文献类型： 外文期刊

作者： Gao, Jinlong ¹ ; Han, Mengwei ⁶ ; Zhang, Dongmei ⁴ ; Ma, Zhanping ¹ ; Zhang, Yongkang ¹ ; Fu, Shuai ¹ ; Liang, Tiangang ¹ ;

作者机构： 1.Lanzhou Univ, State Key Lab Herbage Improvement & Grassland Agro, Lanzhou 730000, Peoples R China

2.Lanzhou Univ, Key Lab Grassland Livestock Ind Innovat, Minist Agr & Rural Affairs, Lanzhou 730000, Peoples R China

3.Lanzhou Univ, Coll Pastoral Agr Sci & Technol, Lanzhou 730000, Peoples R China

4.Chinese Acad Sci, Northwest Inst Ecoenvironm & Resources, Key Lab Cryospher Sci & Frozen Soil Engn, Lanzhou 730000, Peoples R China

5.Lanzhou Univ, Ctr Remote Sensing Ecol Environm Cold & Arid Reg, Lanzhou, Peoples R China

6.Chinese Acad Agr Sci, Inst Agr Resources & Reg Planning, State Key Lab Efficient Utilizat Arid & Semiarid A, Beijing 100081, Peoples R China

关键词： Alpine grassland; Forage quality; UAV remote sensing; Scale transfer; Multi-scale collaboration

期刊名称：EUROPEAN JOURNAL OF AGRONOMY （ 影响因子：5.5； 五年影响因子：5.9 ）

ISSN： 1161-0301

年卷期： 2025 年 170 卷

页码：

收录情况： SCI

摘要： Nitrogen is a key indicator of the nutritional quality and feeding value of forage in alpine grasslands. Therefore, accurately acquiring the spatiotemporal distribution of forage nitrogen over large areas using multi-scale remote sensing data is essential for assessing forage quality and optimizing grazing management strategies. However, insufficient ground observation data and the spatial mismatch between satellite pixels and field measurements pose significant challenges to achieving high-precision monitoring at the regional scale. Unmanned aerial vehicles (UAVs), with their exceptionally high spatial resolution, facilitate innovative approaches to cross-scale monitoring of forage nitrogen in alpine grasslands. This capability helps bridge the spatial scale gap between satellite imagery and ground-based measurements through upscaling. In this study, ground-based observations are integrated with UAV-acquired RGB and multispectral imagery using six machine learning algorithms-support vector machine, extreme learning machine, random forest, extreme gradient boosting, convolutional neural network, and deep neural network-to construct high-precision forage nitrogen estimation models at the plot scale. Various methods were then applied to upscale these results to the regional scale and integrate them with Sentinel-2 imagery to map forage nitrogen across the study area. The results demonstrate that models combining texture features from UAV RGB imagery with spectral features of UAV multispectral data achieve the highest accuracy at the plot level (V-R2 of 0.58-0.67, NRMSE of 14.78-16.76 %). At the regional scale, the proposed ground-UAV-satellite collaborative monitoring framework (V-R2 of 0.65-0.69, NRMSE of 11.92-12.53 %) significantly outperforms the commonly used ground-satellite upscaling strategy (V-R2 of 0.53-0.59, NRMSE of 16.92-17.86 %). Among the evaluated upscaling methods, the dominant class variability weighted method performs significantly better than the traditional local averaging upscaling method in UAV-satellite scale transfer, making it particularly well-suited for application in highly heterogeneous alpine grasslands. In conclusion, this study proposes a cross-scale collaborative monitoring framework in which UAVs function as an intermediate scale to mitigate spatial mismatches between ground-based observations and satellite pixels. The proposed approach offers a promising solution for advancing large-scale, high-precision monitoring of forage quality parameters in alpine grasslands.