Measuring continuous landscape patterns with Gray-Level Co-Occurrence Matrix (GLCM) indices: An alternative to patch metrics?

Abstract

Abstract Characterizing landscape patterns is an important analytical step towards understanding the effects of physical layouts on ecological and social processes. While a continuous representation of landscape structure has great potential as a realistic alternative to traditional patch-based models, its empirical merits have been limited by the lack of measures for quantifying patterns from such continuous surface. This paper examines the utility of Gray-Level Co-Occurrence Matrix (GLCM) indices as spatial metrics applied to the landscape level for measuring underlying spatial properties. Eight GLCM indices (contrast, dissimilarity, homogeneity, energy, entropy, mean, variance, correlation) are compared to most commonly used 18 landscape metrics (LMs) featuring landscape composition, aggregation, dominance, dispersion, and shape complexity, with an application to urban tree canopy landscape. Two different types of map, sub-pixel tree canopy cover percentage map versus binary tree-pixel map, are used to compute GLCM indices and class-level LMs with a moving window approach across 4556 focal points. The data, extracted from the National Land Cover Database (NLCD) and the National Agricultural Imagery Program (NAIP), characterize the city of Columbus and Franklin County, Ohio. Correlation and regression analyses demonstrate that there is a strong and robust analogy between textural traits implied by GLCM indices and patch-based characteristics measured by LMs. Four LM components generated by principal component analysis contribute differently to individual GLCM indices, enabling more nuanced interpretation of GLCM indices in terms of LMs. The identified meanings consist of a unique mix of patch abundance, aggregation, dispersion, large patch dominance, patch size variability, and landscape homogeneity. The prediction of landscape patterns by GLCM indices increases in accuracy with landscape size, to a scale comparable to census tracts, while staying robust to the variation in GLCM bin width. GLCM indices can serve as reliable indicators of spatial configuration, and therefore provide an effective tool for researchers to better utilize continuous landscape models.