No abstract

No abstract

Nitrogen (N) is a primary macronutrient essential for plant structures and metabolic processes, and the deficiency of N leads to critical plant disorders. The spectral reflectance can be used to predict the N status of plants using hyperspectral data. Therefore, the N status of wheat was predicted from hyperspectral data using machine learning techniques. Different derivative pre-processing treatments have been shown to have an impact on the spectral model performance. Therefore, we used different spectral pre-processing techniques (first derivative, deresolve and deresolve plus first derivative) coupled with six machine learning regression models (Support Vector Regression, Random Forest, k-nearest neighbours, Multilayer Perceptron, Gradient Boosting Regression and Partial Least Square Regression) to predict the N status of wheat. The deresolve plus first derivative spectral pre-processing technique along with Random Forest and Gradient Boosting Regression (R 2 > 0.85) were better than the other combination of spectral pre-processing and machine learning models to predict the N status of wheat. The eXplainable Artificial Intelligence (XAI) tool was used to provide the local and global explanations of the model decisions using SHapley Additive explanations (SHAP) values. The important wavelengths predicting N status were between 790 and 862 nm (global model) for Random Forest model. However, these wavelengths varied with the growth stages of wheat. The most important wavelength were 672, 794, 804, 806, 816 and 820 nm during the first six days of wheat growth (local model), 716, 794, 804 and 806 nm after 45 days of wheat growth, 724, 806, 820, 1556 and 1582 after 63-72 days of wheat growth and 718, 720, 724 and 1272 nm after 91-97 days of wheat growth. These results suggest that XAI tools are useful to explain the complex machine learning models related to hyperspectral data for remote monitoring of N status of wheat.

No abstract