Sustained growth of sulfur hexafluoride emissions in China inferred from atmospheric observations

Sulfur hexafluoride (SF6) is an extremely potent greenhouse gas (100-year GWP ~25,000) with a very long atmospheric lifetime (~1,000–3,200 years), meaning current emissions cause near-permanent radiative forcing. SF6 is regulated under the Kyoto Protocol and now the Paris Agreement. Emissions are primarily from high-voltage electrical switchgear, with smaller contributions from magnesium smelting and other industrial uses; natural sources are negligible. Despite reported reductions from UNFCCC Annex-I countries since the 1990s, global SF6 mole fractions and emissions have increased rapidly since the 2000s, likely due to rising emissions from non-Annex-I countries accompanying rapid growth in power demand and renewable energy deployment. China is thought to be a major contributor given its power sector scale. Bottom-up inventories exist (including official UNFCCC submissions by China for selected years), but show large discrepancies across sources (e.g., lower US EPA estimates vs higher EDGAR and recent studies). Existing top-down estimates were based on measurements outside China (South Korea or Japan), limiting sensitivity to western China, leaving a gap in understanding. This study derives China’s SF6 emissions for 2011–2021 using atmospheric observations from nine sites within China and a top-down inversion framework, compares with bottom-up inventories, examines regional patterns including western China, and assesses China’s contribution to global SF6 emissions.

Previous work indicates Annex-I SF6 emissions have declined due to regulations and voluntary measures, while global totals have risen since the 2000s (AGAGE observations). For China, multiple bottom-up inventories exist: EDGAR v7.0, sectoral estimates focused on the power industry, and official UNFCCC-reported values for 2005, 2010, 2012, 2014, 2017, and 2018. Large discrepancies are noted: US EPA estimated ~1.6 Gg yr−1 for 2018, far below EDGAR and other studies (~4–5 Gg yr−1), with China’s UNFCCC-reported 2018 value (3 Gg yr−1) intermediate. Reasons may include incomplete sector coverage (e.g., excluding emissions during equipment manufacture), inaccurate activity data, and underestimated emission factors. Prior top-down studies (e.g., Simmonds et al. 2020; Fang et al. 2014; measurements in Korea/Japan) provided estimates primarily for eastern China and extrapolated nationally via proxies (e.g., population), limiting insights into western regions. Proxies used historically for SF6 emissions include population and GDP; however, power industry activity and nightlights may be better indicators. Global studies suggest growing non-Annex-I emissions linked to expanding power demand and possibly higher emission factors outside Annex-I.

Atmospheric observations: SF6 measurements were collected at nine China Meteorological Administration (CMA) sites spanning China, including western regions: Akedala (AKD), Mt. Waliguan (WLG), Longfengshan (LFS), Shangdianzi (SDZ), Jinsha (JSA), Lin'an (LAN), Jiangjin (JGJ), Shangri-La (XGL), and Xinfeng (XFG). Sampling included weekly/daily flasks and ~hourly in situ measurements at least 10 km from industrial areas. Flask samples and one in situ series (SDZ) used AGAGE Medusa GC/MS (calibrated to SIO-05); another SDZ in situ series used GC-ECD. Precisions were ~0.98% (in situ ECD), 0.4% (in situ Medusa), and 1% (flasks); flask recovery 99.5–100.5% with stable calibrations. Inverse modeling framework: A regional, hierarchical Bayesian inversion estimated annual emissions (2011–2021) using observations, sensitivities (“footprints”), and priors. Sensitivities to surface emissions and boundary conditions were computed with the UK Met Office NAME Lagrangian model over a domain (5°S–74°N; 55°E–192°E), releasing 20,000 particles per hour from each site within ±10 m vertically and running backward 30 days. Meteorology came from the UKMO Unified Model analyses (spatial resolution improving from 0.352°×0.234° to 0.141°×0.094°, 3-hourly). No chemical loss was applied for SF6. Data handling: Observations with >10% of total sensitivity from the surrounding 25 grid cells (stagnant conditions) were excluded. SDZ in situ data preference was given to Medusa; GC-ECD was used where Medusa coverage was lacking (e.g., 2013–2015). In situ data were averaged to 24-hour values to reduce correlated model error and computation, yielding 4,885 measurements post-filtering. Priors and basis functions: A priori annual emissions used EDGAR v7.0 magnitudes, spatially distributed by DMSP/OLS nightlights. Grid cells were aggregated into 150 basis functions via a quadtree algorithm based on a priori contribution (emission×sensitivity), providing higher resolution near sites or high-emission areas. Emission scaling factors had log-normal priors (μ=0.2, σ=0.8). Boundary condition scaling factors on four lateral boundaries had log-normal priors (μ=1, σ=1) using AGAGE 12-box model background mole fractions at nearest grid points. Uncertainties: Measurement errors used instrument precision and, for averaged in situ data, combined RMS measurement error and within-day standard deviation. Model–observation mismatch was treated as a hyper-parameter with a uniform prior (0–20 ppt). Different prior emissions were tested in sensitivity analyses (Supplementary Discussion 1). Inference: A hierarchical Bayesian framework was solved via MCMC: No-U-Turn Sampler (NUTS) for emission and boundary scaling factors, and slice sampling for the model error hyper-parameter. After 125,000 tuning steps, a 250,000-step chain was run with 5,000 burn-in; 200,000 samples were used to derive posterior means and 68% highest posterior density intervals. Annual inversions were performed independently for each year. Performance was evaluated via uncertainty reduction, RMSE improvement, and correlation improvements between modeled and observed mole fractions. Global emissions context: Global SF6 emissions (2011–2021) were updated using the AGAGE 12-box model assimilating background observations from five AGAGE background sites (SMO, CGO, MHD, RPB, THD). A Bayesian approach included systematic uncertainties: transport (1%), lifetime (20%), and calibration (3%).

