The Influence of DEM Accuracy on Topographic Correction of Ikonos Satellite Images

There is no research which specifically investigates the influence of the Digital Elevation Model (DEM) on topographic correction of satellite images. Such an investigation is necessary in view of the very high resolution (VHR) of recent sensors such as Ikonos and QuickBird and the low availability of accurate height information for the derivation of slope and aspect values. A comparative study of multispectral Ikonos images is presented using eight selected interpolation techniques with contour data. The DEM influence was evaluated by analyzing the variance and classification accuracy of topographically corrected images using the different DEMS. These were found to vary widely, with a 30 percent reduction in variance and a 20 percent improvement in overall classification accuracy between the worst and best performing interpolation techniques. Furthermore, smoothing techniques commonly used for the removal of noise in interpolated contour data or in existing DEMs were found to offer no significant improvement in intra-class variance or classification accuracy, when used with a grid resolution compatible with VHR images. However, a more rigorous planar slope function was found to be more effective, and when combined with the best interpolator, the natural neighbor, or Sibson, method, very high classification accuracy was achieved in the topographically corrected images. Introduction The problem of differential terrain illumination on satellite images has been investigated over three decades (Hoffer and Staff, 1975; Holben and Justice, 1980; Shepherd and Dymond, 2003; Xu et al., 2003), and a variety of correction techniques have been developed. Recently, models based on correction for both atmospheric and geometric effects of terrain claim to completely eliminate the effect of topography on image radiance (Sandmeier and Itten, 1997; Richter and Schlaepfer, 2002; Shepherd and Dymond, 2003). These models require the input and processing of both atmospheric parameters to account for differential reflectance effects and terrain parameters to account for illumination differences. Potential sources of error in these solutions include inadequate atmospheric data (Shepherd and Dymond, 2003), the non-Lambertian behaviour of the surfaces being corrected (Smith et al., 1980; Richter and Schlaepfer, 2002) and inaccuracies caused by the DEM used to generate slope and The Influence of DEM Accuracy on Topographic Correction of Ikonos Satellite Images Janet Nichol and Law Kin Hang aspect (Carter, 1992; Ekstrand, 1996; Hale and Rock, 2003). The band ratio technique (Chavez, 1996), which is able to reduce the topographic effect and is the only method not requiring a DEM, results in reduced radiometric resolution. Other techniques which correct image bands individually, require a DEM for the generation of aspect and slope parameters, whereby the inclination of the sun on the terrain can be modeled. These techniques appear to be only partially successful (Ekstrand, 1996; Xu et al., 2003) although both Meyer et al. (1993) and Hale and Rock (2003) who compared a variety of the available techniques concluded that apart from the Cosine correction, which was unsuitable, any correction was better than none. Riano et al. (2003) give an overview of existing techniques for correction of topographically induced radiometric distortion. Recent research in slope modeling has developed techniques for better surface orientation than is available from traditional interpolation techniques. These techniques are especially important since modern very high spatial resolution (VHR) sensors such as Ikonos and QuickBird require a more accurate DEM than can be obtained using standard interpolators. There has been no specific attempt to evaluate the effect of the DEM on the accuracy of topographic correction. Furthermore, most previous research on topographic correction has been concerned with medium resolution sensors such as Landsat and SPOT. Therefore, the main objective of this study is to evaluate the influence of the DEM on topographic correction of Ikonos VHR images. It compares DEMs from different interpolation techniques using two well-known topographic correction algorithms. The project concentrates on the interpolation of DEMs from contour data because, despite modern techniques, contour maps remain the most available form of elevation data.