A model-based approach for geometric correction of optical satellite images

Sammendrag/Abstract: In this report, we describe a method for geometric correction of satellite images generalizing a method developed for NOAA AVHRR images. The method is based on modeling the imaging process, including satellite orbital modeling. We also present an approach for automatic detection of ground control points based on matching of a near-infrared band with watermasks. The methods have been tested on two images: a system-corrected Landsat TM image with pixel size 30 m × 30 m, and an image with pixel size 240 m × 240 m (similar to EOS MODIS) derived from the system-corrected Landsat TM image. The detection procedure gave satisfying results. Using the correction method we obtained root mean square errors of 4.3 pixel units and 0.9 pixel units, respectively, on a set of ground control points. The results are satisfying for EOS MODIS, however, higher accuracy is desired for Landsat TM, and improvements are suggested.