DGRR-Net: Supporting Data and Reproducibility Artifacts for Terrain-Aware Residual Learning for Digital Elevation Model Refinement

This record provides the supporting data and frozen reproducibility artifacts for the manuscript Terrain-Aware Residual Learning for Digital Elevation Model Refinement. DGRR-Net is a dual-scale terrain-aware residual refinement framework for improving coarse-resolution SRTM elevation information on a 1 m target grid using elevation conditioning, explicit terrain derivatives, and high-resolution NAIP optical guidance. The supporting-data archive contains the frozen E2c model checkpoint, normalization statistics, the Protocol-A spatial split, ROI metadata, per-patch evaluation metrics, paired statistical tests, and manuscript benchmark tables. The companion GitHub repository provides the preprocessing, model, training, inference, evaluation, hydrology, and statistical-reproduction implementation. Raw SRTM, USGS 3DEP, and NAIP source products are not redistributed. Their provenance and the information required to reconstruct the study inputs are provided in the accompanying release documentation.