Global mapping of city-level economic growth decoupling from fossil fuels

Urban areas host more than half of the global population and generate about 60% of global GDP and 70% of GHG emissions, making city-level policy central to sustainable development. A key question is whether cities can increase prosperity while reducing dependence on fossil fuel combustion (a focal aspect of green growth). Tracking such progress is difficult because comparable, frequent city-level emissions inventories and subnational economic statistics are scarce and inconsistent, and gridded emissions datasets have limited accuracy at city scales. This study asks whether recent city-level economic growth has decoupled from fossil fuel-intensive activity and proposes a globally consistent, observation-based approach using satellite tropospheric NO₂ as a proxy for combustion-related activity and subnational GDP per capita (PPP) to assess short-term relative decoupling across 5,435 cities (2019–2024).

Previous green growth and decoupling assessments often rely on city or national GHG inventories and GDP, with limited regularity and comparability at the urban scale. Studies have examined decoupling at coarse spatial scales or for selected metropolitan regions, but suffer from data limitations. Satellite-observed pollutants (e.g., NO₂) offer a near-real-time, spatially explicit alternative. NO₂ is short-lived and closely linked to local combustion sources (transport, industry, power) and thus more responsive to changes in energy structure and policy than long-lived gases like CO₂ and CH₄. Existing literature on decoupling (e.g., EKC hypothesis, city-level CO₂ decoupling in China, and national-level analyses) provides context but lacks a comprehensive, globally consistent city-scale assessment integrating both economic and combustion-activity indicators.

• Study scope and spatial units: Urban areas were defined using the Global Urban Polygons and Points Dataset (GUPPD v1), restricting to cities with population ≥100,000, yielding 5,496 urban polygons (metadata include population 2020 and 2025, and names). Analyses focused on 2019–2024.
• NO₂ data and processing: Tropospheric NO₂ vertical column density (TVCD) from Sentinel-5P TROPOMI (level-3, ~2.45 km grid; QA ≥0.75) were obtained (May 2018–May 2025). For each city, daily area-weighted TVCDs were extracted only when ≥80% of the polygon was covered by valid data. Meteorological covariates (from ERA5-Land and ERA5) included wind speed/direction, 2 m air temperature, relative humidity (from T and dew point), surface pressure, downward solar radiation, and planetary boundary-layer height, averaged around the ~13:30 local overpass.
• Trend estimation: Generalized additive models (AirGAM) were fitted to daily NO₂ TVCD time series with meteorological and temporal covariates to derive meteorology-adjusted long-term trend terms and 95% CIs per city. A NO₂ TVCD change factor was defined as the ratio of population-weighted NO₂ TVCD (per 100,000 residents) in Dec 2024 to Jan 2019 (population from GUPPD 2025 and 2020, respectively). Uncertainty was quantified via 10,000-sample Monte Carlo simulations based on trend term CIs to produce 95% CIs of change factors; change was considered significant if CI excluded 1.
• GDP per capita PPP data and QC: A high-resolution gridded GDP per capita (PPP, 2021 international dollars) dataset (admin-2, ~43,501 units; updated from ref. 30; 2014–2024) was used, downscaled from admin-1 using XGBoost with cross-validation. A quality control framework compared subnational annual GDP per capita PPP growth (from DOSE for 2009–2020) against World Bank national GDP per capita growth trajectories to flag implausible regional trends; cities failing (more than 6 out of 11 years outside 95% CI) were excluded (41 cities removed).
• GDP trend extraction and change factor: Annual admin-2 GDP per capita PPP values were interpolated to monthly via modified Akima cubic Hermite interpolation; long-term trend was extracted using singular spectrum analysis (SSA). The GDP change factor was computed as the ratio of the trend value in Dec 2024 to Jan 2019. GDP uncertainty was propagated using a Monte Carlo framework (10,000 samples) based on unit-year standard deviations, producing 95% prediction CIs for each city’s GDP change factor.
• City classification: Among cities with significant NO₂ TVCD change factors and passing GDP QC, the joint signs of NO₂ and GDP change factors determined four categories: cleaner+richer (declining NO₂, increasing GDP), dirtier+richer (increasing NO₂, increasing GDP), cleaner+poorer (declining NO₂, declining GDP), dirtier+poorer (increasing NO₂, declining GDP). Classification is based on start vs. end states (Jan 2019 vs. Dec 2024), not requiring monotonicity.
• Validation and attribution checks: Three validation approaches were used: (1) comparison with national fossil fuel consumption trends, (2) correlation with EDGAR NOₓ emissions, and (3) comparison with long-term surface NOₓ observations across 465 European urban sites. Results show satellites capture direction of surface trends but underestimate magnitudes on average (~16.3% lower). Possible biomass burning influence was assessed by comparing CAMS GFAS biomass burning NOₓ and CAMS anthropogenic NOₓ (2019–2024); for >95% of cities with significant NO₂ trends, median annual fire-to-anthropogenic NOₓ ratios were <6.2%.

