Image requirements
For a bi-temporal or multi-date image series to be subjected to change detection and time series analysis requires that the data area subject to:
Geometric
- Spatial resolution
- Pixel size must be the same
- If not, apply re-sampling algorithm if not (to largest common pixel size)
- Spatial registration between images
- R.M.S.E. between each image must be < 0.5 pixel
- Mis-registration will produce "false" change signals
- Images must have the same datum, projection and coordinates
- Image acquisition look angles must be the same
Spectral
- Ensure spectral band-widths and band-centres between dates match
Radiometric
- Ensure sensor quantisation between dates is identical
- Work with calibrated and corrected image data (radiance, reflectance)
- Ensure processing history of the data is known TEMPORAL
- Solar geometry needs to match
- Fully calibrated and corrected data to remove these effects (illumination, viewing & slope geometry)
- Variations in environmental conditions unable to be corrected for:
- atmospheric
- water bodies
- vegetation
- urban areas
If mis-registration and radiometric errors between images are reduced then change detection can be successful and is then depending on the quality of the individual mapping approach of a specific environmental parameter.
