Press Release

Board View

Combining satellite remote sensing with AI to generate fine dust information

▷ An application incorporating AI technology into satellite-based total atmospheric measurements can estimate surface PM10 and PM2.5 concentrations.

▷ A new approach to air quality information over a wide range of areas, compensating spatial limits of ground-based AQ monitoring networks


On 30 December 2021, the National Institute of Environmental Research (NIER) under the Ministry of Environment started to release imagery of estimated surface PM10 and PM2.5 concentrations converted from aerosol measurements by the Geostationary Environment Monitoring Spectrometer (GEMS) onboard GK-2B, with the help of AI technology.

* Aerosols: Solid or liquid particles suspended in the air. Particulate matter (PM) is a typical type of aerosol.


The Environmental Satellite Center (ESC) of NIER has developed an application that converts GEMS total column aerosol measurements (Fig. 1) into surface PM concentrations (Fig. 2), which directly impacts human health. The application incorporates GEMS data with AI techniques and other sources such as meteorological data. 


[Fig. 1] GEMS observation [Fig. 2] Surface concentrations   (entire atmosphere) (0 ? 10 m height)
 

To better reflect the most recent PM10 and PM2.5, this AI-based method adopted a real-time learning model that uses the previous 30-day data and accordingly regulates the sampling and weighting ratios of low and high concentrations.


The researchers compared estimated surface PM concentrations with the actual measurements by ground-based instruments from January to November this year to verify the accuracy of this application.


The estimated PM10 and PM2.5 concentrations tend to be about 10% lower than the actual measurements. Still, the differences were marginal. From a seasonal perspective, higher correlations were shown in winter and spring, when dust and/or episodic PM events frequently occur, than in summer.

※ Underestimation of monthly average concentrations - 7% (June) to 14% (March) for PM10 and 8% (June) to 14% (November) for PM2.5. Correlation (R) is higher in winter to spring (0.9 or higher) than in summer (0.7 ? 0.8).


In addition, even when there are big differences between satellite- and ground-based measurements, for example, high levels of fine dust passing at high altitudes [Fig. 3(a)] without causing a noticeable impact on the lower atmosphere [Fig. 3(b)], the estimated surface PM concentrations [Fig. 3(c)] showed reliable agreement with ground-based measurements [Fig. 3(b)].


[Fig. 3] An episodic PM event passing at high altitude (22 May 2021, huge gap between satellite- and ground-based measurements)   (a) Satellite-based (AOD*)	(b) Ground-based (PM2.5)	(c) Estimation (PM2.5)
 

* Aerosol Optical Depth (AOD): A quantitative value of aerosols in the atmosphere that shows how much light is scattered or absorbed by aerosols. Higher values indicate higher concentrations of aerosols.


While ground-based monitoring networks provide only point-level air quality data (Fig. 4), GEMS can observe a large part of Asia on an hourly basis. GEMS also enable area-level PM estimates production in synergy with AI technology to produce area-level PM estimates (Fig. 5). 


Therefore, this new approach is expected to contribute to overcoming the spatial coverage constraints of ground-based AQ monitoring networks by providing air quality information of areas with no or few monitoring sites in operation.


[Fig. 4] Ground-based measurements   (point-level data)	[Fig. 5] Satellite-based surface PM estimation (area-level data)
 

The ESC released on its website (https://nesc.nier.go.kr) eight GEMS products* in March and five more** in October 2021. Since November, imagery of aerosol mass flow rate, which combines GEMS data with various techniques and other data sources, also has been provided. 

* Aerosols, total ozone, sulfur dioxide, nitrogen dioxide, effective cloud fraction, vitamin D synthesis index, plant response index, DNA damage index

** Aerosol index (UV/VIS), single scattering albedo, effective cloud pressure, cloud radiance fraction, UV index


The imagery of estimated surface PM concentrations is the second application of GEMS data released by the ESC.


NIER will continue applying this AI-based estimation method to other air pollutants such as nitrogen dioxide.


Eunhae Jeong, Director General of the Climate and Air Quality Research Department of NIER, said, "It is an unprecedented milestone to release surface PM level estimation that combines AI with satellite data. Cutting edge technology has only been tried for research purposes before." She added, "Researchers at NIER are doing their best to provide reliable air quality information to the public using GEMS and other satellite data."


Attachments 

        1. ESC website [Estimated surface PM concentrations]

        2. GEMS data products


For inquiry, please contact 

- Environmental Satellite Center, National Institute of Environmental Research +82-32-560-8430 / 8445 / 8432