Exploring historical maps and spatial data with R and OpenStreetMap

Interactively visualising 19th century proletarian housing in Leuven, the R tidyverse-way

Exploring historical maps and spatial data with R and OpenStreetMap

Interactively visualising 19th century proletarian housing in Leuven, the R tidyverse-way

The combination of spatial open data, standardised spatial web services, and recent innovations in R’s spatial packages ecosystem make it an exiting domain to play with. In this walkthrough I will demonstrate:

  • querying OpenStreetMap for geographic features;
  • processing them with the R sf and dplyr packages;
  • interactively displaying them on historical maps using Leaflet and WMS-webservices.

The geographic features we are interested in are the 19th century “gangen” (Dutch for “passages”) in Leuven, Belgium. Around 1880 Leuven counted about 140 of these gangen, housing 13% of the population of Leuven at that time. They were called passages, because a small passage led you from the public street to terrains behind the ‘regular’ houses. Those terrains were subdivided into small housing units rented by the proletariat and laboring classes in Leuven.

Lets see what spatial open data and webservices can learn us about these gangen today.

Fetch spatial objects from OpenStreetMap

Load the required libraries:

library(osmdata) # fetching OpenStreetmap-data
library(leaflet) # display interactive maps
library(stringr) # optional, tidyverse string-operations
library(dplyr)   # data-manipulation
library(sf)      # spatial features library
library(forcats) # optional, tidyverse factor-manipulation

The gangen were generally called ‘-gang’, e.g. Zandgang, Mathildegang, Blauwputgang, etc., and walking through Leuven you can still find streetnames ending in -gang, denoting the location of historical gangen. To get the spatial location and layout of every street in Leuven ending in -gang, we use OpenStreetMap (OSM).

osmdata is an R package for accessing the data underlying (OSM). It allows us to download all spatial features in a bounding box around Leuven from OSM, and filter on type ‘highway’ to get only the streets:

q.leuven <- opq(bbox = 'Leuven, Belgium') %>% # set bounding box via name
  add_osm_feature(key = 'highway') %>% # get streets
  add_osm_feature(key = 'name') %>% # include the name
  osmdata_sf() # return as a sf-object

## Object of class 'osmdata' with:
##                  $bbox : 50.8242096,4.640295,50.9440707,4.7705305
##         $overpass_call : The call submitted to the overpass API
##                  $meta : metadata including timestamp and version numbers
##            $osm_points : 'sf' Simple Features Collection with 27656 points
##             $osm_lines : 'sf' Simple Features Collection with 5725 linestrings
##          $osm_polygons : 'sf' Simple Features Collection with 109 polygons
##        $osm_multilines : NULL
##     $osm_multipolygons : 'sf' Simple Features Collection with 4 multipolygons

The returned osmdata-object contains about 35.000 spatial features of various types (points, lines, polygons). We are presently only interested in the spatial line representations of the gangen. So we select the osm_lines part of the osmdata-object with the base R $-operator, select() from all the variables returned by OSM only the name-variable, and filter() down the list of spatial lines to only those where the name contains “gang” somewhere.

gangen = q.leuven$osm_lines %>% # use only the lines in the osmdata-object
  select(name) %>% # keep only the name-variabele
  # keep only the lines with 'gang' in the name:
  filter(str_detect(name, 'gang') == TRUE) 

# fix plot issue: https://stackoverflow.com/a/50846998/125085
names(gangen$geometry) <- NULL 

gangen <- gangen %>%
  # Drop unused levels in name-variable w/t forcats::fct_drop():
  mutate(name = fct_drop(name)) %>% 
  arrange(name) # sort alphabetically on the name

