Enhanced future vegetation growth with elevated carbon dioxide concentrations could increase fire activity

The study addresses how rising atmospheric CO2 concentrations affect future wildfire activity and what mechanisms drive these changes. While recent decades have seen increases in fire frequency, size, and fire weather consistent with anthropogenic warming and drying, the balance between physical climate impacts (e.g., heat, drought) and ecological responses (e.g., CO2 fertilization increasing biomass) remains unclear. The authors use seven CMIP6 Earth system models (ESMs) that include fire modules to quantify changes in fire carbon emissions under idealized CO2 increases and to separate radiative (climate) versus biogeochemical (vegetation) effects of CO2. Understanding these drivers is important for anticipating regional fire risks, evaluating natural climate solutions, and informing land-based climate policies.

Prior work shows anthropogenic climate change is projected to enhance fire weather across most burnable land, including the western US, Australia, the Mediterranean, and potentially the Amazon under low mitigation scenarios. Concurrently, rising CO2 has increased terrestrial carbon uptake via CO2 fertilization, contributing to observed global 'greening'. However, its influence on fire is debated: increased biomass could raise fuel loads and fire occurrence, while higher live fuel moisture could mitigate severity. Fire responses vary with regime (fuel-limited vs flammability-limited), and regimes may shift under climate change. Observational fire emission datasets (GFED4.1s, FINNv2.5) have substantial uncertainties. Modeling fire remains challenging due to limited observations, complex human influences, and multiple controlling factors. Previous studies also highlight drought and compound heat-drought events reducing vegetation productivity, especially in the tropics and during extreme summers in mid-high latitudes.

• Models: Seven CMIP6 Earth system models with interactive fire modules: CESM2 (CLM5/CLM-Li), NorESM2 (CLM5), CMCC-ESM2 (CLM4.5 fire), GFDL-ESM4 (FINAL v2), MPI-ESM1-2-LR (JSBACH3.2 with SPITFIRE), MRI-ESM2-0 (HAL-based; limited fire details), CNRM-ESM2-1 (GLOB-FIRM). Most include lightning-caused ignition but use prescribed climatological lightning. Two models (GFDL-ESM4, MPI-ESM1-2-LR) include dynamic vegetation; others prescribe vegetation distribution but simulate vegetation state (e.g., LAI) prognostically. Nitrogen cycle coupling varies by model (GFDL-ESM4 lacks interactive N).
• Experiments: Idealized 1% per year CO2 increase (DECK) for 150 years, with two companion experiments: (i) 1% CO2-bgc (biogeochemical-only: land/ocean biogeochemistry sees rising CO2; radiation fixed at preindustrial), (ii) 1% CO2-rad (radiative-only: atmosphere sees rising CO2 for radiative transfer; land/ocean biogeochemistry sees preindustrial CO2). Land use and population fixed at 1850 in all idealized experiments. Also analyzed SSP3-7.0 scenario for comparison.
• Variables: Primary outcome is fire carbon emissions (fFire). Vegetation variables include net primary productivity (NPP), leaf area index (LAI), and vegetation carbon (cVeg). Climate variables include near-surface air temperature, precipitation, soil moisture, relative humidity, cloud cover, latent and sensible heat fluxes, transpiration, net surface solar radiation, and surface wind speed.
• Evaluation: Historical (2002–2014) extended with SSP5-8.5 to 2021; compared model fFire to GFED4.1s and FINNv2.5 for 12 world regions and global land; assessed climatology, variability, and seasonality.
• Processing: Monthly data interpolated to 2.5° × 2.5° (conservative). Responses estimated as differences between years 100–140 of each experiment and corresponding 40 years from the preindustrial control; also examined 40-year window centered on CO2 doubling (years 50–89). Statistical significance assessed with two-tailed pooled t-tests (90%); multi-model mean (MMM) confidence intervals computed as ±1.65 × standard error. Model agreement on sign assessed via percentage agreement and binomial test (90%).
• Analyses: Spatial maps and regional aggregations of responses; correlations of fFire with vegetation and climate parameters both spatially across grid boxes and temporally at each grid box (including seasonal/lagged relationships). Sensitivity analyses of fFire to vegetation changes (ΔfFire/ΔNPP) and comparison with SSP3-7.0. Consideration of potential nonlinearities and the role of land-use/fire suppression in scenarios.

