Forest structural diversity characterization in Mediterranean pines of central Spain with QuickBird-2 imagery and canonical correlation analysis

Estimating Forest Canopy Cover in Black Locust (Robinia pseudoacacia L.) Plantations on the Loess Plateau Using Random Forest

“…For T B+VIs , the window size of 9 × 9 (equivalent to 21.6 m × 21.6 m), which corresponded to the extent of the field plots, produced higher accuracy than T pan . This result was consistent with that of Wood et al [29] and Gomez et al [59], who suggested that the window size should match the sample plot size to achieve high accuracy.…”

“…The importance of window size has been stressed in evaluations of texture measures [59]. Generally, for the eight GLCM texture measures in the selected seven window sizes, the 15 × 15 window size is optimal for panchromatic image and the 9 × 9 window size is optimal for the spectral variables.…”

Mapping forest leaf area index using reflectance and textural information derived from WorldView-2 imagery in a mixed natural forest area in Florida, US

“…Although existing studies on the comparison of the ability of spectral features extracted from visible bands with that from NIR bands for estimating or correlating some forest structural parameters also demonstrate a similar result that we found here, there are also no explanations on the underlying causality of the result. For examples, Gómez et al (2011) used spectral/spatial features extracted from QuickBird-2 imagery to estimate forest structural diversity and they found that the visible reflectance was more powerful than NIR data. In identifying forest tree species with hyperspectral measurements, Gong et al (1997) demonstrated that the discriminating power of the visible region is stronger than the NIR region.…”

“…The selection of the 13 texture-based features was based on literature review and their potential for estimating and mapping forest structural parameters including LAI from high resolution MS data, which has been demonstrated in many existing studies (e.g., Kraus et al, 2009;Murray et al, 2010;Gebreslasie et al, 2011;Ozdemir and Karnieli, 2011;Gómez et al, 2011Gómez et al, , 2012Gu et al, 2012Gu et al, , 2013Shamsoddini et al, 2013;Zhou et al, 2014). The 1st-order statistical texture measures are derived from the pixel values in a moving window with different window sizes, but do not consider the spatial relationships among pixels within a window.…”

Historical forest biomass dynamics modelled with Landsat spectral trajectories

“…Absolute and relative change of AGB between the rounds of NFI was calculated (Table 2). NFI2 intra-plot structural complexity was evaluated as in Gómez et al (2011a) calculating the median absolute deviation (MAD) of measured D (D MAD ) and H (H MAD ) in each plot: increasing values of the MAD indicate higher structural complexity, and a zero MAD value is possible but unlikely to occur if all trees in a plot have exactly the same dimension. Thirty two plots subject to complete resource extraction between the two NFI rounds were disqualified in support of our assumption of near to natural successional conditions, leaving 573 plots for further analysis.…”