The problem of air pollution in the city leads to serious health and environmental issues. The capability to keep close track of it has however, limited capabilities on the capability of ground sensors and the accuracy of satellite data. The satellite provides considerable information about such pollutants as the PM2.5 and the NO2. Yet, their level of spatial resolution is poor (101 km) and hence will not be as helpful when local policy and population health are in need. This study introduces a structure providing artificial intelligence and machine learning to reduce massive sizes of satellites and generate air quality maps with a resolution greater than a kilometer. The system makes use of satellite data and other data including weather and traffic and land use data. It is based on the integrated model that involves the use of Random Forest, CNN-LSTM, or Graph Neural Networks. The framework will produce precise, context-dependent, and real-time pollution maps by recording the change of things as time and space change. The innovation is based on its capability to unite the data of a number of sources, its ability to compute data and its well-structured and sturdy design. This permits easily scalable and economical implementation in regions where ground-based sensors are few. These are the detailed maps, which serve to draw significant areas, develop the urban planning, and offer the grounding of the environmental health regulation based on data.
Paper
Full text
Downscaling of Satellite based Air Quality Maps using AI ML
Semantic Scholar · 2026
Abstract
The problem of air pollution in the city leads to serious health and environmental issues. The capability to keep close track of it has however, limited capabilities on the capability of ground sensors and the accuracy of satellite data. The satellite provides considerable information about such pollutants as the PM2.5 and the NO2. Yet, their level of spatial resolution is poor (101 km) and hence will not be as helpful when local policy and population health are in need. This study introduces a structure providing artificial intelligence and machine learning to reduce massive sizes of satellites and generate air quality maps with a resolution greater than a kilometer. The system makes use of satellite data and other data including weather and traffic and land use data. It is based on the integrated model that involves the use of Random Forest, CNN-LSTM, or Graph Neural Networks. The framework will produce precise, context-dependent, and real-time pollution maps by recording the change of things as time and space change. The innovation is based on its capability to unite the data of a number of sources, its ability to compute data and its well-structured and sturdy design. This permits easily scalable and economical implementation in regions where ground-based sensors are few. These are the detailed maps, which serve to draw significant areas, develop the urban planning, and offer the grounding of the environmental health regulation based on data.