Fire weakens land carbon sinks before 1.5 °C

As accumulated CO₂ emissions increase and the climate continues to warm, more ecosystem impacts are expected worldwide. Higher temperatures and changing rainfall will alter fire regimes and their impacts on ecosystems. At around 1.26 °C of warming above pre-industrial levels, changes in the intensity and frequency of extreme weather events are already observed, many of which have become more likely due to climate change. Fire regime and biome shifts are occurring across global ecosystems, with models projecting increased transformation as warming increases. Transitions can result from interactions among climate, land use and fire, potentially shifting tropical forests to seasonal forests or savannahs. Fire substantially impacts ecosystems and carbon stores via vegetation mortality, hydrological changes and emissions of greenhouse gases and aerosols, and may drive alternative stable states within similar climates. Given these dynamics, the question is whether limiting warming to 1.5 °C is consistent with avoiding significant ecosystem changes once changes in fire regimes and fire-vegetation-carbon feedbacks are accounted for. Prior studies using fire-weather indicators project increased fire risk at and beyond 1.5 °C, but the impact on land carbon sinks including dynamic fire feedbacks had not been explicitly assessed. This study addresses that gap.

The paper situates its contribution within evidence that: (1) wildfire risk and activity are increasing with warming and extreme weather changes; (2) many CMIP-class models historically lacked realistic fire processes or exhibited biases in fire and vegetation distributions; (3) ecosystems, especially carbon-rich tropical and boreal biomes, may undergo tipping-like transitions influenced by fire, land-use change and climate; and (4) prior IPCC carbon budget assessments acknowledged low confidence regarding fire feedback magnitudes. It references observed declines in burned area in Africa linked to human activities, documented increases in fire weather extremes, Amazon resilience loss signals, and the need to evaluate models against multiple observational datasets for burned area, vegetation cover and carbon.

The study uses the Joint UK Land Environment Simulator (JULES v5.5) with nitrogen limitation and dynamic vegetation (TRIFFID) coupled to the INFERNO fire module, forming the JULES-ES impacts configuration aligned with UKESM land surface. The model includes nine natural and four crop/pasture plant functional types (PFTs), with agricultural area from HYDE3.2 (historical) and LUH2/MAgPIE projections under RCP2.6 and RCP6.0 combined with SSP2 land use and population. Fire processes simulate spatially and temporally varying burned area dependent on fuel load, flammability, land cover and ignitions (from population density and lightning climatology), with fire mortality varying by PFT and reduced burning in cropland. Simulations are on a 0.5° grid, spun up for 10,000 years plus 500 years with fire enabled. The ISIMIP2b protocol provides bias-corrected climate: historical (1860–2006) bias-corrected to EWEMBI observations, and four bias-corrected Earth System Models for 2006–2100: HadGEM2-ES, GFDL-ESM2M, IPSL-CM5A-LR and MIROC. For the main analysis, RCP6.0 is used to ensure all ESMs reach the considered warming levels. Two ensembles of JULES runs were performed: with fire (reduced background mortality; explicit fire emissions and mortality) and without fire. Analyses are framed by Global Warming Levels (GWLs) using a 21-year running mean of global near-surface air temperature (PI baseline 1860–1900) to identify exceedance windows for 1.3, 1.5, 1.7 and 2.0 °C. Tree cover change classification uses 20-year running means and categories (increasing, decreasing, recovering, diminishing) based on sign of change and ratio to PI tree cover. The ‘equivalent impact of tree cover change’ compares with-fire states to without-fire states at 1.5 °C using four criteria to identify equivalent ratios and directions of change and proximity in time; if not found, equivalence is assumed beyond the modeled warming range. Net Biome Productivity (NBP) is computed as NPP minus soil respiration, harvest, wood products and (in fire runs) fire emissions; positive NBP indicates a sink. Statistical significance of fire impacts on NBP versus present day (2010–2019) is tested with an unpaired two-tailed Wilcoxon test on 20-year rolling windows (p<0.05). Model evaluation uses benchmarking (modified FireMIP package) against observations for burned area, fire emissions, vegetation cover, vegetation carbon and recent changes in tree/woody cover. Metrics include normalized mean error (NME; three-step variants removing mean bias and variance) and the Manhattan Metric (MM) for composition, along with assessments of trend overlap with observational uncertainty via probability density overlap. The model reproduces spatial patterns and trends of present-day burned area, vegetation carbon and tree cover within observational constraints, supporting use for future projections.

