Using leaf spectral reflectance to monitor the effects of shading on nicotine content in tobacco leaves

Abstract

The objectives of this study were to determine the relationship between tobacco leaf nicotine content and leaf spectral reflectance, to identify specific regions of the light spectrum that could be used to detect nicotine content, and to develop spectral indices and quantitative models for quick and accurate estimation of nicotine content in flue-cured tobacco leaves under different shade conditions. The flue-cured tobacco cultivar Yunyan 87 was subjected to shade treatments in the field extending from the root elongation stage through the vigorous growth and maturing stages. Three shade treatments of 85% (S1), 65% (S2), and 45% (S3) of full solar radiation (S0, control) were applied and the nicotine, pigment, and nitrogen content and spectral reflectance of tobacco leaves were measured over a time course. Normalized difference vegetation indices (NDVI) and spectral ratio (SR) indices based on leaf reflectance spectra from 350 to 2500nm and correlations with nicotine content were determined. Shading significantly increased the nicotine content during the maturing stage. Significant differences in reflectance were measured for different shade treatments during the vigorous growth stage particularly at 350–700nm in the visible range and 750–1000nm in the near-infrared range. There was a significant correlation between nicotine content and nitrogen and pigment content. The regions of the spectrum that gave the best indication of nicotine content in flue-cured tobacco leaves were 420–750nm in the visible range in combination with 1400–1800nm and 2000–2400nm in the short-wave infrared range. Optimal spectral indices for SR (R450, R500) and NDVI (R2150, R610) were derived from measurements in these ranges. We also established linear models based on the spectral indices derived from field data gathered in 2012 for NDVI (R2150, R610), SR (450, 500), stepwise multiple linear regression (SMLR), and a back-propagation (BP) neural network with R2 values of 0.796, 0.810, 0.842, and 0.968, respectively. The linear models were validated using an independent data set from 2011 with RMSE values of 0.784, 0.958, 0.883, and 0.109 for NDVI (R2150, R610), SR (450, 500), SMLR, and the BP neural network, respectively. The results indicated that hyperspectral remote sensing can be used for quick and accurate monitoring of the leaf nicotine content and shading status of flue-cured tobacco crops.