## Simple feature collection with 58 features and 1 field
## geometry type:  LINESTRING
## dimension:      XY
## bbox:           xmin: 4.686551 ymin: 50.86957 xmax: 4.71337 ymax: 50.88845
## epsg (SRID):    4326
## proj4string:    +proj=longlat +datum=WGS84 +no_defs
## First 10 features:
##                  name                       geometry
## 1           Beursgang LINESTRING (4.692325 50.880...
## 2           Beursgang LINESTRING (4.692272 50.880...
## 3           Beursgang LINESTRING (4.69243 50.8803...
## 4        Blauwputgang LINESTRING (4.711557 50.881...
## 5        Blauwputgang LINESTRING (4.711378 50.882...
## 6        Blauwputgang LINESTRING (4.711617 50.881...
## 7     Bloemenberggang LINESTRING (4.688926 50.882...
## 8  Boghe of Brandgang LINESTRING (4.696834 50.876...
## 9       Busleidengang LINESTRING (4.699747 50.881...
## 10      Busleidengang LINESTRING (4.699971 50.881...

We now have a sf-collection with streets having ‘gang’ in their name. The objectalso doubles as a familiar R-dataframe with two columns/variables: the OSM name of the spatial line and geometry, which holds the spatial definition of the line.

Using the leaflet-package, which is a wrapper for the JS library in R, we can display this sf-object on an interactive map:

leaflet(gangen) %>%
  addTiles() %>%
    # use linestring name to show as a label on mouse-over:
    label = ~name, 
    # optional: highlight linestring in red on mouse-over:
    highlight = highlightOptions(color = "red")) 

The remaining 26 gangen are all in their expected place. There is however an issue: when a gang is not a simple street but a complex of streets, it is represented by multiple spatial strings. That is why there are e.g. three records for Beursgang in the sf-object, and you can also see this clearly if you mouse-over the Matildegang, the gang the most south-east in Leuven (different lines should be highlighted in red).

This is to be expected. The simple features specification implemented in the sf-package and other spatial libaries is build on basic spatial ‘buildingblocks’, such as points (POINT), lines (LINESTRING), and polygons (POLYGON). And osmdata returns you spatial objects in proper buildingblock-form. So if you want to use multiple simple lines to spatially represent a complex of streets, you need to combine lines into a composit spatial set of lines (MULTILINESTRING).

Luckly, the sf-packages nicely integrates into the tidyverse-ecosystem, which allows us to use common dplyr-verbs to do specific spatial operations. We group the records/lines in the sf-objects on the name of the street with group_by() and aggregate the individual records/lines with the same name using summarise(). This gives us a nice sf-dataframe with the expected 26 records of the remaining gangen in Leuven.

gangen.merged <- gangen %>%
  group_by(name) %>%

##  [1] Beursgang          Blauwputgang       Bloemenberggang   
##  [4] Boghe of Brandgang Busleidengang      Cansgang          
##  [7] Daneelsgang        Elzasgang          Hallengang        
## [10] Hanengang          Horengang          Korbeeklogang     
## [13] Mathildegang       Oppendorpgang      Paardengang       
## [16] Peterseliegang     Pioengang          Puttegang         
## [19] Rapengang          Straatjesgang      Theresianengang   
## [22] Valkerijgang       Vestinggang        Wolvengang        
## [25] Wolvenpoortgang    Zandgang           Zongang           
## 27 Levels: Beursgang Blauwputgang Bloemenberggang ... Zongang

Displaying this aggregated sf-object on a map, shows on mouse-over that the lines are properly joined into the expected gangen-complexes.

leaflet(gangen.merged) %>%
  addTiles() %>%
    label = ~name, 
    highlight = highlightOptions(color = "red")) 

Historical maps using WMS-tileservers

Next to contemporary map-layers, it would be interesting to display the (remaining) gangen on historical maps. Conveniently, the regional Agentschap Informatie Vlaanderen makes tile-sets of historical maps available through publicly available WMS-webservices. For a 19th century phenomenon as the gangen, we will use the historical Ferraris map, and the Vandermaelen map.

