Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spatio-Temporal Forecasting of PM2.5 via Spatial-Diffusion guided Encoder-Decoder Architecture

Malay Pandey, Vaishali Jain, Nimit Godhani, Sachchida Nand Tripathi, Piyush Rai

TL;DR

This work tackles spatio-temporal forecasting of PM2.5 by introducing AGNN_GRU, a lightweight encoder-decoder that fuses a TransformerConv-based graph neural network with GRU temporal modeling and Luong attention. It formalizes a directed-spatial graph framework, constructs wind-informed edges, and forecasts across multiple locations without requiring future meteorological inputs. Evaluations on Bihar and China datasets show the model achieving superior spatio-temporal predictive performance, particularly in capturing seasonal patterns and abrupt haze events, while highlighting considerations around model complexity and data length. The approach offers a practical, diffusion-aware solution for air-quality forecasting with potential for real-time deployment on IoT devices and deployment to diverse geographies.

Abstract

In many problem settings that require spatio-temporal forecasting, the values in the time-series not only exhibit spatio-temporal correlations but are also influenced by spatial diffusion across locations. One such example is forecasting the concentration of fine particulate matter (PM2.5) in the atmosphere which is influenced by many complex factors, the most important ones being diffusion due to meteorological factors as well as transport across vast distances over a period of time. We present a novel Spatio-Temporal Graph Neural Network architecture, that specifically captures these dependencies to forecast the PM2.5 concentration. Our model is based on an encoder-decoder architecture where the encoder and decoder parts leverage gated recurrent units (GRU) augmented with a graph neural network (TransformerConv) to account for spatial diffusion. Our model can also be seen as a generalization of various existing models for time-series or spatio-temporal forecasting. We demonstrate the model's effectiveness on two real-world PM2.5 datasets: (1) data collected by us using a recently deployed network of low-cost PM$_{2.5}$ sensors from 511 locations spanning the entirety of the Indian state of Bihar over a period of one year, and (2) another publicly available dataset that covers severely polluted regions from China for a period of 4 years. Our experimental results show our model's impressive ability to account for both spatial as well as temporal dependencies precisely.

Spatio-Temporal Forecasting of PM2.5 via Spatial-Diffusion guided Encoder-Decoder Architecture

TL;DR

This work tackles spatio-temporal forecasting of PM2.5 by introducing AGNN_GRU, a lightweight encoder-decoder that fuses a TransformerConv-based graph neural network with GRU temporal modeling and Luong attention. It formalizes a directed-spatial graph framework, constructs wind-informed edges, and forecasts across multiple locations without requiring future meteorological inputs. Evaluations on Bihar and China datasets show the model achieving superior spatio-temporal predictive performance, particularly in capturing seasonal patterns and abrupt haze events, while highlighting considerations around model complexity and data length. The approach offers a practical, diffusion-aware solution for air-quality forecasting with potential for real-time deployment on IoT devices and deployment to diverse geographies.

Abstract

In many problem settings that require spatio-temporal forecasting, the values in the time-series not only exhibit spatio-temporal correlations but are also influenced by spatial diffusion across locations. One such example is forecasting the concentration of fine particulate matter (PM2.5) in the atmosphere which is influenced by many complex factors, the most important ones being diffusion due to meteorological factors as well as transport across vast distances over a period of time. We present a novel Spatio-Temporal Graph Neural Network architecture, that specifically captures these dependencies to forecast the PM2.5 concentration. Our model is based on an encoder-decoder architecture where the encoder and decoder parts leverage gated recurrent units (GRU) augmented with a graph neural network (TransformerConv) to account for spatial diffusion. Our model can also be seen as a generalization of various existing models for time-series or spatio-temporal forecasting. We demonstrate the model's effectiveness on two real-world PM2.5 datasets: (1) data collected by us using a recently deployed network of low-cost PM sensors from 511 locations spanning the entirety of the Indian state of Bihar over a period of one year, and (2) another publicly available dataset that covers severely polluted regions from China for a period of 4 years. Our experimental results show our model's impressive ability to account for both spatial as well as temporal dependencies precisely.

Paper Structure

This paper contains 20 sections, 8 equations, 4 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dataset
  4. Problem Definition
  5. Our Model: AGNN_GRU
  6. Graph Construction
  7. Related Work
  8. Classical Time-Series Methods
  9. Spatio-Temporal Approaches
  10. Other Related Works
  11. Experiments
  12. Dataset
  13. Baselines
  14. Experimental Settings
  15. Evaluation Metrics
  16. ...and 5 more sections

Figures (4)

  • Figure 1: Architecture of the proposed AGNN_GRU model where the encoder (left block) consists of a GNN (TransformerConv) and decoder (right block) consists of Luong Attention
  • Figure 2: Distribution of AQM stations across the entire state of Bihar. The stations have been clustered region-wise.
  • Figure 3: Spatial Distribution of mean PM$_{2.5}$ concentration for Bihar across seasons. From Left to Right: JJAS, ON, DJF, MA
  • Figure 4: Comparison of true (blue color) and predicted values (orange color) across different locations for Bihar dataset for 12-hour (top row) and 24-hour (bottom row) forecasting settings. Vertical dotted lines define the dataset partitions: training (left), validation (center), and test (right) sets.