Turn spectral measurements into a useful product

AfDevSat’s reported objective is hyperspectral Earth observation. Such an instrument samples many neighbouring wavelength intervals, building a spectrum for an observed area. Different materials can have different spectral patterns. USGS crop research illustrates how these patterns can support classification and why comparison data and error assessment matter. An ambitious observation goal is still separate from a validated delivered product.

More spectral samples do not automatically make ground pixels smaller. Imagine two fictional cameras with the same ground grid: one records ten wavelength bands and another records one hundred. The second samples more spectral information at each grid position. The statement alone tells us nothing about a smaller ground footprint, fewer clouds or a more accurate crop map.

For an invented uncompressed image, use 100 by 100 pixels, twenty bands and two bytes per pixel per band. The array contains 100 times 100 times 20 times 2, or 400,000 bytes. Increasing to forty bands while keeping everything else unchanged gives 800,000 bytes. Metadata, compression and transmission overhead are deliberately excluded from this calculation.

A useful mission plan connects this data volume to storage and downlink limits. Selecting fewer bands or compressing data changes what reaches the user; the effect on the intended analysis needs testing. The photograph shows NACHOS, a separate American compact hyperspectral mission, as a comparison of instrument and spacecraft hardware. It does not depict AfDevSat or establish that the two programs share an optical design.

Count samples across every dimension

Suppose an invented crop classifier is correct on 90 of 100 independently checked plots. Its observed accuracy for that sample is 90 percent. This is not proof of the same performance in another season, region or crop mixture. Keep the reference method, sample coverage and error types with the result.

If ten cloud-covered plots have no valid observation, label them unobserved. Assigning a confident crop class to each would hide missing evidence. A product intended to guide field visits should explain when and where the observation applies, what was measured and how uncertainty affects the suggested action.