MovementGNN terrain mobility classification - derived spatial data (Pacaraima, RR, Brazil)

Derived spatial data accompanying the manuscript "Predictive Entropy in Graph-Based Terrain Mobility Classification for Known Off-Road Vehicle Profiles" (under peer review). Contents of Supplementary_Spatial_Files.zip: Transition-Zone masks per mobility profile (tz_masks_final_run.npz), defined as normalised predictive entropy above the 85th percentile of the final training run (2026-05-24): dismounted 0.4591, motorised 0.4472, mechanised 0.4207, armoured 0.4402; Go / Slow Go / No Go class GeoTIFFs for the four doctrinal mobility profiles; train/validation/test node indices of the spatially blocked 5 km split (split_indices.npz; 483 blocks, 100% block purity, no buffer); internal README with SHA256 checksums. Grid: 3354 x 3598 cells, 30 m resolution, EPSG:4326; study area Pacaraima (RR, Brazil); 12,067,692 terrain nodes. Produced by the MovementGNN software archived at DOI 10.5281/zenodo.21328555 (source repository: https://github.com/D0mP3dr0/movementgnn-terrain-mobility, frozen version MovementGNN_JT_Submission_v1). Input data: Copernicus DEM GLO-30 and Sentinel-2 L2A (open data).