere. The limited absolute difference among the pulse magnitudes studied here (i.e., 180 PgC) explains why the responses were almost identical in the three cases, contrary to what has been found for a larger range of pulse magnitudes .

The results for fire (Fig. b) differed substantially from the fossil fuel pulse results. This time the CO2 injected into the atmosphere came from the land, resulting in decreased land carbon rather than increased land carbon as in the case of fossil fuel. Instead of leading to long-lasting changes, the fire pulses were followed by a gradual return towards the initial equilibrium conditions. Moreover, the responses varied noticeably among the three fire pulses. Finally, fractional changes greater than 1.0 were observed for the atmosphere and land shortly after the pulses because, due to the decomposition of the uncombusted vegetation killed by fire, the net emissions were higher than the gross emissions upon which the magnitude of the pulses were defined. Figure c compares the airborne fraction of the CO2 pulses from fossil fuel vs. fire. All results were similar during ∼25 years following the pulses and for up to ∼50 years for Fire100P and the different fossil fuel pulses. However, the airborne fraction became systematically higher for fossil fuel than for fire after about a century.

These differences then affected the global mean atmospheric surface temperature (Ts, in K), as shown in Fig. a. Fossil fuel CO2 emission pulses caused relatively stable increases in Ts over millennial timescales . In the case of fire pulses, the return of atmospheric CO2 towards pre-fire levels (Fig. b) resulted in smaller warming of much shorter duration. Atmospheric CO2 even decreased below the control level ∼400–500 years after the pulses, which contributed to the observed long-term net cooling effect particularly visible for Fire200P. This slight decrease in atmospheric CO2 came from the long time needed before the ocean returned to the atmosphere all the carbon absorbed following the fire pulses.

Albedo was also involved in the diverging effects of the two processes on Ts (Fig. b). Fossil-fuel-induced CO2 fertilization slightly decreased αL over the whole simulation period, whereas fire noticeably increased αL for decades to centuries. Note that contrary to the situation illustrated in Fig. a, in some northern grid cells tree cover had not fully recovered yet to pre-fire levels 1000 years after the 200 PgC fire pulse. This lasting increase in αL contributed to the net cooling following the fire pulses.

All previous outcomes illustrate that the effects on the global carbon cycle and temperature from fire vs. fossil fuel combustion differ for identical pulse magnitude defined in terms of gross (i.e., combustion only) fire emissions. Now, what if fossil fuel emissions were instead set equal to the net land-to-atmosphere emissions from fire year after year over the entire simulation, a situation where we expect fossil fuel combustion to better mimic the effects from fire? In this case, the impacts on land carbon remained opposite because emissions came from the land for fire but not for fossil fuel; for the atmosphere, however, the CO2 anomalies were more similar (Fig. a vs. Fig. b), though not identical, as can be seen in Fig. b. During the first ∼250 years, these anomalies were systematically lower for fossil fuel because the vegetation absorbed a portion of the emitted CO2, whereas for fire the net emissions already accounted, by definition, for vegetation regrowth, global CO2 fertilization, and all climate–carbon feedbacks. As a result, the ocean absorbed more carbon for fire than for fossil fuel emissions (Fig. a vs. Fig. b).

Based on atmospheric CO2 alone, one would thus expect Ts to be higher for fire than for fossil fuels, yet the opposite was in fact observed (Fig. c) due to the opposite impacts on αL (Fig. d). Note that in the long term, these ΔTs were however much smaller than when fossil fuel emissions were equal to gross fire emissions (Fig. a). The fact that atmospheric CO2 anomalies became slightly lower for fire than for fossil fuel after about 250 years (Fig. b; not visible) can be explained by long-lasting impacts on ocean carbon cycling: compared with fossil fuel, the ocean absorbed substantially more carbon in the initial decades after the fire pulses and then took more time to outgas this carbon when the atmosphere–ocean fluxes shifted sign during the return towards the initial equilibrium conditions.

The previous results were based on single pulses of fire activity; we now turn to stable fire regimes for which the burned area fraction was maintained year after year, instead of being applied only once as in the pulse experiments. Figure  shows that the resulting gross and net emissions had qualitatively similar behaviours for the three stable regimes. Both the gross and net yearly emissions decreased quickly after an initial spike. The yearly net emissions progressively stabilized close to 0, although their mean value was still positive towards the end of the simulations as indicated by the slight positive slope of the cumulative net emissions. The yearly gross emissions, however, stabilized around much higher values because vegetation and soil–litter kept being combusted each year. Contrary to net emissions, the cumulative gross emissions thus increased almost linearly ∼50 years after the onset of fire activity and onwards (results not shown).

