Prediction of N, P, and K Contents in Sugarcane Leaves by VIS-NIR Spectroscopy and Modeling of NPK Interaction Effects

Abstract

Abstract. Methods were studied to predict the N, P, and K contents in sugarcane leaves quickly and accurately at the seedling, tillering, and elongation stages from leaf spectral reflectance. A total of 117 valid leaf samples were used to obtain leaf spectral reflectance with an indoor VIS-NIR spectrophotometer. Using the spectral data processed by CARS-PCA as an independent variable, a six-fold cross-validated PLS model for N, P, and K contents was established. The R2 values of the CARS-PCA-PLS models for N, P, and K prediction were 0.859, 0.677, and 0.932, respectively. Correlation analysis of the predicted N, P, and K contents was performed to explore the interaction effects between N, P, and K. To simulate the interaction effects among the three nutrients, 19 factors were assumed, including possible linear, quadratic, and cubic relationships between N, P, and K, and multi-factor cubic polynomial PLS and MLR regression models were established from those factors. In the modified MLR models, the determinants of N, P, and K were 0.891, 0.802, and 0.944, respectively, which improved the performance of the models by 3.7%, 18.5%, and 1.3%, respectively, compared with the CARS-PCA-PLS models, which were based on the spectral reflectance data. The results showed that application of VIS-NIR spectra combined with interaction effects between the nutrients could effectively predict the N, P, and K contents in the early and middle growth stages of sugarcane.HighlightsCompetitive adaptive reweighted sampling (CARS) was adopted to select wavebands for nutrient prediction.N, P, and K interaction effects were simulated with 19 factors, including linear, quadratic, and cubic relationships.The interaction factors were used in multiple linear regression models, and improved prediction was achieved. Keywords: CARS-PCA, Interaction effect, NPK, Sugarcane, VIS-NIR spectroscopy.