The work, published in the Food Modelling Journal, follows the “farm to fork” paradigm that underpins contemporary food safety risk assessment. Instead of treating a food product as an isolated item, the paradigm demands that researchers follow it through every stage of the chain. Supply chain network models are useful and versatile tools for this purpose, because they can reveal spatial linkages within supply networks that would otherwise remain hidden. Visualisation of such networks provides important insights into their structure and dynamics, and visual analytics enables researchers to explore, interpret and communicate data more rapidly and effectively than tables of numbers ever could.

The new “egg supply chain network model” draws on an a-priori generated “supply chain network data cube”, an excerpt and reimplementation of a dynamic freight flow model previously described by Balster and Friedrich. The data cube represents the German egg supply chain as a directed graph with nodes and edges. Each node carries three attributes: an actor, which can be a producer, wholesale handler, warehouse brand, store brand or consumer; the product, which is fixed as “Eggs”; and a location, given as the names of urban and rural districts in Germany at NUTS-3 level, the finest administrative level used in European statistics, with 402 districts in total and corresponding official location IDs. The edges carry the quantities of eggs handled by the actors. Because the underlying freight model refers to the year 2012, the data cube is based on the same data sources and the same reference year.

To estimate the quantities assigned to the nodes and edges, the researchers used multi-regional input-output analyses as well as optimisation and gravity models implemented in Python. The best fitting distribution for the product “chicken eggs” within the German federal transport infrastructure plan refers to published freight data, and geodata for Germany are sourced from Eurostat. A notable detail is that 28 discrete brands for warehouses and stores are incorporated as separate actors in the data cube, so the network does not just distinguish broad stages of the chain but can, in principle, follow eggs through individual retail and logistics brands.

The egg supply chain network model compiles two data frames from this resource, a node data frame and an edge data frame, with the appendant information about actor, product, location and transported quantities. In general, the model estimates and visualises the quantity of eggs in tons per year per district in Germany handled by a selected actor. To keep the tool easy to apply, the developers defined only two input parameters, “product” and “actor”. Because the underlying data cube covers only chicken eggs, the product parameter is classified as “Constant” and cannot be modified. The actor parameter, by contrast, is fully adjustable: users can choose between producers, wholesale, warehouses, stores, consumers, and consumer subcategories covering wholesale, direct sale and store channels.

Each choice produces a distinct picture of the chain. Selecting “Producers” estimates and visualises egg production per district. Choosing “Wholesale”, “Warehouses” or “Stores” calculates and illustrates the quantities handled by each of these actors, with the 28 discrete brands aggregated in the model, although users who want brand-level detail can select individual warehouse and store brands as separate actor values. The “Consumers” option provides estimates of total egg consumption in tons per year per district, while the channel-specific options break consumption down by how the eggs reached the household. The data frame generated for the user-defined actor is merged with the geodata frame to allow visualisation with an integrated Python script, and the results are written to an output table parameter.

The outputs are twofold. The primary outcome is a data frame with information about product, actor, location and calculated product quantities that geographically characterise the food flows in Germany, saved as a CSV file. When the model is executed with the desktop software FSK-Lab, this file is automatically stored within the FSKX model file after writing it out into a new FSKX file, and online execution in the virtual research environment likewise generates a downloadable FSKX file containing the results. The second output is a choropleth map with a single-hue progression: the darkest hue represents the greatest predicted value and the lightest shade the least, colouring the predefined districts in proportion to the predicted egg quantities within each region.

The default simulation, with “Consumers” as the actor, shows the highest amount of eggs consumed in the district of Berlin, followed by other densely populated areas such as Hamburg, Munich, Hanover, Cologne and Frankfurt on the Main, a pattern that mirrors population density. Switching the actor to “Producers” produces a very different map: the regions with high egg production capacities are district Vechta, followed by Osnabrück and Emsland, all located in the federal state of Lower Saxony, the heartland of German egg farming. The contrast between the two maps captures the essence of the supply chain problem, since the eggs consumed in Germany’s largest cities are overwhelmingly produced hundreds of kilometres away in the northwest.

What makes the model more than a one-off analysis is its packaging. It conforms to the Food Safety Knowledge Exchange (FSKX) format, a standardised file format developed by the German Federal Institute for Risk Assessment to support harmonised annotation, encoding, application and sharing of models. All data and metadata are stored in a single FSKX compliant file, allowing transparent and efficient access to the model as well as reproducibility of simulations and visualisations. The model file was generated with FSK-Lab, an open-source extension to the Konstanz Information Miner (KNIME) that enables KNIME users to work with FSKX models. The model and all required resources can be executed online through a virtual research environment, with a default simulation available at the click of a link and personalised simulation scenarios creatable through the same environment, which requires the Firefox browser version 78 or higher.

The developers see the model as a template for future expansion. Because the underlying data cube is designed as an example network, the product “Eggs” is currently defined as a constant parameter, but provided that further products are incorporated into the data cube, the food flows of these products could be analysed in future versions of the model as well. If data sources referring to time periods other than 2012 are used to regenerate the data cube, the updated data could likewise be analysed by the same network model. One restriction is that the distribution of eggs per district can be estimated and visualised for only one actor per simulation setting, although users who need calculations for several actors can simply create different simulation scenarios. For risk assessors trying to follow a hazard from farm to fork, the model offers something concrete: a transparent, reproducible, standardised picture of where Germany’s eggs come from, where they go, and who handles them along the way.

Subject of Research: Food safety modelling

Article Title: A network model of the egg supply chain in Germany implemented as a FSKX compliant object

Article References: Ganas, P., Fuhrmann, M., & Filter, M. (2021). A network model of the egg supply chain in Germany implemented as a FSKX compliant object. Food Modelling Journal, 2, Article e74171. https://doi.org/10.3897/fmj.2.74171

Image Credits: AI Generated

DOI: 10.3897/fmj.2.74171

Keywords: network, model, supply, chain, Germany, implemented, FSKX, compliant, object, scientific research