The interviews were conducted based on a semi-structured questionnaire comprising questions in three categories: i) use case definition, including the problem context, core idea, objective, development status, etc.; ii) use case mapping, including main business processes targeted, objects addressed, main actors using the envisaged system; and iii) use case information architecture, including the main functionalities/services to be provided to end-users, non-functional requirements, technology components envisioned, reusability, privacy/security, standards usage and available documentation. Subsequently, the researchers designed case-specific control model and technical layer models for the use cases by using the conceptual framework as designed in phase 2. The resulting case specific models were reviewed by the use case leader and iteratively refined by the research analyst. Finally, generic implications were abstracted from the case study findings and incorporated in the framework. The remainder of the paper introduces the results following the research steps as described above.The principles behind the Digital Twin vision origin from the Product Lifecycle Management domain . From this perspective, there is a strong need to integrate all product-related information in a comprehensive product management system that can be accessed by any user in any stage of the Product Life Cycle, for example for through-life performance information, design optimization and manufacturing system improvement . It was proposed to use a digital counterpart of each physical product as a central means to manage product data along the product life cycle. In the beginning of the century, Michael Grieves argued in a presentation to industry that a digital informational construct about a physical system could be created as an entity on its own.

This virtual system contains all information about the physical system and it is linked with that physical system through the entire lifecycle of the system . NASA introduced the concept Digital Twin for this idea , dutch buckets which was an ultra-high fidelity simulation of the space vehicle that would allow the engineers on earth to mirror the precise and actual conditions of the real vehicle during the mission . In essence, Digital Twins are virtual, digital equivalents of physical objects . They are real time and remotely connected to real objects and provide rich representations of these objects and its context. Such Digital Twins go beyond static product designs, like CAD models, but comprise dynamic behaviour . This dynamic nature of Digital Twins may include the representation of current behaviour of real-life objects, but also the simulation or prediction of future behaviour and the recollection of historical behaviour . A Digital Twin can already be created in the design phase of an object’s lifecycle, enhancing the creative phase of inventing new products and elaborating it into a detailed product model . In this stage, a Digital Twin allows early and efficient assessment of consequences of design decisions on the quality and function of products reducing the need to develop costly physical prototypes . After the design phase, a physical stage is entered, in which the Digital Twin comes into existence. A physical object is produced based on the designed Digital Twin, which is updated in case of any deviations. During operational usage, the current and historical state and conditions of a physical product are monitored by using sensors and AutoID devices. Moreover, the Digital Twin can be used to remotely control an object by using actuators. Finally, the disposal phase takes place, in which the physical object is disposed, but the conceptual object may remain for some period e.g. for traceability, compliance and learning. While originating from Product Lifecycle literature, a key technology for realizing Digital Twins is the Internet of Things.

The interaction between real/physical and digital/virtual objects is an essential concept behind the Internet of Things. In the IoT, physical entities have digital counterparts; things themselves become context aware and they can sense, communicate, act, interact with their digital counterparts and others, exchange data, information and knowledge . These counterparts are twins of the physical objects and can be linked to and synchronized with the physical object throughout their lifecycle . The Internet acts as a storage and communication infrastructure that holds a virtual representation of things linking relevant information with the object . As such, Digital Twins serve as central hubs of object information, which combine and update data continuously from a wide range of sources, as illustrated in Fig. 3 . The emergence of Digital Twins in literature has resulted in various definitions both from a Product Life Cycle and an Internet of Things perspective and is discussed below. In the IoT literature, the digital representation of physical objects using sensing technologies is highlighted. In the PLC literature, the emphasis is on mirroring real-life objects across their lifecycle, including simulation of expected object behaviour. In this paper we combine both perspectives and use the following definition: “A Digital Twin is a dynamic representation of a real-life object that mirrors its states and behaviour across its lifecycle and that can be used to monitor, analyze and simulate current and future states of and interventions on these objects, using data integration, artificial intelligence and machine learning.” Farming is a highly complex and dynamic domain. Production processes are inherently dynamic because they depend on natural conditions, such as weather, diseases, soil conditions, seasonability and climate . Moreover, farmers have to deal with critical demands from consumers and society concerning food security, food safety, sustainability and health. As a consequence, farms should not only be very efficient, but also have to meet high quality and environmental standards and should adapt to changing market conditions. This imposes high requirements on the managerial tasks of farmers . They constantly have to reassess production strategies and to reschedule planned activities based on timely monitoring of farm operations in order to achieve their goals. As argued in the introduction section, Digital Twins can significantly enhance the needed control capabilities by enabling the decoupling of physical and information aspects of farm management . However, implementing Digital Twins in farm management is a challenging task for three reasons .

