Use of airborne hyperspectral imagery to determine quality of sorghum crops

Abstract

Remote sensing has shown promise for predicting grain protein content for winter cereals, with satellite image data acquired near flowering, being significantly correlated with grain protein of wheat and barley crops. The use of commercially available satellite or airborne imagery to map grain protein content would allow growers and marketers the ability to evaluate crop performance and provide useful guidelines on harvest logistics and potential segregation to maximise returns. Sorghum is exposed to the air during ripening and this may provide an opportunity to identify grain protein content, a key agronomic indicator of the success or otherwise of nitrogen application to the crop. Our aim was to assess whether airborne hyperspectral imagery could be used to determine grain protein content of sorghum in the northern grains region (Darling Downs) of Australia. The first stage in this process is to determine if variations in the grain crop’s protein content produce detectable variations in image data of the grain crop. The HymapTM sensor was used to acquire a 126 band, 3m pixels data set on 16 April 2004 for several sorghum fields at different growth stages. Availability of concurrent grain protein data restricted the analysis to one of these fields, which was at the end of the grainfilling stage. Grain protein was mapped within four and eight weeks of the HymapTM image by interpolating point samples collected from a near-infrared(NIR) protein sensor mounted on a combine harvester. Preliminary analysis of the image spectral reflectance and field data revealed grain protein content in sorghum was moderately correlated (r=-0.57) with red to near-infrared band (750nm) reflectance. Principal component bands derived from the HymapTM data were weakly correlated (r=0.43) with grain protein. Grain protein content was moderately (r<-0.5) correlated with variations in image spectral reflectance in bands falling between 730nm–1135nm. This information was then used to develop an inverse model, to predict grain protein content from HymapTM image data. A stepwise regression indicated that five bands in the red-edge and NIR regions (750-1150nm) explained the maximum variation in grain protein content (adjusted r2=0.36). The results of this study are exploratory and will be refined in future papers