Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Breaking the Regional Barrier: Inductive Semantic Topology Learning for Worldwide Air Quality Forecasting

Zhiqing Cui, Siru Zhong, Ming Jin, Shirui Pan, Qingsong Wen, Yuxuan Liang

TL;DR

OmniAir tackles the challenge of global air quality forecasting amid regional heterogeneity by learning an inductive semantic topology that generalizes to unseen stations. It encodes invariant environmental attributes into semantic identities, constructs adaptive sparse graphs, and uses air-aware differential propagation to capture diffusion and source-generation processes, achieving linear complexity $O(NK)$. The WorldAir dataset of 7,861 stations enables rigorous global evaluation, where OmniAir outperforms 18 baselines and delivers up to an order-of-magnitude speedup. The approach demonstrates robust cross-region transfer, effective handling of data-sparse regions, and scalable forecasting suitable for real-time global monitoring. These contributions advance global environmental intelligence and offer practical tools for worldwide air quality governance and policy support.

Abstract

Global air quality forecasting grapples with extreme spatial heterogeneity and the poor generalization of existing transductive models to unseen regions. To tackle this, we propose OmniAir, a semantic topology learning framework tailored for global station-level prediction. By encoding invariant physical environmental attributes into generalizable station identities and dynamically constructing adaptive sparse topologies, our approach effectively captures long-range non-Euclidean correlations and physical diffusion patterns across unevenly distributed global networks. We further curate WorldAir, a massive dataset covering over 7,800 stations worldwide. Extensive experiments show that OmniAir achieves state-of-the-art performance against 18 baselines, maintaining high efficiency and scalability with speeds nearly 10 times faster than existing models, while effectively bridging the monitoring gap in data-sparse regions.

Breaking the Regional Barrier: Inductive Semantic Topology Learning for Worldwide Air Quality Forecasting

TL;DR

OmniAir tackles the challenge of global air quality forecasting amid regional heterogeneity by learning an inductive semantic topology that generalizes to unseen stations. It encodes invariant environmental attributes into semantic identities, constructs adaptive sparse graphs, and uses air-aware differential propagation to capture diffusion and source-generation processes, achieving linear complexity . The WorldAir dataset of 7,861 stations enables rigorous global evaluation, where OmniAir outperforms 18 baselines and delivers up to an order-of-magnitude speedup. The approach demonstrates robust cross-region transfer, effective handling of data-sparse regions, and scalable forecasting suitable for real-time global monitoring. These contributions advance global environmental intelligence and offer practical tools for worldwide air quality governance and policy support.

Abstract

Global air quality forecasting grapples with extreme spatial heterogeneity and the poor generalization of existing transductive models to unseen regions. To tackle this, we propose OmniAir, a semantic topology learning framework tailored for global station-level prediction. By encoding invariant physical environmental attributes into generalizable station identities and dynamically constructing adaptive sparse topologies, our approach effectively captures long-range non-Euclidean correlations and physical diffusion patterns across unevenly distributed global networks. We further curate WorldAir, a massive dataset covering over 7,800 stations worldwide. Extensive experiments show that OmniAir achieves state-of-the-art performance against 18 baselines, maintaining high efficiency and scalability with speeds nearly 10 times faster than existing models, while effectively bridging the monitoring gap in data-sparse regions.
Paper Structure (87 sections, 4 theorems, 44 equations, 22 figures, 13 tables, 1 algorithm)

This paper contains 87 sections, 4 theorems, 44 equations, 22 figures, 13 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem Statement
  4. Related Work
  5. Traditional Methods for Air Quality Prediction.
  6. Spatio-Temporal Deep Models.
  7. Methodology
  8. Model Overview
  9. Inductive Semantic Identity Encoder
  10. Inductive Representation Learning
  11. Hybrid Neighbor Graph Construction.
  12. Dynamic Edge Weight Learning.
  13. Adaptive Sparse Edge Learning.
  14. Air Aware Differential Propagation
  15. Multi Step Diffusion Propagation.
  16. ...and 72 more sections

Key Result

Theorem 1.2

As the embedding dimension $M \to \infty$, the kernel $K_{\gamma}(\mathbf{x}, \mathbf{y}) = \langle \gamma(\mathbf{x}), \gamma(\mathbf{y}) \rangle$ converges almost surely to a shift invariant kernel $k(\mathbf{x}-\mathbf{y})$ whose Fourier transform is the spectral density $p(\mathbf{b})$.

Figures (22)

  • Figure 1: (a) Air quality level distribution across pollutants. (b) Regional-monthly PM$_{2.5}$ concentrations revealing cross-regional and seasonal heterogeneity.
  • Figure 2: Overall framework of OmniAir, which models the spatio-temporal evolution of atmospheric pollutants via inductive station identities and adaptive sparse topologies.
  • Figure 3: Proposed encoding converts raw geographic coordinates to semantically rich, high dimensional representations for inductive generalization to unseen stations.
  • Figure 4: Visualization of learned global spatial associations across different continents.
  • Figure 5: Dataset overview. (a) Monthly PM$_{2.5}$ profiles reveal distinct seasonal patterns across five representative regions. (b) Spatial distribution and learned connectivity of global monitoring stations.
  • ...and 17 more figures

Theorems & Definitions (8)

  • Definition 1.1: Fourier Feature Mapping
  • Theorem 1.2: Convergence to Shift Invariant Kernel
  • proof
  • Corollary 1.3: Uncertainty Principle and Bandwidth Control
  • Definition 1.4: $K$-Lipschitz Continuity
  • Theorem 1.5: Stability Bound for Inductive Encoder
  • proof
  • Theorem 1.6: Necessity of Signed Weights