Monitoring accurate global mapping of atmospheric formaldehyde (HCHO) is important because it reflects the oxidation of volatile organic compounds (VOCs) and helps scientists understand photochemical smog, tropospheric ozone (O3) formation, and reactive carbon cycling. Existing satellite products, especially those from sun-synchronous platforms, have already improved global observation, but cloud cover, orbital gaps, and instrumental noise still leave many missing or distorted regions. China’s GF-5B satellite offers valuable morning observations, yet its lower signal-to-noise ratio and striping artifacts have limited its wider use. Based on these challenges, in-depth research is needed on seamless, observation-driven reconstruction of global HCHO fields.
Researchers from the University of Science and Technology of China, the Hefei Institutes of Physical Science, Chinese Academy of Sciences, Anhui University, and collaborating institutions reported (DOI: 10.34133/remotesensing.1043) this advance in the Journal of Remote Sensing, published on 16 March 2026. The study presents a new way to recover continuous global HCHO VCDs from sparse GF-5B satellite observations, addressing a major challenge in atmospheric monitoring: how to extract reliable pollution information from incomplete, noisy satellite data without depending on chemical transport models.
The core innovation lies in combining physical HCHO spectral retrieval with a Spherical Fourier Neural Operator (SFNO), a model designed to learn global geophysical patterns directly on Earth’s spherical surface. Unlike conventional interpolation or model-driven gap filling, this approach is observation-driven and learns to reconstruct missing data from meteorology, emissions, topography, and surrounding satellite signals. It also suppresses the striping noise common in GF-5B retrievals. Validation showed that correlation with ground-based measurements improved from 0.56 for original GF-5B data to 0.75 after reconstruction, even exceeding the performance of TROPOspheric Monitoring Instrument (TROPOMI) data in some regions.
The team trained the model using global GF-5B observations collected from December 2021 to December 2023, together with meteorological variables from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis, emission inventories, population density, and topographic information. A masked reconstruction strategy was used so that the network could learn to recover missing values from incomplete inputs. The model was first trained at coarse resolution and then refined to 0.1° resolution, enabling both global consistency and local detail.
Independent validation used multiaxis differential optical absorption spectroscopy (MAX-DOAS), Pandora spectrometers, and high-performance liquid chromatography (HPLC) measurements. Across global sites, AI-reconstructed HCHO showed clear gains in agreement with observations. At Pandora sites, the average correlation rose from 0.60 to 0.76. In Guangzhou, reconstructed near-surface HCHO also showed a strong correlation of 0.78 with ground-based HPLC data. The reconstructed maps revealed coherent HCHO enhancements linked to tropical vegetation, industrial corridors, and the 2023 Canadian wildfires, including pollution transport toward New York City.
Suggested quote for press release: “This work shows that even noisy satellite observations can be transformed into chemically meaningful global products when combined with AI that learns atmospheric physical and chemical laws form observations. The framework not only improves data quality, but also expands the value of underused satellite missions for air quality tracking, emission analysis, and rapid environmental assessment.”
The researchers retrieved HCHO from the Environmental Trace Gases Monitoring Instrument (EMI) aboard the GF-5B satellite and merged these data with ERA5 meteorological fields, emission inventories, population density, and elevation data. The SFNO model used 62 input channels and was trained through self-supervised masked reconstruction, then fine-tuned with observed same-day satellite values. Ground validation relied on MAX-DOAS, Pandora, and HPLC measurements colocated with satellite overpasses.
This technology could strengthen near-real-time air quality forecasting by providing more continuous HCHO inputs for predicting secondary pollutants such as O₃ and secondary organic aerosol. The framework is also transferable to other satellite platforms and trace gases, making it promising for multisensor global monitoring. Because daily fine-tuning can be completed in only a few minutes, the approach could support operational pollution surveillance, wildfire response, emission policy assessment, and long-term atmospheric chemistry research at a global scale.
###
References
DOI
10.34133/remotesensing.1043
Original Source URL
https://spj.science.org/doi/10.34133/remotesensing.1043
Funding information
This research was funded by the National Natural Science Foundation of China (42225504,42422703, and 42305199), the National Key R&D Program of China (2022YFC3700100 and 2023YFC3710500), the New Cornerstone Science Foundation through the XPLORER PRIZE (2023-1033), the Key Research and Development Project of Anhui Province (2023t07020015), the Youth Innovation Promotion Association of CAS (2021443), the HFIPS Director’s Fund (BJPY2022B07 and YZJJQY202303), and the Hefei Comprehensive National Science Center.
About Journal of Remote Sensing
Journal of Remote Sensing, an online-only Open Access journal published in association with AIR-CAS, promotes the theory, science, and technology of remote sensing, as well as interdisciplinary research within earth and information science.