• Model evaluation: Models reasonably reproduce observed amount, interannual variability, and seasonality of fFire over global land. They tend to overestimate relative to GFED4.1s and underestimate relative to FINNv2.5; the model 90% confidence interval generally includes at least one observational estimate for most regions.
• Global response under 1% per year CO2:
  • Years 100–140: MMM fFire percent increase over global land is 127.7 ± 79.2%; all models increase (36.8% in CNRM-ESM2-1 to 408.6% in GFDL-ESM4). Without GFDL-ESM4: 80.9 ± 20.4%.
  • At CO2 doubling (years 50–89): MMM fFire percent increase is 66.4 ± 38.8%.
• Attribution:
  • Biogeochemical (CO2-bgc): Most of the fFire increase arises from vegetation responses. Years 100–140 MMM percent change: 114.7 ± 101.3% (one model, CMCC-ESM2, shows a small decrease). At CO2 doubling: 60.1 ± 46.9%.
  • Radiative (CO2-rad): MMM global land percent change is not significant: -1.3 ± 16.2% (years 100–140). At CO2 doubling: 1.7 ± 9.4%. Regional decreases dominate in many low-latitude regions; increases occur in high northern latitudes (boreal, Tibetan Plateau).
• Regional responses (years 100–140, 1% CO2): Significant increases in 11 of 12 regions (all except South Asia). Notable increases:
  • United States: 229.1 ± 106.2% (western US 257.6 ± 145.2%; central US 267.7 ± 128.8%). Under CO2-bgc: 196.9 ± 85.1% (range 3.9% to 398.8%).
  • Canada, Europe, central/north Asia also show strong increases. Under CO2-rad: significant decreases in south Africa (-26.8 ± 19.1%), north Africa (-25.3 ± 22.4%), and Australia (-21.0 ± 15.4%); significant increases in Canada (90.9 ± 89.1%) and central/north Asia (53.2 ± 36.8%).
• Vegetation and climate linkages:
  • NPP increases robustly under CO2 and CO2-bgc: global land MMM ΔNPP 651.1 ± 167.9 kgC km⁻² day⁻¹ (74.8 ± 13.8%) for CO2; 673.9 ± 223.8 (77.1 ± 21.3%) for CO2-bgc (all models increase). Under CO2-rad, NPP decreases: -59.4 ± 41.0 (-7.3 ± 4.5%), except increases in high NH latitudes.
  • Spatial patterns of ΔNPP resemble ΔfFire across experiments, indicating biomass production (fuel load) as a key driver of fire emissions changes.
  • Temporal correlations (grid-cell): strongest during JAS for fFire vs soil moisture (MMM r ≈ -0.41 under CO2) and for lagged vegetation-fire links (AMJ cVeg vs JAS fFire r ≈ 0.36 under CO2). Lag correlations between AMJ NPP and JAS fFire improve consistency (MMM r ≈ 0.37 under CO2).
• Biophysical pathways under CO2-bgc: Reduced transpiration due to increased stomatal resistance lowers latent heat flux (-3.8 ± 1.4%), decreases near-surface RH (-1.1 ± 0.38%) and cloud cover (-0.52 ± 0.40%), increases net surface solar radiation (+0.42 ± 0.34%) and sensible heat flux (+5.0 ± 2.1%), and slightly warms land (+0.14 ± 0.19 K), with increased soil moisture (+0.75 ± 0.71%) and decreased wind speed (-1.0 ± 0.78%). These changes can modulate fire risk in addition to the biomass effect.
• Carbon balance: Despite increased fFire, vegetation carbon (cVeg) increases in all models (MMM +2.4 ± 0.51 kgC m⁻² under 1% CO2), implying NPP increases dominate, maintaining an enhanced vegetation carbon sink. Excluding GFDL-ESM4, the increase in fFire is a small fraction (~4%) of the increase in NPP.
• Scenario comparison: SSP3-7.0 produces qualitatively similar but weaker responses than 1% CO2; when normalized by CO2 increase, sensitivities are similar (≈0.04–0.05 kgC km⁻² day⁻¹ ppm⁻¹ after excluding outliers). Land-use changes (e.g., increased crop fraction) and fire suppression in SSPs can dampen fire responses relative to idealized experiments.

