中国农业科学 ›› 2018, Vol. 51 ›› Issue (4): 675-687.doi: 10.3864/j.issn.0578-1752.2018.04.007

• 耕作栽培·生理生化·农业信息技术 • 上一篇    下一篇

基于遥感识别误差校正面积的农作物种植面积抽样高效分层指标研究——以冬小麦为例

杨珺雯1,3,张锦水1,3,潘耀忠2,3,孙佩军2,3,朱爽4

 
  

  1. 1北京师范大学地理科学学部地表过程与资源生态国家重点实验室,北京 100875;2北京师范大学地理科学学部遥感科学国家重点实验室,北京 1008753北京师范大学地理科学学部遥感科学与工程研究院,北京 100875;4北京工业职业技术学院,北京 100042
  • 收稿日期:2017-06-12 出版日期:2018-02-16 发布日期:2018-02-16
  • 通讯作者: 张锦水,E-mail:zhangjs@bnu.edu.cn
  • 作者简介:杨珺雯,E-mail:201211191008@mail.bnu.edu.cn
  • 基金资助:
    国家重点研发计划“粮食丰产增效科技创新专项”子课题(2017YFD0300402-6)、高分辨率对地观测系统重大专项(民用部分)

An Efficient Hierarchical Indicator Based on the Correction Area of Remote Sensing Identification Error for Planting Acreage Sampling——A Case Study of Winter Wheat

YANG JunWen1,3, ZHANG JinShui1,3, PAN YaoZhong2,3, SUN PeiJun2,3, ZHU Shuang4   

  1. 1State Key Laboratory of Earth Surface Processes and Resource Ecology, Faculty of Geographical Sciences, Beijing Normal University, Beijing 100875; 2State Key Laboratory of Remote Sensing Science, Faculty of Geographical Sciences, Beijing Normal University, Beijing 100875; 3Institute of Remote Sensing Science and Engineering, Faculty of Geographical Sciences, Beijing Normal University, Beijing 100875; 4Beijing Polytechnic College, Beijing 100042
  • Received:2017-06-12 Online:2018-02-16 Published:2018-02-16

摘要: 【目的】在遥感与空间抽样相结合进行农作物面积调查中,由于传统的基于面积规模的分层指标设计中缺失对遥感识别结果分类误差的表达,一定程度上影响抽样效率,因此提出基于遥感分类误差校正面积的分层标志——误差校正面积,以期改进农作物种植面积抽样调查效率。【方法】选取北京市通州、大兴区为研究区域,以冬小麦为例,选择16 m分辨率GF-1号影像(获取时间2015年4月4日)为遥感数据源,进行抽样方案的设计。设计与计算高效分层指标,先从像元尺度判断像元相应的错入、错出方向并计算其对应的误差面积,再在抽样单元尺度上统计所有像元的误差面积,将其用于面积规模的校正,校正后结果即为所提出的分层抽样指标——误差校正面积(Scorrect);构建边长为90—300 m的规则正方形格网为抽样框,并完成设置分层层数、确定分层界限方法、样本量分配方式、总体估计方式等空间抽样方案设计。基于设计的抽样方案进行试验,进行研究区冬小麦的区域总量面积反推。以误差校正面积指标和传统分层指标——面积规模为分层指标,进行多次种植面积抽样推断后进行指标有效性分析和精度评价,通过对相关性、典型区域分类错误像元误差分布、总体方差、平均相对误差 、CV值等方面的对比分析,验证所提出指标的可行性与优势。【结果】(1)通过结合原始影像、目标真值分布、遥感分类结果图、分类错误像元误差分布图的对比分析,从像元尺度验证了该指标能校正分类错误像元,从而改善分类结果;在试验抽样框下,误差校正面积的相关系数相较于面积规模略有提高,且数值大于0.7,可保证其与真值较高且稳定的相关性。验证了该指标作为分层指标的有效性。(2)在试验抽样框下,使用误差校正面积作为分层指标进行多次外推面积得到的总体方差在1.70×1013—2.41×1013,面积规模的总体方差为2.05×1013—3.11×1013,误差校正面积在推断稳定性方面高于面积规模;采用误差校正面积作为分层指标得到的 为4.21%—5.00%,面积规模的 为4.87%—5.98%,误差校正面积指标能稳定提高近1%的精度;选择误差校正面积指标作为分层指标进行抽样估算结果的CV值在试验抽样框下始终低于面积规模的推断结果,能稳定减少近0.8%。因此误差校正面积指标在与目标真值相关性、抽样精度、推断稳定性等方面均优于传统面积规模分层指标。【结论】误差校正面积指标可在一定程度上提高种植面积抽样调查精度,保证推断的稳定性,验证了遥感识别误差校正面积指标作为分层标志的有效性,能够提高抽样效率,其相较面积规模指标更具有优势。

关键词: 冬小麦, 种植面积, 分层抽样, 分层指标, 分类误差, 面积规模

Abstract: 【Objective】Remote sensing has a strong superiority in crop planting area acquisition. It is critical to design an efficient hierarchical indicator in a stratified sampling investigation. Traditional method neglects the classifying error of the classification results, which reduces the sampling efficiency. Therefore, a new hierarchical indicator (Scorrect) that can correct the error area is presented in this paper, which is in favor of raising the efficiency of the sampling survey. 【Method】The yield of winter wheat in Tongzhou and Daxing district of Beijing was as a study object. The data source was GF-1 imagery (resolution: 16 m, phase: 2015-04-04) obtained in winter wheat growing season. This paper designed the Scorrect which was based on an algorithm involved the error of omission and the error of commission at pixel scale. This indicator summed all pixels’ correct scale in a sampling unit to correct the area scale indicator (S). A certain plan was devised for Multiple sampling by using Scorrect and the area scale indicator (S). This plan constructed regular square grids (size: 90-300 m) as the sampling frames, set the number of layers, determined the hierarchical boundary method and the allocation method for sample capacity, etc. In order to testify the validity of Scorrect, this paper analyzed the correlation between this indicator and true value, estimated the total winter wheat area by multiple sampling, after that, compared Scorrect with S in sampling variance, average relative error ( ), CV value, etc. The typical corrected regions for pixel with error were selected to compare and analyze for verifying the effectiveness at pixel scale. 【Result】 (1) By comparing and analyzing the typical corrected regions, the error distribution could accord with the actual circumstance for both the error of omission and the error of commission. This indicator could correct error pixel. With the sizes of the sampling unit, Scorrect was better than S in the correlation, and the value was greater than 0.7, which can ensure a high and stable correlation with the truth value. The validity of Scorrect had been verified by experiments. (2) Scorrect was better than S in the stability, accuracy and CV value. Its total variance was 1.70×1013-2.41×1013, and the total variance obtained by using S was 2.05×1013-3.11×1013. Its  was 4.21%-5.00%. The  obtained by using S was 4.87%-5.98%. Therefore, nearly 1% could be enhanced stability in accuracy. Besides, Scorrect could reduce CV value nearly 0.8%. 【Conclusion】The effectiveness and advantages of Scorrect had been proved by the experimental results. The indicator can improve the accuracy of sampling and the stability of inference in a certain extent.

Key words: winter wheat, acreage estimate, spatial stratified sampling, hierarchical indicator, classification error, area scale