浙江农业学报 ›› 2016, Vol. 28 ›› Issue (10): 1790-1795.DOI: 10.3969/j.issn.1004-1524.2016.10.22

• 生物系统工程 • 上一篇    下一篇

基于冠层NDVI数据的北方粳稻产量模型研究

许童羽1, 2, 洪雪2, 陈春玲1, 2, *, 周云成1, 2, 曹英丽1, 2, 于丰华2, 李娜2   

  1. 1.沈阳农业大学 辽宁省农业信息化工程技术中心,辽宁 沈阳 110161;
    2.沈阳农业大学 信息与电气工程学院,辽宁 沈阳 110161
  • 收稿日期:2016-03-16 出版日期:2016-10-15 发布日期:2016-11-20
  • 通讯作者: 陈春玲,E-mail:snccl@163.com
  • 作者简介:许童羽(1967—),男,辽宁义县人,博士,教授,主要从事农业航空技术研究。E-mail:yatongmu@163.com
  • 基金资助:
    国家重点研发项目(2016YFD0200600)

Study on northern japonica rice yield model based on canopy date of NDVI

XU Tong-yu1, 2, HONG Xue2, CHEN Chun-ling1, 2, *, ZHOU Yun-cheng1, 2, CAO Ying-li1, 2, YU Feng-hua2, LI Na2   

  1. 1. Agricultural Information Engineering Technology Center in Liaoning Province, Shenyang Agricultural University, Shenyang 110161, China;
    2. College of Information and Electrical Engineering, Shenyang Agricultural University, Shenyang 110161, China
  • Received:2016-03-16 Online:2016-10-15 Published:2016-11-20

摘要: 以沈阳农业大学试验田为研究区域,将无人机遥感技术与人工结合,采集2015年夏季粳稻生长全过程的冠层NDVI数据。首先,利用二元定距变量相关分析的方法对单天和各旬、各月冠层NDVI与产量进行相关性分析;然后,利用线性回归和Square(或Cubic)曲线分别对相关性较好的单天和各旬与产量建模,并对回归模型进行检验,验证模型精度,同时将效果较好的几个模型进行对比分析。结果表明,单独用一个变量建模,Square(或Cubic)曲线模型优于一次线性回归模型,6月中旬和8月上旬的组合模型是估产最理想的模型,其判定系数(R2)为0.771,相对误差(RE)为4.06%,均方根误差(RMSE)为0.474 t·hm-2,精度较高,具有可行性,据此确定北方粳稻最佳估产时间是6月中旬的分蘖盛期和8月上旬的抽穗期。

关键词: 粳稻, NDVI, 相关性, 回归分析

Abstract: In the present study, experimental field in Shenyang Agricultural University was selected as study region, and unmanned aerial vehicle (UAV) remote sensing technology and manual analysis was combined to collect canopy NDVI data of the whole growth of japonica rice in the summer of 2015. Firstly, dual distance variable correlation analysis was applied to reveal the relationships between NDVI data of single day, ten day or each month and yield. Then, the yield and NDVI data which showed good correlations were adopted to build models via linear regression and Square or Cubic curve, and validation test of the constructed regression model and precision comparison were carried out. It was shown that it was better to build model with Square or Cubic curve than linear regress when only one variable was used. The model consisted of data collected in June 11th to 20th and August 1st to 10th was ideal to predict the yield, of which the determination coefficient (R2), relative error(RE), and root mean square error (RMSE) were 0.771, 4.06% and 0.474 t·hm-2, respectively. It was of high precision and feasibility. Thus, it was suggested that the most suitable time for japonica rice yield prediction in Northern China was June 11th to 20th and August 1st to 10th.

Key words: japonica rice, NDVI, correlation, regression analysis

中图分类号: