AIrborne Hyperspectral Imagery and Yield Monitor Data for Estimating Grain Sorghum Yield Variability

As hyperspectral imagery is becoming more available, it is necessary to evaluate its potential for crop monitoringand precision agriculture applications. In this study, airborne hyperspectral imagery was examined for estimating grain sorghum(Sorghum bicolor (L.) Moench) yield variability as compared with yield monitor data. Hyperspectral images containing128 spectral bands in the visible to near-infrared region were acquired using an airborne hyperspectral imaging system fromtwo grain sorghum fields during the 2000 growing season, and yield data were collected from the two fields using a yieldmonitor. The raw hyperspectral images were geometrically corrected, georeferenced, and resampled to 1 m resolution, andthe raw digital numbers were converted to reflectance. The calibrated image data were then aggregated into images with acell size of 9 m, close to the combines effective cutting width. Correlation analysis showed that grain yield was significantlyrelated to the image data for all the bands except for a few in the transitional range from the red to the near-infrared region.Principal component analysis indicated that the first few principal components of the hyperspectral images accounted for99% of variability in the image data. Stepwise regression analysis based on the first ten principal components revealed thatfive significant components explained 68% and 80% of the variability in grain yield for fields 1 and 2, respectively. Stepwiseregression analysis performed directly on the yield and hyperspectral data identified four optimum bands for field 1 and sevenfor field 2. The best four-band combination accounted for 69% of the variability in yield for field 1, while the best seven-bandcombination explained 82% of the variability for field 2. These results demonstrate that hyperspectral imagery can be a usefulremote sensing data source for crop yield estimation.