GDVI3, GDVI2, NDVI, MSAVI and SAVI were evaluated for their dynamic ranges, the class accuracy of the Vegetation Index (VI) classifications, the effects of shadow delineation on the other land use classes and their applicability in vegetation delineation in Al-Qara Mountains, Oman. Supervised classifications of a SPOT scene by Support Vector Machines (SVM) algorithm were employed. GDVI3 showed the widest dynamic range in all land use types, while GDVI2 also exhibited evidently wider dynamic ranges for arid to semi-arid Al-Qara than NDVI, MSAVI and SAVI. GDVI3 reported the highest accuracies in delineating natural vegetation (dense – 74.80 %, medium-dense- 43.19 %), except for low-dense vegetation (40.51 %). It also performs the best in delineating bare soil and dry grass with over 80 % and 60 % accuracies. The attenuated reflectance created by the shadows results in VI signals in the range of dry grass to bare soil, enabling us to neglect the shadow effect on natural vegetation delineation due to below 9.50 % omissions from the shadows class. GDVI3 also limits shadow delineation better than the other indices, which will enable us to analyze spectral information recovery by the VI with the help of ground truth information under the shadows. For applications such as land degradation assessments, GDVI3 has better prospects over the other indices explored. Saturation at high-vigor vegetation is an issue in GDVI3, GDVI2 and NDVI. Our study also points to a dependency of a VI’s capability to weaken shadows on the number of training data pixels to be utilized in a supervised classification.
ASJC Scopus subject areas