Joint Research Centre
Digital Earth Lab

Comprehensive planning and management of urban areas requires information about building use. However, due to the high cost of the building use surveys, the required data is not available for many cities. In order to address this high cost, we propose a methodology for the reuse of freely available place data for building use mapping purposes.

As part of this activity we investigated the suitability of social media data (especially from Foursquare) as a data source for determining building use. A case study has been conducted in Amsterdam, in an area of 72.12 km2, where 112,567 buildings are located. To each of these buildings, a building use category was assigned that describes the type of the dominant activity that takes place within them. For the estimation of the building use category of each building, three methods were used. For each method a distance value between places and their closest building was taken into consideration. The estimated building use dataset has been compared to the reference building use dataset that has been generated using an official dataset about non-residential building use. As the best estimated building use dataset was selected the one for which the he highest Cohen's kappa coefficient was achieved.

As an outcome of the accuracy assessment, the best results were achieved with a method that assigns the dominant non-residential building use to be the category with the highest number of Foursquare places co-located inside the particular building. For this estimated building use dataset the overall classification accuracy was approximately 54% (see also Figure 6). Regarding the individual building use categories the “exclusively residential” and “hotels, restaurants & cafes” found to have the highest accuracies of approximately 83% and of almost 60% respectively. On the contrary, the methodology failed to identify built-up areas with “industries” or “storage & unclear” building use. We therefore conclude that foursquare data can only be used for the estimation of particular building use categories.

In our future work, in order to examine the reproducibility of the proposed methodology, we plan to repeat this study in other urban areas. Many buildings in the city of Amsterdam have small frontage due to historical reasons. Thus, the repetition of the methodology in other urban areas, that have buildings with different morphological characteristics compared to the buildings of Amsterdam, will enable us to perform a more complete evaluation of the proposed methodology.


Final publication in Environment and Planning B