Data Maps
Mark Padgham
2016-06-01
This vignette extends from the vignette (Basic maps) to demonstrate how osmplotr enables the graphical properties of OpenStreetMap objects to be modified according to user-provided data. Categorical data can be plotted by highlighting defined regions with different colours using add_osm_groups, while continuous data can be plotted with add_osm_surface.
library (maptools) # Needed for this vignette
As in the first vignette, maps produced in this vignette contain data for a small portion of central London, U.K.
bbox <- get_bbox (c(-0.13,51.51,-0.11,51.52))
1. Categorical data: add_osm_groups
The function add_osm_groups enables spatially-defined groups to be plotted in different colours. The two primary arguments are obj, which defines the OSM structure to be used for plotting the regions, and groups which is a list of SpatialPoints objects defining the desired regions. An example of an obj is the SpatialPolygonsDataFrame of buildings downloaded in the first vingette with the following line
dat_B <- extract_osm_objects (key='building', bbox=bbox)
The most direct way to define groups is through specifying coordinates of boundary points:
pts <- sp::SpatialPoints (cbind (c (-0.115, -0.125, -0.125, -0.115),
c (51.513, 51.513, 51.517, 51.517)))
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_groups (map, dat_B, groups=pts, cols='orange', bg='gray40')
print_osm_map (map)
Multiple groups can be defined by passing a list of SpatialPoints objects to the groups argument of add_osm_groups, and specifying corresponding colours.
pts2 <- sp::SpatialPoints (cbind (c (-0.111, -0.1145, -0.1145, -0.111),
c (51.517, 51.517, 51.519, 51.519)))
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_groups (map, dat_B, groups=list (pts, pts2),
cols=c ('orange', 'tomato'), bg='gray40')
print_osm_map (map)
The bg argument specifies the colour of any objects lying outside the boundaries of the specified groups. If this argument is not given, then all objects are assigned to the nearest group, so that the groups fill the entire map.
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_groups (map, dat_B, groups=list (pts, pts2),
cols=c ('orange', 'tomato'))
print_osm_map (map)
1.1 Hulls around groups
add_osm_groups includes the argument make_hull which specifies whether convex hulls should be fitted around the points defining the provided groups, or whether the groups already define their own boundaries (the default behaviour). If a point is added internal to the four points defining the first of the above groups, then the group boundary will connect to that point and create a concave shape.
pts <- sp::SpatialPoints (rbind (coordinates (pts), c (-0.12, 51.515)))
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_groups (map, dat_B, groups=pts, cols='orange', bg='gray40')
print_osm_map (map)
The previous points started in the south-east and ended in the north-east, and thus the concave boundary extends in between the two easterly points. Setting make_hull=TRUE defines groups by the convex hulls surrounding them, which in this case would revert this map to the initial map with the group defined by a regular, convex perimeter.
1.2 Inclusive, exclusive, and bisected polygons
The highlighted regions of the previous maps are irregular because the default behaviour of add_osm_groups is to include within a group only those OSM objects which lie entirely within a group boundary. add_osm_groups has a boundary argument which defines whether objects should be assigned to groups inclusively (boundary>0) or exclusively (boundary<0), or whether they should be precisely bisected by a group boundary (boundary=0). The previous maps illustrate the default option (boundary=-1), while the two other options produce the following maps.
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_groups (map, dat_B, groups=list (pts, pts2), make_hull=TRUE,
cols=c ('orange', 'tomato'), bg='gray40', boundary=1)
print_osm_map (map)
The inclusive option (boundary>0) includes all objects which have any points lying within a boundary, meaning more objects are included resulting in larger regions than the previous default exclusive option. Precisely bisecting boundaries produces the following map.
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_groups (map, dat_B, groups=list (pts, pts2), make_hull=TRUE,
cols=c ('orange', 'tomato'), bg='gray40', boundary=0)
print_osm_map (map)
The ability to combine different kinds of boundaries is particularly useful when highlighting areas which partially contain large polygons such as parks. The parks within the following maps were downloaded with
dat_P <- extract_osm_objects (key='park', bbox=bbox)
(Noting that, as described in the first vignette, Basic maps, both extract_osm_objects and make_osm_map convert several common keys to appropriate key-value pairs, so
osm_structures (structure='park')
## structure key value suffix cols
## 1 park leisure park P #647864FF
## 2 background gray20
reveals that this key is actually converted to key='leisure' and value='park'.)
