Waveband selection using a phased regression with a bootstrap procedure for estimating legume content in a mixed sown pasture

Abstract

Legume content in grass–legume mixtures is a key parameter for deciding the forage quality and the amount of fertilizer application to the pasture due to nitrogen (N) fixation. To estimate legume content in a grass‐white clover (WC) mixed pasture in Hokkaido, we searched for robust hyperspectral wavebands from in situ canopy reflectance spectra over the 400–2350 nm range comparing a phased regression with a bootstrap procedure (PHR‐BS) (Ferwerda et al. 2006) and forward stepwise multiple linear regression (FS‐MLR). Canopy reflectance data and plant samples were obtained from 50 selected sites during two seasons (n = 100); spring (May) and summer (July) 2007. Although selected wavebands were similar in the PHR‐BS and FS‐MLR, PHR‐BS gave a higher predictive accuracy (44–74%) than FS‐MLR (35–73%). Selected wavebands in the final models were blue (400–456 nm) and red bands (659–670 nm) in visible wavelength, red‐edge region (704–724 nm), near infrared regions (813, 937, and 1121 nm), and shortwave infrared regions (2303–2344 nm) that are mainly linked to known biochemical components such as chlorophyll, N, lignin and cellulose. These results suggest that legume content in grass–legume mixtures can be predicted by in situ canopy reflectance, and that the predictive ability of the model can be improved by wavelength selection using the PHR‐BS method.

Genetic algorithm‐based partial least squares regression for estimating legume content in a grass‐legume mixture using field hyperspectral measurements

“…While the hyperspectral wavebands selected in GA‐PLS models did not perfectly match wavelengths previously identified using MLR (Kawamura et al . ), they were within 20 nm of previously known wavelength regions of foliar biochemical properties as reported by other researchers. In the visible region, chlorophyll absorption related to the blue region (400–460 nm) was selected in May.…”

“…The MLR model was developed using five selected wavebands in the previous study (Kawamura et al . ). The spectral band selection has been done by forward stepwise MLR with the maximum number of five selected bands to avoid over‐fitting.…”

“…All data handling and regression calculations were performed using PLS Toolbox version 6.2 (Eigenvector Research Inc., Manson, WA, USA) in Matlab software version 7.12 (Mathworks Inc., Sherborn, MA, USA). The MLR model was developed using five selected wavebands in the previous study (Kawamura et al 2011). The spectral band selection has been done by forward stepwise MLR with the maximum number of five selected bands to avoid over-fitting.…”

“…In our previous study (Kawamura et al 2011) using MLR, we demonstrated the potential of field hyperspectral measurements to estimate legume content in a mixed sown pasture in Hokkaido, Japan. In the current study, using the same dataset, we compared the performance of the GA-PLS with that of standard full-spectrum PLS (FS-PLS) and MLR models for estimating legume content using field hyperspectral data.…”

Estimating Plant Traits of Grasslands from UAV-Acquired Hyperspectral Images: A Comparison of Statistical Approaches

“…For this reason, numerous researchers have focused their attention on Stepwise Multiple Linear Regression (SMLR), which takes advantage of almost all of the hyperspectral information available [7,[29][30][31][32][33][34]. However, SMLR suffers from multicollinearity problems and the extensive spectral overlap of individual biochemical properties [35]. Partial least squares regression (PLSR) has been recognized as an alternative technique to SMLR, and it has been used earlier in the spectral data analysis for wheat [8] and heterogeneous grasslands [36,37].…”