A Satellite-Based Methodology for Harvest Date Detection and Yield Prediction in Sugarcane

An accurate model of yield prediction will benefit many aspects of managing growth and harvest of sugarcane crops. In this study Sentinel-1 and Sentinel-2 time-series were used to automatically detect harvest dates of sugarcane fields in the Far North Queensland of Australia. Harvest date information was further used in combination with weather, soil and elevation data to predict sugarcane yield at different time steps over three consecutive growing seasons using machine learning. Our results suggest that harvest dates could be identified with detection rates of 87% and 91% using Sentinel-1 and Sentinel-2 imagery, respectively. Similarly, sugarcane yield could be predicted using Sentinel-1 and Sentinel-2 satellite imagery in conjunction with other geographical attributes with accuracy of 65% as early as 180 days after the previous harvest.