High-Precision Microscale Particulate Matter Prediction in Diverse Environments Using a Long Short-Term Memory Neural...

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Particulate matter (PM) air pollution poses significant risks to public health.
• Accurate and accessible methods for monitoring PM concentrations are crucial for urban planning and public health initiatives.
• Traditional monitoring methods can be resource-intensive and spatially limited.

Purpose of the Study:

• To develop and evaluate a novel Long Short-Term Memory (LSTM) neural network model for estimating air quality based on particulate matter (PM).
• To leverage color features (HSV: hue, saturation, value) from street images for air quality estimation.
• To assess the model's performance, interpretability, and generalization capabilities across different environments and seasons.

Main Methods:

• Extraction of HSV color features from street view images.
• Development of a novel LSTM neural network model (LSTM-HSV).
• Utilized concentration data for eight ambient PM parameters (PM1.0, PM2.5, PM10, particle number concentration, lung-deposited surface area, equivalent mass concentrations of ultraviolet PM, black carbon, and brown carbon).
• Comparison with other deep learning models (Recurrent Neural Network, Gated Recurrent Unit) and statistically based models.
• Evaluation of model performance, interpretability, and generalization across urban, suburban, village, and harbor environments, and across different seasons.

Main Results:

• The LSTM-HSV model demonstrated superior performance compared to other deep learning and statistically based models.
• The model achieved interpretability rates exceeding 80% for eight key PM metrics.
• Successful application in different seasons and diverse environments (urban, suburban, villages, harbor) confirmed satisfactory generalization capabilities.
• The model accurately estimated various particulate matter parameters.

Conclusions:

• The proposed LSTM-HSV model is a promising and versatile tool for estimating air pollution, specifically particulate matter.
• The model's ability to generalize across temporal and spatial dimensions makes it valuable for real-world applications.
• This approach offers a novel, image-based method for air quality assessment with significant implications for urban planning and public health.