Cloud-Optimized GeoTIFF (COG) is a regular GeoTIFF with its internal structure arranged so that clients can request just the bytes they need over HTTP. That small change has big consequences.
The old way
Traditionally, reading a single tile from a large raster meant downloading the whole file. For a multi-gigabyte scene, that is wasteful — especially when you only want an overview or a small window.
The COG way
With COG, internal tiling and overviews are laid out predictably, and HTTP range requests let a client fetch only the relevant chunks. Combined with object storage like S3 or Spaces, this means:
- Fast overviews without reading full-resolution data.
- Windowed reads that pull just the area of interest.
- No tile server required for many use cases.
Practical takeaway
If you are storing imagery and serving it to maps or analysis tools, converting to COG is usually one of the highest-leverage changes you can make. The tooling — GDAL, Rasterio, and rio-cogeo — makes the conversion straightforward.
