Grocery Access Part 2: Network Analysis in QGIS

Last updated on 2026-10-17 | Edit this page

Estimated time: 75 minutes

Overview

Questions

  • How do we move from grocery store points to an analysis of travel along roads?
  • How do shortest path and service area tools in QGIS answer different questions about access?
  • Why is network distance often better than straight-line distance for access studies?

Objectives

  • Load grocery store, road and study area layers into QGIS from the web
  • Reproject layers to a projected coordinate system so that distances are in meters
  • Calculate shortest paths from one origin to every supermarket
  • Generate service areas that show which streets are within walking distance of a supermarket
  • Recognize data problems that break network analysis

Overview

In Part 1, a team mapped grocery stores in the field with QField. In this episode, we use grocery store points with a road network to ask a new question:

How accessible are grocery stores when movement is limited to the road network?

This episode moves from data collection to analysis. Instead of treating distance as a straight line, we measure routes along streets.

Imagine that a class, a community partner or a student group wants to know how easy it is to reach a supermarket from a residence hall, an apartment complex or a neighborhood. We will:

  1. prepare the grocery store and road data
  2. calculate routes from one starting point to every supermarket
  3. create service areas showing the streets within a chosen distance of a supermarket

The data

Three small files are stored in the data/ folder of this lesson’s repository. You do not need to download them; QGIS reads them straight from the web.

Layer Contents Source
grocery_stores 64 food stores in and around Lafayette and West Lafayette: 17 supermarkets, 2 wholesale stores and 45 convenience stores OpenStreetMap contributors (Open Database License), October 2026
roads 3,466 road segments covering both cities and a margin around them U.S. Census Bureau, TIGER/Line 2024
study_area The city limits of Lafayette and West Lafayette U.S. Census Bureau, 2024 cartographic boundary files

The store points stand in for data collected with QField. In a real project, the team would have verified them in the field first.

Prerequisite

Before you begin

Step 1: Load and inspect the layers

  1. Open QGIS and start a new project.

  2. Go to Layer > Add Layer > Add Vector Layer.

  3. For Source Type, choose Protocol: HTTP(S), cloud, etc. and leave Type as HTTP/HTTPS/FTP.

  4. Paste the first address into URI, click Add, and repeat for the other two:

    https://raw.githubusercontent.com/SpatialTurn/CaseStudies/main/data/grocery_stores.geojson
https://raw.githubusercontent.com/SpatialTurn/CaseStudies/main/data/roads.geojson
https://raw.githubusercontent.com/SpatialTurn/CaseStudies/main/data/study_area.geojson

  5. Click Close. In the Layers panel, drag grocery_stores to the top so the points draw above the roads.

  6. Right-click grocery_stores and choose Open Attribute Table. Look at the fields: store_name, store_type, address, city, osm_id, source and verification.

Before analyzing anything, look for problems:

  • duplicate points
  • stores outside the study area
  • stores with no name or address
  • roads that look disconnected from the rest
Challenge

Quick inspection

Look at the map and the attribute table. What do you notice? Which kinds of data problems would most directly affect a network analysis?

You may notice that six convenience stores, such as a Love’s travel stop, lie just outside the city limits. Stores like these should usually stay in an analysis, because city residents shop there too. (All 17 supermarkets are inside the city limits.) A few convenience stores have no name, and several stores have no address. Two Fresh Thyme points might look like duplicates, but they are two different stores, one in each city; check before deleting anything.

The problems that most directly affect a network analysis are those that affect where a route starts or ends:

  • a store point placed far from any road
  • duplicate stores, which inflate the number of destinations
  • disconnected road segments, which break routes
  • layers in different coordinate systems, which appear in the wrong place

Step 2: Keep only the supermarkets

Convenience stores rarely sell a full range of fresh food, so food access studies usually focus on supermarkets.

  1. Open the Processing Toolbox (Processing > Toolbox).
  2. Search for Extract by attribute and open it.
  3. Set Input layer to grocery_stores, Selection attribute to store_type, Operator to =, and Value to Supermarket.
  4. Click Run. A new layer called Extracted (attribute) appears with 17 supermarkets. Rename it supermarkets (right-click > Rename Layer).

Step 3: Reproject the data

Network analysis should be done in a projected coordinate system, so that distances are measured in meters rather than degrees. The files are stored in latitude and longitude (EPSG:4326). For Lafayette, a good choice is NAD83 / UTM zone 16N (EPSG:26916).

