Towards a Global Food Systems Datahub

In the quest for agricultural sustainability, we face the challenge of feeding the global population under the constraints of finite resources and a delicate ecological balance. The intricate interplay of climate dynamics, socio-economic factors, and environmental stewardship demands an approach to agriculture that is as intelligent and adaptive as it is respectful of our planet’s capacities. Central to this endeavor is the synthesis and utilization of vast, heterogeneous datasets that span from crop genomics to market trends, and from soil health to consumer preferences. Yet, the current paradigm is fragmented, with valuable data isolated across domains, lacking the coherence and accessibility needed for actionable insights. While there is an urgent imperative: to enable the kind of data-driven decision-making that can foster sustainability in agriculture, traditional methods of data analysis are insufficient. They lack the depth and agility required to navigate the complexities of modern agriculture.

For example, precision agriculture technologies have rapidly evolved over the last two decades. Almost all of the machine systems including planters, air-seeders, liquid application systems, and crop harvesters have extended stacks of sensing and control systems to automatically control the application of crop inputs [1]. Farmers can potentially access numerous databases which can be classified into soils (e.g., spatial soil features); machine application (e.g., crop input flow from sensors, and geodata); climate (e.g., temperature, precipitation); remote sensing (e.g., satellite, manned, and unmanned aerial systems with sensors); machine vision (e.g., autonomous scout vehicles, and cameras on large equipment) and above- or below- ground sensors providing near real-time data (e.g., soil moisture, temperature) [1,2].

However, the unprecedented amount of data availability has not radically transformed data-centric technologies like variable rate technologies (seeding, nutrition), site-specific application (chemical selection, pest control, plant nutrition). Among other factors including noisy data, incomplete datasets, missing data layers, and errors in data, the lack of data interoperability has been the biggest obstacle preventing data integration and analytics for decision support [7]. The data sources have been developed by various manufacturers using non-standardized file formats, units, and metadata; have varying level of both spatial and temporal resolutions; need specific softwares to read different data file; and no common platform for data curation, integration, spatial and temporal normalization, and analytics. The magnitude of data gathering, integration, and analytics further magnified due to the transformation of rural landscape, land ownership, gradual aging (the average age of a farmer in the U.S. is 57.5 years), and continuous decline in human capital remaining in rural America available for crop production [2]. Furthermore, while humans may seamlessly transition between different levels of spatial, or temporal, or informational granularity, knowledge representation in AI is by its nature rigid and traditionally forces a priori defined perspectives on the data model.

Consequently, knowledge representation is usually used to solve a singular and specific problem or across a set of very similar purposes. This produces significant overhead when preparing data for subsequent processing or analysis, as the resulting knowledge bases cannot be readily reused. At large scale, and in contexts such as climate-adaptive AI, where relevant data is highly heterogeneous, the cost of data preparation and management, and of assurance of data quality is often prohibitive.

To enable climate-adaptive agriculture, it is necessary to establish modular and multi-granular knowledge representation solutions at large scale that are agnostic towards different levels of spatial information and temporal granularities [3,4,6]. Therefore, the necessity for a coherent and standardized approach to agricultural data is not just a technological imperative, but also a socioeconomic one, as it holds the potential to usher in a new era of efficiency and sustainability in agriculture. One transformative solution would be a Global Food Systems Datahub powered by a knowledge graph backend which would aspire to transcend the limitations listed above by establishing a robust framework for integrating and interpreting data at a granular level. This frame would be beyond an assembly of information; but a living, evolving digital ecosystem that maps relationships and infers patterns, thereby enabling advanced AI applications in the realm of digital agriculture since knowledge representation is a long-standing AI discipline that is currently gaining large-scale adoption in industry in the form of KGs [5].

The need for a knowledge graph-powered hub is multi-faceted. At its core, the knowledge graph offers a semantically rich structure that mimics the interconnectedness of real-world systems, allowing for a nuanced representation of data relationships. This facilitates the application of foundational AI research areas such as multi-modal spatiotemporal deep learning, explainable and trustworthy AI, along with neuro-symbolic question answering. With these technologies, we can analyze and understand the function and malfunctions of agricultural systems, predict outcomes, and make recommendations that are transparent and justifiable.

In practical terms, the Global Food Systems Datahub stands to transform how we approach sustainable intensification – the enhancement of production on existing farmland in harmony with the environment. AI-enabled tools developed from the hub’s knowledge graph can inform strategies that balance crop yield with ecological conservation, such as the spatial targeting of prairie strips or the integration of agrivoltaic systems. High-resolution data can inform landowners’ decisions about land use, optimizing for both production and ecosystem service provision. Moreover, this hub would bridge the gap between the data-rich insights of individual landowners and the broader needs of program managers and policymakers, who often operate with a paucity of detail. It will render visible the hidden layers of agricultural data, thus enabling targeted interventions and informed decision-making at a micro-scale, with macro-scale implications.

The overarching need for the datahub aligns with the pressing global need of sustainable development. The global food systems datahub can play a pivotal role in mitigating climate impacts, managing water-energy-food nexus challenges, and supporting socio-economic and cultural adaptability in agricultural practices. In this scope, for example, falls a recently started large internally funded project at Kansas State University titled “Towards a Global Food Systems Data Hub: Seeding the Center for Sustainable Wheat Production” (Hitzler and McGinty involved) which is set to seed the beginnings of such a data hub by initially focusing on sustainable wheat production. In a similar vein, the National Science Foundation’s Proto-OKN program11 is working on a network of interconnected knowledge graphs, some of which are set to bear significant importance to a future global food systems datahub. Many other related efforts are of course under way world-wide.

In this special issue, several KG centric approaches have been presented by researchers that are also relevant to the overall goal.

With these papers in the special issue, we see that the community is already recognizing the benefits of standardized vocabulary generation, semantic modeling of global food systems data and they may be used for further analyses and reuse by industry and researchers. We hope that this special issue highlights the global food system as one of the most essential yet complex constructs, underpinning not just health and nutrition but also socioeconomic structures, environmental balance, and cultural identities. It embodies an extensive range of data types, from climatological observations, agricultural practices, supply chain logistics, to consumer behavior and nutritional information.

This multidisciplinary combination, however, poses a significant challenge: the integration and interoperability of diverse datasets that remain siloed across different domains. The existing gap hinders the potential for multi-layered analysis, comprehensive understanding, and efficient decision-making, which are critical in addressing pressing issues like food security, sustainability, and resilience against climate change. Advancements in artificial intelligence (AI) and machine learning (ML) offer unprecedented opportunities to bridge this gap by enabling the translation of massive, heterogeneous data into a format that is not only comprehensible across various disciplines but also operable by computational systems. Therefore, the need for a Global Food Systems Datahub (GFSD) with a knowledge graph backend is our envisioned solution as a transformative platform that encapsulates the wealth of global food data into an interconnected, semantically-rich, and AI-ready infrastructure.