Application of narrow-band TBVI in estimating fruit yield in citrus

Abstract

Recently, the two-band vegetation index (TBVI) has been developed as an alternative to the conventional normalised difference vegetation index (NDVI) for assessing various agricultural crop characteristics. The potential of the narrow-band TBVI derived from airborne hyperspectral imagery to predict fruit yield in citrus was examined in this paper. Hyperspectral images in 72 visible and near-infrared (NIR) wavelength bands (from 407 to 898 nm) were acquired three times over a citrus orchard in Japan on May 26, June 21 and July 21, 2005 by an airborne imaging spectrometer for applications (AISA) Eagle system. Spectra for individual trees were obtained by averaging reflectance values of all pixels on each citrus canopy that was recognised from the acquired images. Narrow-band TBVI involving all possible two-band combinations of 72 channels was tested in terms of its potential to predict fruit yield on individual citrus trees. Multiple linear regression (MLR) models based on several significant wavelengths were also evaluated. Results indicate that the hyperspectral image obtained on May 26 demonstrated the highest correlation with citrus yield. This suggests that canopy features at the fast vegetation growth stage provide more relevant information on the yield variability among individual trees. Moreover, the TBVI showed a greater potential than the simple combination of several significant wavelengths in providing an indication of fruit yield in citrus. Based on the hyperspectral image obtained on May 26, the TBVIs calculated for the red and NIR regions were significantly correlated with citrus yield, and the “red edge” region showed a higher relevance than the red region in calculating the TBVI that has a higher correlation with citrus yield. The TBVI based on the 823 nm (NIR) and 728 nm (red edge) wavelengths was found to provide optimal citrus yield information ( R 2 =0.5795, RRMSE=0.6636). Due to the significant effect of the alternate bearing mechanism on individual citrus trees, canopy size cannot be used as a single predictor for citrus yield ( R 2 =0.4516, RRMSE=0.8298). However, by incorporating the TBVI and canopy size into the model, the prediction result was greatly improved ( R 2 =0.6913, RRMSE=0.6071). This suggests that the yield on individual trees is determined by the combined effect of the TBVI and canopy size. This study demonstrates the potential of narrow-band TBVI derived from airborne hyperspectral imagery to predict the fruit yield in citrus. Yield estimates can provide valuable information for forecasting yields, planning harvest schedules and generating prescription maps for tree-specific application of alternate bearing control measures and other management practices.