Nearest neighbour analysis applied to synthetic aperture radar images for the description of urban land cover and land use

Abstract

ABSTRACTThis article presents an analysis method for the creation of an image variable that represents built-up land cover and land use within the urban fabric. The method is inspired by the nearest neighbour analysis and the data are from the synthetic aperture radar systems COSMO-SkyMed (CSK) and Radarsat-2 (RS2). Point features were identified from the extreme high backscattering values for each image and the spatial pattern extracted to represent the proportion of built-up land cover as clustered, random, or dispersed. The coefficient of association between the continuous nearest neighbour ratio image and the land-cover percentage cover in four classes are −0.9 and −0.8 with the CSK and the RS2 images, respectively. Considering a two-class land-cover scheme, the coefficient of association between variables approaches −0.9 for both images. Clustered features highlight individual buildings that are mixed in various neighbouring land-cover and land-use types. Residential land use is particularly well outlined using the CSK image, while large institutional, commercial, and light industry buildings are enhanced through the RS2 cross-polarization nearest neighbour ratio images.