Using Multispectral Imagery and Linear Spectral Unmixing Techniques for Estimating Crop Yield Variability

Vegetation indices derived from multispectral imagery are commonly used to extract crop growth and yield information. Spectral unmixing techniques provide an alternative approach to quantifying crop canopy abundance within each image pixel and have the potential for mapping crop yield variability. The objective of this study was to apply linear spectral unmixing techniques to airborne multispectral imagery for estimating grain sorghum yield variability. Five time-sequential airborne multispectral images and yield monitor data collected from a grain sorghum field were used for this study. Both unconstrained and constrained linear spectral unmixing models were applied to the images to generate crop plant and soil abundances for each image and for all 26 multi-image combinations of the five images. Yield was related to unconstrained and constrained plant and soil abundances as well as to the normalized difference vegetation index (NDVI) and the green NDVI (GNDVI). Results showed that unconstrained plant abundance had better correlations with yield than NDVI for all five images, but GNDVI had better correlations with yield for the first three images. Unconstrained plant abundance derived from the fourth image provided the best overall correlation with yield (r = 0.88). Moreover, multi-image combinations generally improved the correlations with yield over single images, and the best three-image combination resulted in the highest overall correlation (r = 0.90) between yield and unconstrained plant abundance. These results indicate that linear spectral unmixing techniques can be a useful tool for quantifying crop canopy abundance and mapping crop yield.