Recent reductions in aerosol emissions have increased Earth's energy imbalance

Earth’s Energy Imbalance (EEI) is defined as the difference between net downward shortwave and outgoing longwave radiative fluxes at the top of the atmosphere. EEI drives changes in the heat content of the oceans, cryosphere, land, and atmosphere, with the oceans absorbing ~90% of the excess heat. Satellite observations from CERES and in situ ocean heat content estimates show a positive EEI trend over the past two decades, consistent with an anthropogenic forcing signal that cannot be explained by internal variability alone. Recent model–observation comparisons indicate models can reproduce patterns of TOA flux changes but tend to underestimate the magnitude of the recent EEI trend, especially in shortwave (SW). At the same time, commonly used forcing datasets may not capture recent declines in anthropogenic aerosol emissions (e.g., SO₂ in China), and aerosol effective radiative forcing (ERF) trends have reversed sign, implying a recent strengthening of net radiative forcing and a potential temporary acceleration of warming. The study aims to quantify the anthropogenic contributions—particularly from aerosol emission reductions—to the observed EEI trend by using climate models forced with observed SST and sea-ice, updated aerosol emission inventories, and targeted experiments that isolate aerosol ERF from the total forcing and radiative response.

Recent estimates place EEI at 0.89 ± 0.26 W m⁻² for 2010–2022, up from 0.79 ± 0.27 W m⁻² for 2006–2018 (IPCC AR6). CERES satellite data reveal a positive EEI trend of about 0.50 ± 0.47 W m⁻² decade⁻¹ (2005–2019), consistent with ocean heat content–derived trends. Prior single-model studies suggest anthropogenic forcing plus climate response is needed to explain the observed trend. However, many earlier modeling studies did not incorporate recent changes in anthropogenic aerosol emissions or used SST/sea-ice datasets and forcing assumptions that could bias trends. In CMIP6 emissions, the decline of SO₂ over China after ~2007 is not well represented. Several lines of evidence indicate a reversal toward more positive aerosol ERF trends (both ERFari and ERFaci), implying a temporary acceleration of warming. Nevertheless, aerosol–cloud interaction effects remain uncertain, and many models struggle to reproduce satellite-observed cloud property changes. These issues motivate updated simulations and targeted experiments to attribute the EEI trend components.

Data and observations: TOA radiative fluxes are from CERES EBAF Edition 4.2 (monthly mean net, SW, LW at 1°×1°, Jan 2001–Dec 2022). MODIS (Terra+Aqua) Collection 6.1 Level-3 provides aerosol optical depth (AOD at 550 nm) and cloud properties. Ocean heat uptake (0–2000 m) for 2005–2019 is used for comparison. Trend uncertainties for CERES are taken as 0.20 W m⁻² decade⁻¹ (95%), converted to 0.168 W m⁻² decade⁻¹ (90%).

Models and experiments: Four atmospheric GCMs with prescribed SST/sea-ice were used with multiple ensemble members: CESM2-CAM6 (20), ICON-HAM (4), HadGEM3-GC3.1-LL (3), NorESM2 (6). GFDL AM4 simulations from a prior study are included for comparisons where applicable. All models use AMIP-style observed monthly SST and sea-ice boundary conditions for 2000–2019. Three core experiments were performed per model:

• BASE: All anthropogenic and natural forcings evolve historically and with SSP2-4.5 after 2014/2015 as applicable; anthropogenic aerosols from CEDS (April 2021) to 2019 (HadGEM3 uses CMIP6 CEDS/SSP2-4.5). Biomass burning, well-mixed greenhouse gases, ozone, and solar follow CMIP6 historical then SSP2-4.5.
• AERO2000: Same as BASE, except anthropogenic aerosol (precursor) emissions are fixed at representative year 2000 (2014 in HadGEM3) throughout; SST/sea-ice remain transient as in BASE. In NorESM2, biomass burning aerosols were also fixed; CESM2 sensitivity tests indicate biomass burning changes have negligible impact on the EEI trend.
• ALL2000: All anthropogenic and natural forcings are fixed to year 2000 levels (1850 in HadGEM3) while prescribed SST/sea-ice evolve. GFDL AM4 provides BASE-equivalent (“AM4 PSST ERF”) and ALL2000-equivalent (“AM4 PSST”) simulations; an AERO2000-equivalent run is not available. Additional RFMIP-style GFDL simulations using preindustrial SST/sea-ice are used to compare ERF trends.