Gross emissions thus appear highly inadequate to assess the cumulative impacts of fire regime shifts. Indeed, yearly gross emissions towards the end of the simulations were higher for Fire100S than for Fire200S, even though the outcome was obviously the opposite for the cumulative net emissions (Table ). The lower land carbon density caused by more frequent fires has previously been reported to result in a “saturation effect” of gross emissions ; here, this effect was so large that gross emissions ended up being lower for Fire200S than for Fire100S about 50 years after the onset of fire activity. A similar saturation effect clearly affected the cumulative net emissions, which were only twice as large for Fire200S compared to Fire20S, whereas the equilibrium yearly burned area was 12 times larger for Fire200S vs. Fire20S (Table ). This slightly supra-linear scaling in burned area (e.g., 12 times instead of 10 times larger for Fire200S vs. Fire20S) among stable fire regimes was caused by fire-induced changes in vegetation composition. The input prescribed burned area in each fire cell (see Sect. ) actually corresponds to a gross value that is reduced to account for the PFT-specific unburned islands occurring within burn perimeters . More frequent fires led to increases in grass cover at the expense of trees and shrubs, thereby increasing the net burned area.

Even for fossil fuel emissions that were equal to the net emissions from stable fire regimes, the effects from the two processes differed once again. Figure a shows the distribution of net cumulative emissions (i.e., from year 0 until the specific year considered) from fossil fuel among the active carbon pools. This splitting was similar to the one following a single fossil fuel pulse (Fig. a), except that the maximum land uptake was proportionally lower and the ocean took a little longer to become the main carbon sink. For fire (Fig. b), land carbon rather decreased (with a fractional change equal to -1.0 as the net emissions were, by definition, equal to the total change in land carbon) and the uptake of carbon by the ocean had to be substantially higher than for fossil fuel.

The airborne fraction of the net emissions from stable fire regimes was initially higher than for the same amount of emissions from fossil fuel, but the anomalies in atmospheric CO2 progressively became more similar (Fig. a). This should have caused Ts to be higher for fire than for fossil fuel, yet once again the opposite was observed (Fig. b). Cumulative fossil fuel CO2 emissions led to Ts increases that were relatively stable over thousands of years . For fire, in contrast, the initial increase in Ts after the onset of fire activity was followed ∼50–100 years later by a gradual decrease in Ts. As was the case for the pulse simulations (see Sect. ), this different effect on Ts came from opposite changes in land albedo, which substantially increased for fire due to changes in vegetation cover but slightly decreased for fossil fuel due to CO2 fertilization (Fig. c).

In this study, we have shown a consistent pattern of fundamental differences between the effects on the carbon cycle and climate per unit of CO2 emitted by non-deforestation fire vs. fossil fuel combustion. These discrepancies ultimately came from the net addition of CO2 to the three active carbon pools by fossil fuel combustion (contrary to fire), as well as the differences in the average lifetime of the atmospheric CO2 increase and in the non-CO2 climatic impacts. First, the sources of CO2 emissions are qualitatively distinct: fire simply reshuffles carbon among the active pools, whereas fossil fuel combustion entails a net carbon transfer from the geological to the active pools over millennial timescales . Second, the terrestrial pools (vegetation plus soil–litter) cannot respond in the same way to the atmospheric CO2 anomalies created by fire vs. fossil fuel emissions. The only direct effect (i.e., excluding climate change) of fossil fuel emissions on land carbon storage occurs through the CO2 fertilization effect. Fire, however, gives rise to a much more dynamic land carbon response. Fire activity not only leads to CO2 emissions through the combustion of land carbon and the further decomposition of killed but uncombusted vegetation, but it also decreases the amount of vegetation that can instantaneously be fertilized by the fire-induced increase in atmospheric CO2. Subsequently, however, vegetation regrowth and the associated soil–litter build-up in the burned patches act as strong carbon sinks. Third, these contrasting effects on terrestrial vegetation mean opposite changes in land albedo: fire-induced decrease in vegetation cover increases αL, whereas fossil-fuel-induced CO2 fertilization decreases αL through dynamic vegetation changes like increased shrub and tree cover in tundra and generally higher leaf and stem area index for the vegetation already in place . This divergence in αL responses implies unequal Ts changes, which then feed back to affect the carbon cycle itself. Therefore, the effects on carbon cycling and temperature are incongruent even when fossil fuel emissions are equal to the net emissions from fire.

