Evaluation of Satellite and Reanalysis Products for Surface PM₂.₅ Using Regulatory-Grade Observations in the Kansai Region of Japan

Reliable surface monitoring of airborne particulate matter with an aerodynamic diameter smaller than 2.5 μm (PM₂.₅) remains limited in many regions around the world. In many countries, particularly developing regions, air-quality assessment relies on sparse or low-cost sensor networks with limited unit or data quality, or is even absent due to the high costs of installation and maintenance. Therefore, satellite observations are often proposed as an alternative option or complementary source for PM₂.₅ information. However, the extent to which satellite-based estimates can reliably represent surface PM₂.₅ concentrations relative to regulatory-grade ground-based measurements remains insufficiently quantified.

This study addresses this gap by evaluating satellite- and model-based PM₂.₅ estimates by comparing them with high-quality ground observations in the Kansai region of Japan. Kansai has a dense network of approximately 270 regulatory-grade PM2.5 monitoring stations, operated under Japan’s Air Pollution Control Act, in combination with urban, coastal, and topographic environments. That is why it is an ideal location and benchmark for conducting this study. Monthly PM₂.₅ observations for 2025 were used as reference data. This study utilizes model-based PM₂.₅ fields from the Copernicus Atmosphere Monitoring Service (CAMS) reanalysis. At the same time, satellite-derived aerosol information was retrieved from the Moderate Resolution Imaging Spectroradiometer (MODIS) Multi-Angle Implementation of Atmospheric Correction (MAIAC) aerosol optical depth (AOD) at 550 nm. The satellite- and model-based products were evaluated using the Pearson correlation coefficient, mean bias, and root mean square error (RMSE).

The results show that after quality control, 2,146 station–month pairs were available for evaluation. CAMS PM₂.₅ product shows only moderate agreement with ground-based measurements with Pearson r at 0.36. It shows a tendency to overestimate surface PM₂.₅ with a positive mean bias of 4.8 µg m⁻³ and an RMSE of 6.7 µg m⁻³. This result shows that CAMS captures large-scale variability, but does not fully represent local PM₂.₅ conditions at monitoring locations. By comparison, MODIS MAIAC AOD has a stronger correlation with observed PM₂.₅ (r = 0.56). This result indicates that changes in satellite-observed aerosol loading are more closely linked to variations in surface PM₂.₅. Using a simple linear model, AOD was able to explain about 31% of the monthly PM₂.₅ variability and substantially improved prediction accuracy, reducing the RMSE to 2.2 µg m⁻³. Seasonal analysis of the MODIS MAIAC AOD also reveals that there is a higher correlation with observed PM₂.₅ during winter and spring, with an r value of around 0.6–0.7. This is due to more stable atmospheric conditions and a lower boundary-layer height, allowing surface particles and atmospheric aerosols to be more closely linked. Meanwhile, in summer, the relationship weakened (r < 0.4), likely because a higher boundary layer and increased humidity reduce the direct connection between column aerosol measurements and surface PM₂.₅. Overall, these results provide quantitative evidence on the strengths and limitations of satellite and reanalysis products as alternative sources of air-quality information, particularly for regions with sparse or no surface monitoring.