Use of hyperspectral imagery for mapping grape varieties in the Barossa Valley, South Australia

Abstract

The wine industry is important to Australia's economy. With the advances in remote sensing there has been increasing interest in its potential application for vineyard varietal, condition and health mapping. CASI (Compact Airborne Spectrographic Imager) data was obtained over a vineyard in the Barossa Valley, South Australia in an attempt to discriminate between the grape cultivars (Vitis vinifera) , Cabernet Sauvignon and Shiraz. Statistical analysis of sample spectra from the two varieties in the CASI imagery showed that the significant differences in the visible region. Maximum likelihood classification was employed to map the two grape varieties present on the site. Classification was performed using 12 visible and near infrared CASI bands and repeated using a spectral subset of seven bands shown to be most significant in separating the varieties. Discrimination between Cabernet Sauvignon and Shiraz was successful with 91.5% of vine rows correctly classified. Spectral subsetting did not improve classification and led to under classification of vine pixels.