Attribution framework: EEI is decomposed into radiative response (αΔT, estimated via ALL2000) and total ERF (BASE−ALL2000). ERF is further split into aerosol ERF (ERFaero: BASE−AERO2000) and other forcings (ERFother: AERO2000−ALL2000). Clear-sky versus cloud radiative effect (CRE) components are analyzed.

Trend analysis: Monthly anomalies are deseasonalized (subtract 2001–2019 climatology for each month). Linear trends are from least squares on monthly anomalies; maps use Mann–Kendall and Theil–Sen. Increased downward radiation is positive. CRE = all-sky − clear-sky fluxes for total regions. Sensitivity CESM2 experiments with fixed year-2000 SST/sea-ice (60-year cycles) for 2001 vs 2019 aerosols test robustness of ERFaero estimated from transient differences, showing consistency with the transient approach.

• Observed EEI trend: CERES shows a significant positive global EEI trend of 0.47 ± 0.17 W m⁻² decade⁻¹ for Jan 2001–Dec 2019 (0.51 ± 0.17 W m⁻² decade⁻¹ through June 2023). Models reproduce interannual variability well (r > 0.7 for all) but underestimate the trend by ~10–40%.
• Model mean EEI trend and aerosol contribution: Multi-model mean EEI trend (BASE) is 0.38 W m⁻² decade⁻¹ (2001–2019), within CERES uncertainty but biased low. When anthropogenic aerosol emissions are held constant (AERO2000), the mean EEI trend drops by 0.20 W m⁻² decade⁻¹ (intermodel 0.15–0.28) to 0.18 W m⁻² decade⁻¹, well outside CERES’ uncertainty. Thus, recent aerosol emission reductions strengthened the EEI trend by about 0.2 ± 0.1 W m⁻² decade⁻¹, approximately doubling the modeled EEI trend.
• Radiative response: With all forcings fixed (ALL2000), the model mean EEI trend is −0.16 W m⁻² decade⁻¹ (−0.08 to −0.25), indicating that evolving forcings are required to produce the observed positive EEI trend.
• SW and LW decomposition: Observed SW downward trend is +0.67 ± 0.17 W m⁻² decade⁻¹ and LW downward trend is −0.20 ± 0.17 W m⁻² decade⁻¹; models generally underestimate the magnitudes (HadGEM3 performs best). The large positive SW trend that drives EEI arises from roughly equal contributions of aerosol ERF and total radiative response.
• ERF trends: Total ERF trend from IPCC AR6 is ~0.57 W m⁻² decade⁻¹, similar to models. The SW ERF trend is entirely due to aerosols (ERFaero), whereas LW ERF is dominated by other forcings (especially increased well-mixed greenhouse gases). Multi-model mean ERFaero trend is 0.21 W m⁻² decade⁻¹ (0.15–0.28), comprising ~38% (30–46%) of total ERF trend.
• Intermodel differences: GFDL shows weaker SW/EEI trends and smaller ERFaero trend (~0.12–0.13 W m⁻² decade⁻¹ in RFMIP-style setup), consistent with its relatively weak (less negative) baseline aerosol ERF sensitivity among CMIP6 models.
• Regional patterns: Models reproduce observed AOD decreases over Eastern US, Europe, and China and increases over India; positive SW clear-sky trends align with regions of aerosol reduction north of 30°N. Discrepancies include SH ocean AOD trends and biomass burning regions (Canada, Siberia, Amazon), partly due to emissions climatologies post-2014.
• Causes of model underestimation: Missing/incorrect SW clear-sky trend south of 30°S (likely linked to Southern Ocean/sea-ice/surface albedo changes) and underestimated negative LW cloud trend (linked to cloud feedback biases) contribute to the 10–40% underestimation of global EEI trend.