The Ferraris map was created between 1770 and 1778, and is the first systematic, large scale mapping in present-day Belgium and Western Europe. The map reflects the detailed state of the Southern Netherlands just before the start of the Industrial Revolution and the end of the Ancien Régime, when the landscape was still similar to that of the Middle Ages.

Directly in Leaflet, we use addWMSTiles() to request the Ferraris map tiles from the WMS-webservice:

leaflet() %>%
  setView(4.69839084, 50.880089, zoom = 14) %>% # center on Leuven
    baseUrl = "https://geoservices.informatievlaanderen.be/raadpleegdiensten/histcart/wms",
    layers = list("ferraris"), # request Ferraris-tiles from available WMS tilesets
    options = WMSTileOptions(format = "image/png"),
    attribution = "Source: GIS Geoservices Informatie Vlaanderen") %>%
  addControl("Leuven on the Ferraris map (1777)", position = "topright")

We can do the same for the Vandermaelen map. This “Carte topographique de la Belgique” of Belgium was published by Belgian cartographer Philippe Vandermaelen between 1846 and 1854. With the relatively late industrialisation of Leuven, the map provides an nice background to contextualise the location of the gangen in the second half of the century.

leaflet() %>% 
  setView(4.69839084, 50.880089, zoom = 14) %>%
    layers = list("vandermaelen"),
    options = WMSTileOptions(format = "image/png"),
    attribution = "Source: GIS Geoservices Informatie Vlaanderen") %>%
  addControl("Leuven on the Vandermaelen map (1846-1854)", position = "topright")

Very visible on the Vandermaelen map is the grid-pattern of (intended) streets in the eastern part of Leuven, laid out in the 1838 “plan-Laenen” of city architect François-Henri Laenen.

Overlay OSM-data on historical maps

We can create a single Leaflet-map with all three layers, the Ferraris map, the Vandermaelen map, and the OSM-map of Leuven. Warning: the projection of both historical maps (EPSG:31370) is different from the default projection used by Leaflet, but specifying that falls outside of the scoop of this walkthrough.

m.leuven <- leaflet() %>% 
  setView(4.69839084, 50.880089, zoom = 14) %>%
    group = 'Ferraris-map (1777)',
    layers = list("ferraris"),
    options = WMSTileOptions(format = "image/png", transparent = FALSE),
    attribution = "Source: GIS Geoservices Informatie Vlaanderen") %>%
    group = 'Vandermaelen-map (1846-1854)',
    layers = list("vandermaelen"),
    options = WMSTileOptions(format = "image/png", transparent = FALSE),
    attribution = "Source: GIS Geoservices Informatie Vlaanderen") %>%
  addTiles(layerId = 'Current map of Leuven') %>% # default OSM-tiles
    position = 'bottomright',
    baseGroups = c(
      'Current map of Leuven',
      'Vandermaelen-map (1846-1854)', 
      'Ferraris-map (1777)'))

As a fourth layer, we add the OSM-data representing the gangen, as filtered above:

m.leuven %>%
    data = gangen.merged,
    label = ~name, 
    highlight = highlightOptions(color = "red")) %>%
  addControl("Contemporary 'gangen' on the Ferraris (1777), Vandermaelen (1846-1854) and current map of Leuven", position = "topright")

By switching layers (bottom-right), this final map allows us to explore the (remaining) Leuvense gangen in the context of different (historical) maps.

For instance, remaining gangen in the south of Leuven such as the Elzasgang, are indicative of the location of traditional medieval poverty-housing patterns in which gangen were located at the edge of the city next to the entrence roads (Naamsestraat). Gangen in the north-west mainly grew in the second half of the 19th century with the late industrialisation of Leuven. It could be in newer areas, or in in streets already existing in medieval times such as the Zandgang in the Fonteinstraat. In the east of Leuven, gangen also grew throughout the 19th century inside the wide streetgrids of the 1838 plan-Laenen.

Maarten Hermans
Senior research associate