Identifying Important Canopy Hyperspectral Reflectance Features for Estimating Crop Biophysical Traits.

Modern hyperspectral sensors permit reflectance measurements of crop canopies in hundreds of narrow spectral wavebands.  While full-spectrum data describes the plant canopy in greater detail compared to broad-band methods, it also suffers from issues with data redundancy and spectral autocorrelation.  Data mining techniques that extract important spectral features, including reflectance band ranges and spectral derivative peaks, will aid the development of novel sensors for estimating specific plant traits.  The objective of this work was to develop a range-operator-enabled genetic algorithm to extract canopy spectral reflectance features for estimating durum wheat canopy traits.  Field experiments at Maricopa, Arizona during the winters of 2010-2011 and 2011-2012 tested six durum wheat cultivars with six split-applied nitrogen fertilization rates.  Destructive biomass samples were collected four times in each growing season and were used to estimate leaf area index, canopy dry weight, and plant nitrogen content.  Canopy spectral reflectance data in 701 narrow wavebands from 350 nm to 1050 nm were collected weekly over each treatment plot using a field spectroradiometer.  First and second spectral derivatives were calculated using Savitzky-Golay filtering.  The narrow-band data was also used to estimate reflectance in broader wavebands, as typically collected by two commercially-available multi-band instruments having either four or eight channels (Cropscan).  Partial least squares regression (PLSR) was used to compare the ability of each spectral data set (broad-band, narrow-band, and derivative data) to estimate each durum wheat canopy trait.  A range-operator-enabled genetic algorithm was developed to select the type of spectral feature, starting wavelength, and waveband width for extracting key spectral features for trait estimation.  Results demonstrated that PLSR models using spectral features selected by genetic algorithm could estimate durum wheat canopy traits better than PLSR models based on either broad-band data or full-spectrum data.