AirFlow: Context Preserving and Multi-Rate State Modeling for Air Quality Forecasting

2026-08-10

AI Summary

Air pollutants exhibit heterogeneity, multi-scale dynamics, and distributional drift, posing significant challenges for conventional modeling approaches. To address these issues, this work proposes AirFlow, a novel framework that models multivariate station observations through a pollutant-aware dual-stream architecture without relying on graph propagation or predefined signal decomposition. The core innovations include a statistics-guided normalization routing mechanism enabling pollutant-adaptive normalization, a hierarchical dual-stream state space model to capture multi-rate dynamics, and a gated bidirectional cross-attention module for fusing multi-scale temporal representations. Evaluated on real-world data across multiple cities, AirFlow achieves state-of-the-art performance in 34 out of 36 metrics, reducing RMSE by up to 11.11% compared to the strongest baselines, while requiring only 0.0483M parameters and 0.0215G FLOPs.

Abstract

Accurate air quality forecasting is essential for public health and urban environmental management, but remains challenging because pollutant channels differ in periodicity and distribution drift, while their concentration trajectories contain both multi-scale dependencies and rapid changes. Recent methods have improved spatial dependency learning and meteorological covariate modeling. However, pollutant channels are still passed through the same normalization rule and temporal backbone, using a shared latent representation for channel-specific distributions and changes at different rates. To address this limitation, we propose AirFlow, a pollutant-aware dual-stream framework that operates on station multivariate observations without additional graph propagation or predefined signal decomposition. Specifically, AirFlow designs two novel blocks: (1) a statistic-guided normalization routing mechanism that selects a normalization path for each pollutant according to its 24-hour autocorrelation and distribution drift; and (2) a hierarchical dual-stream state model that combines multi-scale state space propagation with learnable response coefficients, where gated bidirectional cross-attention exchanges information and adaptively fuses the resulting representations. Experiments on real-world data from multiple cities show that AirFlow achieves the best performance in 34 of 36 metrics comparisons, with reductions of up to 11.11% root mean square error over the state-of-the-art baseline. AirFlow also requires only 0.0483M parameters and 0.0215G FLOPs, achieving high forecasting accuracy with low computational overhead.

Problem

• air quality forecasting
• pollutant channels
• distribution drift
• multi-scale dependencies
• temporal modeling

Innovation

• statistic-guided normalization
• dual-stream state model
• multi-scale state space
• gated cross-attention
• pollutant-aware forecasting