Non-Geo Relief Visualization

Preparing non-georeferenced DSM/RGB data (frescoes, artifact scans) as web-ready COGs

For non-georeferenced data — frescoes, artifact scans, excavation surfaces — there's no real lat/lng grid to put the data on, but the app's terrain pipeline (and every visualization mode built on it) still expects a georeferenced, tiled raster-dem/COG source. The trick is a "fake-geo" grid: pick an arbitrary origin and a small pixel size in a real CRS (EPSG:3857 here), assign it to the data with gdal_translate -a_gt/-a_srs, and scale elevation (-a_scale) so the z-range stays usable once the tiny real-world pixel size is reinterpreted as a much larger one. This page is the complete DSM/RGB COG-processing guide for that workflow — see Bring Your Own Data for how to load the resulting COG into the app.

0. Where to get the tools

These commands need gdal_translate/gdal_calc.py on your PATH:

Windows: install OSGeo4W (pick the "Express Install" and select the gdal package), which puts gdal_translate.exe in C:\OSGeo4W\bin. Run commands from the "OSGeo4W Shell" it installs, or add that bin folder to your PATH.
Any OS via conda: conda install -c conda-forge gdal (or mamba install -c conda-forge gdal) inside a fresh environment.
QGIS: the QGIS desktop installer bundles its own GDAL — on Windows its gdal_translate.exe ships under <QGIS install dir>\bin, usable even if you don't open QGIS itself.

1. Check current resolution

gdalinfo input.tiff | findstr "Pixel Size"

Shows your actual pixel size (e.g., 0.0005m = 0.5mm for photogrammetry data).

2. Calculate scale factor

Real GSD: 0.0005m (0.5mm/pixel from photogrammetry)
Target GSD: 2m/pixel (chosen for web maps)
Why 2m? Web Mercator at zoom 20 has ~0.15m/pixel resolution at the equator. Using 2m/pixel is conservative and works well across zoom levels 15-20. Could as well be e.g. 20cm or 0.2m.
Z-scale factor: 2m ÷ 0.0005m = 4000 (multiply z-elevations by 4000 to compensate the xy rescaling)

DSM command (elevation data)

gdal_translate -of COG -a_gt 0.01 2.0 0 -0.01 0 -2.0 -a_scale -4000 -a_nodata nan -co BIGTIFF=YES -a_srs EPSG:3857 -co BLOCKSIZE=256 -co TILING_SCHEME=GoogleMapsCompatible -co RESAMPLING=BILINEAR -co OVERVIEW_RESAMPLING=NEAREST -co COMPRESS=DEFLATE -co PREDICTOR=3 -co OVERVIEWS=IGNORE_EXISTING -co ADD_ALPHA=NO -b 1 -colorinterp_1 gray DuraEuropos_Synagogue_W_0.5mm_DSM.tiff DuraEuropos_Synagogue_W_0.5mm_DSM_fakegeo-3.cog.tiff

DSM parameters:

-of COG — Cloud Optimized GeoTIFF (web-optimized with internal tiling/overviews)
-a_gt 0 2 0 0 0 -2 — Geotransform: origin (0,0), 2m pixels, no rotation
-a_scale 4000 — Multiply z-values by 4000 (compensates 0.5mm→2m pixel scaling)
-a_nodata nan — Numeric nodata value (RC uses NotANumber as NoData)
-co BIGTIFF=YES — Handle files >4GB
-a_srs EPSG:3857 — Web Mercator projection
-co BLOCKSIZE=256 -co TILING_SCHEME=GoogleMapsCompatible — Recommended by geomatico/maplibre-cog-protocol
-co RESAMPLING=BILINEAR -co OVERVIEW_RESAMPLING=NEAREST — Bilinear for base tiles, nearest for pyramids (avoids nodata bleed into overviews while keeping a reasonable middle ground — see the Average-vs-Bilinear note below)
-co COMPRESS=DEFLATE -co PREDICTOR=3 — Lossless compression (preserves elevation precision); float elevation requires predictor 3 for continuous elevation data compression
-co OVERVIEWS=IGNORE_EXISTING -co ADD_ALPHA=NO — Overwrite overviews, don't add alpha
-b 1 — Extract only the elevation band (drop RGB+Alpha if present)
-colorinterp_1 gray — Mark as grayscale elevation (required for hillshade to work)

Choice of AVERAGE vs BILINEAR (or other) for DSM overviews? AVERAGE resampling properly excludes nodata pixels from calculations, preventing nodata from bleeding into valid elevation data in overviews — but it produces a poorer, grid-like structure in the data. Good middle ground: bilinear for base-tile resampling, nearest for overview resampling.

