Hyperspectral data mining to identify relevant canopy spectral features for estimating durum wheat growth, nitrogen status, and grain yield

Abstract

While hyperspectral sensors describe plant canopy reflectance in greater detail than multispectral sensors, they also suffer from issues with data redundancy and spectral autocorrelation. Data mining techniques that extract relevant spectral features from hyperspectral data will aid the development of novel sensors for plant trait estimation. The objectives of this research were to (1) compare broad-band reflectance, narrow-band reflectance, and spectral derivatives for estimation of durum wheat traits in the field and (2) develop a genetic algorithm to identify the most relevant spectral features for durum wheat trait estimation. Experiments at Maricopa, Arizona during the winters of 2010–2011 and 2011–2012 tested six durum wheat cultivars with six split-applied nitrogen (N) fertilization rates. Durum wheat traits, including leaf area index, canopy dry weight, and plant N content, were measured from destructive biomass samples on four occassions in each growing season. Grain yield and grain N content were also measured. Canopy spectral reflectance data in 701 narrow wavebands from 350nm to 1050nm were collected weekly using a field spectroradiometer. First- and second-order spectral derivatives were calculated using Savitzky-Golay filtering. The narrow-band data were also used to estimate reflectance in broad wavebands, as typically collected by two commercial multispectral instruments. Partial least squares regression (PLSR) compared the ability of each spectral data set to estimate each measured durum wheat trait. A genetic algorithm was developed to mine narrow-band canopy reflectance and spectral derivative data for spectral features that improved estimates of durum wheat traits. Multispectral data in 4 broad bands estimated leaf area index, canopy dry weight, and plant N content with root mean squared errors of cross validation (RMSECV) between 33.0% and 67.6%, while hyperspectral data in 701 narrow bands reduced RMSECV to values between 19.3% and 36.3%. Use of the genetic algorithm to identify less than 25 relevant spectral features further reduced RMSECV to values between 15.1% and 30.7%. Grain yield was optimally estimated from canopy spectral measurements between 110 and 130days after planting with RMSECV less than 7.6% using the genetic algorithm approach. The timing corresponded to anthesis and early grain fill when presence of wheat heads likely affected canopy spectral reflectance. By using a genetic algorithm to mine hyperspectral reflectance and spectral derivative data, durum wheat traits were optimally estimated from a subset of relevant canopy spectral features.