Jd. Wickham et al., LANDSCAPE CONTAGION IN RASTER AND VECTOR ENVIRONMENTS, International journal of geographical information systems, 10(7), 1996, pp. 891-899
Raster land cover data has been used to construct adjacency matrices b
y scanning the pixel edges and calculating proportions according to th
e land cover types joined. Like joins (e.g., forest to forest) are mos
t common because pixel size is typically much smaller than the average
patch size. The dominance of like joins (the matrix diagonal) of the
adjacency matrix has led to one interpretation of the matrix as landsc
ape contagion (the tendency of land cover to cluster into a few, large
patches). Construction of the adjacency matrix in a vector environmen
t results in a different measurement. In a vector environment, land co
ver patches are not sub-divided into pixels. Therefore, the main diago
nal of the adjacency matrix is zero. Construction of an adjacency matr
ix in a vector environment estimates the evenness of distribution of e
dge types, not contagion. The same edge type evenness metric is estima
ted in a raster environment by ignoring the matrix diagonal and re-sca
ling the off-diagonal elements so that they sum to one (1).