.png){kind=icon}
.png){kind=icon}
Return
A Spatiotemporal graph attention network for PM2.5 forecasting using multi-source data
.png){kind=icon}
PRE
AI
DOI：10.1007/s11869-025-01818-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
PM2.5 is a highly hazardous air pollutant that threatens public health, environmental sustainability, and SDGs progress. Accurate spatiotemporal prediction of its concentrations enables effective air quality management and early warnings. This study presents a novel deep learning framework for PM2.5 prediction, which integrates a spatiotemporal graph convolutional network with multi-level attention mechanisms. The proposed model incorporates both satellite remote sensing data and meteorological variables to comprehensively represent the factors affecting PM2.5 dynamics. A weighted spatial graph is constructed to model inter-station dependencies, and a hybrid architecture combining Graph Attention Networks (GAT) and Gated Recurrent Units (GRU) is employed to jointly capture spatial and temporal patterns. Furthermore, a global attention module is introduced to enhance the learning of long-range spatiotemporal dependencies. The model is evaluated using PM2.5 observations from monitoring stations in the Beijing–Tianjin–Hebei region. Experimental results demonstrate that the proposed approach significantly outperforms conventional baseline models, highlighting the advantages of incorporating attention-based spatiotemporal learning and multi-source data fusion for fine-grained air quality prediction.
Keywords:
PM 2.5 forecasting
Spatiotemporal modeling
Multi-source data
Multi-head self-attention mechanism
Journal
A
IF:
2.9
Papers:
74
Citations:
4.1K