DSM command — raw altitude (no fake-geo z-rescale)

The DSM command above intentionally distorts elevation by -a_scale 4000 so that the fake-geo xy rescale (0.0005m pixels reprojected as if they were 2m pixels) doesn't also compress the z-range into an unusable range for the web viewer's hillshade/color-relief. That's the right choice for viewing, but it means the resulting COG's pixel values are not real-world altitude — don't measure elevation directly off it in QGIS or elsewhere.

To get a COG whose pixel values are true altitude (real elevation units, still georeferenced with the same fake-geo xy grid so it aligns with the fake-geo RGB/DSM above), drop -a_scale entirely and keep the source's native z-units as-is:

gdal_translate -of COG -a_gt 0.01 2.0 0 -0.01 0 -2.0 -a_nodata nan -co BIGTIFF=YES -a_srs EPSG:3857 -co BLOCKSIZE=256 -co TILING_SCHEME=GoogleMapsCompatible -co RESAMPLING=BILINEAR -co OVERVIEW_RESAMPLING=NEAREST -co COMPRESS=DEFLATE -co PREDICTOR=3 -co OVERVIEWS=IGNORE_EXISTING -co ADD_ALPHA=NO -b 1 -colorinterp_1 gray -ot Float32 input_DSM.tif output_DSM_raw-altitude.cog.tiff

If you already produced an -a_scale 4000-encoded COG and need to recover real altitude from it later (e.g. to measure a wall height in QGIS), either re-run gdal_translate from the original source without -a_scale, or use gdal_calc.py -A encoded.cog.tiff --outfile=real_altitude.tif --calc="A/4000" to invert the scale factor on the encoded output directly.

RGB orthophoto command

gdal_translate -of COG -a_gt 0.01 2.0 0 -0.01 0 -2.0 -co BIGTIFF=YES -a_srs EPSG:3857 -co BLOCKSIZE=256 -co TILING_SCHEME=GoogleMapsCompatible -co COMPRESS=JPEG -co QUALITY=95 -co RESAMPLING=BILINEAR -co OVERVIEW_RESAMPLING=BILINEAR -co PREDICTOR=2 -co OVERVIEWS=IGNORE_EXISTING -co ADD_ALPHA=NO DuraEuropos_Synagogue_W_0.5mm_RGB.tiff DuraEuropos_Synagogue_W_0.5mm_RGB_fakegeo.cog.tiff

RGB parameters:

-of COG — Cloud Optimized GeoTIFF
-a_gt 0 2 0 0 0 -2 — Geotransform: origin (0,0), 2m pixels, no rotation
-co BIGTIFF=YES — Handle files >4GB
-a_srs EPSG:3857 — Web Mercator projection
-co BLOCKSIZE=256 -co TILING_SCHEME=GoogleMapsCompatible — Recommended by geomatico/maplibre-cog-protocol
-co COMPRESS=JPEG -co QUALITY=95 — Lossy compression (better file size for imagery) + JPEG quality
-co RESAMPLING=BILINEAR — Smooth interpolation for base tiles
-co OVERVIEW_RESAMPLING=BILINEAR — Smooth interpolation for pyramids (prevents artifacts)
-co PREDICTOR=2 -co OVERVIEWS=IGNORE_EXISTING -co ADD_ALPHA=NO

No -a_nodata, -colorinterp, or -a_scale — not needed for RGB imagery.

RGB with alpha channel (RGBA → RGB with mask)

If your orthophoto has transparency (4 bands):

# Be careful, command not tested
# gdal_translate -of COG -co BIGTIFF=YES -co COMPRESS=JPEG -co QUALITY=85 -co RESAMPLING=BILINEAR -co OVERVIEW_RESAMPLING=BILINEAR -b 1 -b 2 -b 3 -mask 4 -a_srs EPSG:3857 -a_gt 0 2 0 0 0 -2 input_orthophoto.tiff output_orthophoto.cog.tiff

Additional RGBA parameters:

-b 1 -b 2 -b 3 — Select only RGB bands (drop alpha)
-mask 4 — Convert band 4 (alpha) to internal mask (prevents transparency issues in overviews and allows JPEG compression)

Why drop alpha for JPEG? JPEG doesn't support transparency. Converting alpha to an internal mask preserves transparency information while allowing JPEG compression of the RGB data.