Optimum narrow-band indices for estimation of vegetation water content using hyperspectral remote sensing considering soil background

Developments in the field of hyperspectral remote sensing have provided the possibility of having new indices for estimation of vegetation biochemical and biophysical properties. Information about vegetation water content and water stress has widespread utility in agriculture, forestry and hydrology and support management of the natural resources. The objective of this study was first to explore sensitive spectral bands that are most suitable for estimation of vegetation water content and second to investigate if soil type affects in selecting the best narrow band index and optimum bands for them in estimation of vegetation water content. The study takes advantage of using a dataset collected during a controlled laboratory experiment. Water content was destructively acquired for four species with different leaf size and shape and different treatments. The spectral measurements have been carried out by using a GER spectroradiometer. Two groups of narrow band vegetation indices, namely ratio based and soil based were compared for estimating vegetation water content by using linear regression model. All two band combinations involving 584 wavelengths between 400 and 2400 nm were used for calculation of narrow band vegetation indices (RVI, NDWI, TSAVI and SAVI2). for pool (n=95), dark soil(n=48) and light soil (n=47) dataset. The predictive performances of hyperspectral indices were then determined and compared using cross validated R 2 and RMSE between measured and estimated water content. However in pool data set, the selected narrow-band in all indices showed a high correlation in estimation of water content, highest correlation were observed for SAVI2 and RWI with water content. The coefficient of determination (R 2) between water content and optimum narrow band RWI, NDWI, SAVI2 and TSAVI using pool data set were 0.85, 0.81, 0.86, and 0.80 respectively. In soil type dataset, the RWI and NDWI were the best indices in light soil and RWI and SAVI2 in dark soil. The result indicates the better performance of narrowband SAVI2 almost in all data set. The least variation was depicted in SAVI2 when the soil type was changed. The result highlighted the role of background effect in selecting the best vegetation index and optimum spectral region for indices.