The study set out to determine how much recent anthropogenic changes—especially reductions in aerosol emissions—have contributed to the observed positive trend in Earth’s energy imbalance. By forcing multiple atmospheric GCMs with observed SST and sea-ice and isolating forcing components through targeted experiments, the analysis shows that decreases in aerosols since 2001 have substantially strengthened the EEI trend, primarily via enhanced net downward shortwave radiation. Quantitatively, holding aerosols constant reduces the modeled EEI trend by ~0.20 W m⁻² decade⁻¹, indicating that aerosol reductions account for about half of the modeled EEI trend and have approximately doubled it over 2001–2019. The remaining contribution arises from the radiative response (feedbacks) and other forcings (especially increased greenhouse gases, which dominate LW ERF).

These results are consistent with the hypothesis that recent air-quality-driven aerosol emission reductions have decreased planetary albedo over polluted regions, increasing absorbed solar radiation and amplifying EEI beyond what greenhouse gases alone would produce. The models also reproduce observed interannual variability, reinforcing the robustness of the attribution. However, model biases—particularly in Southern Ocean shortwave clear-sky trends and LW cloud feedbacks—lead to an underestimation of the observed trend magnitude. Intermodel spread in aerosol ERF sensitivity (e.g., weaker in GFDL) highlights remaining uncertainty in aerosol–radiation and aerosol–cloud interactions.

Implications are that continued aerosol emission reductions projected in many scenarios will likely further strengthen EEI in the near term, contributing to accelerated surface warming atop ongoing greenhouse gas forcing.

This study provides multi-model, multi-ensemble evidence that recent reductions in anthropogenic aerosol emissions have significantly increased Earth’s energy imbalance over 2001–2019 by roughly 0.2 W m⁻² decade⁻¹, approximately doubling the modeled EEI trend. The large positive shortwave trend driving EEI is explained by comparable contributions from aerosol ERF and the radiative response, while longwave trends are dominated by greenhouse gas–related ERF. Incorporating time-evolving aerosols is essential to reproduce observed EEI trends and variability.

Given that many future scenarios project rapid aerosol and precursor emission declines due to air quality policies, aerosol reductions are likely to continue strengthening EEI and accelerating near-term warming. Future research should improve representation of Southern Ocean processes and surface albedo changes, constrain cloud feedbacks and LW cloud trends, refine biomass burning emission histories and SH ocean aerosol loads, and narrow uncertainty in aerosol–cloud interactions and aerosol ERF across models, including through coordinated protocols like CERESMIP.

• Models underestimate the observed EEI trend by ~10–40%, linked to missing/incorrect SW clear-sky trends south of 30°S (likely tied to Southern Ocean/sea-ice/surface albedo changes) and an underestimation of the negative LW cloud trend (cloud feedback biases).
• Aerosol ERF remains uncertain across models; GFDL exhibits weaker aerosol sensitivity, contributing to intermodel spread in ERFaero trends.
• Emission inventories differ (e.g., CEDS April 2021 vs CMIP6 versions); while one CESM2 test suggests negligible impact on EEI trend, broader uncertainties in recent regional emissions (e.g., Chinese SO₂ decline, biomass burning after 2014) remain.
• Biomass burning emissions are climatological after 2014 in several setups, limiting realism for regions like Canada, Siberia, and the Amazon.
• Satellite AOD products are inconsistent over parts of the Southern Hemisphere ocean, complicating evaluation.
• Land ice is fixed in the models, potentially contributing slightly to the underestimated SW trend.
• Prescribed SST/sea-ice boundary conditions introduce structural uncertainty related to dataset choice; GFDL lacked a dedicated fixed-aerosol (AERO2000) experiment within the same setup, requiring cross-study comparisons.