Estimation of leaf area index at the late growth stage of crops using unmanned aerial vehicle hyperspectral images

Abstract

Estimation of leaf area index (LAI) is of vital importance to improve the prediction accuracy of crops quality and yield. However, it is more difficult to precisely assess LAI at the late growth stages of crops due to the influences of leaf senescence and soil background. Unmanned aerial vehicles (UAVs), with hyperspectral sensors onboard, can acquire high spatial and spectral resolution images and provide detailed information of fields, and consequently, are widely used for monitoring the biophysical parameters of crops in precision agriculture. The aim of this study was to evaluate the potential of UAV-based hyperspectral data in LAI estimation for sunflower and maize at the milk-filling stage, with machine learning regression algorithms (MLRA) for data analyses. Three algorithms including linear regression (LR), partial least square regression (PLSR) and kernel ridge regression (KRR) were used with the individual vegetation index (VI), VI-combination and spectral reflectance of full wavelengths as input variables. Results indicate that from the perspective of accuracy of estimation models, the PLSR based on VI-combination derived from hyperspectral images outperformed the LR based on individual VI and KRR based on VI-combination or spectral reflectance, which was proven to be the most suitable for the LAI estimation for both maize and sunflower at late growth stage, with 68% and 64% of the variation in LAI were explained, respectively. From the perspective of VIs tested, the modified triangular vegetation index (MTVI1) and improved soil-adjusted vegetation index (MSAVI) were found to be the best LAI estimators for maize and sunflower. Meanwhile, the contributions of the two VIs were also superior over other VIs tested in developing estimation models based on the PLSR method.