Start with the grid
Imagine a civilian lake survey receiving two image files, both described as ten-metre data. That shared label does not establish that they contain the same spatial detail or place the shoreline equally accurately. Before comparing them, the survey needs to know what the number describes. This is a hypothetical example of reading a product description.
ESA's Earth Observation guide defines ground sample distance as the ground distance between neighbouring pixel centres. That describes sampling. Spatial resolution concerns the detail distinguishable in the image, while geolocation accuracy concerns where that evidence is placed. Product documentation may use resolution as shorthand for a grid size, so a familiar label still needs its definition.
Read the product row, including both directions
Copernicus gives a concrete radar example in its Sentinel-1 product documentation. For Interferometric Wide Swath, Ground Range Detected, High Resolution products, the summary table lists 20 by 22 metres of resolution and 10 by 10 metres of pixel spacing. The two pairs describe the same product. They are separate entries because they answer different questions.
The table's notes matter: those resolution values are mid-range figures at mid-orbit altitude, averaged over swaths. They are not a promise that every image location or every Sentinel-1 product has identical characteristics. The two directions are range and azimuth. Reducing the row to a single ten-metre headline loses both the resolution distinction and the product-specific qualification.
A finer output grid can come from processing
The US Geological Survey provides an optical and thermal example. Its Landsat band table lists the Landsat 8 and 9 thermal infrared bands at 100 metres, resampled to 30 metres during Level-1 product processing. A reader examining the delivered grid must therefore keep the acquisition information alongside the output spacing.
Resampling places existing information onto another grid; the smaller output cells do not establish that the instrument independently observed new detail at every cell. In our lake example, exporting a coarse image onto a finer grid would not, by itself, supply newly measured shoreline features. That is a distinction about evidence, not an objection to resampling or to the usefulness of the resulting product.
Detail and location need different checks
USGS describes how Landsat's precision and terrain correction uses ground control points and elevation information. Its processing guidance also warns that products with different corrections can be misaligned when stacked. The location of a feature consequently depends on the product's geometric processing and reference information, not simply on the dimensions of its pixels.
For the hypothetical lake survey, an apparent shift between two shoreline maps could require investigation of image alignment before being described as physical change. A visually sharp boundary and many decimal places in its coordinates do not settle that question. The useful evidence includes the processing level, the stated location quality and whether the two products were suitable for the comparison.
Name the quantity behind an accuracy claim
Measurement precision is another separate question. ESA describes radar interferometry as using phase information from multiple acquisitions to measure effects such as ground deformation. Such a displacement estimate is a different quantity from the spatial detail resolved in an image. A small reported displacement does not mean the picture resolves objects of that same size.
When a report claims accuracy, ask accuracy of what: location, displacement, temperature or a mapped boundary? Then look for validation and uncertainty for that particular result. Pixel spacing alone cannot supply them. Our lake survey needs measurements appropriate to its question, with their limits explained. The documented capabilities of these civilian products do not establish the performance of an unrelated sensor.
Official ESA, Copernicus and USGS documentation reviewed on 11 October 2026. The lake survey is hypothetical. Sentinel-1 figures apply only to the named product class and the table's averaging conditions; Landsat figures apply to the named thermal bands. ESA's introductory guide is used for definitions and the distinction between deformation measurement and image detail, not for its historical commercial-market or constellation claims.
Read the evidence.
- 01Newcomers Earth Observation GuideEuropean Space Agency
- 02S1 ProductsESA / Copernicus SentiWiki
- 03What are the band designations for the Landsat satellites?US Geological Survey
- 04Landsat Levels of ProcessingUS Geological Survey