Fire decline in dry tropical ecosystems enhances decadal land carbon sink

Fire is a key disturbance in terrestrial ecosystems, especially in dry tropical regions, and is tightly linked to vegetation, climate, biogeochemical cycles, and human activities. While climate regulates fuel loads and conditions for ignition and spread, human activities increasingly shape fire regimes through ignition, suppression, fuel management, agricultural expansion, and landscape fragmentation. Satellite observations indicate a substantial global burned area (BA) decline over recent decades, with the largest decreases in African savannas. Concurrently, the airborne fraction of anthropogenic CO2 emissions decreased during 2002–2014 despite rising emissions, implying a strengthening of ocean and/or land sinks. The Global Carbon Project’s process-based land and ocean sink estimates still leave an unexplained budget imbalance of about 0.6 PgC yr^-1 in the last decade. Although direct fire emissions are typically included in carbon budgets (particularly for deforestation and peat burning), the indirect impacts of changing wildfire regimes on ecosystem carbon cycling have not been well quantified, and many land surface models are not constrained by observed BA trends. Under dynamic equilibrium, fire-induced carbon losses are eventually offset by regrowth; however, shifts in fire regime can alter steady states of carbon pools, leading to long-term carbon gains or losses. This study quantifies how the observed global BA decline between 2001–2007 and 2008–2014 affected the terrestrial carbon cycle, including both direct emissions reductions and indirect feedbacks on productivity and respiration.

Prior work highlights climate controls on tropical and subtropical fire variability and the increasing role of human activities in shaping fire regimes. Global analyses reported a roughly 25% BA decline from 1997–2015, correlated with agricultural expansion and fragmentation, particularly in African savannas. The Global Carbon Budget provides process-based land and ocean sink estimates but leaves an additional ~0.6 PgC yr^-1 required to reconcile the airborne fraction in the past decade. Land-use change emissions include some fire components (e.g., deforestation, peat), but prognostic fire modules show large spread and are not necessarily constrained by observed BA changes. Theoretical and empirical studies suggest that while individual fire-recovery episodes can be carbon-neutral over long timescales, shifts in fire frequency or severity can alter steady-state carbon pools. Historical reconstructions indicate biomass burning has declined in some regions since the mid-20th century, with modeling suggesting reduced BA since the 1930s due to population and cropland increases. However, the indirect, lagged effects of fire regime changes on net ecosystem exchange (NEE) have been underexplored in global assessments.

The study uses the CARbon DAta-MOdel fraMework (CARDAMOM) to retrieve terrestrial carbon cycle states and fluxes over 2001–2014 constrained by multiple satellite and inventory datasets. Fire occurrences are prescribed by satellite-derived burned area (GFED4s as reference, with GFED4 and ESA-CCI used for sensitivity), and fire carbon emissions are modeled as a function of BA, live and dead carbon pools, and combustion factors, with additional fire-induced mortality transferring live biomass to dead pools. CARDAMOM simulates six carbon pools (foliar, labile, wood, fine roots, litter, soil) and plant-available water, with GPP driven by meteorology and foliar carbon, autotrophic respiration as a function of GPP and temperature, allocation to live pools, turnover, and heterotrophic respiration dependent on temperature and moisture. Key parameters (photosynthesis, phenology, allocation, turnover, combustion factors, fire resilience) and initial states are optimized per 4°×5° grid using a Metropolis-Hastings MCMC approach. Assimilated constraints include: fire CO emissions inferred from MOPITT CO inversions via GEOS-Chem with biome-specific CO:total carbon emission ratios; satellite LAI; GPP variability inferred from SIF; spatial distribution of aboveground biomass; and global soil organic carbon (HWSD). To quantify impacts of BA decline, the authors compare an Observed BA scenario against a Fixed BA scenario where 2008–2014 BA is held at the 2001–2007 mean, with all else equal. Differences in simulated fire emissions (FIRE), gross primary productivity (GPP), terrestrial ecosystem respiration (TER), and net ecosystem exchange (NEE = TER − GPP) between scenarios over 2008–2014 represent the effect of BA decline; legacy effects are explored beyond 2014 by repeating meteorological forcing. The study further converts the estimated AFIRE and ANEE impacts into equivalent adjustments to the airborne fraction using Global Carbon Project components (EFF, ELUC, SOCEAN, SLAND, GATM), and evaluates bias and RMS reductions relative to observations.

