Co-Registration Accuracy

[reseadmin]

In our acquisitions we found co-registration errors under two conditions:
        -A co-registration problem between two acquisition dates
        -A problem with the PARGE projection at the edge of the flight line, probably due to calculation instabilities related to an unreliable estimate of the aircraft roll.
But it is not impossible that these co-registration problems may be a consequence of a problem of projection of Hyperspectral data on Lidar data, Our hypothesis to resolve this difference would be to reduce the Hyper-spectral/Lidar co-registration error to a sub-pixel error.


Co-registering the data over forest areas is very challenging and can indeed cause problems which are not always solvable. Here are some thoughts:
- To make sure that the software is working properly, you'd have to check if the boresight alignment is up to date and has been done accurately for the processing
- Using the DSM for the geometric correction of forests always is affected by data intrinsic problems: how did you create the DSM, was it first pulse or was it some average? Radiometrically, first pulse is not corresponding to the radiometrically visible portion of the forest. At the edge of the forest this will be causing across track distortions. Sub-pixel errors are almost impossible at maximum resolution for this reason.
- if it was a roll problem, you'd see the distortions on all positions across track and not only at the edges,
- Something which also would need to be checked is if the height reference of the DSM and the imagery agree; a height offset causes distortion at the edges of the image only,
- the same appplies if the sensor model was not defined accurately.
- merging sensors on the raw data is not necessarily improving the situation,
- if it is about coregistration between VNIR and SWIR, there's also the problem of different spatial resolution and PSF - a perfect coregistration is not possible due to this reason.

So, if absolute coregistration is a must for your analyses, you'd have to reduce the resolution of your imagery to a resolution corresponding to the intrinsic accuracy of the data acquisition after considering all possible error sources in the budget.