Mapping the retail food environment at high resolution across Europe
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Aim
Neighbourhood food outlet availability influences dietary behaviours and nutrition related health outcomes. However, food environment research across Europe remains limited by inconsistent spatial data, heterogenous food outlet classifications, and the lack of publicly available high-resolution retail food data. We developed and evaluated a scalable framework to acquire, classify, validate, and analyse food-related points of interest (POIs) from Google Maps across 39 countries in Europe and Turkey. We also developed a method for characterizing the joint co- occurrence of multiple outlet types as neighbourhood-level food environment exposures.
Methods
We harmonized ∼3.26 million food-related POIs. Outlets were classified into five main categories (cafés and bakeries, restaurants, fast-food and snack outlets, groceries and food retail, and bars and pubs) and 25 subcategories. We assessed data quality against official register s in five regions (the Netherlands, United Kingdom, Denmark, Madrid (Spain), and Brno (Czech Republic)), testing agreement in positional accuracy, completeness, spatial clustering (Nearest Neighbour Index), and spatial density (kernel density estimation (KDE)). To enable cross-country comparisons, POIs were aggregated to national and city scales, with food outlet density calculated as outlets per 1,000 residents and standardised using z-scores. To characterize neighbourhood-level co-exposures, we applied Principal Component Analysis (PCA) followed by Latent Class Analysis (LCA) to z- standardized outlet densities across 500 m hexagonal grid cells in eight major European cities.
Results
Google Maps data showed high positional accuracy, with 91% of outlets located within 30m of registry records. Completeness varied by region, while KDE comparisons showed moderate-to-strong spatial agreement with city-specific variation. PCA identified two components explaining 68.4% of the variance (PC1: 51.3%, PC2: 17.1%), with PC2 differing between café/bar and fast-food/supermarket densities. LCA identified four neighbourhood classes: low-access, moderate-mixed, dining-out, and high-density-mixed.
Conclusion
Our harmonized high-resolution framework supports cross-national monitoring of retail food environments and spatial epidemiological research across European settings.
