The filling stage of winter wheat is crucial for grain formation. Precise irrigation during this period can significantly enhance both grain yield and water productivity, especially in arid regions. This study introduces a method for precise irrigation decision-making of winter wheat at the filling stage based on UAV hyperspectral inversion of leaf water content (LWC). Through the relationship between soil water content (SWC) and LWC, the optimal irrigation amounts at the filling stage are determined. We utilized two-year field irrigation experiments (2022–2023). The successive projection algorithm (SPA) was applied to select sensitive bands of LWC. Partial least squares regression (PLSR) and random forest (RF) were employed to establish an LWC inversion model. The SPA-RF model was found to be the most effective, with determination coefficients (R²) of 0.95 and 0.96, root mean square errors (RMSE) of 3.00 % and 2.70 %, and normalized root mean square errors (NRMSE) of 6.47 % and 6.01 %, respectively. The SPA algorithm also improved the inversion efficiency of LWC. A significant positive correlation between SWC and LWC during the filling stage was observed, and a conversion model was developed for the pre-, mid-, and late-filling stages. The R² values for pre-, mid-, and late-filling stages were 0.75, 0.80, and 0.73, respectively, with corresponding RMSE values of 28.79 m³/ha 17.26 m³/ha, and 37.35 m³/ha. The results indicate a high consistency between the SWC estimated via hyperspectral inversion and the irrigation quota based on measured SWC, making the proposed method a valuable tool for optimizing irrigation during this critical growth phase. The method for estimating irrigation amounts during the filling stage, based on UAV hyperspectral imagery proposed in this study, offers valuable support for achieving precise irrigation decisions for winter wheat.

中文翻译：

---

基于无人机高光谱叶片含水量反演的冬小麦灌浆期精准灌溉决策


冬小麦的灌浆阶段对于籽粒的形成至关重要。在此期间，精确灌溉可以显著提高粮食产量和水生产力，尤其是在干旱地区。本研究介绍了一种基于无人机高光谱叶片含水量反演 （LWC） 的冬小麦灌浆期精确灌溉决策方法。通过土壤含水量 （SWC） 与 LWC 之间的关系，确定灌溉阶段的最佳灌溉量。我们利用了为期两年的田间灌溉实验 （2022-2023）。应用连续投影算法 （SPA） 选择 LWC 的敏感条带。采用偏最小二乘回归 （PLSR） 和随机森林 （RF） 建立 LWC 反演模型。发现 SPA-RF 模型是最有效的，决定系数 （R²） 分别为 0.95 和 0.96，均方根误差 （RMSE） 分别为 3.00 % 和 2.70 %，归一化均方根误差 （NRMSE） 分别为 6.47 % 和 6.01 %。SPA 算法还提高了 LWC 的反演效率。在充填阶段观察到 SWC 和 LWC 之间存在显著的正相关关系，并开发了充填前、中和后期阶段的转换模型。灌浆前、中期和后期阶段的 R² 值分别为 0.75、0.80 和 0.73，相应的 RMSE 值为 28.79 立方米/公顷、17.26 立方米/公顷和 37.35 立方米/公顷。结果表明，通过高光谱反演估计的 SWC 与基于测量的 SWC 的灌溉定额之间具有高度一致性，使所提出的方法成为在这个关键生长阶段优化灌溉的宝贵工具。 本研究中提出的基于无人机高光谱图像的灌溉量估算灌溉量的方法为实现冬小麦的精确灌溉决策提供了有价值的支持。