• Fire impacts on global carbon storage become statistically significant at a central estimate of 1.07 °C above pre-industrial (range 0.8–1.34 °C), implying that fire is already weakening land carbon sinks.
• Remaining carbon budgets are reduced when fire feedbacks are included: mean reductions of about 15 GtCO₂ for 1.3 °C, 25 GtCO₂ (≈−5%) for 1.5 °C, 37 GtCO₂ for 1.7 °C, and 64 GtCO₂ (≈−4–6%) for 2.0 °C relative to IPCC AR6 estimates (from a 2020 start); all four driving ESMs agree on reductions.
• Burned area generally increases at 1.5 °C and increases further at 2.0 °C across most GFED regions. Notable increases occur in Europe (+15% at 1.5 °C; +25% at 2.0 °C) and boreal North America (+12%; +20%). Africa shows decreases in burned area in many fire-dominated areas, consistent with observed declines and increased fire suppression associated with population changes.
• Vegetation cover responses at 1.5 °C: tree cover tends to increase in northern high latitudes (e.g., boreal zones, parts of Canada and Europe) and the Congo, while decreasing in tropical South America and Asia, with shifts toward heat- and drought-tolerant C4 grasses in parts of southern Amazonia. Regional outcomes vary by driving ESM, driven by differing precipitation projections.
• Accounting for fire often causes equivalent levels of tree cover change to be reached at lower global temperatures (earlier in time) relative to runs without fire, especially in North America (HadGEM2, IPSL), southern Brazil, West Africa and Western Australia. Some regions experience delayed impacts due to fire mortality slowing gains (e.g., parts of Russia, central Australia, western India), and in some areas the without-fire impact level is not reached within the simulation when fire is included.
• Global NBP increases with temperature due to high-latitude warming and CO₂ fertilization enabling northward treeline expansion and increased NPP, but fire reduces NBP in most regions (exceptions with slight increases at lower GWLs include MIDE, NHAF, SHAF due to declining burned area). The negative impact of fire on NBP is generally smaller under MIROC than under other ESMs.
• Regions with large forests and high fire activity—temperate North America (TENA), Northern Hemisphere South America (NHSA) and Southern Hemisphere South America (SHSA)—show more frequent switches between sink and source in extreme years when fire is included. Across these highlighted regions, significant fire impacts on NBP occur over 0.58–2.02 °C global warming.
• Overall, inclusion of fire-vegetation feedbacks weakens land carbon sinks before reaching 1.5 °C, advances or amplifies biome shifts in some regions and reduces the remaining carbon budget, complicating mitigation pathways and any recovery from overshoot of 1.5 °C.

By explicitly representing fire-vegetation-carbon feedbacks, the study shows that significant weakening of land carbon sinks occurs at warming levels already reached or imminent, addressing the central question of whether 1.5 °C is consistent with avoiding substantial ecosystem change once fire is considered. Projections reveal earlier onset of tree cover loss in several carbon- and biodiversity-rich regions, increased burned area in many GFED regions, and greater frequency of sink-to-source transitions during extreme years in temperate North and South American forests. Although global NBP benefits from high-latitude greening and CO₂ fertilization, fire largely counteracts carbon uptake regionally, narrowing the margin of safety embedded in remaining carbon budgets. The quantified reductions to the remaining carbon budget (e.g., −25 GtCO₂ for 1.5 °C; −64 GtCO₂ for 2.0 °C) indicate that previously assessed budgets may be optimistic if fire feedbacks are omitted. These findings underscore that 1.5 °C is not a strict threshold between safe and unsafe outcomes; risks rise with each increment of warming. The study highlights the need for integrated adaptation (e.g., fire management, land-use practices) and rapid mitigation to limit warming, reduce the likelihood of biome shifts, and maintain land carbon sink effectiveness, especially given potential inertia and hysteresis in ecosystem recovery.

The paper provides the first explicit quantification of fire feedback impacts on remaining carbon budgets and demonstrates that land carbon sinks weaken before 1.5 °C when fire is included. Burned area increases across most regions at higher GWLs, many areas experience earlier tree cover losses or gains when fire is accounted for, and regional NBP is reduced, with some forests intermittently switching to net carbon sources in extreme years. The mean reductions in remaining carbon budgets are estimated at about 25 GtCO₂ for limiting warming to 1.5 °C and 64 GtCO₂ for 2.0 °C. These results imply that overshoot-and-return pathways become more challenging and that risk assessments and policies should integrate fire feedbacks. The authors suggest repeating the analysis with multiple fire-enabled DGVMs as ISIMIP-3 data become available, exploring lightning change impacts, and refining land-use and population assumptions. Despite the added challenges, strong emissions reductions can still limit the worst impacts, but some regions may be closer to ecological thresholds than previously recognized.

• Model diversity: The analysis uses a single fire-enabled land-surface model (JULES-INFERNO). Results may differ across other DGVMs; repeating with multiple models (e.g., under ISIMIP3) is recommended.
• Forcing uncertainties: Despite bias correction, the four driving ESMs differ in future trends because bias adjustment corrects mean offsets over the observational period, not trends; this propagates into projections.
• Process assumptions: Land-use and fire assumptions (e.g., suppression from population change, agricultural expansion) follow global trends and RCP/SSP scenarios that may differ under alternative pathways or policy decisions.
• Ignitions and lightning: Lightning is prescribed from a climatology (1995–2014) scaled for cloud-to-ground strikes; projected lightning changes are not included.
• Spatial/temporal resolution and spin-up: 0.5° resolution may miss fine-scale processes; while long spin-ups are used, some slow ecosystem processes and hysteresis may still be imperfectly captured.
• Observational constraints: Benchmarking acknowledges uncertainties in observations (burned area, emissions, NBP, vegetation cover/carbon), especially in regional trends; evaluation relies on multiple datasets and statistical overlap rather than definitive ground truth.
• Baseline differences: Small differences in PI baselines (1850–1900 in AR6 vs 1860–1900 here) and the 2020 budget start, with additional emissions since 2020, affect direct comparability of remaining budget figures.