What Satellite Change Detection Can—and Cannot—Show

A difference on the map is a question, not the answer

A hillside looks greener in one satellite image than it did in another. Has forest recovered? Was a field planted? Did the images catch different seasons, or did haze and cloud shadow affect one of them?

Change detection compares observations from different times to find places that may have changed. It is a useful way to narrow attention across a large area. But a mapped difference is not, by itself, an explanation. To make a sound claim, the comparison has to account for what each image measured, when it was collected and how it was processed.

Start with the question

Before choosing images or an index, decide what you need to know. Are you looking for new built surfaces, a change in vegetation, a shift in open water, or damage after a particular event? These are different questions and may call for different bands, dates and methods.

A broad label such as “environmental change” is not yet something an image can test. Write down the place, the time period and the kind of change you expect to detect. Also decide what would count as evidence against that interpretation. This keeps the analysis tied to a real question instead of producing a colourful map first and a story afterward.

Know what the sensor can resolve

Satellite images are measurements in bands of light, not photographs in the ordinary sense. Sensors record reflected energy in selected parts of the spectrum. Analysts combine those measurements to examine features such as vegetation or water.

The spatial detail varies by sensor and band. The European Space Agency describes Sentinel-2 as a 13-band mission, with bands at 10, 20 and 60 metre resolution.[1] The U.S. Geological Survey lists Landsat 8 multispectral bands at 30 metres and its panchromatic band at 15 metres.[2] Those figures describe the ground sampling of particular bands; they do not mean every feature of that size can be identified reliably. A narrow stream, small clearing or mixed land parcel may occupy only part of a pixel.

When comparing dates, use compatible bands and processing where possible. If sensors or products differ, record the difference and consider whether it could explain part of the apparent change.

Make the dates comparable

A pair of images taken in different seasons can show different vegetation, soil moisture, river levels or snow cover even if the land itself has not changed in the way you are investigating. In Nepal, where climate and terrain vary sharply, matching the month alone may not make two observations comparable. This is a practical reason to check local conditions and the purpose of the analysis, rather than relying on a fixed calendar rule.

Clouds and their shadows hide the surface. Haze and atmospheric correction can also affect pixel values. Landsat surface-reflectance products are processed to support analysis, but USGS documentation describes known issues in some conditions, including around clouds and over bright snow or ice.[3][4] A processing label is not a guarantee that every pixel is suitable for every comparison.

Look at the quality information supplied with the imagery. Mask clouds, shadows and invalid observations where appropriate, and inspect the images themselves around places where the result matters.

Separate the signal from the explanation

A change map can highlight pixels whose measurements differ between dates. A classification can group those differences into categories chosen by the analyst. Neither step automatically establishes what happened or why.

For example, a vegetation signal may change because of a land-cover transition, seasonal growth, harvest, fire, drought, image alignment, or processing differences. The evidence needed to distinguish among these explanations depends on the place and question. A result should therefore be described at the level the analysis supports: “these pixels were flagged for further review” is a narrower claim than “forest was cleared.”

Field observations, local records, higher-resolution imagery or other independent sources can help resolve important ambiguities. Field visits are not a universal cure: their value depends on where and when observations are collected, and whether they address the uncertainty in the map.

A short check before you trust the map

Before using a change-detection result in a report or decision, ask:

• What specific change was the analysis designed to detect?
• Which sensors and bands were compared, and what are their spatial resolutions?
• Are the dates comparable in season, surface conditions and event timing?
• How were clouds, shadows, missing pixels and processing issues handled?
• Could image alignment or mixed pixels create an apparent boundary or change?
• What independent evidence supports the interpretation, and what remains uncertain?
• Does the map’s scale match the decision it is being used to inform?

Keep the source images, dates, processing steps and uncertainty with the map. If another analyst cannot understand how the result was produced, it will be difficult to review or update it.

Use the map to guide the next question

Satellite imagery is strongest when it helps people see where to look, compare conditions over time and ask better questions. It can make patterns visible across a wide area, while field evidence and local knowledge can help explain what those patterns mean.

A responsible change map says what was compared, what the method detected, and what still needs checking. That is enough to make it useful. It does not need to claim more than the evidence can show.