Acta Agrestia Sinica ›› 2026, Vol. 34 ›› Issue (11): 4181-4190.DOI: 10.11733/j.issn.1007-0435.2026.11.025

Previous Articles    

Estimation of Forage Oat Aboveground Biomass Based on Multi-Temporal Multi-Source UAV Data

WU Guan-zhou2, YANG Bo1, MA Lin1, LIU Ya-ling3, WANG Xue-min1, MA Yun-tao2   

  1. 1. Institute of Animal Science, Chinese Academy of Agricultural Sciences, Beijing 100193, China;
    2. College of Land Science and Technology, China Agricultural University, Beijing 100193, China;
    3. Inner Mongolia Pratacultural Technology Innovation Center Co., Ltd., Hohhot, Inner Mongolia 010051, China
  • Received:2026-06-03 Revised:2026-08-10 Published:2026-11-04

基于无人机多时相多源数据的饲用燕麦地上生物量估算

吴冠洲2, 杨博1, 马琳1, 刘亚玲3, 王学敏1, 马韫韬2   

  1. 1. 中国农业科学院北京畜牧兽医研究所, 北京 100193;
    2. 中国农业大学土地科学与技术学院, 北京 100193;
    3. 内蒙古草业技术创新中心有限公司, 内蒙古 呼和浩特 010051
  • 通讯作者: 王学敏,E-mail:yuntao.ma@cau.edu.cn;马韫韬,E-mail:wangxuemin@caas.cn
  • 作者简介:吴冠洲(2002-),男,汉族,河北邢台人,硕士,主要从事数字农业研究,E-mail:S20243213672@cau.ed.cn

Abstract: To improve the estimation efficiency of aboveground biomass (AGB) of forage oat (Avena sativa L.), a total of 1460 forage oat germplasm accessions from Zhangjiakou were investigated. Based on four-data unmanned aerial vehicle (UAV) multi-source data collected from July to August 2025. red-green-blue (RGB) imagery, multispectral imagery, UAV photogrammetry-derived plant height, and vegetation indices were fused to develop a multi-temporal and multi-source deep learning model for AGB estimation. The performance of the deep learning model was compared with that of multiple linear regression (MLR), random forest (RF), and extreme gradient boosting (XGBoost). Following the exclusion of 50 samples with inadequate image quality, 1410 plot samples were retained for model construction. The results of the independent test set showed that the deep-learning model achieved a coefficient of determination (R2) of 0.86, a mean absolute error (MAE) of 37.43 g, and a root mean square error (RMSE) of 47.17 g, outperforming the three comparison models. Residual analysis indicated a systematic underestimation of AGB for high-biomass samples. The classified spatial maps could qualitatively reproduce the primary spatial patterns of observed AGB but could not replace plot-level error assessment. These results indicated that the combination of multi-temporal multi-source remote sensing and deep learning could provide technical support for AGB monitoring and field-scale spatial mapping of forage oat.

Key words: Forage oat, Aboveground biomass, UAV remote sensing, Multi-temporal multi-source data, Deep learning

摘要: 为提高饲用燕麦(Avena sativa L.)地上生物量估算效率,本研究以张家口地区1460份饲用燕麦种质资源为对象,基于田间采集的4个时相无人机多源数据,融合可见光(Red green blue,RGB)、多光谱(Multispectral image,MSI)、摄影测量株高(UAV photogrammetry-derived plant height)和植被指数(Vegetation indices,VIs),构建多时相多源深度学习地上生物量估算模型,并与多元线性回归(Multiple linear regression,MLR)、随机森林(Random forest,RF)和极端梯度提升(Extreme gradient boosting,XGBoost)模型进行比较。剔除50份影像质量不合格样本后,以1410份资源小区样本开展建模。独立测试集结果表明,深度学习模型的决定系数R2为0.86,平均绝对误差(Mean absolute error,MAE)和均方根误差(Root mean square error,RMSE)分别为37.43 g和47.17 g,优于3种对比模型。残差分析显示,高生物量样本仍存在低估趋势;分级空间图能够定性再现实测地上生物量的主要空间格局,但不能替代逐小区误差检验。研究表明,多时相无人机多源遥感与深度学习相结合,可为饲用燕麦地上生物量估算和田间尺度空间制图提供技术支撑。

关键词: 饲用燕麦, 地上生物量, 无人机遥感, 多时相多源数据, 深度学习

CLC Number: