Guan'gai paishui xuebao (May 2021)
Estimating Leaf Area Index of Peanut Using Hyperspectral Remote Sensing
Abstract
【Background】 Leaf area index (LAI) is one of the most important plant traits. It modulates both photosynthetic rate and transpiration, and understanding its variation in response to environmental change at different growth stages is hence important to improve crop management and safeguard agricultural production. 【Objective】 The purpose of this paper is to present the results of an experimental study on the relationship between hyperspectral reflectance and LAI of peanut. 【Method】 The experiment was conducted in 2019 at Zhengzhou Agricultural Meteorological Experimental station, in which hyperspectral data of the peanut canopy at different growth stages was measured using the ASD HandHeld 2. In the meantime, we also measured LAI of the peanut using the LAI 2200 canopy analyzer. We used both the logarithm of reciprocal of the hyperspectral data and the derivatives of the hyperspectral data to estimate the LAI. Different models including models using a single parameter derived from the hyperspectral data and multiple stepwise regression models were developed to estimate the LAI. Their accuracy and reliability were compared and tested again the measured LAI. 【Result】 Exponential model using a single parameter derived from the hyperspectral data is reliable with R2 being reasonably high for most tested examples. In particular, when using the parameter VI1=Rg/Rr (where Rg and Rr are the reflectance spectra of the green and red wavelength respectively), VI2=(Rg-Rr)/(Rg+Rr), or Rr, their associated R2 was all greater than 0.68. The rooted mean of square error of VI2 was the least, followed by VI1. 【Conclusion】 Using a single parameter, exponential mode using VI2 was most accurate to estimate peanut LAI at different growth stages. Analysis revealed that the most sensitive spectral indices for estimating the LAI were VI2 and VI1, and the multiple stepwise regression models using multiple spectral parameters were more accurate than the single-parameter models but they are more computationally complicated.
Keywords