Use of hyperspectral images from UAV-based imaging spectroradiometer to estimate cotton leaf area index

Abstract: Remote sensing technology can be used to estimate the leaf area index (LAI) value of crops rapidly and harmlessly. This study collected the hyperspectral images of cotton field using a new type of imaging spectroradiometer (UHD185), which has 138 channels from visible to near-infrared band range with high spatial resolution. The weight of UHD185 is less than 500 g and it can be easily mounted on a multi-rotor unmanned aerial vehicle (UAV), which makes it possible to acquire hyperspectral ortho images of crops at different spatial scales in any growth stage according to the requirements of crop growth monitoring. Cotton LAI values of 80 sampling points were measured by the SUNSCAN canopy analyzer during the UAV campaign, and the spectral information of cotton canopy at each point was extracted from the hyperspectral images after atmospheric and radiometric correction. The aim was to seek a new method to build the hyperspectral estimation model of cotton LAI using hyperspectral images. In the processing of multispectral image data, the bands used to calculate vegetation index (VI) are usually fixed. However, there are many other bands that can be also used to calculate VIs in hyperspectral images and these bands may contain important information about the ground objects. In this research, the extremal reflectance in the range of certain bands was chosen to calculate the extremum vegetation index (E_VI), intending to get the most significant vegetation feature of each pixel. Four VIs i.e. ratio vegetation index (RVI), difference vegetation index (DVI), normalized differential vegetation index (NDVI), and green-band normalized differential vegetation index (GNDVI), which are closely related to vegetation coverage, were selected to estimate the cotton LAI value. Original spectral reflectance, selected spectral reflectance by successive projections algorithm (SPA), fixed-band vegetation index (F_VI) and E_VI were taken as independent variables respectively to build hyperspectral cotton LAI estimation models using least squares and partial least squares (PLS) regression methods. The results indicated: 1) When using spectral reflectance to estimate cotton LAI, the SPA successfully simplified the model and improved the accuracy at the same time; 2) Models containing VIs had better predictive ability than those which used spectral reflectance as independent variables; (3) RVI was the best VI to estimate cotton LAI when using single VI as parameter; (4) E_VI improved the accuracy of LAI estimation models significantly compared with F_VI; (5) The LAI - E_VIs - PLS model, which takes multiple E_VIs as independent variables and uses PLS regression method, had the highest accuracy and best predictive ability (R2=0.85, RMSE=0.02). The cotton LAI estimation map was made by resolving each pixel in the cotton hyperspectral images using the LAI - E_VIs - PLS model. The result was validated by the measured LAI and showed good precision (R2=0.88, RMSE=0.29). Therefore, this new method to monitor the LAI of crops is feasible