Stratified diagnosis of cotton canopy spectral characteristics based on CWT-SPA and its relationship with moisture, nitrogen, and SPAD values

Abstract

ABSTRACT Cotton is an economically valuable crop worldwide, and accurate and rapid leaf moisture content, nitrogen content, and soil plant analysis development (SPAD) value estimations are crucial for cotton growth. In this study, high-resolution spectral data were collected for the Tahe 2 cotton variety in Alar City, Xinjiang. The raw spectra were preprocessed using four methods: first derivative, standard normal variate transformation, second derivative, and multiplicative scatter correction. Wavelet coefficients at multiple scales were generated from the raw spectra using a continuous wavelet transform (CWT) and CWT with first derivative (CWT-FD). A quantitative detection model combining CWT and CWT-FD with successive projection algorithm – partial least squares (SPA-PLS) was developed to estimate the leaf moisture content, nitrogen content, and SPAD values. The performances of the different scales and methods for estimating these parameters were compared. Six stratification methods were employed to determine the optimal modelling results and decomposition scales for the parameters under various stratification schemes. The results demonstrated that the CWT and CWT-FD methods significantly improved the accuracy of estimating cotton canopy spectral characteristics related to the parameters, surpassing that of conventional spectral processing methods. The mexh, db5, and db5 wavelet basis functions, along with scale 3, were identified as the optimal modelling choices for moisture content, nitrogen content, and SPAD values throughout the cotton plant growth period. Among the six vertical stratification structures of cotton, the CWT and CWT-FD methods, along with the three wavelet basis functions, achieved the best results for the three parameters in the upper layer. The optimal relative percent deviation values for the three parameters of typical leaves ranged from 1.603 to 2.016, indicating satisfactory modelling performance. These findings provide a theoretical foundation for accurately estimating leaf moisture content, nitrogen content, and SPAD values using spectral technology, enabling the simultaneous extraction and monitoring of multiple physiological indicators.