Submitted to: Field Crops Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: April 25, 2013
Publication Date: June 30, 2013
Citation: Singh, S.K., Hoyos-Villegas, V., Ray, J.D., Smith, J.R., Fritschi, F.B. 2013. Quantification of leaf pigments in soybean (Glycine max (L.) Merr.) based on wavelet decomposition of hyperspectral features. Field Crops Research. 149:20-32. Interpretive Summary: Accurate prediction of leaf pigments (chlorophyll) from light reflected from leaves (spectral reflectance) is important because it allows the non-destructive, rapid assessment of crop-nitrogen status under field conditions. Reflectance and the concentration of selected leaf pigments were measured on 385 field grown soybean lines during flowering and seed development stages each in both 2009 and 2010. Different techniques and methods were used to develop various prediction models to evaluate the relationship between reflectance and pigment concentration. The models showed potential for estimation of pigment concentrations using soybean reflectance data. A general-purpose model was developed that predicted pigment concentrations with high accuracy. The methods used in this study are suitable to develop models to predict leaf pigment concentration based on canopy reflectance. Such models make it easier to measure the nitrogen status of crops without expensive and destructive measurements.
Technical Abstract: Accurate prediction of leaf pigments from spectral reflectance is important because it allows non-destructive, rapid assessment of crop-N status under field conditions. Canopy reflectance and leaf pigments (chlorophyll and carotenoids concentration) were measured on 385 field grown soybean genotypes during flowering and seed development stages each in 2009 and 2010. Spectral features related to pigments were extracted based on several known spectral indices and using a number of analytical methods to develop prediction models with single waveband (single-band), two (simple-ratio) or more (multiple linear regression, MLR) wavebands. Among the tested methods, fitness and accuracy (measured as coefficient of determination, R2; root mean square error, RMSE; and relative error, %RE) of the prediction models developed using MLR was greatest. The accuracy of known indices such as the Maccioni-index and canopy chlorophyll content index showed potential for estimation of pigment concentrations using soybean canopy reflectance data. Though, models developed using transformed spectra outperformed the original reflectance spectra irrespective of the analytical method used. In general, the validation of the MLR models revealed limited accuracy across sampling dates and types of spectra used. Continuous wavelet transformed spectra using ‘mexican hat’ wavelet family (CWT-mexh) produced the best model with the highest accuracy. The selected wavebands in the models primarily consisted of the visible (400 – 750 nm) as compared to the NIR (750 – 1350 nm) spectrum. A general-purpose MLR model using CWT-mexh spectra that predicted pigment concentrations with high accuracy (R2 = 0.86, RMSE = 2.12 and RE = 12.5%; chlorophyll and R2 = 0.83, RMSE = 0.56 and RE = 12.7%; carotenoids) was developed. The analytical and transformation methods employed in the current study are suitable to develop models to predict leaf pigment concentration based on canopy reflectance.