GDAL, DuckDB, QGIS and the browser

GDAL command line
DuckDB
QGIS
In the browser

GDAL command line

The gdal command (GDAL 3.11 and later) reads the COGs from a /vsicurl/ URL. Set GDAL_DISABLE_READDIR_ON_OPEN=EMPTY_DIR so GDAL doesn’t list the remote folder:

export GDAL_DISABLE_READDIR_ON_OPEN=EMPTY_DIR
COG=/vsicurl/https://data.source.coop/ausantarctic/ghrsst-mur-v2/2025/06/01/20250601090000-JPL-L4_GHRSST-SSTfnd-MUR-GLOB-v02.0-fv04.1_analysed_sst.tif

# Size, grid, tiling, overviews, scale and offset
gdal raster info $COG

# The value at a point: both the stored integer and the scaled value (deg C)
gdal raster pixel-info --position-crs WGS84 -f CSV $COG 147,-45

# A region, as it is (xmin,ymin,xmax,ymax)
gdal raster clip --bbox 130,-55,150,-42 $COG tasman.tif

# A region in a local projection, at 1 km
gdal raster reproject -d "+proj=laea +lon_0=147 +lat_0=-45" \
  --bbox -600000,-600000,600000,600000 --resolution 1000,1000 -r bilinear \
  $COG tasman_laea.tif

# The whole world at about 0.25 degree, read from an overview
gdal raster resize --size 1440,0 $COG world_quarter_degree.tif

These need GDAL 3.11 or later; the classic utilities (gdalinfo, gdal_translate, gdalwarp, …) work with the same inputs on older versions.

A stack of days as the bands of one virtual raster, read lazily:

for d in 2024-12-29 2024-12-30 2024-12-31; do
  ymd=${d//-/}
  echo "/vsicurl/https://data.source.coop/ausantarctic/ghrsst-mur-v2/${d:0:4}/${d:5:2}/${d:8:2}/${ymd}090000-JPL-L4_GHRSST-SSTfnd-MUR-GLOB-v02.0-fv04.1_analysed_sst.tif"
done > days.txt
gdal raster stack @days.txt days.vrt
gdal raster pixel-info --position-crs WGS84 -f CSV days.vrt 147,-45

Inside AWS us-west-2, read through S3 instead:

export AWS_NO_SIGN_REQUEST=YES AWS_REGION=us-west-2
gdal raster info /vsis3/us-west-2.opendata.source.coop/ausantarctic/ghrsst-mur-v2/2025/06/01/20250601090000-JPL-L4_GHRSST-SSTfnd-MUR-GLOB-v02.0-fv04.1_analysed_sst.tif

DuckDB

The STAC GeoParquet index is an ordinary Parquet file with one row per day. In the DuckDB shell:

-- How many days of each status, and their date ranges
SELECT "ghrsst:status" AS status, count(*) AS days,
       min(datetime) AS first, max(datetime) AS last
FROM 'https://data.source.coop/ausantarctic/ghrsst-mur-v2/ghrsst-mur-v2.parquet'
GROUP BY ALL ORDER BY first;

-- The analysed_sst URLs of every final day in 2024
SELECT datetime, assets.analysed_sst.href AS href
FROM 'https://data.source.coop/ausantarctic/ghrsst-mur-v2/ghrsst-mur-v2.parquet'
WHERE "ghrsst:status" = 'final'
  AND datetime >= '2024-01-01' AND datetime < '2025-01-01'
ORDER BY datetime;

-- Days made or re-made since the status fields were added, and when
SELECT datetime, updated, "processing:lineage"
FROM 'https://data.source.coop/ausantarctic/ghrsst-mur-v2/ghrsst-mur-v2.parquet'
WHERE updated IS NOT NULL
ORDER BY datetime DESC LIMIT 20;

The same queries run from Python, R, or anything else with a DuckDB binding, and any Parquet reader (Polars, pandas, arrow, Spark) reads the file.

QGIS

Add a COG as a raster layer from its URL: Layer > Add Layer > Add Raster Layer, source type Protocol: HTTP(S), cloud, etc., and paste the https://data.source.coop/...tif URL. QGIS reads the overviews as you zoom out, so a whole-world view is quick. The scale and offset are in the layer’s Information panel (degrees Celsius for analysed_sst).

In the browser

COGs can be read in a web page with HTTP range requests, no server needed.

ortho-cog-viewer

ortho-cog-viewer draws COGs in any projection, reprojecting on the GPU as you pan. Everything is in the link:

For another date, change the date in the url= parameter. The layer URL is percent-encoded because it carries its own styling after a # (min, max and cmap set the colour scale, in degrees Celsius).

playproj

playproj shows how rasters are stored and read. Its Tiles and overviews page reads a COG’s header (no pixels) and draws its pixel grid, its tiles and its overview levels on the globe; click a tile for its byte range in the file. That’s the structure every tool on these pages relies on, and that the virtual stores index:

A MUR analysed_sst COG’s tiles and overviews

Its Rasterize page can use a COG as the grid too, and sums the bytes of the tiles a polygon touches: how much of a file a region query actually reads.

Your own page

geotiff.js reads a window or an overview of a COG in JavaScript:

import { fromUrl } from "geotiff";

const url = "https://data.source.coop/ausantarctic/ghrsst-mur-v2/2025/06/01/" +
  "20250601090000-JPL-L4_GHRSST-SSTfnd-MUR-GLOB-v02.0-fv04.1_analysed_sst.tif";
const tiff = await fromUrl(url);
// The whole world at 720 by 360, from an overview; values are the stored integers
const [raw] = await tiff.readRasters({ bbox: [-180, -90, 180, 90], width: 720, height: 360 });
const degC = Array.from(raw, (v) => (v === -32768 ? NaN : v * 0.001 + 25));