- China’s SF6 emissions doubled over the decade: from 2.6 (2.3–2.7) Gg yr−1 in 2011 to 5.1 (4.8–5.4) Gg yr−1 in 2021; increase of 2.6 (2.2–2.9) Gg yr−1 (~100%). - Top-down estimates align with EDGAR v7.0 and Guo et al. bottom-up inventories and sectoral power-industry estimates, but are substantially higher than US EPA and earlier Chinese UNFCCC submissions; the gap narrowed for post-2014 UNFCCC reports. - Significant emissions occur outside eastern China; western and other less-populated regions contribute materially to national totals and growth. Contributions to the national emissions increase (2011–2013 vs 2019–2021) include: north (10%), northwest (7%), central (12%), south (14%), and southwest (11%). - Provincial emissions correlate strongly with power industry size (electricity generation/consumption; r=0.83, p<0.01) and nightlights (r=0.86), but weakly with magnesium production (r=0.11) and less strongly with population (r=0.66) or GDP (r=0.73), indicating the power sector as the dominant source and a better proxy. - China’s share of global SF6 emissions averaged ~46% over 2011–2021, rising from 34% (2011) to 57% (2021). China’s increase of 1.91 (1.69–2.16) Gg yr−1 between 2011–2013 and 2019–2021 exceeds the global increase of 1.04 (0.76–1.33) Gg yr−1, implying an offset of approximately −0.9 Gg yr−1 reductions elsewhere. - The expansion of China’s electricity generation and consumption contributed ~60% of the global increase in electricity over 2011–2021; China accounts for a larger fraction of global SF6 emissions (~46%) than of global power generation/consumption (~20–33%), suggesting higher average emission factors in China. - In 2021, SF6 emissions reached 125 (117–132) Mt CO2-eq yr−1 (100-year GWP=24,300), ~1% of China’s total CO2 emissions and comparable to national CO2 totals of several countries (e.g., Netherlands, Nigeria). The annual SF6 increase rate of 6.5 (5.7–7.2) Mt CO2-eq yr−1 equals ~11% of CO2 reductions from renewable energy deployment.

The results demonstrate that China is the dominant driver of global SF6 emissions growth, with emissions likely concentrated in the power sector, including significant contributions from western regions supporting power generation and long-distance transmission. The finding that China’s emissions growth surpasses the global total increase implies decreasing emissions elsewhere are being offset. Stronger correlations with power industry metrics and nightlights, relative to population or GDP, underscore the importance of electrical infrastructure as the main source. The alignment of top-down estimates with EDGAR and recent bottom-up work, and divergence from US EPA and early UNFCCC reports, suggest earlier inventories likely undercounted sectors or used low emission factors; improved agreement post-2014 indicates methodological or data enhancements. China’s higher fraction of global SF6 emissions than of electricity output suggests higher emission factors or practices compared to the global average. Given SF6’s long lifetime, continued growth risks long-lasting climate impacts, potentially undermining decarbonization gains from renewable energy unless leakage is minimized and substitutes adopted. Enhancing measurement networks, especially in western China, is critical to track emissions, evaluate controls, and refine inventories.

This study provides a decade-long, observation-constrained top-down estimate of China’s SF6 emissions using nine in-country sites and a hierarchical Bayesian inversion. Emissions doubled from 2011 to 2021, with significant contributions from western regions linked to the power sector and transmission infrastructure. China’s emissions account for a growing share of the global total, with national increases sufficient to explain all global growth and offset reductions elsewhere. The work reconciles discrepancies among inventories, indicating improved but still lower official UNFCCC-reported values after 2014. Policy implications include prioritizing SF6 leakage mitigation, adoption of SF6-free technologies or substitutes, and broader regulatory frameworks. Future research should expand high-frequency measurements in under-sampled western regions, improve sector-specific activity data and emission factors (especially in manufacturing and equipment handling), and refine inversion systems and priors to reduce uncertainties and inter-annual variability.

- Top-down annual estimates exhibit inter-annual variability that may arise from changes in observational coverage, model meteorology, model–observation mismatch, or inversion artifacts. - Regional constraints are weaker in subregions with fewer measurements, potentially affecting spatial detail; subregional trends were averaged to reduce such effects. - Results depend on prior emissions (e.g., EDGAR spatial distributions via nightlights) and boundary condition assumptions; alternative priors lead to some differences (tested in sensitivity analyses). - The inversion treats model error as a bulk hyper-parameter and cannot fully resolve all transport uncertainties. - Global context for non-Annex-I countries beyond China remains poorly constrained due to limited data, limiting attribution of rest-of-world trends.