Exploring Spatial Influence of Remotely Sensed PM2.5 Concentration Using a Developed Deep Convolutional Neural...

Area of Science:

• Environmental Science
• Data Science
• Geospatial Analysis

Background:

• Increasing volumes of remote sensing data necessitate innovative spatial analysis methods, particularly for big data applications.
• Existing methods may not fully capture complex spatial relationships in environmental big data.

Purpose of the Study:

• To propose and validate a deep convolutional network (CNN) model for spatial analysis of remotely sensed big data.
• To investigate the spatial influence of population, GDP, terrain, and land-use/land-cover (LULC) on PM2.5 concentrations over China.

Main Methods:

• Development and application of a deep convolutional network (CNN) model for spatial feature extraction.
• Comparative analysis against geographically weighted regression (GWR) to evaluate model accuracy and performance.
• Quantification of the spatial influencing magnitude of various factors on PM2.5 concentrations.

Main Results:

• The deep CNN model demonstrated high accuracy in analyzing remotely sensed big data, significantly outperforming GWR.
• Population, GDP, terrain, and LULC collectively explained 97.85% of the spatial distribution of PM2.5 annual concentrations.
• Terrain and LULC were identified as dominant factors, individually influencing PM2.5 by 50.07% and 40.91%, respectively, and together explaining 96.65% of the spatial pattern.

Conclusions:

• Deep CNN models are highly effective for spatial analysis of remotely sensed big data.
• Terrain and LULC are critical determinants of PM2.5 spatial distribution in China.
• The developed CNN approach offers a powerful tool for environmental big data analysis and policy-making.