Wavelet-based coupling of leaf and canopy reflectance spectra to improve the estimation accuracy of foliar nitrogen concentration

Abstract

The leaf or canopy reflectance spectra of vegetation have been widely employed in estimating foliar nitrogen (N) concentration; however, they alone may not actually reflect the spectral and detailed information at a sampling plot. In this study, the potential spectral details of Carex (C. cinerascens) at a plot scale were derived using discrete wavelet transform, in which a simple operation of addition was employed to combine the reconstructed leaf and canopy reflectance at the fourth decomposition level (named “leaf-canopy d4 reflectance”). Partial least squares regression (PLSR), successive projections algorithm-based multiple linear regression (SPA-MLR) and random forest regression (RFR) models with leaf, canopy and leaf-canopy d4 reflectance were established and validated for foliar N estimation, respectively. The results showed that the PLSR (R2CV=0.718, determination coefficient of cross-validation; R2Val=0.743, determination coefficient of independent validation; RPD=1.91, residual prediction deviation), SPA-MLR (R2CV=0.709, R2Val=0.747, RPD=1.97) and RFR (R2CV=0.714, R2Val=0.783, RPD=2.16) models with leaf-canopy d4 reflectance outperformed their corresponding models with leaf or canopy reflectance. We conclude that the wavelet-based coupling of leaf and canopy reflectance spectra has great potential in the accurate estimation of foliar N concentration. This proposed strategy helps to understand the spectral details of vegetation at a plot scale, providing the potential for improving the plot-based estimation of plant nutrients in grassland, precision agriculture or forestry.