Coupling continuous wavelet transform with machine learning to improve water status prediction in winter wheat

Water is one of the essential factors for crop growth and development. Rapid and non-destructive monitoring of winter wheat water status is crucial for water diagnosis and irrigation management. Wavelet analysis has been widely used to process hyperspectral reflectance data for remote sensing monitoring of crop water status. However, most studies utilized only a single wavelet feature, and the potential of combined wavelet features remains unclear. This study aimed to use the combination of machine learning algorithms with wavelet analysis to make full use of the spectral information and improve the performance of wavelet analysis in winter wheat water status monitoring. Field experiments under four water and two nitrogen treatments were carried out from 2020 to 2022. Crop water status indicators and canopy reflectance spectra of winter wheat were acquired and analyzed. Wavelet index models and prediction models based on two-band wavelet features and multi-sensitive wavelet features were constructed by employing machine learning algorithms including multiple linear regression (MLR), random forest (RF), and support vector machine (SVM). The results showed that the best prediction of canopy water content, plant water content, and canopy equivalent water thickness was produced by RF (R-2 = 0.92, RMSE = 2.39%), SVM (R-2 = 0.93, RMSE = 2.12%) and SVM (R-2 = 0.79, RMSE = 99.31 mu m) based on multiple sensitive wavelet features, respectively. In addition, the wavelet index model also showed a good monitoring ability (R-2 = 0.71-0.82). The machine learning models based on multiple sensitive wavelet features after feature selection avoided collinearity between features, made full use of spectral information, and improved the monitoring performance of wavelet transform for winter wheat water status. The findings will be helpful to the water stress diagnosis and accurate irrigation management of winter wheat.

Automated summary

This study aims to improve the performance of wavelet analysis in winter wheat water status monitoring by combining machine learning algorithms. The results showed that the random forest and support vector machine models based on multiple sensitive wavelet features achieved the best prediction accuracy for canopy water content, plant water content, and canopy equivalent water thickness. This research is significant for water stress diagnosis and precise irrigation management of winter wheat.