Plotting buildings inclusively within each group and parks bisected by the group boundaries produces the following map:
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_groups (map, dat_B, groups=list (pts, pts2), make_hull=TRUE,
cols=c ('orange', 'tomato'), bg='gray40', boundary=0)
col_park_in <- rgb (50, 255, 50, maxColorValue=255)
col_park_out <- rgb (50, 155, 50, maxColorValue=255)
map <- add_osm_groups (map, dat_P, groups=list (pts, pts2),
cols=rep (col_park_in, 2), bg=col_park_out, boundary=0)
print_osm_map (map)
Bisection divides single polygons to form one polygon of points lying within a given boundary and one polygon of points lying outside the boundary. The two resultant polygons are often separated by visible gaps between locations at which they are defined. Because the layers of a plot are progressively overlaid, such gaps can be avoided by initially plotting underlying layers using add_osm_objects prior to grouping objects:
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_objects (map, dat_P, col=col_park_out)
map <- add_osm_groups (map, dat_P, groups=list (pts, pts2),
cols=rep (col_park_in, 2), bg=col_park_out, boundary=0)
map <- add_osm_groups (map, dat_B, groups=list (pts, pts2), make_hull=TRUE,
cols=c ('orange', 'tomato'), bg='gray40', boundary=0)
print_osm_map (map)
Bisections with boundary=0 will only be as accurate as the underlying OSM data. This example was chosen to highlight that bisection may be inaccurate if actual OSM points do not lie near to a desired bisection line. The larger a map, the less visually evident are likely to be any such inaccuracies. Finally, note that the plot order was changed to allow the building within the park to be overlaid upon the grass surfaces. Plot order, whether controlled manually or with make_osm_map, may often have to be tweaked to appropriately visualise all objects.
The boundary argument has no effect if bg is not given, because in this case all objects will be assigned to a group and there will be no boundaries between groups and other, non-grouped objects.
1.3 Adjusting colours with adjust_colours
The adjust_colours function allows different groups to be highlighted with slightly different colours for different kinds of OSM objects. For example, the following code adds highways to the above map in slightly darkened versions of the highlight colours (using boundary=1, so any highways with any points lying within the bounding box are included in the groups):
dat_HP <- extract_osm_objects (key="highway", value="primary", bbox=bbox)
dat_H <- extract_osm_objects (key="highway", value="!primary", bbox=bbox)
# darken colours by aboud 20%
cols_adj <- adjust_colours (c ('orange', 'tomato'), adj=-0.2)
map <- add_osm_groups (map, dat_HP, groups=list (pts, pts2), make_hull=TRUE,
cols=cols_adj, bg=adjust_colours ('gray40', adj=-0.4),
boundary=1, size=2)
map <- add_osm_groups (map, dat_H, groups=list (pts, pts2), make_hull=TRUE,
cols=cols_adj, bg=adjust_colours ('gray40', adj=-0.2),
boundary=1, size=1)
print_osm_map (map)
(And of course adjust_colours ('gray40', adj=-0.2) is nothing other than ‘gray32’, and adj=-0.4 gives ‘gray24’.)
1.4 Dark-on-Light Highlights
A particularly effective way to highlight single regions within a map is through using dark colours upon otherwise light coloured maps.
map <- osm_basemap (bbox=bbox, bg='gray95')
map <- add_osm_groups (map, dat_B, groups=pts, cols='gray40', bg='gray85',
boundary=1)
map <- add_osm_groups (map, dat_H, groups=pts, cols='gray20', bg='gray70',
boundary=0)
map <- add_osm_groups (map, dat_HP, groups=pts, cols='gray10', bg='white',
boundary=0, size=1)
print_osm_map (map)
1.5 Visualising clustering data
One of the most likely uses of add_osm_groups is to visualise statistical clusters. Clustering algorithms will generally produce membership lists which may be mapped onto spatial locations. Each cluster can be converted to a data frame of SpatialPoints, and the groups specified as a list, each item of which represents one cluster. A general approach is illustrated here with groups defined by single, randomly generated points.
