Yield prediction with machine learning algorithms and satellite images

• Published in: Journal of the science of food and agriculture Vol. 101; no. 3; pp. 891 - 896
• Main Author: Sharifi, Alireza
• Format: Journal Article
• Language: English
• Published: Chichester, UK John Wiley & Sons, Ltd 01.02.2021; John Wiley and Sons, Limited
• Subjects: Agricultural production; Agricultural products; Agriculture; Algorithms; Barley; Crop yield; Food security; Gaussian process; Gaussian process regression; Harvesting; Hordeum - chemistry; Hordeum - growth & development; Iran; Learning algorithms; Machine Learning; Meteorological data; normal distribution; Predictions; Remote sensing; Remote Sensing Technology - methods; Satellite imagery; Satellites; Seasons; yield forecasting; yield prediction; Iran; Gaussian process regression; remote sensing; barley; yield prediction; machine learning
• ISSN: 0022-5142; 1097-0010; 1097-0010
• DOI: 10.1002/jsfa.10696

BACKGROUND Barley is one of the strategic agricultural products available in the world, and yield prediction is important for ensuring food security. One way of estimating a product is to use remote sensing data in conjunction with field data and meteorological data. One of the main issues surrounding this comprises the use of machine learning techniques to create a multi‐resource data‐based estimation model. Many studies have been conducted on barley yield prediction from planting to harvest. Still, the effect of different time intervals on yield prediction has not been investigated. Furthermore, the effect of different periods on yield prediction has not been investigated. RESULTS In the present study, the whole growth period was divided into three parts. Using one of the major barley production areas in Iran, the performance of the proposed model was evaluated. In the first step, a model for integrating field data, remote sensing data and meteorological data was prepared. The results obtained show that, among the four machine learning methods implemented, the gaussian process regression algorithm performed best and estimated yield with r2 = 0.84, root mean square error = 737 kg ha−1 and mean absolute = 650 kg ha−1, 1 month before harvest. CONCLUSION It was found that the estimation results change depending on different agricultural zones and temporal training settings. The findings of the present study provide a powerful potential tool for the yield prediction of barley using multi‐source data and machine learning. © 2020 Society of Chemical Industry