  1. In the Processing Toolbox, search for Reproject layer and open it.
  2. Set Input layer to supermarkets and Target CRS to EPSG:26916.
  3. Click Run, and rename the output supermarkets_utm.
  4. Repeat for roads (rename the output roads_utm) and study_area (study_area_utm).
  5. Set the project coordinate system too: click the CRS button in the bottom-right corner of the QGIS window (it shows something like EPSG:4326), search for 26916, select NAD83 / UTM zone 16N, and click OK.
Caution

Why set the project CRS as well?

QGIS’s network tools report travel cost in the units of the project’s coordinate system. If the project is still in EPSG:4326, a 4 km route is reported as a cost of about 0.036 (degrees). With the project in EPSG:26916, the same route has a cost of about 4,000 (meters).

Do not assume that layers are ready for analysis just because they draw in the same place. Check the CRS of each layer (right-click > Properties > Information).

Step 4: Do not clip the roads too tightly

It is tempting to clip the road layer to the city limits to make it smaller. Here, that is a mistake. The road file already stops a short distance outside the two cities, and the city limits are irregular: they do not even meet each other across the Wabash River. Clipping the roads to them removes the bridges and splits the network into pieces.

Two maps of Lafayette and West Lafayette. Left: with the full road layer, orange routes run from the origin star in West Lafayette to all 17 supermarkets. Right: with roads clipped to the irregular, dashed city limits, the network is split at the Wabash River and only the 7 supermarkets in West Lafayette are reached; the 10 in Lafayette are marked with red crosses.
Clipping the roads to the city limits breaks the network.

With the full road layer, all 17 supermarkets can be reached. With roads clipped to the city limits, only 7 can. When you do clip a network, clip it to a generous rectangle or buffer around your study area, never to an irregular boundary.

Step 5: Choose a starting point

Our origin is the Purdue Memorial Union, in the center of campus. You will set it inside the shortest path tool in the next step, either by clicking on the map or by typing its coordinates:

507575, 4474900 [EPSG:26916]
Discussion

Why start with one origin?

Why might it be better to start with a single origin before moving on to many homes or neighborhoods?

A single origin keeps the first analysis easy to read. You can inspect the routes, compare stores and find problems with snapping or connectivity before moving to a more complex workflow with many origins.

Step 6: Shortest paths to every supermarket

  1. In the Processing Toolbox, search for Shortest path (point to layer) and open it.
  2. Set the parameters:
    • Vector layer representing network: roads_utm
    • Path type to calculate: Shortest
    • Start point (x, y): click the … button, then click the Purdue Memorial Union on the map, or paste the coordinates from Step 5
    • Vector layer with end points: supermarkets_utm
    • Shortest path: leave as a temporary layer
    • Non-routable features: leave as a temporary layer
  3. Click Run.

You now have a line layer, Shortest path, with one route from the origin to each supermarket. Its cost field is the route length in meters.

Callout

When a store cannot be reached

If a store cannot be reached through the network, its route has an empty (NULL) cost. Stores that cannot be connected to the network at all are listed in the Non-routable features layer. Both usually point to a data problem: a store far from any road, a gap in the network, or a network clipped too tightly.

Step 7: Compare network and straight-line distance

  1. Style the Shortest path layer with a bold, bright line, the supermarkets with a distinct point symbol, and the roads in light gray.
  2. Open the attribute table of Shortest path and click the Field Calculator button (the abacus icon).
  3. Create a new field route_km (Decimal number) with the expression "cost" / 1000.
  4. Create another field straight_km with the expression distance(start_point($geometry), end_point($geometry)) / 1000. This is the straight-line distance between the two ends of the route.
  5. Create a third field detour with the expression "route_km" / "straight_km".
  6. Close the Field Calculator, turn off editing (save the edits), and sort the table by route_km.
Two maps of Lafayette and West Lafayette with an origin star at the Purdue Memorial Union and 17 black supermarket points. Left: purple straight lines radiate from the origin to each supermarket, averaging 5.0 kilometers. Right: orange routes follow the streets to the same stores, averaging 5.7 kilometers, crossing the Wabash River on the bridges near downtown Lafayette.
Straight-line distance compared with network distance.

The nearest supermarket by road is about 1.3 km away. On average, routes are about 14% longer than straight lines, and some are much longer: the Walmart Supercenter on Veterans Memorial Parkway is 8.0 km away in a straight line but about 10.1 km by road, a detour of about 1.3. The river, the few bridges, railroads and cul-de-sacs all force detours.

Challenge

Interpreting route results

A store looks close on the map, but its route is much longer than expected. Name two possible reasons.

  • The street network requires a detour, for example because there is no bridge or direct road
  • The store or the origin snapped to a different road segment than you expected
  • Barriers such as rivers, railroads, highways with few crossings, or cul-de-sacs
  • The store point is slightly off its true entrance, for example in the middle of a large parking lot

Step 8: Service areas around supermarkets

Shortest paths answer the question: how do I get from one place to each store?

