Applying hyperspectral remote sensing technology to the prediction of soil moisture content (SMC) during the growth stage of soybean emerges as an effective approach, imperative for advancing the development of modern precision agriculture. This investigation focuses on SMC during the flowering stage under varying nitrogen application levels and film mulching treatments. The soybean canopy’s original hyperspectral data, acquired at the flowering stage, underwent 0–2-order differential transformation (with a step size of 0.5). Five spectral indices exhibiting the highest correlation with SMC were identified as optimal inputs. Three machine learning methods, namely support vector machine (SVM), random forest (RF), and back propagation neural network (BPNN), were employed to formulate the SMC prediction model. The results indicate the following: (1) The correlation between the optimal spectral index of each order, obtained after fractional differential transformation, and SMC significantly improved compared to the original hyperspectral reflectance data. The average correlation coefficient between each spectral index and SMC under the 1.5-order treatment was 0.380% higher than that of the original spectral index, with mNDI showing the highest correlation coefficient at 0.766. (2) In instances of utilizing the same modeling method with different input variables, the SMC prediction model’s accuracy follows the order: 1.5 order > 2.0 order > 1.0 order > 0.5 order > original order. Conversely, with consistent input variables and a change in the modeling method, the accuracy order becomes RF > SVM > BPNN. When comprehensively assessing model evaluation indicators, the 1.5-order differential method and RF method emerge as the preferred order differential method and model construction method, respectively. The R² for the optimal SMC estimation model in the modeling set and validation set were 0.912 and 0.792, RMSEs were 0.005 and 0.004, and MREs were 2.390% and 2.380%, respectively. This study lays the groundwork for future applications of hyperspectral remote sensing technology in developing soil moisture content estimation models for various crop growth stages and sparks discussions on enhancing the accuracy of these different soil moisture content estimation models.

Winter wheat production in the extremely arid oasis region of Xinjiang relies heavily on irrigation and fertilization, but conventional high-input management can induce luxury transpiration and non-productive water consumption, limiting the coordinated improvement of grain yield, water use efficiency (WUE), and economic benefits. To identify the threshold at which water–fertilizer inputs shift from efficient use to inefficient water consumption and to define a robust management range, a two-year field experiment was conducted in southern Xinjiang during the 2022–2023 and 2023–2024 growing seasons. Four irrigation levels, corresponding to 60%, 80%, 100%, and 120% of crop evapotranspiration (ET_c), and four fertilization levels were established to evaluate the effects of water–fertilizer interactions on canopy development, leaf gas exchange, evapotranspiration, yield, WUE, and economic benefits. Appropriate water and nutrient supply promoted canopy establishment and maintained higher photosynthetic capacity, thereby increasing grain yield, WUE, and net return. However, excessive inputs weakened yield gains and failed to synchronously improve WUE and economic benefits. Linear plateau models revealed clear thresholds in both the crop-stand scale evapotranspiration (ET)–dry matter accumulation (DM) relationship and the leaf-scale transpiration rate (Tr)–net photosynthetic rate (Pn) relationship. The seasonal ET thresholds were 504.59 and 553.87 mm in the two growing seasons, respectively, and the Tr threshold was 4.83 mmol m⁻² s⁻¹. Beyond these thresholds, additional water consumption was not effectively converted into photosynthetic assimilation or biomass accumulation, indicating luxury transpiration. Year-specific response surface analysis and TOPSIS evaluation showed that I3F3, namely 100% ET_c combined with 210–195–75 kg ha⁻¹ N–P₂O₅–K₂O, together with its adjacent range, sustained high grain yield, WUE, and economic benefits, with I3F3 achieving the best overall performance in both years. The intersection of the two-year high-performance regions further defined a robust interannual feasible range with an irrigation amount of 506.21–545.09 mm and a total fertilizer input of 369.54–628.33 kg ha⁻¹. Overall, maintaining water and fertilizer inputs within the I3F3-adjacent range can reduce non-productive water consumption and luxury transpiration risk while synergistically improving grain yield, WUE, and economic benefits in winter wheat.

Vertical heterogeneity in soybean canopies may lead to differences in the relationships between SPAD values at different canopy positions and canopy hyperspectral information, thereby affecting prediction performance. To characterize these spectral relationships and compare SPAD predictability, field experiments were conducted during the soybean flowering stage in Yangling, Shaanxi, China, in 2022 and 2023, with nine irrigation–nitrogen combinations and a rainfed control. A total of 120 paired observations of plant-level SPAD and canopy reflectance spectra were obtained from 20 fixed plots. Upper-layer SPAD, lower-layer SPAD, and their equal-weight arithmetic mean were used as estimation targets. Difference indices (DIs), soil-adjusted vegetation indices (SAVIs), and their combined feature set (ZU) were derived from fractional-order differentiated spectra and used with random forest (RF), support vector regression (SVR), and backpropagation neural network (BPNN) models to compare prediction performance among the three targets. The optimal differentiation orders and wavelength combinations varied among SPAD targets. Upper-layer SPAD showed its strongest association at an order of 0.4 (maximum |r| = 0.737), whereas lower-layer and mean SPAD showed their strongest associations at an order of 1.2, with maximum |r| values of 0.733 and 0.810, respectively. Across the spectral input–algorithm combinations, mean SPAD generally achieved higher prediction accuracy. Its best out-of-fold R² was 0.636, with an RMSE of 2.450 SPAD units, compared with maximum R² values of 0.461 and 0.510 for upper- and lower-layer SPAD, respectively. Further evaluation showed that mean-SPAD models maintained relatively favorable performance under repeated grouped validation and bidirectional cross-year testing. The SAVI-RF model achieved a repeated-validation R² of 0.597 ± 0.064 and cross-year R² values of 0.655 and 0.611 in the two directions. These findings indicate that SPAD predictability from canopy hyperspectral data varies with the sampling target, with the equal-weight mean of upper- and lower-layer SPAD showing relatively higher and more stable predictive performance within the current experiment. The study highlights the importance of vertical canopy sampling in hyperspectral SPAD estimation and provides evidence to inform sampling design and hyperspectral monitoring of soybean at flowering.