DIVERGE: 39,731-residual exhaustive parameter sweep of decomposition × upsampling methods on a Fairfield County, Ohio LiDAR DEM

This deposit accompanies the RESIDUALS prior-art exhaustive run. It contains a structured catalog of 39,731 residual outputs produced by applying 593 distinct decomposition configurations (gaussian, bilateral, wavelet, morphological, top-hat, polynomial, DoG, LoG, anisotropic diffusion, rolling-ball, guided, and others) crossed with 25 upsampling methods (bicubic, lanczos, B-spline, FFT zero-pad, sinc-windowed, edge-directed, etc.) applied to a 1500×375 Digital Elevation Model derived from LiDAR over Fairfield County, Ohio. Contents: data/catalog.parquet — 39,731 entries (decomposition family, parameters, upsampling method, scale, file hash). data/signatures.parquet — 40-dim signature vector per residual (32-bin radial-FFT log-power spectrum + 8 statistical moments). data/umap_coords_2d.parquet, umap_coords_3d.parquet — UMAP embeddings of the signatures (n=39,716; 15 div-by-zero edge cases dropped). data/thumbnails.zarr.zip — 39,716 cropped 256×256 float32 thumbnails centered on a feature-rich region of the DEM. documentation/prior_art.md, method_parameters.json, hashes.csv — full method parameter spaces and SHA-256 hashes of all 39,731 outputs (proof-of-existence for the full 4.28 TB dataset). visualizations/ — algorithm zoo grid, UMAP scatter, parameter sweep videos, atlas mosaic preview. The full 4.28 TB raw output set is not deposited here (Zenodo size limits) but is fully reproducible from the source DEM + the RESIDUALS code (github.com/bshepp/RESIDUALS). All file hashes are included for verification. This deposit serves a dual purpose: (1) prior-art documentation to prevent exclusive claims on these specific signal-processing combinations applied to DEMs, and (2) a reusable benchmark for algorithm-fingerprinting and feature-extraction research.