Diffuse reflectance spectroscopy for field scale assessment of winter wheat yield

Abstract

The objective was to evaluate the ability of visible and near-infrared (NIR) spectroscopy to predict winter wheat grain yield, according to the performance of different prediction models. In situ reflectance measurements (350–1050 nm) were acquired from winter wheat flag leaves grown under nine mineral nitrogen (N) fertilization treatments (0–300 kg N ha−1), during stem extension developmental stage. Linear statistical models (MLR—multiple linear regression, PLSR—partial least squares regression) and non-linear prediction (ANN—artificial neural networks) were generated to estimate grain yield, based on derived variables from hyperspectral data as input features (first derivative of reflectance in form of principal components—PCs and vegetation indices—VIs). The expected influence of variable N fertilization on agronomic and spectral variables was recorded. The red and NIR reflectance contributed most to development of PCs, while VIs were calculated from 704 nm (λRED) and 785 nm (λNIR). Very strong positive relationship was determined between grain yield and VIs. ANN models were the most efficient in capturing the complex link between grain yield and leaf reflectance compared to the corresponding VIs, MLR and PLSR models, indicating good learning performance. In terms of N stress and non-N-limited environment, it can be concluded that the prediction methods used in this study can provide in-season estimates of winter wheat yield at a field scale based on hyperspectral data. Key spectral features and algorithms defined in this study should help to support site-specific and real-time yield forecasting in winter wheat production.