Locating facilities that
supply the most demand points
The success of any organization, whether private or public, hinges on where its facilities are located. This impacts how easily they can attract customers and manage transportation costs. In this exercise, I dove into location-allocation analysis, using the network dataset from the previous two exercises (see earlier coursework: Building a network). The goal was to find the best spots for facilities and effectively assign demand to them. However going into this exercise, it was important to remember that the ideal location varies based on factors like whether it's a private or public entity and the types of products or services offered. For instance, a police or fire station should be positioned to serve a wide range of citizens, regardless of their background, while on the other hand, a retirement home might be best placed near areas with a larger elderly population.
Again working with Västra Götaland County in Sweden, I chose to look into two hypothetical scenarios using two different types of location-allocation analyses:
Type of location-allocation analysis: Maximize Market Share
The scenario: The Willy’s supermarket chain is considering adding one more store location in Gothenburg city. They have 4 candidate locations but want to find out which one captures the most demand or market share, aiming at customers within a 15-minute walking distance of the new facility.
Type of location-allocation analysis: Maximize Coverage, Minimize Facilities
The scenario: Västtrafik, Gothenburg’s public transportation company, wants to reduce the maintenance costs of their bus stations. In this sense, a station represents two or more stops, for example, on opposite sides of a street. Therefore they want to find stations that capture the most demand within a 10-minute walking distance as well as find the ones that are potentially redundant candidate stations for removal.
The general workflow for choosing the best location with the ‘Location-Allocation’ network analysis tool in ArcGIS Pro was first to import, examine, select, and clip the relevant data to the area of interest. Using Points of Interest (POI) data from Open Street Map, I selected the facility of interest and imported them into the Location-Allocation layer. Furthermore, since Location-Allocation is about finding the best location based on supply and demand, demand points are needed, which are typically based on population. For this, I needed to have some form of population data aggregated into points. Depending on the desired level of accuracy, this can probably be done in several ways. But in this case, it was enough to use population data generalized into grids of 1x1km squares for rural areas, and 250x250m squares for urban areas, which were then converted to point features centered in each square. With this, I could import the aggregated population point data as the demand points to the ‘Location-Allocation’ layer.
The 4 candidate facilities in the supermarket scenario were determined by visualizing the combination of where the population is concentrated and areas that are seemingly underserved by currently operating supermarkets. The problem type at hand was Maximize Market Share. This is an appropriate approach for the analysis because this problem type chooses a specific number of facilities in a way that allotted demand is maximized while being among competitors.
Before running an analysis with a ‘Maximize coverage, Minimize facilities,’ problem type, I had to import the demand points as outlined above, but also the facilities, which required a few extra steps. This is because I wanted to ensure that stations near important places would be required in the solution and not considered candidates for removal. In ArcGIS Pro, using ‘Select by Attribute’ on the POIs layer, I selected graveyards, hospitals, kindergartens, malls, museums, parks, schools, sports centers, stadiums, and universities, which resulted in 1,064 places in total. Then, I used ‘Select by Location’ to find all the bus stations within 250m of those places, which resulted in 427 facilities. With that selection, I imported those as ‘Required’ into the Location-Allocation solver, then switched the selection and added the rest as ‘Candidates’.
The final result shows the final number of candidate parent stations for removal was 265 out of a total of 790 stations, while still maintaining about 92% of total demand coverage within a 10-minute walking distance. However, industrial sites in the city (where there is a low population) are hubs of activity and therefore need to accommodate workers going to and from various industries, but have been unreasonably left out of the solution due to little or no demand points residing there. Thus, in hindsight, I should have perhaps included stations near industrial sites as ‘Required’ as well.