Developing PM_2.5 and PM₁₀ prediction models on a national and regional scale using open-source remote sensing data

Area of Science:

• Environmental Science
• Remote Sensing
• Atmospheric Science

Background:

• Air quality is crucial for public health, with particulate matter (PM2.5 and PM10) being key pollutants.
• Satellite remote sensing, particularly the Sentinel-5P TROPOMI mission, is vital for atmospheric monitoring.
• Accurate remote sensing of PM2.5 and PM10 remains challenging, often relying solely on ground stations.

Purpose of the Study:

• To estimate PM2.5 and PM10 concentrations in Croatia using Sentinel-5P and Google Earth Engine data.
• To develop and validate seasonal air quality models for heating and non-heating periods.
• To assess the feasibility of using remote sensing for accurate particulate matter monitoring.

Main Methods:

• Utilized Sentinel-5P and open-source remote sensing data via Google Earth Engine.
• Employed ground-based data from the National Network for Continuous Air Quality Monitoring as ground truth.
• Trained seasonal air quality models using a random forest algorithm with machine learning.

Main Results:

• Developed models demonstrated moderate to high accuracy in estimating PM2.5 and PM10.
• Seasonal variations in particulate matter concentrations were visually mapped and analyzed.
• The approach successfully estimated air quality using integrated remote sensing and ground-truth data.

Conclusions:

• The proposed machine learning approach effectively estimates PM2.5 and PM10 air quality.
• Satellite remote sensing combined with ground data offers a viable method for monitoring particulate matter.
• The study highlights the potential for improved air quality assessment through advanced data integration.