RegionGCN: Spatial-Heterogeneity-Aware Graph Convolutional Networks

Modeling spatial heterogeneity in the data generation process is essential for understanding and predicting geographical phenomena. Despite their prevalence in geospatial tasks, 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, several approaches have been proposed to incorporate spatial heterogeneity into neural networks. Current geographical weighting approaches, however, are ineffective on graph neural networks, yielding no significant improvement in prediction accuracy. We assume the crux lies in the overfitting risk brought by a large number of local parameters. Accordingly, we propose to model spatial process heterogeneity at the regional level rather than at the individual level, which largely reduces the number of spatially varying parameters. We further develop a heuristic optimization procedure to learn the region partition adaptively in the process of model training. Our proposed spatial-heterogeneity-aware graph convolutional network, named RegionGCN, is applied to the modeling of county-level vote share in the 2016 U.S. presidential election based on socioeconomic attributes. Results show that RegionGCN achieves significant improvement over the basic and geographically weighted GCNs. We also offer an exploratory analysis tool for the spatial variation of nonlinear relationships through ensemble learning of regional partitions from RegionGCN. Our work contributes to the practice of geospatial artificial intelligence in tackling spatial heterogeneity.