文献类型： 外文期刊

作者： Jia, Yonglin ¹ ; Li, Yi ¹ ; He, Jianqiang ¹ ; Biswas, Asim ³ ; Siddique, Kadambot. H. M. ⁴ ; Hou, Zhenan ⁵ ; Luo, Honghai ⁵ ; Wang, Chunxia ⁶ ; Xie, Xiangwen ² ;

作者机构： 1.Northwest A&F Univ, Northwest Agr & Forestry Univ, Coll Water Resources & Architectural Engn, Minist Educ,Key Lab Agr Soil & Water Engn Arid & S, Yangling 712100, Shaanxi, Peoples R China

2.Xinjiang Acad Agr Sci, Inst Soil Fertilizer & Agr Water Saving, Urumqi 830091, Peoples R China

3.Univ Guelph, Sch Environm Sci, Guelph, ON N1G 2W1, Canada

4.Univ Western Australia, UWA Inst Agr, Perth, WA 6001, Australia

5.Shihezi Univ, Coll Agr, Shihezi 832003, Xinjiang, Peoples R China

6.Shihezi Univ, Coll Water Conservancy & Architectural Engn, Shihezi 832003, Xinjiang, Peoples R China

关键词： Precision agriculture; UAV; Nitrogen requirement; Nitrogen diagnosis; Cotton

期刊名称：FIELD CROPS RESEARCH （ 影响因子：6.4； 五年影响因子：6.6 ）

ISSN： 0378-4290

年卷期： 2025 年 321 卷

页码：

收录情况： SCI

摘要： Context or problem: Remote sensing, particularly through unmanned aerial vehicles (UAVs), has emerged as a pivotal tool in precision agriculture, especially for nitrogen (N) management. Traditional methods, while effective in quantifying crop N status using the Nitrogen Nutrition Index (NNI), fall short in providing quantitative fertilization strategies. Methods: This study bridges this gap by developing a comprehensive method that leverages multispectral remote sensing data from UAVs to refine N fertilizer management in cotton cultivation within arid environments. By integrating both field observations and UAV-derived multispectral data, we established robust models capable of estimating both leaf and overall cotton nitrogen contents (CNC-leaf and CNC-all), as well as NNI, throughout the growing season. This facilitated real-time calculation of required N fertilizer doses in cotton fields. Results: We uniquely applied covariance diagnosis and full subset screening techniques, underscoring the efficacy of vegetation index categories (VIs) in enhancing prediction accuracy. The Random Forest (RF) model exhibited superior performance in predicting plant nitrogen content, particularly in CNC-leaf prediction (Calibration: R2=0.92, RMSE=7.7 g m-2, MAE=5.5 g m-2; Validation: R2=0.60, RMSE=16.5 g m-2, MAE=12.2 g m-2) as opposed to CNC-all prediction (Calibration: R2=0.78, RMSE=117.0 g m-2, MAE=154.3 g m-2; Validation: R2=0.34, RMSE=138.4 g m-2, MAE=190.4 g m-2). The RF model also demonstrated optimal performance in NNI prediction (Calibration: R2=0.93, RMSE=0.05, MAE=0.04; Validation: R2=0.73, RMSE=0.12, MAE=0.10), surpassing the predictions for CNC. Conclusions: Utilizing CNC and NNI estimates derived from the optimized RF model, this study succeeded in generating a comprehensive map detailing the N fertilizer requirement across cotton Fertilization zones were established for different treatments, revealing that biochar application levels primarily determine nitrogen fertilizer needs. As biochar application increases, nitrogen fertilizer demand decreases. Moreover, nitrogen application rates typically increase when irrigation levels reach either 120 % ETc or 60 % ETc. Implications or significance: This innovative approach not only empowers farmers with intuitive and accurate tools for real-time cotton N management but also fosters enhanced agricultural practices by integrating advanced remote sensing technologies with sophisticated data analysis methods. The findings of this study have significant implications for sustainable and efficient agricultural practices, particularly in arid regions, setting a new precedent in precision nitrogen management.