What Sentinel-2 gives you for free
The two Copernicus Sentinel-2 satellites image every piece of land on Earth roughly every five days, in thirteen spectral bands, with an open archive back to 2017. Four bands (blue, green, red, near-infrared) come at 10 m, six at 20 m. That is unmatched for monitoring change over time, and far too coarse to see a field boundary, a building footprint or a road width clearly.
What super-resolution actually does
A super-resolution model is a neural network trained on pairs of coarse and fine imagery of the same places. Given a new 10 m scene, it predicts what the same scene would look like on a finer grid, using everything it learned about how edges, textures and spectral signatures relate across scales. The output is a raster with 1 m pixels in which roads, parcel edges, buildings and vegetation patterns are much sharper than the input.
Novascope applies this to all ten useful bands at once and keeps values on the original reflectance scale, which matters: indices such as NDVI, NDWI or NDBI computed on the output behave as they do on native Sentinel-2.
What it cannot do
Output is model-enhanced: the 1 m grid is produced by a deep super-resolution model from 10 m and 20 m measurements. Structure finer than the source resolution is reconstructed and must be validated before survey-grade or legal use.
- It does not see through cloud. A cloudy input pixel stays cloudy; choose a clear scene, or use an all-weather model built for that purpose.
- It does not add information that was never measured. A one-metre object that has no trace at 10 m will not appear. Independent checks against 3 m reference imagery show real detail resolved to about 5 m; finer texture is model-generated.
- It is not a substitute for survey-grade, legal-evidence or safety-critical imagery, where a native very-high-resolution sensor is the right tool.
How to validate the output
- Compare against anything you already trust for a small part of the area: a drone flight, a cadastral map, a recent commercial image, or a site visit.
- Check that features you know exist are where they should be and have plausible sizes; look especially at straight edges and regular patterns, where models are strongest.
- Compute your usual index on both the 10 m input and the 1 m output and confirm the area statistics agree; they should, because reflectance is preserved.
- Keep the scene ID with your results so the exact acquisition is traceable.
When to use it, and when not to
Use enhanced Sentinel-2 when you need sharper context over a large area or across many dates and the cost of tasking a commercial satellite for each one is not justified: crop and field monitoring, construction progress, urban planning, mining and energy sites, insurance screening, environmental surveys. Use native sub-metre imagery when a measurement must stand on its own, or when the object of interest is smaller than a few metres.
How Novascope delivers it
Draw or upload your area, pick the exact scene from the archive while looking at its real 10 m pixels over your area, pay per square kilometre from €2.00, and download cloud-optimised geotiff, 10 bands, 1 m grid, utm, uint16 reflectance × 10,000; plus 8-bit true-colour and ndvi geotiffs by the next working day. Every file is produced by the AOS team on the scene you chose; nothing is pre-rendered or interpolated between dates.