Monitoring of Soil Moisture Content of Winter Oilseed Rape (Brassica napus L.) Based on Hyperspectral and Machine Learning Models

Timely acquisition of soil moisture content (SMC) in the root zone of field crops is crucial for achieving precise irrigation. In this study, high-spectral technology was employed to collect both SMC data at different soil depths during the bolting stage of winter oilseed rape (Brassica napus L.) and corresponding hyperspectral data through two consecutive years of field experiments (2021-2022). Two categories of spectral parameters were established, including "three-edge" spectral parameters such as blue, yellow, and red edge areas, as well as empirical vegetation indices that showed strong correlations with previous research and crop parameters. The spectral parameters with the highest correlation coefficients with SMC at various soil depths were selected. Subsequently, the selected vegetation indices were utilized as inputs for modeling. Support vector machine (SVM), random forest (RF), back propagation neural network (BPNN), and extreme learning machine (ELM) were employed to construct SMC estimation models for different soil depths during the jointing stage of winter oilseed rape. The results indicated that the majority of the three-edge parameters and empirical vegetation indices displayed higher correlation coefficients with SMC in the 0-20 cm soil depth compared to the 20-40 cm and 40-60 cm depths. The RF model proved to be the optimal modeling method for SMC estimation, particularly for the 0-20 cm soil depth. The validation set of the estimation model for this depth achieved a high coefficient of determination (R-2) of 0.944, a root-mean-square error (RMSE) of 0.005, and a mean relative error (MRE) of 3.074%. The findings of this study provide a basis for high-spectral monitoring of SMC in the root zone of winter oilseed rape, offering valuable insights for the rapid assessment of crop growth under water stress conditions.