Estimating the Canopy Chlorophyll Content of Winter Wheat Under Nitrogen Deficiency and Powdery Mildew Stress Using Machine Learning

Abstract

As an important indicator of the photosynthetic capacity of crops, canopy chlorophyll content (CCC) is nondestructively estimated by reflectance obtained using various spectrometers. Crop growth is often severely affected by N deficiency and diseases, and the compatibility between different stress data to develop unified estimation models needs further clarification. In this study, in field experimental conditions, hyperspectral data of the wheat canopy for nitrogen deficiency and powdery mildew stress were collected, along with canopy chlorophyll content. Comparative analysis of hyperspectral remote sensing data input features (original reflectance (OR), spectral index (SI) and wavelet features (WF)) was conducted. A combination of feature selection and machine learning was used to determine the best estimation mode for the accurate inversion of CCC under two stress conditions. The results showed that the canopy spectra of nitrogen deficiency and powdery mildew stress changed in the same trend. For nitrogen deficiency, the wavelengths sensitive to CCC mainly reflected canopy structure characteristics, followed by pigments. For powdery mildew stress, the wavelengths sensitive to CCC mainly reflected pigment characteristics, followed by canopy structure. Eight input features (two reflectance wavelengths, two spectral indices and four wavelet features) were determined using competitive adaptive reweighted sampling (CARS) and variance inflation factor (VIF) methods. Machine learning (ML) showed better estimation for both stress samples. For CCC estimation under nitrogen stress, random forest regression (RFR) was the more suitable (R 2 =0.828) and showed better monitoring accuracy in both the calibration and validation sets. For CCC estimation under powdery mildew stress, support vector machine regression (SVR) was more suitable (R 2 =0.787), especially when OR and WF data were used as input features. For the unified estimation of CCC under both stresses, SVR was slightly higher than RFR, where the WF-SVR Model R 2 was 0.846. The results demonstrated that the CWA-CARS-ML mode was feasible for CCC estimation under two different stresses, which provides an ideal reference and technical support for the evaluation of photosynthetic potential and precise management of crops.

Keywords: Chlorophyll content, Different stress, Remote sensing, Wavelet, Machine learning