Sensitivity Evaluation of Spectral Vegetation Indices Using Sensitivity Functions for Stand Volume Estimation

Identification of the best vegetation indices (VIs) for use in quantitative analyzes of vegetation is one of the important issues for ecologists. The objective of this study was sensitivity evaluation of vegetation indices to a stand volume for identification of the best vegetation index (VI) in stand volume estimation using a statistical sensitivity method. Also, another objective of this study was to compare results of stand volume estimation using sensitivity function and the best subset regression. We also evaluated sensitivity of VI relative to another VI using relative sensitivity function. In this study, 99 plots 60m×60m each were used with systematic cluster sampling method. In each plot, data on tree species, diameter at breast height, stand height and geographic coordinates of each plot center were recorded. The vegetation indices were created using Landsat ETM+ data. In order to analyze the relationship between stand volume and vegetation indices, average digital number of pixels within 2×2 pixels window were extracted from vegetation indices. The result of the sensitivity function showed that NDWI and Greenness had high sensitivity compared to DVI, RAI and GEMI in stand volume estimation, respectively. Therefore, NDWI and Greenness were selected for estimating stand volume using satellite data. Also, the result of the best subset regression analyses showed that DVI and NDWI were best for estimation of the stand volume. The regression model with NDWI and Greenness could better predict stand volume (adjusted R2=55.4%) compared to DVI and NDWI (adjusted R2=43.5%). This is a 12% increase in adjusted R2. The results showed that relative sensitivity of NDWI to GEMI and Greenness is high. Generally, the sensitivity function expresses the change in sensitivity of a VI through the range of allometric characteristics, it is irrelevant of the unit or magnitude of vegetation indices and it tests the significance of the sensitivity with t-or-z statistic is useful for evaluation of sensitivity analysis of vegetation indices and identifying the best vegetation indices in quantitative and qualitative assessment of forested area.