DuneCOAST Paper 1 reproducibility package: UAS imagery, LiDAR-derived canopy height, training data, reference maps, and Colab workflow for coastal dune vegetation classification

This record contains the reproducibility package associated with the manuscript Beyond Spectral Indices: Z-Aware Post-Classification Refinement for Heterogeneous Coastal Environments. The package contains the clean canonical DuneCOAST Jupyter/Google Colab notebook, processed co-registered RGB and multispectral imagery, LiDAR-derived canopy-height data, expert-curated training samples, expert-digitized reference maps, feature-audit materials, trained-model and classification outputs, manuscript figures and source tables, software-environment information, and documentation for three test areas within the Long Branch, New Jersey foredune system. DuneCOAST maps four surface states—exposed sand, shadow, active vegetation (AV), and dormant vegetation (DV)—using a Random Forest classifier applied to optical predictors followed by targeted post-classification refinement using LiDAR-derived canopy height. LiDAR-derived canopy height (Z) is intentionally excluded from the first-order Random Forest feature profiles and introduced during the Pair-Z structural-refinement stage. The Test Area A, B, and C reproduce-output archives contain the fixed profile-specific products and source tables used for the manuscript figures and numerical results. The direct Pair-Z classification is the POST product used in the reported PRE–POST analyses; EdgeBound is retained as an optional downstream boundary diagnostic and is not included in the primary quantitative reclassification audit. See README_RUN_ME_FIRST.md and DuneCOAST_User_Manual.pdf for repository organization, execution instructions, validation conventions, and manuscript-figure reproduction guidance.