The RegionGCN paper is now published in Annals of the American Association of Geographers

A recent research article titled "RegionGCN: Spatial-Heterogeneity-Aware Graph Convolutional Networks" has just been published in the Annals of the American Association of Geographers, the flagship journal of Geography. This work is a key extension of GCN-based spatial regression to spatial regionalization, marking a new milestone for our direction of spatially-explicit spatial modeling in graph neural networks.

Neural network models usually assume spatial stationarity, which could limit their performance in the presence of spatial process heterogeneity. By allowing model parameters to vary over space, the risk of overfitting is increased by a large number of local parameters. In this work, we propose a graph neural network, called RegionGCN, to model spatial process heterogeneity at the regional level rather than at the individual level, which largely reduces the number of spatially varying parameters.

The model is evaluated in the modeling of county-level vote shares in the 2016 U.S. presidential election data. Results show that RegionGCN achieves significant improvement over the basic and geographically weighted GCNs


Guo, H., Wang, H., Zhu, D., Wu, L., Fotheringham, A. S., & Liu, Y. (2025). RegionGCN: Spatial-Heterogeneity-Aware Graph Convolutional Networks. Annals of the American Association of Geographers. https://doi.org/10.1080/24694452.2025.2558661