What to Check on the Ground After a Satellite Map Flags Change

A field visit should answer a specific question

A satellite map marks a patch of forest as changed. A team drives out, takes a photograph, and sees trees standing. Does that settle the question? Not necessarily. The image may be from months earlier. The location may fall near a boundary. Or the map may flag a change that is difficult to recognize from one point on the ground.

Field verification is most useful when it tests a specific uncertainty in a map. Before planning a visit, ask what you need to check: whether a change occurred, which land-cover class fits, whether the mapped boundary is in the right place, or whether the map’s interpretation is plausible. One observation cannot answer all of those questions.

A site visit and an accuracy assessment are different jobs

If the purpose is to investigate a few surprising locations, a targeted visit can help explain why the map and other evidence disagree. It may improve the interpretation of those locations. It does not show how often the map is right across a whole municipality or region.

A formal accuracy assessment has a broader job: estimate how well the map performs over a defined area. That requires a sampling design suited to the question. Olofsson and colleagues recommend probability sampling for assessing land-change maps and estimating area, along with a clear protocol for assigning reference labels and an analysis that respects the sample design.[1]

This distinction matters. A convenient sample of accessible road-side sites can be useful for troubleshooting. It cannot, by itself, support a representative accuracy percentage. If you need a statistical estimate, plan the sample before going to the field and report its limits and uncertainty.

Choose observations that can challenge the map

For a formal assessment, decide what the map’s categories mean and what unit is being checked: a point, a pixel, a plot, or a larger patch. Then select reference locations independently of the data used to build the classification. Checking only places the map already labels as changed can miss changes it failed to detect.

The sampling plan should reflect the decision. If the question concerns change, observations need to relate to both the mapped change and the relevant comparison period. A current visit can describe current conditions; it cannot directly show what was on the same ground several years earlier. Historical imagery, dated records, or interviews with appropriate consent may be needed to interpret past conditions.

The UN/FAO handbook, drawing on established land-change assessment guidance, emphasizes reference data collected independently from the classification and consistent class definitions for map and reference labels.[2] A small exploratory visit can still be worthwhile, but describe it as exploratory rather than presenting it as a formal accuracy assessment.

Match the observation to the image

Timing and location both matter. If vegetation, water, construction, or damage can change quickly, a field visit long after the satellite pass may describe a different condition. Record the image date and the observation date; do not treat them as interchangeable.

A coordinate also has uncertainty. The device, canopy, terrain, and way the observer records a location can affect where a point falls. Compare the observed feature with the map at an appropriate scale, and note when a boundary or mixed pixel makes a direct match ambiguous. A field note such as “tree cover present at this point” may be sound; “the full mapped polygon is forest” needs evidence across that polygon.

Record evidence another person can review

At each site, record the observation date, location and its stated accuracy, the method used, the feature observed, and the relationship between the observation and the mapped unit. Photographs can help if they are dated, oriented or described, and linked to the location. Record access limits, permissions, and anything that prevented a clear observation.

Keep the map label separate from the field observer’s label. If they disagree, preserve both and investigate the reason. The difference may reveal a map error, a timing mismatch, a classification definition that needs clarification, or a location problem. Do not force a match just to make the table look tidy.

Say what the check supports

A useful field report explains how locations were chosen, how many were visited, when observations were made, and what evidence was used to assign labels. If the sample was designed for a statistical accuracy estimate, report the method and uncertainty in terms that match that design. Published guidance recommends documenting the sampling and reference procedures, reporting error information, and quantifying uncertainty for estimates.[1]

If the work was a targeted set of visits, say so. Report what those observations helped clarify and what they cannot establish. “We checked six accessible sites and found two discrepancies” is more informative than claiming that a map has been validated, when the sampling did not support that conclusion.

A practical field-check list

Before leaving the office, make sure you can answer:

• Which uncertainty is this visit meant to address?
• What map unit and land-cover definitions will observers use?
• Are the chosen locations representative, or are they targeted for diagnosis?
• Does the observation date match the period shown in the imagery?
• How will location accuracy, photographs, permissions, and access limits be recorded?
• What will count as an unresolved or ambiguous observation?
• What conclusions will the sample support—and which will it not?

A field visit does not make a map reliable simply by happening. Its value comes from a clear question, observations suited to that question, and an honest account of what the evidence can support.