Read the image sampling

INPE’s product catalogue lists Amazonia-1 WFI images on a 64-metre grid, with an 850-kilometre swath. These describe different properties: one concerns sampling on the ground and the other the width covered. A broad view is useful for regional patterns, but cannot be treated as a close-up of every small object.

In an ideal square-grid exercise, a 64 m by 64 m cell covers 4,096 square metres, or 0.4096 hectares. A 640 m by 640 m square aligned to the grid contains 100 such cells. This arithmetic describes the sampled area, not the certainty with which a boundary can be located. A narrow road can influence a pixel’s signal while remaining mixed with surrounding surfaces; a visible contribution does not mean its exact width is resolved.

WFI samples blue, green, red and near-infrared bands. The catalogue’s true-colour example assigns bands 3, 2 and 1 to the display’s red, green and blue channels. Near-infrared is outside ordinary human colour vision; including it in a displayed composite requires an explicit mapping to visible colours. A display colour is therefore not enough to identify the wavelength measured.

Consider an original toy pixel with normalized red response 0.2 and near-infrared response 0.6. Their normalized difference is (0.6 − 0.2) ÷ (0.6 + 0.2) = 0.5. That calculation summarizes a spectral contrast. It does not directly count trees, determine a legal land-use category or show when a clearing happened. Calibration, atmosphere, mixed cover and the purpose of an analysis affect what conclusions are justified.

Ground sampling is different from display size

The catalogue specifies 10-bit quantization, giving 2 to the power of 10, or 1,024 possible codes. This is a numerical encoding property, separate from the pixel’s ground size. The number of codes does not by itself state physical accuracy, measurement uncertainty or the units of a later processed product.

For an original comparison, imagine two otherwise identical images of the same 64-metre cells. One stores an intensity using 256 possible codes and the other uses 1,024. The second offers finer numerical steps in that encoding, but it does not divide each cell into four smaller ground areas. Enlarging either image on a screen likewise adds display pixels without acquiring new measurements of the ground.