Spatial Validation of Food Environment in Berkeley Using 3D Street View Data

INTRODUCTION: For built environment studies, the locations of different food environment factors matter. Issues associated with using secondary databases such as misclassification and geospatial inaccuracy hinder researchers from assessing food environment precisely; yet field audits are resource intensive and time-consuming. The purpose of this study is to illustrate and assess the feasibility and geospatial accuracy of using 3D street view data to collect food environment data.

METHODS: The names and geographic coordinates of food establishments on 4 major streets (72 blocks) that have the highest food establishment densities in Berkeley, California, were sampled via 3D street view data, and field audits were conducted to validate this data using GPS units. We then compared the agreements on counts of food establishments and their geospatial accuracy, and calculated inter-rater reliabilities.

RESULTS: We found high levels of agreement between field audits and the 3D street view data: >80% agreement in counts, the average distance between food establishment locations obtained via 3D street view data and ground-truthing was 6 meters, and >80% of the food establishments obtained by two methods were within 10 meters of each other. There were also nearly perfect inter-rater reliabilities (>90% agreement and kappa statistics of >0.80) for both methods.

DISCUSSION: Our results indicated the reliability of using 3D street view data for food environment audits. Other built environment factors such as walkability and other land use characteristics, can also be assessed using the same approach as it has the potential to significantly reduce the cost of field data collection.