A METHOD FOR ACCURATE GEOMETRIC CORRECTION OF NOAA AVHRR HRPT DATA

Although different methods for geometric correction of NOAA AVHRR data have already been established, the multitemporal character of most st udies has motivated the development of quick and nearly automatic corr ections, paying little attention to the development of accurate method s for correction of particularly interesting images. In the case of fi eld experiments (when comparing AVHRR data with local ground measureme nts) or when combining AVHRR data with other high resolution satellite data, geometric accuracy turns out to be very important, and we need a more precise method for geometric registration of the data. In the p roposed method, a Keplerian orbital model for the NOAA satellites is c onsidered as a reference model, and orbital elements are given as inpu t to the model from ephemeris data (orbit extrapolation from the last previous TBUS message). A few ground control points (GCP's) are then u sed to improve the determination of the orbital elements by comparing the two values of the instantaneous satellite-Earth's center distance that can be obtained from two independent relations for each GCP. A le ast-squares fitting algorithm is used to get the optimal determination of the five parameters that define the osculating orbit, by means of an iterative procedure. Satellite position within the orbit (mean anom aly) for each time instant is then determined, considering time differ ence between ephemeris data and image reported time (satellite interna l clock). After precise determination of nominal attitude angles for e ach instant, geometric correction is done for GCP's, and residual erro rs are interpreted as attitude angle variation effects (modeled as a s econd-degree polynomial in time), and in this way corrected. All attit ude angle deviations (in pitch, roll, and yaw) are simultaneously corr ected by applying to two reference vectors (the vector normal to the s canning plane and the vector that defines the instantaneous viewing di rection of the first pixel of each line), a three-axis rotation. In th e resulting algorithm, only three vectors are needed to define the geo metry at each time instant. All the parameters can then be stored in n ine main arrays of M elements (M is set to make the time resolution ju st the scanline time). A separate program perfoms the geometric correc tion, applying the orbital model to each point of the desired output g eographical area. An application of this method is presented, in which AVHRR data is registered over a 1:25 000 topographic map with subpixe l accuracy, allowing the use of AVHRR data in a nearly local scale in combination with other high resolution data such as LANDSAT TM or SPOT .