The Application of the Concept of Indicative Neighbourhood on Landsat ETM + Images and Orthophotos Using Circular and Annulus Kernels

Abstract

We calculated the mean and standard deviation from Landsat ETM+ red and panchromatic channels, and from red tone and lightness values of orthophotos within several kernel radii, in order to recognize three different variables: multinomial—forest stand type (deciduous, coniferous, mixed), numerical—forest stand coverage, binomial—the presence/absence of orchid species Epipactis palustris. Case-based iterative weighting of observations and their features in the software system Constud was used. Goodness-of-fit of predictions was estimated using leave-one-out cross validation. Cohen’s kappa index of agreement was applied to nominal variables, and RMSE was used for stand coverage. The novel aspect is the inclusion of additional information from particular neighbourhood zones (indicative neighbourhood) using annulus kernels, and combining those with focal circular ones. The characteristics of neighbourhood in conjunction with local image pattern enabled more accurate estimations than the application of a single kernel. The best combinations most often contained a 10ldots25 m radius focal kernel and an annulus kernel having internal and external radii ranging from 25 to 200 m. The optimal radii applied on the Landsat image were usually larger than those for the orthophotos. The optimal kernel size did not depend on either reflectance band or target variable.