Satellites Reveal Steep Decline in Carbon Monoxide Emissions Across Central Asia

Carbon monoxide pollution across Central Asia has long been one of the blind spots of global air-quality monitoring. Ground stations are sparse, emission inventories are often outdated, and the region’s sources range from sprawling coal-fired power plants to smoldering underground coal fires that burn for years without any smokestack to measure. A new study published in Atmospheric Chemistry and Physics has now built a six-year, satellite-driven picture of carbon monoxide emissions across the region, and its central finding is striking: average emissions fell by 37.9 percent between 2019 and 2024, while the intensity of the region’s major pollution hotspots dropped by 40.8 percent over the same period.

The research team, led by Ye Feng and Jason Blake Cohen of the China University of Mining Technology, used observations from the TROPOspheric Monitoring Instrument, or TROPOMI, flying aboard the Sentinel-5 Precursor satellite. Launched in October 2017, TROPOMI measures carbon monoxide total columns in the shortwave infrared absorption band near 2.3 micrometers, covering the entire globe once daily with a swath 2600 kilometers wide and a nadir resolution of 5.5 by 7 kilometers. The researchers combined these carbon monoxide retrievals with concurrent formaldehyde columns and wind fields from the ERA5 reanalysis to drive a modeling system called the Model-Free Inversion Estimation Framework, which derives surface emissions directly from satellite data without requiring a pre-existing emission inventory or computationally intensive chemical transport model simulations.

The study domain spans 35 to 50 degrees north and 70 to 100 degrees east, encompassing Xinjiang in northwestern China together with Kazakhstan, Kyrgyzstan, Uzbekistan, and parts of several neighboring countries. This is a region built on coal. Xinjiang hosts extensive coal mining, coal-dependent heavy industry, and power plants that generate electricity exported to eastern China. Kazakhstan, Kyrgyzstan, and Uzbekistan add their own energy and mineral industries to the regional mix. The area also contains rural zones where residential solid-fuel burning and occasional biomass burning shape the seasonal pollution cycle, and it sits atop persistent underground coal fires that can ignite through spontaneous oxidation of exposed seams and intensify when mining fractures the coal.

Technically, the inversion rests on a mass-conservation equation for the carbon monoxide column. Emissions act as a source term that increases the column, chemical decay acts as a sink, formaldehyde serves as a proxy for secondary carbon monoxide production from the oxidation of volatile organic compounds, and wind-driven transport terms capture both horizontal advection and pressure-induced convergence or divergence of the atmosphere itself. The framework fits weighting coefficients for each 0.05 by 0.05 degree grid cell on a month-by-month basis using least-squares regression against all overlapping daily satellite and meteorological data, discarding fits that fail statistical significance tests or yield physically implausible coefficients. Sampling the resulting probability distributions one thousand times per grid cell yields both a daily mean emission estimate and a measure of parameter-induced variability.

What sets this study apart is its unusually candid treatment of uncertainty. The team deliberately perturbed the retrieved carbon monoxide and formaldehyde columns within conservative bounds of plus or minus 10 percent, and separately plus or minus 20 percent, then propagated these perturbations through the entire inversion to see how much the emission estimates moved. The result was sobering: roughly 69 percent of the daily grid-cell emission estimates fell outside the adopted uncertainty bounds, meaning that atmospheric variability, dust loading, small sources, and retrieval noise can dominate the apparent signal at the scale of individual grid cells. Under the stricter 10 percent filter, only 31.3 percent of grid cells remained within the uncertainty bound over the six-year period, rising to just 33.7 percent under the 20 percent threshold. The authors argue this challenges a common assumption that reducing observational uncertainty automatically produces more precise emission results, noting that both very high and very low emission estimates can be distorted by noise.

Dust posed its own problem. Mineral dust absorbs light differently across wavelengths than black carbon, so the team used the ratio of absorption aerosol optical depth at 443 nanometers to that at 880 nanometers, measured by NASA’s Multi-angle Imaging SpectroRadiometer, to flag dust-dominated areas. Grid cells with ratios above 4.5 were excluded, removing roughly 14 percent of the study area, mostly in the deserts of western China and southern Kazakhstan, where absorbing aerosols would otherwise corrupt the gas retrievals and produce spurious emission signals.