set.seed (2)
ngroups <- 12
x <- bbox [1,1] + runif (ngroups) * diff (bbox [1,])
y <- bbox [2,1] + runif (ngroups) * diff (bbox [2,])
groups <- cbind (x, y)
groups <- apply (groups, 1, function (i)
sp::SpatialPoints (matrix (i, nrow=1, ncol=2)))
Having generated the points, a map of corresponding clusters can be generated by the following simple code.
map <- osm_basemap (bbox=bbox, bg='gray95')
map <- add_osm_groups (map, dat_B, groups=groups,
cols=rainbow (length (groups)))
print_osm_map (map)
Although individual groups will generally be defined by collections of multiple points, this example illustrates that they can also be defined by single points. In such cases, the bg option should of course be absent, so that all remaining points are allocated to the nearest groups.
This map also illustrates the kind of visual mess that may arise in attempts to specify colours, particularly because the sequence of colours passed to add_osm_groups will generally not map on to any particular spatial order, so even if a pleasing colour scheme is submitted, the results may still be less than desirable. Although it may be possible to devise pleasing schemes for small numbers of groups, manually defined colour schemes are likely to become impractical for larger numbers of groups.
map <- osm_basemap (bbox=bbox, bg='gray95')
map <- add_osm_groups (map, dat_B, groups=groups, borderWidth=2,
cols=heat.colors (length (groups)))
print_osm_map (map)
Note the submitting any positive values to the additional borderWidth argument causes add_osm_groups to drawn convex hull borders around the different groups. Even this is not sufficient, however, to render the result particularly visually pleasing or intelligible. To overcome this, add_osm_groups includes an option described in the following section to generate spatially sensible colour schemes for colouring distinct groups.
1.6 The Colour Matrix: Colouring Several Regions
An additional argument which may be passed to add_osm_groups is colmat, an abbreviation of ‘colour matrix’. If set to true (the default is FALSE), group colours are specified by the function colour_mat. This function takes a vector of four or more colours as input, wraps them around the four corners of a rectangular grid, and spatially interpolates a chromatially regular grid between these corners. To visual different schemes, it has a plot argument:
cmat <- colour_mat (plot=TRUE)
This grid illustrates the default colours, rainbow (4). The two-dimensional colour field produced by colour_mat may also be rotated by a specified number of degrees using the rotate argument.
cmat <- colour_mat (n=c(4,8), rotate=90, plot=TRUE)
This example also illustrates that the size of colour matrices may also be arbitrarily specified. Using the colmat option in add_osm_groups enables the previous maps to be redrawn like this:
map <- osm_basemap (bbox=bbox, bg='gray95')
map <- add_osm_groups (map, dat_B, groups=groups, borderWidth=2, colmat=TRUE,
cols=c('red','green','yellow','blue'), rotate=180)
print_osm_map (map)
Note both that when add_osm_groups is called with colmat=TRUE, then cols need only be of length 4, to specify the four corners of the colour matrix, and also that the rotate argument can be submitted to add_osm_groups and passed on to colour_mat.
1.7 Bounding areas within named highways
As explained in the first vignette, Basic maps, the function connect_highways takes a list of OSM highway names and a bounding box, and returns the boundary of a polygon encircling the named highways. This can be used to highlight selected regions simply by naming the highways which encircle them, producing maps which look like this:
highways <- c ('Monmouth.St', 'Short.?s.Gardens', 'Endell.St', 'Long.Acre',
'Upper.Saint.Martin')
highways1 <- connect_highways (highways=highways, bbox=bbox)
highways <- c ('Endell.St', 'High.Holborn', 'Drury.Lane', 'Long.Acre')
highways2 <- connect_highways (highways=highways, bbox=bbox, plot=T)
highways <- c ('Drury.Lane', 'High.Holborn', 'Kingsway', 'Great.Queen.St')
highways3 <- connect_highways (highways=highways, bbox=bbox, plot=T)
Note the use of the regex character ? in the first list of highway names, denoting the previous character as optional. This is necessary here because there are OSM sections named both “Shorts Gardens” and “Short’s Gardens”.
class (highways1); length (highways1); length (highways2); length (highways3)
## [1] "SpatialPoints"
## attr(,"package")
## [1] "sp"
## [1] 55
## [1] 47
## [1] 55
connect_highways returns a list of SpatialPoints representing the shortest path that sequentially connects all of the listed highways. (Connecting all listed highways may not necessarily be possible, in which case warnings will be issued. As described in the first vignette, Basic maps, connect_highways also has a plot option allowing problematic cases to be visually inspected and hopefully corrected.)
