Read optical pixels and missing data

An optical product stores measurements on a grid. A displayed color can combine several wavelength bands, including bands outside human vision. False color is a deliberate assignment of measurements to display channels; it needs a legend and cannot be interpreted as the color a person would see from a window.

In an idealized square grid with 16-meter spacing, a cell covers 16 × 16 = 256 square meters. A 160-by-160-meter patch spans ten cells along each side, or one hundred cells. This exercise uses the spacing reported for Gao Fen 1-01’s wide-field imagery, but it does not imply that every object inside a cell is separately resolved.

A cloud or its shadow can change the recorded signal without a change in the land beneath it. A cloud mask flags affected samples so that they are not silently treated as clear ground measurements. Masking does not reveal the hidden surface. Another date or another suitable sensor may be needed.

A Gaofen-1 study tested cloud and shadow detection on 108 scenes acquired from 2013 to 2016. Its authors discuss mistakes involving bright surfaces and thin clouds, and uncertainty in manually drawn reference masks. A favorable overall score therefore does not guarantee a correct label for every pixel in a new scene.

Suppose an original exercise contains one hundred land cells, forty unusable because of cloud or shadow. An algorithm labels twelve of the sixty usable cells as changed. The observed-sample fraction is 12/60 = 20 percent. It does not establish that twenty percent of the entire area changed; the hidden cells might have a different pattern.

A useful report gives the observed area, acquisition dates, mask method and unobserved area alongside the result. For comparisons, align the grids and consider seasonal, illumination and processing differences. Resampling onto smaller cells changes the grid; it does not recover ground detail that the instrument never recorded.

Keep missing data out of the denominator