Service areas answer a different question: which streets are within a given distance of a store?

  1. In the Processing Toolbox, search for Service area (from layer) and open it.
  2. Set the parameters:
    • Vector layer representing network: roads_utm
    • Vector layer with start points: supermarkets_utm
    • Path type to calculate: Shortest
    • Travel cost: 800 (meters, about a 10-minute walk)
  3. Click Run.

The output, Service area (lines), shows the parts of the street network you can reach within 800 m of a supermarket.

Map of Lafayette and West Lafayette roads in light gray, with green street segments around each of the 17 supermarkets showing the streets within 800 meters along the network. The green areas are small clusters around stores, mostly along the commercial corridors; most streets, including downtown Lafayette and much of the Purdue campus area, are gray.
Streets within 800 meters of a supermarket.

Only about 11% of the street length inside the two cities is within an 800 m walk of a supermarket. Most supermarkets sit on commercial corridors at the edges of the cities, designed for people who drive.

Step 9: Interpret the results

Service areas support questions such as:

  • Which parts of the study area appear well served by the current supermarkets?
  • Which neighborhoods appear poorly connected?
  • How does the answer change with a travel cost of 400 m, 800 m or 1,600 m?

Combine this with the census map of households without a car, made in notebook 1c of the Data Collection module’s Census episode:

Two choropleth maps of the share of households without a vehicle by census tract, in five classes from light yellow (0 to 3 percent) to dark red (15 to 41 percent). Left, all of Tippecanoe County: rural tracts are light and many are hatched as low reliability. Right, a zoom on Lafayette and West Lafayette: the darkest tracts are near downtown Lafayette and the Purdue campus, such as tracts 4, 55 and 105. Tract 104, on campus, is gray because it has no households.
Households without a vehicle by census tract in Tippecanoe County.

Of the six census tracts with the highest share of households without a car, four (tracts 4, 7, 105 and 112, in central Lafayette and around the Purdue campus) have almost no streets within 800 m of a supermarket. That is exactly where a food access problem would be expected.

Remember that the result is only as good as the network and the assumptions behind it:

  • A road centerline layer does not show sidewalks, crosswalks or footpaths
  • Highways appear walkable in the network, but are not
  • Verified store points improve the destinations, but cannot fix a poor road layer
Discussion

Reflection

What parts of real-world access are still missing from this analysis, even after using a network instead of straight-line distance?

Possible answers include sidewalk quality, crosswalks, traffic safety, bus routes, store hours, prices, food quality, disability access, and whether a road is really walkable.

Challenge

Change the distance

Run Service area (from layer) again with a travel cost of 1600 (about a 20-minute walk). How does the picture change?

The share of city streets within reach rises from about 11% to about 40%, and the green areas begin to join along the main corridors. But some areas are still left out: tract 105, beside the Purdue campus, where more than 40% of households have no car, has no streets within 1,600 m of a supermarket along the network. Walking 20 minutes each way with groceries is a real burden, which is why access studies also consider car ownership and public transit.

Step 10: Save your outputs

Temporary layers disappear when QGIS closes. To keep them, right-click each one and choose Export > Save Features As…, and save it to a GeoPackage (for example grocery_access.gpkg) with a different Layer name for each:

  • supermarkets_utm
  • roads_utm
  • shortest_paths
  • service_area_800m

Then save the project (Project > Save). These outputs can be reused to join service areas to census data, to design a map layout, or to compare neighborhoods.

Suggested pacing: loading and inspection 10 minutes; extracting and reprojecting 15; shortest paths 15; comparing distances 10; service areas 15; discussion and challenges 15.

Teaching strategy: keep the lesson focused on one idea: network distance differs from straight-line distance. Connect it to the Data Collection census map of households without a car (shown in Step 9).

Common learner problems:

  • the project CRS is left in EPSG:4326, so costs appear in degrees
  • layers are not reprojected consistently
  • the road network is clipped too tightly (show the clipping figure)
  • learners confuse service areas with buffers (a buffer ignores roads)
  • outputs are left as temporary layers and lost

Extensions: compare shortest and fastest paths; build a layer of several origins (residence halls or apartment complexes); convert service area lines to polygons with the Concave hull tool; join service areas to census tracts.

Key Points
  • Grocery store points become far more useful for access questions when combined with a road network
  • Reproject layers and set the project CRS to a projected system so that costs are in meters
  • Shortest path tools compare routes; service area tools show which streets are within reach
  • Network distance is longer than straight-line distance, sometimes by much more than expected
  • Clipping a network to an irregular boundary can break routes; check for unreachable destinations