GEOSCAN Search Results: Fastlink


TitleDeriving percent crop cover over agriculture canopies using hyperspectral remote sensing
AuthorPacheco, A; Bannari, A; Staenz, K; McNairn, H
SourceCanadian Journal of Remote Sensing vol. 34, 2008 p. S110-S123,
Alt SeriesNatural Resources Canada, Contribution Series 20181953
PublisherInforma UK Limited
Mediapaper; on-line; digital
File formatpdf
Subjectsgeophysics; remote sensing
ProgramCanada Centre for Remote Sensing Divsion
Released2014 06 02
AbstractThe objective of this study is to investigate the potential of hyperspectral remote sensing for providing green percent crop cover information over agricultural canopies for use in precision farming. Ground measurements and airborne hyperspectral Probe-1 data were acquired over two agricultural sites in July 1999 near Clinton, Ontario, Canada, and in July 2000 near Indian Head, Saskatchewan, Canada. A manual endmember selection and extraction approach was used to unmix the reflectance cubes. Linear constrained and weakly constrained unmixing was conducted to determine crop endmember fractions. Results indicate that the two spectral unmixing algorithms were equally successful. If prior knowledge of the fields exists, a manual endmember selection technique is a suitable approach for the selection of endmembers in an image scene. Correlations between ground data and Probe-1 crop fractions showed very good results (root mean square error (RMSE) of ±11.06% and ±11.20% and index of agreement (D) of 0.91 and 0.90 for constrained and weakly constrained unmixing, respectively) when the endmembers were adjusted for their impurity. This highlighted the importance of finding pure or the purest pixels for each endmember in the image scene. The use of spectral unmixing to quantify within-crop and within-field percent crop cover showed mixed success. Canola and wheat crops (RMSE of ±10.67% and ±11.77%; D of 0.94 and 0.85) revealed better correlations than the bean and corn crops (RMSE of ±10.83% and ±12.46%; D of 0.64 and 0.59). The somewhat poor results for pea crops (RMSE of ±9.70%; D of 0.58) are possibly due to the ground vertical photograph classification process where it was difficult to distinguish pea stems and residue components.

Date modified: