The models of the use case ‘Added value weeding data’ are described in more details as an illustrative example. Table 4 provides an overview of the applied control models. Point of departure in all use cases is a particular farm crop or animal, i.e. potato, tomato, lettuce, cow or pig. These objects are nested in high-level objects, such as fields, greenhouses and stables. The plant use cases all also predict the expected output of the farming process, i.e. potatoes, tomatoes or lettuce to be harvested. Two use cases also virtualize equipment used, i.e. a weeding machine or truck. All use cases analysed combine multiple types of Digital Twins, starting with monitoring the actual state of objects and then predicting future states e.g. expected yields or animal health. Most of the use cases also include intelligence to advice interventions. The crop farming use cases also process these advices into prescriptive Digital Twins, e.g. by defining task maps. Most use cases focus on the usage phase of a lifecycle and do not include imaginary or recollection Digital Twins. Only the use case ‘Added value weeding data’ applies a recollection Digital Twin for optimisation of machine settings based on historical data about machine behaviour. None of the use cases have implemented yet autonomous Digital Twins. Table 5 lists the main technical components that are used to implement the layers of the Digital Twin technical architecture. It shows that in the Device Layer all use cases apply domain-specific sensors and three use cases also use specific actuators. Most technologies for technical communication are based on standardized protocols of both conventional technology such as wired networks and recent wireless IoT networks such as LoraNet. The use case ‘Happy cow’ has chosen to apply a custom-built network consisting of distributed access points that enable communication up to several kilometers. Also in the IoT Service layer a combination of technologies is used.
Process-based orchestration of services is not yet addressed. Only the first use case ‘Within-field management zoning’ includes some Modelling Services. In the Digital Twin Management Layer all use cases provide services that combine and store data from diverse data sources and represent harmonized virtual entities. These services also include intelligence for simulation or decision support dependent on the supported control functions . In the application layer,hydroponic bucket all use cases provide dedicated dashboards for the interaction with users and three of the use cases also integrate with existing farm management systems . Finally, all use cases comprise some generic technical functions for the service organisation, security and management. When growing organic vegetables, weeding is one of the most important and frequent activities to control both the quality of the field and its produce . In recent years, automated intra-row weeding machines have entered the market, enhancing the weeding process significantly. The most advanced weeding machines use machine vision applications to distinguish crops from weeds. These camera data can not only be used for automated control of the weeding task, but also as a valuable information source for farm management. This use case uses these location-specific camera data of a weeding machine as a main data source to provide actual insights into the number of lettuce heads growing on the field, the plants’ growth status, weed prevalence and best harvesting moment. As such it creates Digital Twins of a field, plants and weeds to monitor crop growth and to predict the crop weight and size of lettuce. The applied control model of the use case ‘Added value weeding data’ is shown in Fig. 10. The main farming processes are sourcing and planting young lettuce plants, producing lettuce in the field, harvesting lettuce which is ready for consumption and delivering it to the market. The main physical objects involved are planting machines and young plants, fields containing weeds and growing lettuce, weeding machines, harvesting machines and harvested lettuces. The Digital Twins of this use case are used for monitoring weed pressure and crop growth, controlling the weeds to be removed and predicting the optimal moment of harvesting. To do so, the sensor function uses processed camera images to calculate crop parameters such as size.
Furthermore, crop growth sensing adds weather data and field properties, including temperature, relative humidity, wind speed and direction, solar radiation and soil moisture . The data acquisition function also includes external weather data. These data are then transformed into Digital Twins. The virtualisation in this use focusses on the field, which implies that the main Digital Twin is a high-precision and actual heat map of a field. A field map comprises weed density and the number and size of crops , and the expected final weight and crop size of the lettuce . Planting seedlings are excluded. The Digital Twins of the individual lettuce crops and weeds are used by farmers during the weeding activity and afterwards the calculated parameters of every plant in the field are also available remotely. Furthermore, Digital Twins of the weeding machine is used to optimize machine settings afterwards . The discriminator function uses the Digital Twins of weeds and growing lettuces to monitor weed pressure and crop growth, i.e. crop size and crop distance. The decision maker function translates the weed pressure into a planning of the weeding activities.The user sets a target value for crop weight and then the optimal harvest moment is determined. Based on this information the optimal moment of harvesting is determined and the harvesting is planned. For lettuce, growers get paid by lettuce head in the right weight class. Finally, the effector function executes the planned weeding task. The weeds are automatically removed, controlled by the actuators in the weeding machine that apply machine instructions based on Digital Twin of the weeds. Because of the high-precision weed density maps, fields can be weeded partially, only where needed. Also the planned harvesting activities are executed by harvesting machines but they do not use customised machine instructions. The control cycle partly takes place on-site within the weeding machine. Camera data are directly processed into local Digital Twins that distinguish crops and weeds. These Digital Twins are then instantly translated to actuator instructions and the weeds are removed without human involvement. However, all other control activities are done remotely by farmers who interact with the Digital Twins via cloud-based systems.
