Operationalizing digital twins in agriculture with machine learning

Summary

Agricultural applications produce data at a fast rate and decision support systems are called to extract insights from them. However, current agricultural decision support systems rely on static models, they are task specific, and they lack automation. This hinders a more advanced type of decision support that modern agricultural systems require. At the same time, the paradigm of digital twins is becoming more prominent in other disciplines. Digital twins seem to be able to overcome the aforementioned limitations and offer benefits that have yet to be realized in agriculture. Despite their success in other disciplines, the potential of digital twins has not been actualized in agriculture. In this thesis we investigate ways to operationalize digital twins, first by enabling them to make predictions when the data are insufficient in amount or temporal resolution, and second by considering their ability to be transferred to diverse conditions.