Graph-Based Physics-Guided Urban PM2.5 Air Quality Imputation with Constrained Monitoring Data

Abstract: This work introduces GraPhy, a graph-based, physics-guided learning framework for high-resolution and accurate air quality modeling in urban areas with limited monitoring data. Fine-grained air quality monitoring information is essential for reducing public exposure to pollutants. However, monitoring networks are often sparse in socioeconomically disadvantaged regions, limiting the accuracy and resolution of air quality modeling. To address this, we propose a physics-guided graph neural network architecture called GraPhy with layers and edge features designed specifically for low-resolution monitoring data. Experiments using data from California's socioeconomically disadvantaged San Joaquin Valley show that GraPhy achieves the overall best performance evaluated by mean squared error (MSE), mean absolute error (MAE), and R-square value (R2), improving the performance by 9%-56% compared to various baseline models. Moreover, GraPhy consistently outperforms baselines across different spatial heterogeneity levels, demonstrating the effectiveness of our model design.

• The paper demonstrates that integrating graph neural networks with physics-based models significantly improves PM2.5 imputation accuracy under data constraints.
• It introduces the GraPhy framework designed to simulate diffusion and convection processes, yielding robust performance over sparse monitoring networks.
• Empirical results from California's San Joaquin Valley show improvements in MSE, MAE, and R-square, underscoring the model's practical impact.

Graph-Based Physics-Guided Urban PM2.5 Air Quality Imputation

The study, "Graph-Based Physics-Guided Urban PM2.5 Air Quality Imputation with Constrained Monitoring Data," presents a novel framework named GraPhy—designed to improve air quality modeling in urban areas that suffer from sparse monitoring networks. The focus of this research is on enhancing PM2.5 estimation in socioeconomically disadvantaged regions with constrained air quality monitoring resources.

Key Contributions

GraPhy integrates graph neural networks (GNNs) with physics-guided principles to address challenges inherent in sparse monitoring data. It is significant in its ability to deliver high resolution and accurate air quality predictions even when data is limited. The framework uniquely combines the following components:

1. Graph Neural Networks (GNNs): Utilized for flexibility in modeling both individual sensor data and inter-sensor relationships, offering an efficient approach to spatial modeling where sensor density is low.
2. Physics-Guided Learning: Incorporates domain knowledge from Convection-Diffusion equations into the GNN architecture. This integration aims to bolster model efficacy in the presence of high spatial heterogeneity.
3. Custom Architecture Design: GraPhy layers are crafted with specific functionalities to overcome issues like sparse data through modules that simulate diffusion and convection processes, allowing dynamic fusion of information based on input data distribution.

Numerical Results

Evaluations conducted using air quality data from California’s San Joaquin Valley reveal that GraPhy significantly improves prediction accuracy over various baseline models. The model achieves superior performance in terms of mean squared error (MSE), mean absolute error (MAE), and R-square values. Specifically, it shows enhancements ranging from 9% to 56% relative to established benchmarks and maintains robustness across different levels of spatial heterogeneity.

Implications and Future Directions

The practical implications of GraPhy are profound, particularly for urban environments where socioeconomic factors restrict comprehensive monitoring systems. By leveraging a graph-based approach, GraPhy can potentially lead to more equitable and effective air quality management strategies. The framework can be further adapted to integrate additional sources of air quality data and extend its functionality across diverse urban landscapes.

Theoretically, this research paves the way for more sophisticated integration of physical laws with machine learning models—enabling smarter and more localized environmental predictions. Future advancements could focus on expanding this hybrid modeling approach to incorporate more complex environmental factors, including seasonal variations and additional pollutant types.

Overall, GraPhy exemplifies innovative strides in urban air quality modeling by offering a viable solution for regions with constrained monitoring infrastructure while laying the groundwork for continued exploration in physics-guided machine learning.