<p>Air pollution in densely populated, industrialized cities makes accurate air quality prediction essential for mitigation efforts. Air quality data, often structured as geo-tagged time series, form a large-scale spatiotemporal dataset. A growing body of research has focused on developing air quality prediction models using deep learning techniques to improve air pollution control. However, conventional deep learning models often struggle to capture the complex spatiotemporal dependencies inherent in air quality data. In response, hybrid deep learning models incorporating Graph Convolutional Networks (GCNs) have emerged as a promising solution. These models represent Air Quality Monitoring Stations (AQMSs), which are distributed in non-Euclidean space, as graph structures to more effectively capture both spatial and temporal relationships. This study presents a systematic review of hybrid deep learning models that leverage GCNs for spatiotemporal air quality prediction. Using the PRISMA methodology, relevant studies were identified from the Web of Science and Scopus databases. The review examines approaches for spatiotemporal feature extraction and graph construction within AQMS networks. It also compiles a comprehensive repository of air quality datasets and auxiliary data sources, offering a valuable reference for future research. Lastly, the study discusses current challenges and provides recommendations for advancing GCN-based air quality prediction models.</p>