Wavelength-dependent spatial variation in the reflectance of ‘homogeneous’ ground calibration targets

Introduction Remotely sensed data are most useful if calibrated to spectral reflectance of known features. One simple method of calibration is regression of remote data on the reflectance of several ground targets as measured in the field, the so called empirical line method (ELM). The ideal situation would be one where a range of ground targets representing all the features of interest in the remote image were available for ground measurements (Lawless et al., 1998). The identification of suitable ground targets is constrained by several limitations, such as their size (to minimise edge effects), their absolute reflectance (to represent spectral characteristics of the image) and their effective spatial variability (to extract reflectance characteristics representative of the target). The size of a ground target is dependent on the spatial resolution of the image that must be calibrated (Justice & Townshend, 1981) and the number of observations needed to represent features in the image has been suggested to depend upon the spatial resolution of the remotely sensed image (Justice & Townshend, 1981) and on the spatial variability of the ground target (Harlan et al., 1979; Curran & Williamson, 1986). Although ground targets used for calibration should be spectrally “bland” and spatially uniform by definition (Clark et al., 1999), it is sometimes very difficult to find such places available for calibrating remotely sensed images. When surfaces that apparently satisfy these conditions are available in suitable size, their sampling needs to be designed to optimise representation of the whole surface and available resources (e.g., effort and time). Surfaces that look spatially uniform by eye may actually contain spatial variation, and this spatial variation may depends on wavelength (Atkinson & Emery, 1999). Such variability can be detected using geostatistics, which is concerned with issues such as spatial correlation and analyses of spatial data. Geostatistical tools have been used in a variety of studies and the variogram has been applied in remote sensing and ecology to design optimal sampling strategies for variables sampled in space (Atkinson, 1991; Rossi et al., 1992) and time (Salvatori et al., 1999). This study investigates the spatial variability of potentially suitable ground calibration targets (GCT) using a geostatistical approach, which gives results that can be used to design optimal sampling strategies for such surfaces. The targets were selected from an area where an Itres Instruments Compact Airborne Spectral Imager (casi) with ground resolution of about 1.5 metres was flown at the same time as ground data were acquired.