• Coverage and significance: NO₂ TVCD change factors were computed for 5,435 cities (>100,000 population). Of these, 2,103 showed significant decreases, 413 significant increases, and 2,919 had no significant change (~54%). East Asia and Europe had the most reductions; China led globally with significant NO₂ TVCD reductions (719 out of 728 cities with significant NO₂ change; 66.3% of all Chinese cities >100k). In the USA, 31.6% of cities >100k showed significant decreases; Germany 98%, France 80.5%, UK 68.5%, Japan 52.7%. Increases were most common in the Middle East, Central Asia and India: Iran (84.9% of cities >100k), Uzbekistan (60.7%). Russia 26.9% and India 15.3% of cities showed significant increases.
• Decoupling typology (n=2,475 cities with significant NO₂ change and passing GDP QC): • Cleaner + richer (green): ~80% (1,974 cities). Largest shares in East Asia (40%), Southern Asia (13%), Western Europe (6%). China contributed 719 green cities, including Beijing, Shanghai, Guangzhou, Suzhou, Chengdu, Jieyang. Many European capitals (Paris, Berlin, Rome, Amsterdam) and North American cities (e.g., San Jose, Denver, Seattle, Austin, Las Vegas; Montreal, Vancouver) also showed decoupling. • Dirtier + richer (brown): 390 cities (16%), concentrated in India (138, 35.4%), Iran (68, 17.4%), Russia (35, 9.0%) and Central Asia (e.g., Uzbekistan). Examples include Riyadh, Moscow, Tashkent, Izmir, Abu Dhabi—suggesting growth driven by automobile-dependent transport, heavy industry and fossil fuel-based power. • Cleaner + poorer (gray): 93 cities (~4%), notably in Sudan (19, 14.6%), Afghanistan (14, 15.1%), Yemen (9, 9.7%), Lebanon (6, 6.5%). Examples: Khartoum, Kabul. Likely reflect economic contraction or deindustrialization rather than deliberate air-quality policy. • Dirtier + poorer (red): 18 cities (~1%), mainly in Iran (5, 27.8%), Libya (5, 27.8%), Angola (2, 11.1%) and India (2, 11.1%). Rising NO₂ with stagnant/declining GDP suggests worsening fossil fuel dependence amid weak economic performance.
• Interpretation highlights: Decoupling evident across many Chinese and European cities, potentially reflecting stringent emissions controls, transport electrification, and energy transitions. Brown and red cities cluster in regions with fossil-fuel-dependent infrastructure and weaker regulation. Urban NO₂ trends align with the environmental Kuznets curve: many green cities appear on the declining arc, whereas brown/red cities remain on the rising arc.

Findings indicate that short-term relative decoupling between economic growth and fossil fuel-dependent activity is already occurring in a large share of cities, especially across East Asia (notably China) and Europe, and in parts of North America, pointing to the efficacy of policies such as emissions controls, low-emission zones, cleaner transport and energy transitions. Conversely, many cities in South Asia, the Middle East and parts of Central Asia remain tightly coupled to fossil fuels, indicating risks of pollution-intensive growth pathways where rising incomes have not yet translated into cleaner combustion or stricter regulation. The typology aligns with the environmental Kuznets curve: green cities exemplify the declining-pollution phase enabled by technology adoption and regulation; brown/red cities lie on the rising arc where growth stimulates increased NO₂. Gray cities show reductions associated with economic stagnation or conflict rather than deliberate policy. The satellite-based, globally consistent framework offers a practical monitoring and benchmarking tool for policy evaluation and international support targeting urban sustainability transitions.

This study introduces a globally consistent, observation-based framework to assess city-level relative decoupling between fossil fuel-dependent activity (proxied by satellite NO₂ TVCD) and economic growth (GDP per capita PPP). Applying it to 5,435 cities (2019–2024) reveals that most cities with significant NO₂ trends and reliable GDP data exhibit cleaner + richer trajectories, especially in China and Europe, while a smaller share (notably in South Asia and the Middle East) remain dirtier + richer or even dirtier + poorer. The approach enables repeatable, comparative tracking of urban green growth and policy effectiveness. Future work should extend the time horizon beyond six years, refine attribution to specific sectors and policies, improve urban-scale economic datasets with robust uncertainties (especially in low- and middle-income regions), and integrate additional environmental and social indicators to broaden assessments beyond NO₂ and GDP.

• Indicator scope: NO₂ TVCD is an imperfect proxy for fossil fuel combustion and not a direct measure of CO₂ emissions or comprehensive air quality; results indicate decoupling of combustion-related activity, not absolute reductions in total emissions or overall environmental impact. Pollution embodied in imports is not considered.
• Time horizon and events: The 2019–2024 period is short and overlaps COVID-19, potentially affecting trends and limiting inference about long-term structural shifts.
• Column–surface representativeness: Tropospheric columns generally reflect boundary-layer NO₂ over land, but satellites tend to underestimate surface trend magnitudes; dynamics and lifetime variations can introduce discrepancies. City-wide averaging masks intra-urban variability.
• Biomass burning and attribution: While biomass burning NOₓ contributions are small for most cities, they can affect some regions (e.g., central Africa). Emissions outside administrative boundaries but affecting cities may be misattributed.
• Economic data uncertainties: Subnational GDP products involve modeling and downscaling uncertainties, especially where national statistics are sparse. GDP change factor CIs can be large; QC removed 41 cities with implausible trends. Population uncertainties in GUPPD were not propagated into NO₂ change factor uncertainty.
• Methodological choice: Classification relies on start vs. end state change factors and may obscure intermediate variability; however, trajectories rarely crossed quadrants.