European Journal of Sustainable Agroecosystems

Background: The digitization of agriculture has resulted in the creation of large, varied datasets that include soil features, climate conditions, plant growth parameters and data about farm management. However, traditional ML approaches require these data to be compiled, raising serious questions regarding privacy, property rights, and legislation for individual farms and organizations. 
Goal: The objective of this study is to review contemporary literature on Federated Learning (FL) as a decentralized concept that will allow for secure agricultural decision support systems without moving relevant information to a common database. 
Framework: The literature review is based on peer-reviewed articles published mainly during the period from 2017 to 2026. and connects them to the larger context of precision agriculture driven by the Internet of Things and edge computing. 
Findings: Based on the available literature, FML architectures find applications in crop yield forecasting, identifying diseases and pests in plants, irrigation planning, nitrogen management, and assessing climate risks, with techniques providing privacy protection such as differential privacy, secure aggregation, and trust protocols based on blockchain technology acting as supplementary measures. Comparative studies reviewed in this paper generally claim that FML has similar performance in prediction to centralized models while being able to reduce data exposure risk significantly and lessening communication costs in a number of cases. Among the challenges found in the literature are the problem of statistical heterogeneity (the issue of data being non-identically and non-independent), low connectivity in the countryside, difficulties with understanding aggregated models, and absence of unified standards governing the use of data in agriculture. 
Significance: The review contributes to the existing body of knowledge by providing the audience with a structured synthesis of scientific literature on the issue of how FML can help achieve trustworthy digital agriculture. 
Conclusion: Federated machine learning shows scientific promise that is reflected in extensive research on the issue.