Predicting grain yield and protein content using canopy reflectance in maize grown under different water and nitrogen levels

Predicting grain yield and protein content of maize using spectral reflectance data is very important for improved agricultural production. In this study, we predicted the grain yield and protein content of maize grown under different irrigation and nitrogen levels in 2018 and 2019 based on canopy spectral reflectance measurements at the V6 (sixth leaf), VT (tassel), and R2 (blister) stages. We developed a predictive approach, namely, spectral reflectance-physiological parameters-productivity, to predict grain yield and protein content on maize crop. First, the quantitative relationships between grain yield and protein content and physiological parameters (canopy chlorophyll content [CCC], leaf carbon accumulation [LCA], leaf nitrogen content [LNC], and leaf nitrogen accumulation [LNA]) were analysed. Then, vegetation indices (VIs) and wavelet features based on spectral reflectance were used to establish estimation models for physiological parameters. The physiological parameters were used as a bridge to connect the spectral reflectance data with grain yield and protein content. The purpose was to establish spectral inversion models to indirectly estimate grain yield and protein content. Results showed that grain yield had a significant linear relationship with CCC and LCA. In addition a grain protein content and LNC and LNA were also significantly related under different water, and nitrogen availability. The physiological parameter models with ratio vegetation indices (RVI), biorthogonal 3.3 (bior3.3), and reverse biorthogonal 1.5 (rbio1.5) were reliable in terms of predictability and applicability. Independent data verification suggested that grain yield was predicted using RVI769,758 (R-2 = 0.773, RMSE = 2.509) in water availability, rbio1.5(781,29) (R-2 = 0.744, RMSE = 2.850) in nitrogen availability, and rbio1.5(772,11) (R-2 = 0.506, RMSE = 2.297) in water-nitrogen availability. In addition, the grain protein content was predicted using RVI793,757 (R-2 = 0.704, RMSE = 0.744) in water availability, bior3.3(743,19) (R-2 = 0.717, RMSE = 0.957) in nitrogen availability, and RVI492,418 (R-2 = 0.715, RMSE = 1.224) in water-nitrogen availability. Therefore, maize grain yield and protein content can be accurately predicted using our modelling approach.

Automated summary

Predicting grain yield and protein content of maize using spectral reflectance data is crucial for improved agricultural production. This study developed a predictive approach based on canopy spectral reflectance measurements at different growth stages to establish models for estimating physiological parameters and ultimately grain yield and protein content. Results showed significant relationships between grain yield and protein content with physiological parameters, and models using vegetation indices and wavelet features were reliable for prediction of grain yield and protein content under various water and nitrogen availability conditions.