Developing sugar cane yield prediction algorithms from satellite imagery

Abstract

THE RESEARCH PRESENTED in this paper discusses the accuracies of remote sensing and GIS as yield prediction tools at both a regional and crop scale over three Australian cane growing regions; Bundaberg, Burdekin and the Herbert. For the Burdekin region, the prediction of total tonnes of cane per hectare (TCH) produced from 4999 crops during the 2011 season was 99% using an algorithm derived from 2010 imagery (green normalised difference vegetation index) and average yield (TCH) data extracted from 4573 crops. Similar accuracies were produced for the Bundaberg region during 2010 (95.5% from 3544 blocks) and 2011 (91.5% for 3824 crops) using a Bundaberg specific
algorithm derived from 2008–2010 imagery and yield data. The Bundaberg algorithm was also accurate in predicting yield at specific in-crop locations (91.5% accuracy;
SE = 0.028).