Submitted to: International Conference on Precision Agriculture Abstracts & Proceedings
Publication Type: Proceedings
Publication Acceptance Date: August 25, 2006
Publication Date: November 26, 2006
Citation: Yang, C., Everitt, J.H., Bradford, J.M. 2006. Using hyperspectral imagery and spectral unmixing techniques for mapping grain sorghum yield variability. International Conference on Precision Agriculture Abstracts & Proceedings. CDROM.
Interpretive Summary: Spectral unmixing is an image processing technique used to quantify canopy abundance within each pixel and has the potential for mapping crop yield variability. This study applied linear spectral unmixing to hyperspectral imagery to generate crop plant and soil abundance images. These abundance images were significantly related to grain yield monitor data and had better correlations with yield than the majority of the normalized difference vegetation indices derived from the hyperspectral imagery. These results indicate that linear spectral unmixing techniques can be used alone or in conjunction with vegetation indices for quantifying crop canopy cover and mapping yield variability.
Vegetation indices derived from remotely sensed imagery are commonly used to estimate crop yields. Spectral unmixing techniques provide an alternative approach to quantifying crop canopy abundance within each pixel and have the potential for mapping crop yield variability. The objective of this study was to apply linear spectral unmixing techniques to airborne hyperspectral imagery for estimating grain sorghum yield variability. Airborne hyperspectral imagery and yield monitor data collected from two grain sorghum fields were used for this study. Both unconstrained and constrained linear spectral unmixing models with 28 plant and soil spectrum pairs as endmembers were applied to the hyperspectral imagery to generate crop plant and soil abundance images. Statistical analysis showed that yield was significantly related to plant and soil abundances. Moreover, unconstrained plant abundance provided essentially the same r-values with yield among the 28 endmember pairs. Although unconstrained soil abundance and constrained plant/soil abundance provided better r-values for some endmember pairs, the r-values were sensitive to the choice of endmember spectra. For comparison, all 5,151 possible narrow-band normalized difference vegetation indices (NDVIs) were calculated from the 102-band images and related to yield. Results showed that the best plant and soil abundances provided higher r-values than 96.3% and 99.9% of all the NDVIs for fields 1 and 2, respectively, though the best NDVIs had higher correlations than the best abundances. Like NDVIs, spectral unmixing is another tool to reduce a hyperspectral image to a single layer image. Since plant abundance can better represent yield variability than most narrow-band NDVIs, it can be used as a relative yield map especially when yield data are not available. These results indicate that spectral unmixing applied to hyperspectral imagery can be a useful tool for mapping crop yield variability.