Other variables than carbon pools and αL were affected by these different changes in Ts and amplified them. Sea ice area, for example, often diverged noticeably between corresponding fossil fuel and fire simulations. For FF100P-G and FF200P-G, there was a small (∼2 and ∼4%, respectively) but permanent decrease in global sea ice area that did not occur in the corresponding fire simulations. For FF100P-N and FF200P-N, sea ice area also decreased a little for a few centuries at least before gradually returning toward initial levels. (For FF20P-G and FF20P-N, the changes in global sea ice area were indistinguishable from internal variability.) For fire pulses, however, the substantial ΔαL-based cooling over the Northern Hemisphere due to extensive land masses slightly increased Arctic sea ice area; note that ΔαL had a much smaller absolute influence on Antarctic sea ice, for which the changes were highly variable spatially. Such transfer of αL-induced cooling to the surrounding ocean has also been observed following deforestation simulations, along with an additional decrease in atmospheric temperature over most latitudes resulting from the lower ocean temperature . In our simulations of stable fire regimes and the corresponding fossil fuel experiments, changes in sea ice area were much larger due to higher net CO2 emissions. For fossil fuel, sea ice area was permanently reduced in all simulations. For fire, the ΔαL-based cooling was not strong enough this time to prevent major losses of both Arctic and Antarctic sea ice, because the atmospheric CO2 anomalies were larger and longer-lasting than following a single fire pulse. However, the increase in αL helped maintaining lower temperatures for the stable fire regimes than for the corresponding fossil fuel simulations, and global sea ice area progressively recovered to the control level, albeit with spatial differences between the Arctic and Antarctic that matched the hemispherical changes in atmospheric temperature.

The outcomes of our study should be interpreted with five caveats in mind. First, we developed idealized fire regimes in order to obtain substantial fire impacts while facilitating the comparison of results across the different magnitudes of pulses or stable regimes. Our fire regimes were therefore more severe than the current situation on Earth, as seen with our equilibrium results of ≥0.9 Ghayr-1 for burned area and ≥7.3 PgCyr-1 for gross emissions under stable regimes (Table ) vs. current values of 0.3–0.5 Ghayr-1 and 1.5–3 PgCyr-1 , respectively. Moreover, our “equal” spatial fire patterns (i.e., same burned area fraction in each fire cell) gave much more weight to fires in extra-tropical regions compared with the current fire distribution . Despite the differences in vegetation regrowth and fire-caused changes in albedo among regions, the impacts on atmospheric CO2 and Ts did not seem overly sensitive to changes in the distribution of burned area fraction among fire cells following a single fire pulse (Fig. ).

Second, we neglected all non-CO2 emissions from fire and fossil fuel. Accounting for the short-term post-fire surface blackening caused by char would reduce the albedo cooling effect. In contrast, explicitly tracking all the patches created by individual fire events, instead of representing their average grid-level effect as we did here, would increase the simulated albedo cooling effect over boreal forests at least , although the impact would likely be minor for the Fire200P and Fire200S simulations in which the burned area fraction was close to 90% in each fire cell. Furthermore, the fire-caused emissions of aerosols and non-CO2 greenhouse gases into the atmosphere would have a much stronger impact on Ts than changes in surface albedo; however, the magnitude and even the sign of the climatic effect from these non-CO2 atmospheric emissions remain highly uncertain . Future studies on the differences in the carbon cycling and temperature impacts between fire and fossil fuel would nevertheless benefit from considering the effects of non-CO2 emissions.

Third, the UVic ESCM does not currently simulate the non-trivial exchanges of carbon between land and ocean or between inland waters and the atmosphere , which are also impacted by fire. For example, the land-to-ocean flux of all particulate and dissolved pyrogenic carbon could be as high as ∼50–100 TgCyr-1 . More research is therefore needed to accurately represent the highly variable and poorly quantified fate of such exchanges of pyrogenic carbon; meanwhile, their influence on our results is speculative but is unlikely to challenge the main outcomes we obtained.

Fourth, the quantitative results we obtained were dependent upon the specific features of the UVic ESCM. For example, the simulated post-fire vegetation regrowth appeared too slow in northern grid cells (Fig. a), thereby overestimating the duration of both the αL-based cooling and CO2-based warming following fire. The carbon–concentration feedback parameters from the UVic ESCM are close to the mean from other fully coupled climate–carbon models, but its carbon–climate feedback parameters are on the high end , meaning that the atmospheric CO2 levels were more affected by temperature changes than would have occurred in most other models. Once again, these factors should not challenge the main outcomes we obtained.

Fifth, our study addressed only non-deforestation fires after which the natural vegetation is free to recover. One might argue that our stable fire regimes are similar to deforestation fires because, over large spatial scales, both fire types decrease terrestrial carbon storage and vegetation cover. However, our non-deforestation fires affected equally all fire cells, whereas deforestation fires are deemed exclusive to tropical regions . Given that fire-induced changes in terrestrial carbon density and albedo vary substantially among regions, we caution against the direct extrapolation of our results to deforestation fires. In fact, when neglecting non-CO2 emissions, deforestation fires are conceptually more similar to other sources of LULCC than to non-deforestation fires. Note that previous global-scale climatic studies of LULCC (see , for an extensive list) have represented all LULCC sources in the same way. Yet the variations in delayed CO2 fluxes between fire and other LULCC sources matter for carbon cycling and, as mentioned previously, non-CO2 emissions could have a dominant impact on the climate. Consequently, studies dedicated to deforestation fires that specifically represent their delayed CO2 fluxes and go beyond CO2 emissions would allow for a more refined understanding of their climatic impacts.