Cotton yield estimation based on hyperspectral remote sensing in arid region of China

The aim of this study was to reveal the correlation between cotton yield and the canopy spectral index, and to build yield estimation models based on hyperspectral imaging, so as to promote the application of the hyperspectral technique in monitoring cotton growth and estimating yield. Combined with growth and development characteristics of cotton, the spectral reflectivity of the cotton canopy at different growth stages was calculated. The hyperspectral vegetation index was defined according to spectral curve characteristics. The spectral reflectivity and yield were analyzed based on estimation of the ground spectrum and yield from the budding peak stage (BPS) to later boll opening stage (BOS). Statistical analysis showed that cotton yield was negatively correlated with spectral reflectivity in the visible light bands, positively correlated in the near-infrared (NIR) wavebands, and negatively correlated in the short-wave infrared bands during all growth stages. The normalized differences of the spectral index were determined with the following bands 560 nm, 670 nm, 670 nm, 890 nm, 890 nm, 980 nm, 920 nm, 980?nm, 857 nm, 1?210 nm, 820 nm, 1?650 nm and 820 nm, 2?200 nm according to cotton canopy hyper-spectrum, which were significantly or especially significantly correlated with yield at all growth stages. At the same time, the VARI_700 ARVI spectral index was especially significantly correlated with cotton yield. Remote sensing models of cotton yield at different growth stages can provide a basis for monitoring cotton growth and forecasting production.