MULTIVARIATE SYSTEM ANALYSIS OF MULTISPECTRAL IMAGERY

Remote sensing, particularly from a satellite, is potentially an effective and economic means to gather information about natural resources. Routine applications, however, have been fraught with many difficulties and disappointing results. A system analysis of remote sensing as a source of information has revealed the sources of many problems. More importantly, these analyses have shown methods for data preprocessing which should greatly improve results. The application of ERTS data to rangeland management problems was the ultimate goal of this research. Several noise sources (causing random fluctuations of radiance values) were identified as significantly degrading the quality of ERTS data. These included source (scene) noise, atmospheric propagation changes, radiometric errors, electronic system noise, data processing noise, and data analysis errors. Systems analysis suggested several data preprocessing methods to improve data quality. Preprocessing of data included: (1) cleaning of data, (2) ratioing of variables, (3) transformation of variables, and (4) filtering. Multivariate analysis methods were used to evaluate variables, preprocessing results, and classification accuracy. The most significant variables for rangeland analysis were the ratio of ERTS band 7 to bank 5 and band 5. Cleaning of training data greatly reduced classification errors by more accurately determining class signatures. Filtering reduced classification errors for vegetation types from 18.8 percent to 3.6 percent. The two-dimensional moving average filter was particularly effective in reducing atmospheric fluctuations and noise introduced by the satellite sensor system. Canonical transformation of variables eliminated correlation between variables, concentrated between groups variance in the first canonical variates, but did not significantly improve classification results.

• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/00991112
• Corporate Authors:

  American Society of Photogrammetry

  105 North Virginia Avenue
  Falls Church, VA  United States  22046
• Authors:
  • Maxwell, E L
• Publication Date: 1976-9

Media Info

• Features: Figures; References; Tables;
• Pagination: p. 1173-86
• Serial:
  • Photogrammetric Engineering and Remote Sensing
  • Volume: 42
  • Issue Number: 9
  • Publisher: American Society for Photogrammetry and Remote Sensing
  • ISSN: 0099-1112

Subject/Index Terms

• TRT Terms: Imagery; Information processing; Multivariate analysis; Noise; Remote sensing
• Identifier Terms: Earth Resources Technology Satellite
• Uncontrolled Terms: Multispectral imagery
• Subject Areas: Data and Information Technology; Design; Environment; Highways;

Filing Info

• Accession Number: 00142586
• Record Type: Publication
• Files: TRIS
• Created Date: Dec 15 1976 12:00AM