AI-powered algorithms enhance satellite monitoring of air pollution

The research team, led by Professor Shi Chong of the Institute for Aerospace Information (AIR), at the Chinese Academy of Sciences, has made great strides in satellite-based air quality monitoring, working with Japanese scientists.

They have developed Airtrans, an innovative algorithm that greatly improves accuracy and efficiency to acquire key aerosol properties from multispectral satellite observations. This study is published in the Journal Remote Sensing of Environment.

Particle particles suspended in the atmosphere – play a key role in air pollution, human health, and climate change. Two key indicators for studying aerosols are the aerosol light thickness (AOT), which measures the concentration of particles in the air, and the fine mode fraction (FMF), which indicates the proportion of small particles, such as vehicle discharges and industrial processes.

Traditional satellite search algorithms that rely on pre-computed lookup tables typically use only two or three spectral channels. This limitation reduces the details and accuracy of the results. The best method of estimating Fullphysics provides more accuracy by leveraging additional spectral information, but often struggles with complex aerosol types that are too slow for real-time applications.

To address these challenges, researchers developed Airtrans, a hybrid algorithm that combines physical modeling and machine learning techniques. This algorithm employs a radiative transfer model called RSTAR to simulate satellite spectral reflectance across a variety of aerosol types and surface properties, and creates an extensive database that pretrains neural network models.

This model is fine-tuned using ground-based measurements to increase accuracy and adaptability to a variety of real-world scenarios. Additionally, AirTrans explains hourly changes in surface reflectance and background AOT, further improving its accuracy.

Airtrans is specially designed for use in multispace equipment such as the advanced Himawari Imager (AHI) Himawari-8, which observes the East Asian region every 10 minutes. This feature changes the aerosol’s real-time high-frequency tracking throughout the day. This is a difficult task to achieve by traditional methods.

When tested against ground measurements, Airtrans achieved a root mean square error of just 0.132 in AOT and 0.146 in FMF, representing a significant improvement of about 40% and 49%, respectively, compared to the official Himawari-8/Ahi product.

In particular, AirTrans resolved the systematic underestimation of FMF found in existing search products. During real-world events such as Dust Storm and Haze episodes, Airtrans successfully acquired spatio-temporal dynamics of aerosol concentration and particle size, highlighting the effectiveness of contamination monitoring and early warning purposes.

This study shows that Airtrans is not only accurate and fast, but also widely applicable. It can be adapted for use with other multispectral satellite sensors, and can support accurate air pollution prevention and climate change efforts, particularly in the context of carbon neutrality goals.