Heterogeneous Community Surveillance–Driven Physics-Informed Reformulation of Fine-Scale Convection–Diffusion Air Pollution Distribution

Taher Alzahrani¹, Saima Rashid^2,*
1 College of Computer and Information Sciences, Imam Mohammad Ibn Saud Islamic University (IMSIU), Riyadh, Saudi Arabia
2 Department of Mathematics, Government College University, Faisalabad, Pakistan
* Corresponding Author: Saima Rashid. Email: email

Computer Modeling in Engineering & Sciences https://doi.org/10.32604/cmes.2026.076957

Received 16 October 2025; Accepted 20 March 2026; Published online 09 April 2026

Download PDF

Abstract

Air pollution poses a serious public health threat in developing countries such as Pakistan, where rapid urbanization and industrialization have intensified atmospheric contamination. Although mobile sensing deployed on public transportation expands spatial coverage beyond fixed monitoring stations, accurate high-resolution pollution mapping remains constrained by sparse observations, computational burden, neglected pollutant interactions, and limited interpretability. To address these challenges, this study proposes a unified physics-informed deep learning framework for fine-grained air pollution map reconstruction and joint multi-pollutant estimation. The framework integrates mobile and stationary monitoring data with atmospheric dispersion principles to enhance physical consistency under limited observational coverage. A physics-guided air pollution map reconstruction model (Phy-APMR) is developed to jointly estimate ozone (O₃), particulate matter (PM2.5), and PM₁₀, while explicitly modeling cross-pollutant interactions through a deep interaction module. A tanh-based self-attention mechanism adaptively weights heterogeneous inputs, and a physics-constrained loss function enforces consistency with atmospheric transport behavior. An adaptive short-time update sampling strategy is further introduced to accelerate convergence and enable high-frequency updates under dynamic transportation conditions. Experiments conducted in major urban regions of Pakistan during 2023–2024 demonstrate that the proposed framework consistently outperforms conventional reconstruction and single-pollutant estimation approaches. The model achieves cross-validation R2 values of 0.92 for O₃, 0.90 for PM_2.5, and 0.86 for PM₁₀, while reducing convergence time by over 80%. Interpretability analysis identifies formaldehyde, carbon monoxide, hydroxyl radicals, and temperature as dominant contributors. Spatial and seasonal analyses indicate pronounced summer O₃ pollution and wintertime particulate accumulation, particularly in Punjab and Sindh. These findings demonstrate the framework’s capability for robust multi-pollutant reconstruction and its potential to support coordinated air quality management and evidence-based environmental policy in Pakistan.