An independent geographical dispatch · Remote observation, grounded interpretation

Map desk / Frequently asked questions

Questions from the map desk

Short answers to the questions that often sit behind an apparently simple map.

What does AI add to satellite image analysis?

AI can organise spectral, spatial and temporal patterns into candidate classes, objects or changes. It does not independently establish the geographical meaning or cause of those patterns. Human decisions still shape the target, labels, evaluation and interpretation. Start with the observation guide.

Is a smaller pixel always better?

No. Useful resolution depends on the feature and question. Spectral information, observation frequency, quality and historical consistency can matter more than pixel size. Resampling an image onto a finer grid does not create newly observed detail.

Can satellite analysis see through cloud?

Optical imagery is obstructed by cloud. Radar such as Sentinel-1 can observe through cloud, but measures a different surface response and needs different interpretation. Radar also has geometric effects and noise characteristics that must be considered.

Does high model confidence mean the label is correct?

No. Model scores may be poorly calibrated and can be confidently wrong on unfamiliar landscapes or systematic label errors. Independent evaluation is still needed. Our classification dispatch explains why confidence and correctness are different claims.

Does NDVI measure ecosystem health?

Not directly. NDVI summarises a contrast between near-infrared and red reflectance. It can describe vegetation response but does not by itself establish biodiversity, species composition or ecological health. Soil, canopy structure, season and atmospheric effects can influence interpretation.

Can a change map explain what caused the change?

Usually not alone. A detected surface change may have several explanations. Attribution requires additional context such as weather records, management information or field evidence. Read Following change through time for a comparison workflow.

What makes reference evidence independent?

It has not been reused to train or tune the model being evaluated and is interpreted under a documented protocol suited to the target location, time and spatial support. Keeping reference interpretation separate from the model output can also reduce confirmation bias. See the ground-checks guide.

Are the diagrams live maps or measured results?

No. The diagrams are explanatory schematics. They do not represent live observations or measured datasets for real locations. Place names in the articles provide geographical context, not a claim that this site conducted a study there.

Can I upload imagery or send a message through the tool?

No imagery upload is provided. The contact tool prepares a plain-text draft locally for copying or download. Send it separately through your own email application. The tool does not transmit your draft to us. Visit the field desk.