Fast Near-Road Pollutant Dispersion Modeling: A Physics-Structured AERMOD-RLINE Surrogate

High-resolution dispersion modeling is essential for capturing the steep spatial gradients of near-road air pollution, yet the regulatory AERMOD-RLINE system remains computationally prohibitive for regional-scale applications due to its complex numerical integration. To address this efficiency bottleneck without compromising fidelity, this study proposes PS-XGB-RLINE, a physics-structured XGBoost surrogate model for RLINE. By integrating a "line source decomposition" strategy with physics-based partitioning, the framework transforms complex network dispersion into standardized unit tasks that preserve physical consistency. Validation across geographically independent meteorological data—training in Texas and testing in California and Shanghai—demonstrates that the model achieves predictive performance comparable to the RLINE ( R 2 >0.98, MAE < 1.0 μg/m 3 ), while delivering a 15-fold computational speedup (< 0.6 s/hour for 1 km 2 regional road networks). This approach effectively resolves the conflict between physical rigor and operational speed, enabling the efficient implementation of high-resolution traffic pollution simulations over a full year at a regional scale. • PS-XGB-RLINE resolves the conflict between physical rigor and operational speed. • Standardized source decomposition ensures geometric generalization across cities. • Physics-based partitioning guarantees consistency under diverse turbulence regimes. • Achieves >15-fold speedup with high fidelity ( R 2 >0.98) against regulatory models. • Enables efficient year-long, high-resolution regional traffic pollution simulations.