First of all, the highly dynamic production system in agriculture poses requirements that go beyond many other sectors concerning the capabilities of Digital Twins to mirror dynamic behaviour. In such a dynamic environment, it is really challenging to get seamless access to object data while ensuring the integrity of data and respecting usage rights, safety and security. Furthermore, real-time synchronization can be complicated in rural areas, which often have limited coverage and bandwidth. Second, agricultural products are living objects that inherently are diverse and are characterized by complex behaviour. Moreover, farms don’t have one Digital Twin of concern for smart farming, but they are composed of a large variety of interrelated objects . Main objects are i) inputs including seeds, feed, fertilizers or pesticides, ii) throughputs including objects in production and resources including fields, stables, machinery and personnel, and iii) agricultural output including harvested crops, animals ready to be slaughtered, etc. Digital Twins of a fine granularity level, e.g. up to individual plants or animals, would add more value, but are also more difficult to implement, which results in higher costs. In case of a fine granularity, a key challenge is to manage the interdependences between Digital Twins at different granularity levels. Third, farms are part of a dynamic network and share data with many stakeholders including customers, input suppliers, farmer cooperatives, advisors, contractors, and certification and inspection organizations . These stakeholders may also have access to the farmer’s Digital Twins, but limited to the information that they need. This implies that there must be interoperable solutions for providing external access to specific views on Digital Twins in a secure and trusted way. Vice versa, external stakeholders can enrich farm Digital Twins with a wealth of archives such as historical and forecasted meteorological data, satellite data, soil, water and air analyses, etc. There should be proper mechanisms in place to dynamically integrate these data in farm Digital Twins. Digital Twins can be seen as a new phase in smart farming. It is building upon existing technologies especially for precision farming, Internet of Things and simulation. As a consequence, there are multiple applications in the agricultural domain, although often not framed as Digital Twins. However, most of these applications are still rather basic forms of Digital Twins, focusing for example on digital representation in a cloud dashboard. More advanced applications, including e.g. predictive and prescriptive capabilities across the lifecycle, are still in an early stage of development.

In our literature review, we only found a few explorative studies and some case studies that frame an IoT-based system as a Digital Twin, without a detailed motivation or definition of the concept. These papers, which are discussed below, are all congress papers, except one book chapter. To the best of our knowledge, Verdouw and Kruize were the first who explored the application of Digital Twins in farm management. The paper considers Digital Twins from an Internet of Things perspective,grow bucket in which physical objects have virtual, digital equivalents that are real-time and remotely connected. Illustrated by six cases of the FIWARE Accelerator program, the paper shows that Digital Twins are already implicitly used in smart farming, but existing applications mostly focus on basic monitoring capabilities. The authors argue that these capabilities establish a basis for optimization, simulation and decision support based on on-line Digital Twins. Jo et al. conducted a feasibility study on using Digital Twins in pig farms to improve animal welfare. They introduce thre so-called Digital Twin platforms and proposed a design smart livestock farming system using such a Digital Twin platform. The paper states that a Digital Twin is the digital replica of the real world, but it does not further elaborate on the concept, neither it describes how the designed system supports Digital Twins. Monteiro et al. present a technical IoT-based model and a prototype to implement Digital Twins in vertical farming. Digital Twins are defined as digital mirrors of physical objects. The Digital Twin designed in the paper is envisioned to support the lifecycle of planning, operation, monitoring, and optimization of vertical farms.Alves et al. developed an IoT-based prototype to sense field conditions including soil moisture, air temperature and humidity, and to visualize this information in a dashboard. This prototype is called a Digital Twin, in which data flows automatically between a physical and a digital object. The authors argue that a Digital Twin enables farmers to make better decisions and to decrease the environmental impact in water, land and soil resources. In the research of Kampker et al. , a Digital Twin is an artificial potato, which is planted in the field and harvested just as real potatoes. The ‘digital’ potato is equipped with sensors that measure its treatment, especially shocks, blows and rotation speed, etc. This information is used to adjust the settings of the harvesting machine, thus minimizing damage to the potatoes. Subsequently, the digital potato is synchronized with a cloud platform to enable smart services like potato price and field revenue estimations . Linz et al. applied Digital Twins in the development of field robots, for example for phenotyping and crop treatments in vineyards. They simulate the autonomous behaviour of robots in a 3D environment using real-time data and mirror the simulated Digital Twin to operate the real-life robot. This results in shorter lead times of development, better evaluation of sensor behaviour and reducing the needed field experiments to evaluate phenotypes or test the effects of crop treatments. Skobelev et al. propose a multi-agent approach to development of Digital Twins of plants. A plant Digital Twin is defined as “a computer model that imitates its life cycle and synchronizes with the living plant using examinations by agronomists and data on environmental conditions ”. Finally, Sreedevi and Santosh Kumar argue that there are relatively few studies about Digital Twins in agriculture compared to other domains. Furthermore, they discuss the potential contribution of Digital Twins in hydroponics farming, especially for predicting probable failures and optimizing the whole farming system, including the management of nutrients, pH values, pathogens and weeds.