strobpr
commented
3 months ago We first need to agree within CEOS what 'high' resolution means. There are as many schemes as there are interpretations of processing levels ;)
Here is the one we use in Copernicus:
Furthermore we need to distinguish geolocation uncertainty (absolute) from geometric co-registration (relative - in theory divided into intraband, intrapath [difference in viewing geometry <3(?)deg], and crosspath [different viewing geometry]).
As a benchmark 'pixel' should be replaced, as it is rather random, by 'GSD' (ground sampling distance - usually at sea level), which is a sensor characteristic.
In the range of HR2 and coarser it is reasonable to expect (less than) 0.5 GSD (absolute) geolocation uncertainty (CE95) and same or better in co-registration terms.
For HR1 or finer I would recommend to fix the requirement at 5m absolute and 0.5-1.0 GSD relative, as threshold for intraband and intrapath.
Need to agree a solution to the issues related to high resolution EO datasets and the 0.5 pixel rRMSE sub-pixel accuracy requirement of the Surface Reflectance PFS, and make a formal decision at LSI-VC-15 on the way forward. Discussions should consider the impact of this requirement on the applicability of CEOS-ARD to ‘New Space’ and industry generally; the core motivations for CEOS-ARD (stackable data, time series analyses); the stop-gap downsampling approach used by KARI and whether this is something that would be officially recommended to other VHR satellite operators; and the balance of scientific rigour versus inclusivity.