Turn observations into a defensible map

The Xichang fire figure combines observations from January, April and May 2020, including Gao Fen 1-01 and three other satellites. Read the embedded dates before comparing colors. The panels document different stages of an event, not four simultaneous views through identical instruments.

For an original mapping task, build a small evidence table: sensor, date, bands or radar mode, pixel grid, processing level and quality flags. Check whether the measurements concern the same physical quantity. Then align corresponding features. A mismatch along a road or shoreline can produce an apparent change even when the ground has not changed.

The QXS-SAROPT research team paired Gaofen-3 radar observations with optical images for three port cities. They describe selecting matching points, registering subregions and manually reviewing patches. The work illustrates why downloading two pictures of the same city is not enough to make a reliable combined dataset.

The authors also state that coverage of only three cities limits the dataset. In a new original exercise, suppose a method is trained on port scenes but requested for a mountainous farming district. Test it on independent, representative examples from that district. Good results on familiar ports do not settle performance under different terrain or land-cover conditions.

Imagine a map marks one hundred locations as flooded. Independent checks find eighty truly flooded and twenty not flooded. Those checks describe the reliability of positive map labels: 80/100 = 80 percent. They do not reveal how many flooded locations the map missed; that needs evidence beyond the positive labels. These numbers are invented for practice, not a reported Gaofen result.

Deliver the map with its date, provenance, processing version, validation sample, missing coverage and limitations. A decision maker may combine it with field reports and local knowledge. The final judgment should be traceable to the observations and their uncertainty, rather than to how crisp or colorful the map looks.