The findings demonstrate that under increasing atmospheric CO2, fire carbon emissions increase across most land regions, primarily due to biogeochemical effects of CO2 on vegetation (fertilization-driven increases in biomass and fuel loads). Radiative effects alone do not produce a significant global increase and often reduce fire activity in the tropics while increasing it in boreal regions, reflecting climate impacts on productivity and fuel moisture. The strong similarity between spatial patterns of NPP and fFire changes and positive lagged correlations between spring vegetation growth and summer fire emissions support vegetation as a key mediator of future fire activity. Biophysical feedbacks associated with altered vegetation physiology (reduced transpiration, atmospheric drying, increased sensible heating) further influence fire risk, potentially amplifying biomass-driven increases in some regions while offsetting in others via changes in soil moisture. Despite the substantial rise in fire emissions, the terrestrial biosphere remains a net carbon sink because NPP gains exceed fire losses in these experiments, though the increased fire activity partially offsets the sink. The results underscore the need to integrate ecological drivers into fire risk mitigation and to consider the implications for land-based climate solutions, where added biomass could inadvertently heighten fire risk in some regions.

This study shows that idealized increases in atmospheric CO2 substantially increase global fire carbon emissions in CMIP6 Earth system models, with the dominant contribution arising from biogeochemical effects on vegetation (CO2 fertilization enhancing fuel loads). Radiative effects alone yield minimal or regionally mixed fire responses. Vegetation–climate–fire interactions, including biophysical feedbacks from altered plant water use, shape regional outcomes. Although vegetation remains a net carbon sink, increased fire activity can erode part of the CO2-driven carbon uptake. Policy and management strategies—especially those centered on increasing biomass (e.g., reforestation/afforestation)—should account for the potential for elevated fire activity. Future work should improve fire process representation (including ignition, suppression, and agricultural fires), dynamic vegetation responses and competition, the coupling with the nitrogen cycle, lightning modeling, and evaluation against multiple observational datasets. Including realistic land-use changes and human-fire interactions in scenario analyses will better constrain future fire projections.

• Large inter-model spread in fire responses and sensitivities; one model (GFDL-ESM4) is an outlier with very strong vegetation–fire sensitivity.
• Most models lack fully dynamic vegetation; only GFDL-ESM4 and MPI-ESM1-2-LR simulate vegetation type shifts, limiting representation of vegetation–fire feedbacks.
• Lightning flash rates are prescribed from observations, precluding feedbacks of climate change on lightning ignitions.
• Observational constraints on fire emissions (GFED4.1s, FINNv2.5) have substantial uncertainties (order of a factor of two) and varying fire-type coverage; models differ in representation of agricultural and peat fires.
• Idealized experiments fix land use and population at 1850, omitting historical and future changes in human ignitions, suppression, and land management that strongly affect fires.
• Statistical attribution via correlations may miss nonlinearities and regime shifts (fuel- vs flammability-limited), and spatial correlations are modest in some analyses.
• Differences in nitrogen cycle coupling and soil/litter decomposition schemes affect vegetation responses and fuel dynamics across models.
• Burned area was not analyzed due to limited availability; fFire alone does not fully characterize fire behavior/severity.