 The paper proposes a deep learning framework called GTLSTM for predicting various air pollutants in urban areas. The model combines graph convolutional networks and long short-term memory networks based on a strategy with temporal sliding to capture spatial and temporal dependencies of different types of air pollution. Experiments show that the proposed GTLSTM model can achieve higher accuracy and stability than state-of-the-art baselines, providing decision support capabilities for mitigating air quality issues. This article was authored by Wenjing Mao, Lamine Zhao, Wei Lin Wang, and others.