Hyperspectral assessment of leaf nitrogen accumulation for winter wheat using different regression modeling

Real-time non-destructive monitoring of nitrogen accumulation by hyperspectral remote sensing is important for crop nitrogen management. In this study, winter wheat field experiment incorporating several varieties and exogenous nitrogen treatments was performed at multiple sites. Using hyperspectral readings of the experimental crops, the continuum removal method was used to expand the chlorophyll absorption characteristic band. The correlation among the spectral reflectance of the wheat canopy, the continuum removal spectrum, and leaf nitrogen accumulation (LNA) were systematically analyzed. The correlations between LNA and spectral parameters (e.g., original spectral reflectance, two-band combination parameters, and common vegetation indices) and continuum-removed absorption feature parameters were all compared. Three nonlinear modeling methods were considered (partial least squares regression, SVM regression, and random forest regression) and their relative ability to predict LNA was compared. Continuum removal treatment significantly improved the correlation between the continuum-removed spectra of the chlorophyll absorption regions (550-750 nm) and LNA. Results also show that RSI (NBDI743, NBDI703) could be used to estimate LNA using univariate linear regression (R-2 and root mean square error were 0.806 and 1.231 g m(-2), respectively). The SVM regression was found to be the most accurate regression model when chlorophyll absorption characteristic band reflectivity values normalized by the continuum removal process were taken as an input (R-2 and root mean square error values were 0.895 and 0.903 g m(-2), respectively). This approach was able to predict LNA of wheat using continuum-removed absorption features through hyperspectral measurements, which provide technical support for nitrogen diagnosis and precise crop production management.

Automated summary

Real-time non-destructive monitoring of nitrogen accumulation through hyperspectral remote sensing is crucial for crop nitrogen management. This study compared different parameters and modeling methods for predicting leaf nitrogen accumulation, with SVM regression model found to be the most accurate. Continuum removal treatment improved the correlation between chlorophyll absorption region spectra and leaf nitrogen accumulation.