Improving the geolocation accuracy of VHR satellite imagery: orthorectification based on Copernicus DEM and its change to an up-to-date 5m DEM

The increasing amount of geo-spatial data goes in-line with the increasing number of Earth observation satellites in space. The integration of this spaceborne data in large storage data cubes allows the analysis of continental to global phenomena with machine-learning techniques but also requires consistent and accurate geolocation of the various input data. The unprocessed imagery contains distortions of various magnitudes depending on the off-nadir viewing angle but also the local topography imaged by the sensor. The radial distortion of nadir looking MR and HR sensors is correlated with the increasing distance of the image pixel from the scene center and must be corrected before integration in data cubes. The same applies to agile VHR sensors with their oblique viewing capability as well as side-looking SAR sensors. The final geolocation accuracy therefore depends on a detailed knowledge of the sensors’ viewing angle/direction but also the digital elevation data used on the ground segment processing chain (i.e. for the orthorectification process step).

 

The 30m and 90m instances of Copernicus DEM (GLO-30 & GLO-90) can be accessed via a free-and-open data policy from ESA and both represent a consistent and accurate Digital Surface Model (DSM) with input data acquired between December 2010 and January 2015. However, Earth surface is changing due to anthropogenous and environmental processes and relevant height data for the orthorectification process must be kept up-to-date but also upgraded in terms of resolution to keep pace with the improvement in spatial resolution of recent and future satellites.

 

This presented study will compare various orthorectification results of VHR Pléiades and Pléiades Neo satellite imagery in plain but also mountainous topographic terrain. In addition, a differentiation between urban and rural environments is applied when presenting the results of an absolute geolocation study and corresponding visual analysis. The DEM data used for orthorectification is the Copernicus DEM in its 30m version (CopDEM GLO-30; acquired 2010- to 2015) as well as more current WorldDEM Neo data (5 meters; input data acquired since 2017 until at least 2025) stem from the TanDEM-X mission. A major improvement is the rapid availability of an error-free DEM version after raw data acquisition which allows alignment to the acquisition date of the satellite-based VHR imagery. A tailoring of WorldDEM Neo DSM for orthorectification purposes is also presented.