The next section elaborates on how this is technically implemented. Digital Twins can be seen as a new phase in smart farming. Using Digital Twins as central means for farm management enables the decoupling of physical flows from its planning and control. As a consequence, farmers can manage operations remotely based on real-time digital information instead of having to rely on direct observation and manual tasks on-site. This allows them to act immediately in case of deviations and to simulate the effect of interventions based on real-life data. The main contribution of the paper is that it has proposed a conceptual framework for designing and implementing Digital Twins for smart farming. The framework builds on an analysis of literature and a clarification of the concept of Digital Twins, which is still developing. An important novelty of the framework is that it adds a typology of Digital Twins based on the life cycle phases of the objects being virtualised. Depending on the perspective, the emphasis is currently often on monitoring or predictive Digital Twins. However, Digital Twins can already be created in the design phase of a life cycle and support the creation of its physical, real-life sibling. During operational usage, Digital Twins can not only be used to monitor and simulate the effects of interventions, but also to remotely control an object by using actuators. Finally, Digital Twins are also very valuable after disposal of a physical object e.g. for traceability, compliance and learning. So far, these distinct Digital Twin types are not explicitly addressed in the literature, which results in conceptual confusion. This paper has contributed to avoid this by introducing a typology and by defining the distinct control capabilities of each type in a control model. The case studies show that there are already applications in the agricultural domain that are not framed as Digital Twins. This is not surprising, since Digital Twins are building upon existing technologies especially for precision farming, internet of things and simulation. However, especially more advanced applications, including e.g. predictive and prescriptive capabilities across the lifecycle, are still in an early stage of development. The designed framework was useful to explicitly describe and analyze how Digital Twins are used in practice. As such, it has provided a new perspective on the cases that originally focused on the innovative application of Internet of Things technologies to farming. It also showed the value of not yet applied Digital Twin types, which inspired the use cases about potential redesign scenarios.
For this reason, we expect that applying the Digital Twin concept, as described in our framework, can accelerate the development and adoption of Digital Twin solutions for smart farming. However, future research is needed to provide evidence for this hypothesis. Furthermore, the implementation model of our framework only deals with implementing the enabling information technology. We did not take into account organisational and behavioural issues, such as the impact on supply chain collaboration, data ownership and governance, stackable planters the potential emergence of disruptive business models based on Digital Twins, ethical considerations, and so forth. We would like to encourage researchers in these disciplines to also study Digital Twins, since these non-technical issues might be decisive for the success of Digital Twins. Our intended follow-up work is related to the further development of the framework. In particular, we plan to elaborate the conceptual framework into an information architecture framework, which will comprise a consistent set of architectural viewpoints for modelling Digital Twin-based software systems . This architectural framework will be the basis for developing Digital Twin applications that cover the entire life cycle. Many farming systems in Europe are struggling with substantial challenges resulting from fundamental changes in their economic, technological, demographic, ecological and social environment . The resilience of farming systems, i.e. their ability to cope with and respond to shocks and stresses, has therefore become a major concern . The Covid-19 pandemic and the measures for its containment – e.g. lockdowns, travel restrictions and border closures – were expected to add another shock to farming systems. Using 11 indepth case studies, this paper investigates the extent to which different farming systems across Europe were affected by the crisis, which resilience strategies they adopted, and which characteristics enabled or constrained their resilience abilities. This paper contributes to a fast-growing literature on impacts of the Covid-19 pandemic on different parts of agricultural and food systems, e.g. food value chains, marketing channels, trade patterns and food security . Impacts on different farming sectors, e.g. due to production and demand distortions, have also been discussed . Others have reflected on the resilience of food systems at large in the light of Covid-19 . However, a systematic assessment how characteristics of farming systems have enabled or constrained their responses to the Covid-19 crisis is missing. By using an elaborate framework to assess and compare the resilience of farming systems before and during the pandemic, this paper aims to enhance our understanding how different farming systems were exposed to the crisis, which resilience capacities were revealed and how resilience was enabled or constrained by the farming systems’ social and institutional environment. Section 2 explains the SURE-Farm framework to assess the resilience of farming systems and the special data collection on Covid-19. Results are presented in Section 3, followed by discussion and conclusions in Section 4. Following the social-ecological tradition of resilience thinking , we define the resilience of a farming system as its ability to ensure the provision of its desired functions in the face of often complex and accumulating economic, social, environmental and institutional shocks and stresses, through anticipating, coping and responsive capacities . The resilience of a farming system is affected by specific system characteristics, and by the enabling or constraining environment, in particular institutional arrangements and resource availability .