# Elevation product selection evidence — method

This edition compares 20 named elevation products using 20 product attributes. Primary product pages, technical manuals and use conditions were inspected on 24 October 2026 in America/Denver. Exact UTC source observations are retained. These are documented specifications and conditions; no raster was downloaded, transformed, aligned, sampled or independently checked for accuracy.

## Product units

Products are separated when their geometry, coverage, access conditions or processing differ: USGS resolutions and Alaska DTM/DSM, Copernicus GLO-30/GLO-90/EEA-10, and ArcticDEM/REMA strips versus mosaics. They are related products, not independent measurements. The source version identifies a documented release or edition, which is not the acquisition date of every source pixel.

Grid spacing is reported in the unit used by the product. Arc-seconds are angular spacing, so a nominal metre label is approximate and longitude ground distance varies with latitude. DSM, DTM and mixed radar/optical-derived surfaces remain distinct. Horizontal and vertical references, vertical units and pixel registration are separate fields; a missing value is not silently assigned from a related product.

## Accuracy and datum limits

Accuracy entries retain the provider’s metric and its scope. LE90, registered-mosaic uncertainty, source-lidar quality level and a local validation result are not interchangeable. The <4 m Copernicus specification is not a promise for a selected cell. REMA’s mosaic uncertainty statement is not applied to unregistered strips. GEBCO spacing is not seafloor survey density. The ASTER catalogue’s north-bound inconsistency is disclosed rather than resolved by assumption.

USGS vertical references vary by region and product; an actual tile’s metadata controls. WGS84 ellipsoid heights, EGM96, EGM2008 and NAVD88 are not treated as identical. A horizontal CRS conversion alone is not evidence of a successful vertical transformation. No conversion accuracy, seam quality, nodata treatment or resampling result is claimed.

## Rights and access

Licence labels summarize the inspected product terms with links and dates, and short excerpts identify some conditions. They do not constitute legal clearance. JAXA research-data notification, CGIAR redistribution restrictions, MERIT’s dual-licence/full-copy conditions, FABDEM’s non-commercial share-alike terms and Copernicus instance-specific conditions are retained. A free download is not assumed to permit every commercial use or redistribution. Source imagery rights are not transferred to the reader by a public DEM release.

## Not measured

GDAL and its Python bindings were unavailable in the checked environment. No GDAL command or raster benchmark ran. Two candidate USGS 3DEP 1-metre tiles were identified through official National Map product metadata only; their bytes, hashes, CRS, extent, alignment and vertical metadata remain unverified. They are modern GeoTIFF candidates, not evidence of a successful legacy SDTS import.

`Not established in inspected sources` is an explicit unknown. `index.json` is canonical; `index.csv` is a flattened view, `source-ledger.json` records source observations, and `hashes.txt` contains SHA-256 checksums. The original compiled annotations are **CC BY 4.0**; source data and documentation retain their own terms. No elevation files are redistributed and no ranking or row is sponsored.

## E1 terms and typography correction

The `licence` field names the licence or documented status. Detailed scope remains separate, and shared USGS, Copernicus, PGC, GEBCO and NASA policy wording is consistent across the two compilations where the same terms apply. Unknown blanket rights stay unknown. Dated publisher evidence supports the labels; this edit does not establish legal clearance or execute data acquisition. Existing source observations remain dated as originally opened.

Letter–digit errors came from authored Python string literals in the P3 builder, then propagated through JSON into tables. The E1 renderer reads the corrected canonical records. Its validation checks every prose field and rendered table against a documented exact-token allowlist; it does not insert spaces into URLs, identifiers, versions or quoted source text. The earlier generator and runs remain unchanged.