Automatic Noaa Avhrr Image Registration in Mountains Area

This article presents a new and novel approach to accomplish AVHRR data navigation with errors of less than 0.9 km in the middle latitude regions. The AVHRR channel 4 counts and GTOPO30 elevation data and image correlation method are used. This approach is suitable for both day and night passes. It can also be used for other satellites with thermal sensing radiometers. A two steps approach is implemented. The first step is to obtain the first approximate navigation (initial navigation). The second step is to refine the initial navigation. The first step uses a physical elliptical orbit model, the Brouwer-Lyddane orbit prediction package, to determine satellite position and velocity, and uses a generic algorithm for computing earth location values to obtain longitude and latitude for pixels in the AVHRR image. The second step uses an automatic ground control points (GCPs) method to adjust the first approximation to obtain highly accurate geo-registration for this AVHRR image. In the automatic GCPs method, the templates with fixed geo-locations are extracted from AVHRR image and the correlation coefficients between every template and the elevation image (target window) are calculated. The template and target window is the best matched at the location where the correlation coefficient has the maximum value. The difference between longitude and latitude of the central pixel of the template and the matched sub-window of the target window is the error of the initial navigation. Subtract the error from the initial navigation, the AVHRR image navigation is refined. To quantitatively test the refining results. The high resolution coastlines from GMT (The General Map Tools, developed and maintained by Paul Wessel and Walter H. F. Smith.) package are superimposed onto the AVHRR image. The navigation errors are obtained from comparing the location difference of the coastlines from GMT and from AVHRR image. NOAA12 AVHRR images of early, middle and end of each month form January to August 1999 are used to test the performance of this algorithm. This paper also discusses the efficiencies of different correction methods and the effects of template’s size on the accuracy. The accuracy of this method is less than 1 km in the middle latitude areas.