Understand two image scales and fusion

The April 2023 prelaunch design account describes a 16-metre panchromatic product and 32-metre multispectral imagery with six spectral bands. These are announced design descriptions, not independently measured flight performance. The 2022 commercial announcement had used hyperspectral wording; it should not be silently combined with the later account into an invented instrument specification.

A panchromatic measurement covers a broad wavelength interval. Multispectral bands divide the measurements into separate wavelength ranges. Finer spatial sampling and more spectral bands answer different questions: where a boundary lies and how surfaces differ spectrally are related, but neither automatically supplies the other.

Use a simplified square grid to explore the announced scales. A 32 by 32 metre area corresponds to four 16 by 16 metre cells, because each side is divided in two. This is a geometric teaching example, not a reconstruction of Taifa-1’s detector layout. A 320-metre-wide feature spans ten 32-metre samples or twenty 16-metre samples along one dimension.

Smaller cells alone do not guarantee that a small object is identifiable. Optical blur, contrast, pointing, processing and ground location also matter. A map exported with one-metre pixels does not mean the satellite measured independent one-metre detail; the output grid and the original information must be documented separately.

Finer cells do not invent spectral measurements

Pansharpening combines spatial information from a finer panchromatic image with coarser multispectral information. Digital Earth Australia’s worked example distinguishes resampling onto a shared grid from the later fusion operation. Merely enlarging a multispectral file does not perform that fusion or create new measurements.

For an invented mapping workflow, compare the fused image with the original bands and independent reference information. Check alignment, boundaries and the quantities used by the application. A visually crisp result can still give a misleading classification if the transformation changes relevant values. Keep the original measurements and processing version so that the result can be examined and reproduced.