"ARES: Benchmark Dataset for Adaptive Region Extraction in Heterogeneous Spatial Domains"

"This dataset supports the experiments in the paper \"ARES: Structurally Constrained Adaptive Region Extraction in Heterogeneous Spatial Domains.\" It includes three components: (1) US48 \u2014 a national-scale raster dataset derived from the global PPP product covering the 48 contiguous U.S. states, with a road-network structural prior from BTS; (2) Colorado \u2014 a state-level spatial subset with pronounced local density heterogeneity; (3) Synthetic-MS \u2014 a multi-source synthetic benchmark covering six provinces in central China, using VIIRS nighttime light radiance as the attribute intensity field and OpenStreetMap-derived road skeletons and built-up land-cover masks as the structural prior and ground truth, respectively. These datasets are designed to evaluate unsupervised cohesive pattern extraction under spatial non-stationarity."