From two images to a change claim

Two images of a region are not automatically ready for subtraction. They need compatible locations, usable observations and an understood processing history. Otherwise, a shifted boundary or a cloud shadow can appear to be a surface change.

In a classroom exercise, a field boundary moves by one cell because one image is misregistered. A simple difference map highlights a strip along the boundary even though the field itself has not changed. Checking alignment addresses that alternative explanation.

A difference could reflect a real surface change, seasonal conditions, illumination, an instrument difference or a processing choice. The task is to test plausible explanations using dates, metadata and independent evidence. A classification label is the result of an analysis, not a property that arrived automatically from orbit.

CBERS imagery can support environmental monitoring, but a candidate change is not by itself a complete account of cause, intent or responsibility. An observation can direct further investigation without answering every question about the event.

Build a change claim

Imagine a 1,000-cell study region with 800 valid cells on both dates. An analysis flags 40 of those valid cells. The flagged share is 5 percent of the observed cells; the other 200 cells remain unassessed. Calling it a verified 5 percent change across the whole region would ignore the missing coverage.

Record the input scenes, masks, alignment, method and validation results. Someone reviewing the map should be able to distinguish a preliminary alert from a confirmed interpretation and see which parts of the region were actually compared.