Research on Quality Control Method of Surface Temperature Observations Based on Spatiotemporal Graph Neural Network

This article proposes an enhanced quality control (QC) method based on Spatiotemporal Graph Convolutional Networks (STGCN) to identify potential outliers in surface temperature observations. The STGCN model employs a graph structure to simultaneously capture temporal and spatial dependencies, with the adjacency matrix constructed using spatial distances and topography-assisted elevation priors. Compared to baseline methods, experimental results indicate that STGCN achieves superior overall performance across evaluation metrics, effectively balancing Type I and Type II errors. The findings demonstrate that the proposed framework is an effective QC method for detecting observational anomalies in surface temperature datasets.