Top 5 This Week

Unveiling Future Trends in Agri-Food using semi automated horizon scanning

Introduction

Sustainability and climate change represent major challenges for the agri-food sector in the upcoming decades and will significantly alter the requirements for nutrition and food production. The Food and Agriculture Organization (FAO) estimates that food systems contribute over one-third of global greenhouse gas emissions. Conversely, the agricultural sector is among the most vulnerable to the impacts of climate change. This dual role of the food sector as both a significant contributor to and victim of climate change necessitates urgent and concerted efforts to develop resilient and sustainable practices.

The article explores how semi automated horizon scanning—a technique for identifying emerging trends and risks—can be applied to the agri-food sector. By utilizing advanced machine learning and natural language processing, key trends are identified that could contribute to a more resilient food system in the context of climate change and evolving consumer behaviors. This scientifically based and methodologically advanced search for early signals of technological and social change seeks to assist actors in the agri-food system in proactively addressing weak signals, leveraging emerging opportunities, and building resilience against transformations.

Trends as Indicators of Change

Trends are crucial indicators of change, enabling stakeholders to better understand the evolving landscape and act proactively rather than reactively. Early recognition of trends aids in risk assessment and mitigation as well as long-term strategic planning, ensuring better resource allocation, investment decisions, and policy formulation. In the agri-food sector, obvious trends like digitalization, artificial intelligence, and climate change are already well-known. However, less obvious early signals should also be considered, as they have significant long-term potential to tackle the addressed challenges above.

Strategic Foresight provides tools to identify and evaluate early trend signals (“weak signals”) and systematically analyze future developments and their interdependencies. By monitoring and analyzing trends, stakeholders can make informed decisions, foster innovation, and build adaptable systems. Horizon Scanning and Sensemaking, in particular, provide valuable insights as essential instruments of strategic foresight.

Looking for the bigger picture

To identify new trends at an early stage, it is necessary to broaden the perspective and look beyond one’s own industry for changes and innovations with disruptive potential. Semi-automated horizon scanning offers a helpful tool for searching and analyzing large volumes of data. It allows a systematic investigation of potential future threats, opportunities, and trends to inform strategic planning and decision-making, often using various data sources and analytical methods. Automated horizon scanning employs AI and data analytics to continuously monitor and analyze vast data sources, identifying emerging trends and potential future risks or opportunities. Semi-automated scanning helps quickly analyze a large collection of documents on current topics and potential changes. For example, news articles, patents, media sources, scientific papers, or industry reports can be used. Extracted from various websites, these documents are processed with machine learning techniques specifically for natural language processing (NLP). This web scraping involves automatically downloading texts from predefined websites, and the collected data along with their metadata are analyzed using ML techniques.

In the last decade, advancements in NLP have significantly improved text analysis capabilities, going beyond word counting to understand semantics and external knowledge. Topic models, a statistical method, identify abstract topics in a corpus by grouping words that frequently appear together. These models, developed by David Blei[1], assist in topic identification but require interpretation. Word embeddings, based on research by Mikolov et al. [2], represent words as vectors in a multidimensional space, capturing their semantic relationships. This enables complex computations, such as associating the vector for ‘king’, subtracting the vector for ‘man’, and adding ‘woman’. The result is extremely close to the vector for ‘queen’. This approach also works, for example, for capital cities and their countries. Building on this, in 2018 a team from Google developed BERT[3] (which stands for Bidirectional Encoder Representations from Transformers), which essentially scaled up this concept. In its basic form, this model was trained on an unlabeled corpus of more than 3.2 billion words, taken from the Google Books corpus and the English Wikipedia. This approach significantly enhances NLP tasks. Today, many more models have been created which are essentially the baseline for any NLP task, since they vastly outperform any other approach. As such, the development of BERT and similar models represents a major advancement in the field of natural language processing, enabling more accurate and efficient understanding and continuously expanding the possibilities of managing large datasets to extract weak signals and drivers of change.

Trends and Signals in the Agri-Food System

In the context of the agri-food system, three primary thematic areas emerge as critical despite the multitude of underlying trends and signals: changing consumption behavior, resilient food systems, and the interaction between the food sector and climate change. Developed nations’ agri-food systems are characterized by dominant players, extensive trade networks, and year-round food availability, often driven by price mechanisms. Concurrently, consumer awareness of food quality is increasing, with a growing demand for nutritious, healthy food produced under strict environmental and social standards. The COVID-19 pandemic highlighted the need for resilient food systems crucial for sustainability objectives. Technological advancements like digitalization, real-time data analytics, decision support software, and AI are transforming the agri-food landscape, making food systems more efficient and adaptable to disruptions.

Our Horizon scanning identified several trends contributing to sustainable food systems, for example: the rise of alternative proteins, new forms of food production, and the emergence of local food circles. The increasing global population necessitates sustainable protein sources like plant-based products and cultured meat. Consumer behavior shifts towards plant-based diets are also crucial, with meat consumption per capita in the Western world expected to peak by 2030.

Innovative food production methods, such as urban horticulture, vertical farming, and aquaponics, complement traditional agriculture, enhancing food security and resilience against climate change. Local food circles promote regionally grown food, supporting sustainable practices and local farmers. Innovations like seasonal food box subscriptions, online farm shops, and community-supported agriculture reflect a growing interest in direct sales from farmers to consumers, fostering decentralized food production systems. While these trends offer promising pathways towards sustainable food systems at first glance, each has complexities and potential drawbacks. For example, alternative proteins may reduce CO2 emissions but increase water usage. Urban agriculture faces limitations due to the small proportion of built-up areas relative to total agricultural land. The benefits of local food systems depend on regional factors and seasonal variations.

In conclusion, sustainability and climate change are major challenges for the agri-food sector in the coming decade. Addressing these issues requires a multifaceted approach, integrating technical and social innovations across the value chain. Collaboration among all stakeholders is essential to implement and support these innovations, ensuring a resilient and sustainable future for global food production and consumption.

[1] Blei, David M. “Probabilistic topic models.” Communications of the ACM 55.4 (2012): 77-84.

[2] Mikolov, Tomas, et al. “Efficient estimation of word representations in vector space.” arXiv preprint arXiv:1301.3781 (2013).

[3] Devlin, Jacob, et al. “Bert: Pre-training of deep bidirectional transformers for language understanding.” arXiv preprint arXiv:1810.04805 (2018).