Models for Radar Scatterer Density in Terrain Images

Statistical models for the density of strong scatterers detected in high resolution radar images of rural terrain are presented. The probability distribution of the density of these natural terrain detections was found to be negative binomial. The variance of the negative binomial depended strongly on the window size used to measure the density. This dependence indicates that these detections, like those of a Poisson process, are locally uncorrelated, but have a slowly varying mean density whose correlation distance is 1 km or more. Negative binomial parameters were computed using over 200 km2 of terrain image for densities measured using windows sized from 75 m × 75 m to 375 m × 375 m. Average terrain detection densities of 10-3 and 10-4 per resolution cell were evaluated on images with resolutions of 7 and 28 ft.