Abstract: Eco-unmanned farms integrate ecological principles, information science, and agricultural engineering to achieve precise, intelligent, and sustainable production through unmanned operations. However, these farms are inherently complex systems encompassing heterogeneous elements such as soil, crops, meteorology, hydrology, and machinery. The dynamic interactions among these components exhibit strong nonlinearity and spatiotemporal variability, posing significant challenges to traditional management approaches that rely on isolated sensing and static models. Digital twin technology offers a transformative solution by constructing high-fidelity virtual replicas that map, simulate, and interact with physical farms in real time. This paper presents a comprehensive review of digital twin applications in eco-unmanned farms, covering frameworks, key technologies, core functionalities, closed-loop mechanisms, challenges, and future directions. The study first elaborates on a three-layer conceptual framework comprising the physical entity layer, the virtual model layer, and the data interaction and service layer. The physical entity layer encompasses intelligent agricultural machinery (e.g., unmanned tractors, plant protection UAVs, and spraying robots), an air-space-ground integrated sensing network (including satellite remote sensing, UAV remote sensing, and ground-based IoT sensors), and farmland infrastructure. This layer serves as both the data source and the execution terminal for control commands. The virtual model layer establishes a multi-scale model library that integrates mechanistic models (e.g., crop growth models and soil water-salt transport models), data-driven models (e.g., convolutional neural networks and machine learning regression models), and hybrid models that combine physical principles with artificial intelligence. The data interaction and service layer enables seamless data flow and closed-loop decision-making through edge computing, cloud computing, standardized interfaces, and "What-If" scenario simulation capabilities. Building upon this framework, the paper elaborates on the construction methodologies for digital twins in eco-unmanned farms. The air-space-ground integrated data platform integrates multi-source sensing data from satellites, UAVs, and ground sensors to establish a three-dimensional perception network that covers the entire farm at multiple scales. The core model library is developed by integrating mechanistic models with physical interpretability and data-driven models with high predictive accuracy, forming a hybrid modeling approach that balances precision and computational efficiency. The core functionalities of digital twins are systematically analyzed from three perspectives. First, "What-If" ecological simulation enables virtual experimentation of different management strategies, such as irrigation scheduling, variable-rate fertilization, and pest control, providing predictive decision support and reducing the cost and risk of field trials. Second, visualization and interaction technologies, including 3D reconstruction based on UAV LiDAR and multi-view photography, neural radiance fields, virtual dashboards, and VR/AR interfaces, enhance user immersion and operational efficiency. Third, operational optimization and collaborative management achieve global scheduling of resources, equipment, and workflows through multi-objective optimization, predictive maintenance, and energy-carbon footprint management, elevating digital twin applications from single-point optimization to farm-wide collaborative control. The closed-loop application mechanisms are further examined through the lens of bidirectional driving and data feedback. The "virtual-to-physical" driving mechanism translates optimized decisions generated in the virtual space into precise execution commands, such as variable-rate prescription maps and path planning instructions for unmanned machinery. Conversely, the "physical-to-virtual" driving mechanism feeds real-time monitoring data back to the virtual model, enabling dynamic state updates and parameter calibration. Data feedback and model optimization mechanisms, including data assimilation algorithms, online learning systems, and adaptive control strategies, empower digital twins with continuous self-evolution capabilities. Despite significant progress, several key challenges hinder the widespread adoption of digital twins in eco-unmanned farms. These include the trade-off between model accuracy and computational complexity, the technical bottleneck of multi-source heterogeneous data fusion, the dilemma of ensuring real-time performance and reliability in complex field environments, and the absence of full-lifecycle model evolution mechanisms. Addressing these challenges requires future research to focus on four directions. By overcoming these challenges and pursuing these directions, digital twin technology can fully realize its potential to revolutionize the intelligent management and sustainable development of eco-unmanned farms.