After all the filtering, the numbers tell a consistent story of decline. Regional mean emissions dropped from 4.93 micrograms per square meter per second in 2019 to 3.06 in 2024, a statistically significant seasonally adjusted trend of minus 0.33 micrograms per square meter per second per year, with a 95 percent confidence interval running from minus 0.52 to minus 0.15. An independent seasonal Mann-Kendall test corroborated the decrease with a slope of minus 0.20 and a p-value below 0.001. Eleven emission hotspots, defined as areas exceeding twice the regional mean flux of 6.72, were identified between roughly 40 and 47 degrees north, and their average intensity fell from 9.51 to 5.63 micrograms per square meter per second. The probability distribution of emissions shifted distinctly toward lower values, and the spatial variability of estimates weakened as well, with the peak of the standard-deviation distribution dropping from 1.66 to below 1.30.

The seasonal fingerprint is exactly what coal-dependent economies would suggest. Emissions peaked in winter and early spring, consistent with residential heating and the industrial production push that ends around the Chinese New Year, and dipped in late spring and summer when hydroxyl radical abundance rises and carbon monoxide is oxidized faster. Monthly regional means fell from an exceptionally high 9.25 in January 2019 to roughly 2.5 to 3.0 after 2022, though wintertime rebounds occurred, including a value of 5.07 in December 2023. Category-level analysis of a fixed sample of continuously operating facilities showed declines everywhere: city-center locations fell 42.2 percent, underground coal-fire sites 41.0 percent, steel plants 34.4 percent, and coal-fired power plants 38.9 percent between 2019 and 2024.

Some of the most intriguing findings concern sources that conventional inventories miss entirely. At four persistent underground coal-fire sites, the satellite-derived emissions were clearly enhanced while the EDGAR bottom-up inventory showed almost nothing, because such fires are not represented as a separate source category in standard anthropogenic inventories. At the Wugonggou coal-fire site, the carbon monoxide column showed weaker seasonal variation than nearby urban and industrial locations, consistent with the relentless, year-round nature of underground combustion, and peak emissions there occurred a year later than at industrial and urban sites. Meanwhile, comparisons with EDGAR in urban centers revealed the opposite pattern: the inventory exceeded the satellite-derived estimates by factors of roughly 2.0 in Urumqi, 2.4 in Almaty, and a remarkable 63 in Jiayuguan, pointing to systematic differences in how bottom-up inventories allocate and time their emissions.

The study also probed the link between coal burning and carbon monoxide directly. For eleven coal-fired power plants in 2023, coal consumption ranged from about 57 to 206 kilotonnes per year while grid-scale emissions ranged from roughly 2.8 to 4.0 thousand tonnes of carbon monoxide per year, showing a moderate positive association with a correlation coefficient of 0.59 that narrowly missed conventional significance. The emission-to-coal-consumption ratio, ranging from 14 to 53 grams of carbon monoxide per kilogram of coal, fell steeply as consumption rose, though the authors caution this inverse relationship is partly an artifact of mathematical coupling. The researchers are careful about limits: no independent ground-based or facility-level measurements exist for validation, grid-cell values include all sources within the cell rather than individual facility stacks, and the observed changes may reflect emission controls, industrial restructuring, meteorology, and sampling effects that cannot be fully separated. Even so, the work demonstrates that observation-driven inversion can characterize regional pollution in data-sparse regions and flag the sources, from coal fires to shifting industries, that inventories overlook.

Subject of Research: Satellite-based estimation of carbon monoxide emissions and their uncertainty over Central Asia from 2019 to 2024

Article Title: Quantifying and attributing CO emissions over Central Asia using TROPOMI and explicit observational uncertainty

Article References: Feng, Y., Cohen, J. B., Li, X., Lu, L., Wang, S., Liu, Z., Wang, L., Liu, J., & Qin, K. (2026). Quantifying and attributing CO emissions over Central Asia using TROPOMI and explicit observational uncertainty. Atmospheric Chemistry and Physics, 26(20), 15139-15159. https://doi.org/10.5194/acp-26-15139-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-15139-2026

Keywords: carbon monoxide, Central Asia, TROPOMI, satellite observations, emission inversion, air pollution, uncertainty analysis, coal fires, EDGAR, Xinjiang, coal consumption, emission inventories