• Burned area declined by 34 Mha yr^-1 (−9%) between 2001–2007 and 2008–2014 in GFED4 and by 52 Mha yr^-1 (−10%) in GFED4s; ESA-CCI indicates a −23 Mha yr^-1 (−6%) decline. The largest BA reductions occurred in savanna (−8%), woody savanna (−7%), and grassland (−21%).
• CARDAMOM-estimated global fire carbon emissions decreased from 2.1 ± 0.1 PgC yr^-1 (2001–2007) to 1.8 ± 0.1 PgC yr^-1 (2008–2014), an average decline of −1.5% yr^-1 over 2001–2014, with a faster −1.8% yr^-1 decline during 2007–2014. Between the two periods, fire emissions decreased by 0.2 ± 0.1 PgC yr^-1.
• Regional contributions to reduced fire emissions (mean ± SD across approaches) include: North Africa (−70 ± 9 TgC yr^-1, −16%), southern South America (−60 ± 20 TgC yr^-1, −30%), northern South America (−53 ± 9 TgC yr^-1, −36%), Southeast Asia (−55 ± 22 TgC yr^-1, −24%), Australia (−23 ± 4 TgC yr^-1, −16%), and Europe (−7.5 ± 0.8 TgC yr^-1, −30%).
• Indirect ecosystem feedbacks from BA decline increased GPP and live biomass, raising TER but by about half the GPP increase, yielding an additional NEE reduction (enhanced land sink) of about −0.4 ± 0.2 PgC yr^-1 during 2008–2014 relative to Fixed BA.
• The indirect NEE effect is roughly twice the magnitude of the direct fire emission reduction, producing a combined sink enhancement comparable to the ~0.6 PgC yr^-1 Global Carbon Budget imbalance.
• Legacy effects: ANEE builds during 2008–2014 as cumulative BA differences grow, then decays quasi-exponentially after BA perturbations cease. Average ANEE in the first 5 years post-perturbation is similar to the in-window effect (~−0.4 ± 0.1 PgC yr^-1), declining to <50% in years 6–10 and <10% in years 11–20. Global net biome exchange (including fire) returns to neutrality within ~18 years on average, faster in savanna-dominated regions.
• Spatial patterns: Largest ANEE and GPP enhancements align with regions of BA and fire emission reductions (Sahel/North Africa, dry subtropics of South America), consistent with observed increases in EVI/NIRv, reports of woody encroachment in African savannas, and satellite-based biomass gains.
• Incorporating AFIRE and ANEE adjustments into the airborne fraction reduces the mean bias of process-based AF estimates by 86% and RMS by 53% relative to observations, indicating fire decline significantly contributes to explaining the recent increase in the land carbon sink.

The analysis demonstrates that observed declines in burned area—primarily in dry tropical ecosystems—reduced direct fire emissions and, critically, triggered indirect ecosystem responses that enhanced the terrestrial carbon sink on decadal timescales. The increased GPP and live biomass following fewer fires, offset only partially by higher respiration, led to a net reduction in NEE roughly double the direct emissions decrease. These effects together closely match the magnitude of the unexplained ~0.6 PgC yr^-1 sink in the Global Carbon Budget, substantially improving the reconciliation of the airborne fraction when included. The results suggest that recent land sink strengthening cannot be fully attributed to CO2 fertilization, climate variability, or land-use change alone; reductions in wildfire activity contribute substantially, particularly in savanna regions with short carbon residence times. The legacy behavior implies benefits persist for years after reduced burning, but asymptotically diminish as ecosystems approach new steady states. The findings underscore the importance of fire regime changes—driven by human land use and management—in modulating the global carbon cycle and climate feedbacks.

Observed declines in global burned area during 2008–2014 led to a direct reduction of fire carbon emissions by 0.2 ± 0.1 PgC yr^-1 and an additional indirect land sink enhancement of 0.4 ± 0.2 PgC yr^-1 through ecosystem feedbacks, together comparable to the ~0.6 PgC yr^-1 Global Carbon Budget imbalance. These effects are concentrated in dry tropical regions and persist for years, though they decline as ecosystems approach new steady states. Incorporating these fire-related effects markedly improves agreement between process-based sink estimates and the observed airborne fraction. The study highlights fire management as a key lever for enhancing terrestrial carbon storage and mitigating climate change. Future research should integrate nutrient limitations, grazing, vegetation succession, and model structural uncertainties, and assess how projected climate-driven increases in fire risk and drought may modulate or reverse these benefits.

• Nutrient limitations (e.g., nitrogen, phosphorus) and grazing impacts were not explicitly represented in the estimated GPP response, potentially biasing sink enhancements high.
• Model structural uncertainties in CARDAMOM and the representation of fire-ecosystem interactions were not fully explored; the study focuses on parametric uncertainty via MCMC.
• Sensitivity experiments isolate BA changes between 2001–2007 and 2008–2014; legacy effects of fires prior to 2001–2007 are implicitly included in inferred parameters but not explicitly analyzed.
• Spatial resolution (4°×5°) may obscure finer-scale heterogeneity in fire behavior and ecosystem responses, particularly for small fires and land management practices.
• While multiple BA datasets were considered, GFED4s was used as the reference; uncertainties in BA detection (especially small fires) and emission factors propagate to fire emission estimates.
• The extrapolation of post-2014 legacy effects uses repeated meteorological forcing, which may not capture real future climate variability and extremes.