Physics-based deep learning reveals rising heating demand heightens air pollution in Norwegian cities

TL;DR

This work tackles the intertwined effects of climate factors and air quality, focusing on Norway's three major cities. It contrasts traditional regression and ML-based feature selection with Physics-Based Deep Learning (PBDL) and LSTM for predicting NOx and PM2.5 using a decade of daily traffic and weather data, incorporating an Ordinary Differential Equation (ODE) framework to encode physical processes. The study finds Heating Degree Days (HDD) strongly correlate with higher pollution, especially PM2.5, and shows PBDL generally improves predictive accuracy over LSTM, notably reducing PM2.5 RMSE in Trondheim by about 6.39 μg/m³. These results demonstrate the value of physics-informed forecasting for informing data-driven climate and air quality policies, while highlighting city-specific dynamics and policy interactions. The methodological blend—panel models, ML-based feature selection, and PBDL with ODE losses—offers a robust blueprint for policy-relevant AQ forecasting in data-rich environments with strong physical underpinnings.

Abstract

Policymakers frequently analyze air quality and climate change in isolation, disregarding their interactions. This study explores the influence of specific climate factors on air quality by contrasting a regression model with K-Means Clustering, Hierarchical Clustering, and Random Forest techniques. We employ Physics-based Deep Learning (PBDL) and Long Short-Term Memory (LSTM) to examine the air pollution predictions. Our analysis utilizes ten years (2009-2018) of daily traffic, weather, and air pollution data from three major cities in Norway. Findings from feature selection reveal a correlation between rising heating degree days and heightened air pollution levels, suggesting increased heating activities in Norway are a contributing factor to worsening air quality. PBDL demonstrates superior accuracy in air pollution predictions compared to LSTM. This paper contributes to the growing literature on PBDL methods for more accurate air pollution predictions using environmental variables, aiding policymakers in formulating effective data-driven climate policies.

Figures (23)

• Figure 1: The methodological flow chart. Initially, we conducted feature engineering on the dataset to identify the primary climate factors influencing air pollution. Subsequently, we employed a panel regression model, along with three machine learning algorithms: K-means, Hierarchical clustering, and Random Forest, to explore and compare the feature selection outcomes of these models. Next, we evaluated the predictive performance of air pollution using PBDL and LSTM models. Finally, we integrated the feature selection and prediction outcomes to draw our conclusions.
• Figure 2: Monthly levels of two pollutants across the three cities.
• Figure 3: The correlation results from the panel models at individual city and combined scale (a) for NOX and (b) for PM2.5. Note: The full form of variables is presented in Table 1. The text in the image may appear small and may require zooming in for better visibility.
• Figure 4: Cluster plots generated using K-means. Note: The text in the image may appear small and may require zooming in for better visibility.
• Figure 5: Hierarchical clustering plot after data normalization with no quadratic and cubic variables. The full form of the variables can be inferred from Table 1.