Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spatio-Temporal Field Neural Networks for Air Quality Inference

Yutong Feng, Qiongyan Wang, Yutong Xia, Junlin Huang, Siru Zhong, Yuxuan Liang

TL;DR

The paper reframes air quality inference as reconstructing a Spatio-Temporal Field G with gradient F=∇G over continuous space-time, rather than relying solely on discrete graphs. It introduces Spatio-Temporal Field Neural Networks (STFNN) that learn the gradient field via Implicit Neural Representations and couples it with a local Spatio-Temporal Graph through a Pyramidal Inference framework to fuse global and local information. Empirical results on a nationwide 2018 dataset for PM2.5 in China show STFNN achieving state-of-the-art performance, with strong ablations validating the gradient-focused design and the effectiveness of the ring-based gradient estimation. The work further demonstrates interpretability via curl analysis and extends its generalizability to NO2 inference, indicating broad applicability to multiple pollutants and real-world deployment potential.

Abstract

The air quality inference problem aims to utilize historical data from a limited number of observation sites to infer the air quality index at an unknown location. Considering the sparsity of data due to the high maintenance cost of the stations, good inference algorithms can effectively save the cost and refine the data granularity. While spatio-temporal graph neural networks have made excellent progress on this problem, their non-Euclidean and discrete data structure modeling of reality limits its potential. In this work, we make the first attempt to combine two different spatio-temporal perspectives, fields and graphs, by proposing a new model, Spatio-Temporal Field Neural Network, and its corresponding new framework, Pyramidal Inference. Extensive experiments validate that our model achieves state-of-the-art performance in nationwide air quality inference in the Chinese Mainland, demonstrating the superiority of our proposed model and framework.

Spatio-Temporal Field Neural Networks for Air Quality Inference

TL;DR

The paper reframes air quality inference as reconstructing a Spatio-Temporal Field G with gradient F=∇G over continuous space-time, rather than relying solely on discrete graphs. It introduces Spatio-Temporal Field Neural Networks (STFNN) that learn the gradient field via Implicit Neural Representations and couples it with a local Spatio-Temporal Graph through a Pyramidal Inference framework to fuse global and local information. Empirical results on a nationwide 2018 dataset for PM2.5 in China show STFNN achieving state-of-the-art performance, with strong ablations validating the gradient-focused design and the effectiveness of the ring-based gradient estimation. The work further demonstrates interpretability via curl analysis and extends its generalizability to NO2 inference, indicating broad applicability to multiple pollutants and real-world deployment potential.

Abstract

The air quality inference problem aims to utilize historical data from a limited number of observation sites to infer the air quality index at an unknown location. Considering the sparsity of data due to the high maintenance cost of the stations, good inference algorithms can effectively save the cost and refine the data granularity. While spatio-temporal graph neural networks have made excellent progress on this problem, their non-Euclidean and discrete data structure modeling of reality limits its potential. In this work, we make the first attempt to combine two different spatio-temporal perspectives, fields and graphs, by proposing a new model, Spatio-Temporal Field Neural Network, and its corresponding new framework, Pyramidal Inference. Extensive experiments validate that our model achieves state-of-the-art performance in nationwide air quality inference in the Chinese Mainland, demonstrating the superiority of our proposed model and framework.
Paper Structure (22 sections, 7 equations, 6 figures, 2 tables)

This paper contains 22 sections, 7 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Methodology
  4. Global View: Spatio-Temporal Field
  5. Local View: Spatio-Temporal Graph
  6. Hybrid Framework: Pyramidal Inference
  7. Implementation
  8. Spatio-Temporal Encoding
  9. Ring Estimation
  10. Neighbor Aggregation
  11. Experiments
  12. Experimental Settings
  13. Datasets
  14. Baselines for Comparison
  15. Hyperparameters & Setting
  16. ...and 7 more sections

Figures (6)

  • Figure 1: Spatio-Temporal Graph vs. Spatio-Temporal Field.
  • Figure 2: (a)-(d): User interface of our STFNN system for air quality inference. (e): An illustration of Ring Estimation.
  • Figure 3: Paradigms for air quality inference. (a) A spatio-temporal graph perspective. (b) A spatio-temporal field perspective.
  • Figure 4: Implementation of STFNN
  • Figure 5: (a) ablation study (b) the variation of curl
  • ...and 1 more figures