Modelling Digital Agriculture: a Method and AI-Infused Tool for Living Labs

Digital technologies and Artificial Intelligence (AI) are increasingly transforming agricultural practices by introducing advanced sensing, data-driven monitoring, and decision support systems. These innovations reshape technical infrastructures, farm processes, and stakeholder roles within the agri-food value chain, while also having implications for the natural environment. Understanding these transformations is therefore essential for designing digital solutions that contribute to more sustainable agrifood systems. Requirements Engineering (RE) provides well-established techniques for gathering and representing stakeholder needs through visual models. In particular, semi-formal notations are widely used to capture different perspectives, such as domain, goals, and process. However, there is still little empirical evidence on the application of requirements modelling in agriculture and, more broadly, in participatory innovation environments. Moreover, the use of RE approaches in these contexts raises an End-User Development (EUD) challenge: enabling domain experts to actively shape and appropriate the modelling artefacts. This thesis explores how modelling with semi-formal notations can enhance knowledge exchange in agricultural living labs, i.e., networks of heterogeneous stakeholders addressing common challenges through participatory activities. The research involved 20 European living labs undergoing digital transformation. In these settings, stakeholders needed structured representations to illustrate how farm processes change after the introduction of digital technologies, thereby supporting communication and analysis. The research employed a design science methodology, spanning iterative cycles of design and validation within the living labs. The practical challenges arising from this context guided the literature review and informed the design of the proposed artefacts grounded in RE and EUD. Specifically, the thesis tackles four challenges: the use of semi-formal notations in agricultural living labs; the engagement of living lab stakeholders in both the creation and interpretation of models; the design of support for novice and non-technical modellers in alignment with their capabilities; and the role of AI-based support within the requirements modelling landscape for digital agriculture. Two main artefacts were developed and evaluated within the living labs: PoMeLO (PrOcess ModElling in Living labOratories), a RE method for modelling socio-technical transformations in agricultural living labs, and ModeLLer, an AI-infused EUD modelling tool designed to support novice modellers. The thesis contributes empirical insights into requirements modelling in participatory contexts and provides one of the first in-depth studies in the agricultural domain. It bridges RE and EUD, adopts a Human-Centred AI perspective, and delivers reusable artefacts that can support further research as well as the adoption of requirements modelling practices in sustainable digital innovation. The thesis is organised into four main parts, moving from problem investigation to the design and evaluation of the two main research artefacts. Part I — Problem Investigation Chapter 1 — Introduction Introduces the research problem, objectives and research questions. Chapter 2 — Literature Review Presents background and related work on Requirements Engineering, End-User Development, modelling methods, digital agriculture, and AI-infused systems. Part II — PoMeLO: Requirements Engineering Method Chapter 3 — Overview of Part II Summarises the research gaps and questions addressed by the method. Chapter 4 — Method Design and First Validation Introduces PoMeLO, its methodological foundations, and the first validation round. Chapter 5 — Consolidated Method and Field Experiment Presents the refined version of PoMeLO and its second validation through a two-phase field experiment. Chapter 6 — Implementation and Evaluation Results Complements the evaluation of PoMeLO by reporting results from its implementation and use in the CODECS context. Part III — ModeLLer: End-User Modelling Tool Chapter 7 — Overview of Part III Summarises the research gaps and questions addressed by the tool. Chapter 8 — Tool Design and Cognitive Walkthrough Introduces ModeLLer and reports a cognitive walkthrough evaluation conducted on a prototype. Chapter 9 — User Study within a Computational Thinking Framework Reports the validation of ModeLLer through a user study grounded in Computational Thinking. Chapter 10 — AI-Infused ModeLLer and Expert Evaluation Presents the improved AI-infused version of the tool, including a technical evaluation and a Wizard of Oz study with domain experts. Part IV — Conclusions Chapter 11 — Contributions and Limitations Discusses the main scientific and practical contributions, as well as the limitations of the research. Chapter 12 — Future Work Outlines possible extensions of the proposed artefacts and results.