Optimization of Second-Order Grey-Level Texture in High-Resolution Imagery for Statistical Estimation of Above-Ground Biomass

Abstract

In this paper, part of the Mt. Kenya forest with mixed vegetation biophysical characteristics was selected for grey-level co-occurrence matrix (GLCM) optimization and comparison based on semivariogram modeling from high spatial resolution QuickBird imagery. The results were applied to demonstrate the role of GLCM-textures in the estimation of Above-Ground Biomass (AGB) for: the dominant afromontane (camphor) trees, tea, young and old planted pine trees from QuickBird imagery. The texture optimization results were compared and combined with spectral (near-infrared) information for AGB estimation. To quantify the significance of GLCM-textures in AGB estimation, regressions between the field-AGB estimates and estimates from the NIR band and the tested GLCM-textures as independent variables, and their integration as dependent variables were compared. As independent variables, NIR and variance-texture bands gave the best results for the dominant camphor trees, with accuracies of 72% and 67.34% respectively. Variance and mean textures gave the best results upon combination with NIR, showing an improvement of 4.33% and 4.82% respectively over the NIR estimates. For tea, the combination of NIR with homogeneity, entropy and second moment textures gave the best and equal results (R2 = 0.684). For the young pine trees, correlation texture gave the overall best results (R2 = 0.741), and for the older pine trees, contrast texture gave the best results (R2 = 0.753) as independent variables. We conclude that the role of texture type and optimal window in AGB estimation depends on the: size (height), age, species, inherent spatial structure (natural or planted) and crown size of the vegetation species.

Keywords: Above-Ground Biomass (AGB), GLCM-texture optimization, QuickBird, semivariance, vegetation-trees