These lists of highway coordinates can then be used to highlight the areas they encircle. First group the highways and establish a colour scheme for the map:
groups <- list (london$highways1, london$highways2, london$highways3)
cols_B <- c ('red', 'orange', 'tomato') # for the 3 groups
cols_H <- adjust_colours (cols_B, -0.2)
bg_B <- 'gray40'
bg_H <- 'gray60'
And then plot the map.
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_objects (map, dat_P, col=col_park_out)
map <- add_osm_groups (map, dat_B, groups=groups, boundary=1,
bg=bg_B, col=cols_B)
map <- add_osm_groups (map, dat_H, groups=groups, boundary=1,
bg=bg_H, col=cols_H)
map <- add_osm_groups (map, dat_HP, groups=groups, boundary=0,
cols=cols_H, bg=bg_H, size=1)
print_osm_map (map)
These encircling highways are included in the london data provided with osmplotr.
2. Continuous data: add_osm_surface
The add_osm_surface function enables a continuous data surface to be overlaid on a map. User-provided data is spatially interpolated across a map region and OSM items coloured according to a specified continuous colour gradient. The data must be provided as a data frame with three columns, ‘(x,y,z)’, where ‘(x,y)’ are the coordinates of points at which data are given, and ‘z’ are the values to be spatially interpolated across the map.
A simple data frame can be constructed as
n <- 5
x <- seq (bbox [1,1], bbox [1,2], length.out=n)
y <- seq (bbox [2,1], bbox [2,2], length.out=n)
dat <- data.frame (
x=as.vector (array (x, dim=c(n, n))),
y=as.vector (t (array (y, dim=c(n, n)))),
z=x * y
)
head (dat)
## x y z
## 1 -0.130 51.5100 -6.696300
## 2 -0.125 51.5100 -6.439063
## 3 -0.120 51.5100 -6.181800
## 4 -0.115 51.5100 -5.924512
## 5 -0.110 51.5100 -5.667200
## 6 -0.130 51.5125 -6.696300
And then passed to add_osm_surface
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_surface (map, dat_B, dat=dat, cols=heat.colors (30))
print_osm_map (map)
At present, add_osm_surface generates an warning if it is applied more than once to any one kind of Spatial object (polygons or lines), as illustrated in the following code (in which both dat_H and dat_HP are of class SpatialLinesDataFrame:
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_surface (map, dat_HP, dat=dat, cols=heat.colors (30))
map <- add_osm_surface (map, dat_H, dat=dat, cols=heat.colors (30))
## Scale for 'colour' is already present. Adding another scale for
## 'colour', which will replace the existing scale.
This is because add_osm_surface creates new ggplot2 aesthetic schemes for each kind of object, and these schemes are not intended to be modified or replaced within a single plot. The above map may still be printed, but the warning means that the last provided colour scheme will be applied to all objects of that class. This means that osmplotr can only overlay two distinct colour schemes: one for all objects of class SpatialLines, and a potentially different one for all objects of class SpatialPolygons.
Of course, any number of additional objects may be overlaid with add_osm_objects, for example,
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_surface (map, dat_B, dat=dat, cols=heat.colors (30))
cols_adj <- adjust_colours (heat.colors (30), -0.2)
map <- add_osm_surface (map, dat_HP, dat=dat, cols=cols_adj, size=1.5)
map <- add_osm_objects (map, dat_P, col=rgb (0.1,0.3,0.1))
map <- add_osm_objects (map, dat_H, col='gray60')
print_osm_map (map)
2.1 Colourbar legends for data surfaces
A colourbar legend for the surface may be added with add_colourbar. As with add_axes, this function is provided separately to allow colourbars to be overlaid only after all desired map items have been added. The only parameters required for add_colourbar are the limits of the data (zlims) and the colours (along with the map, a modified version of which is returned).
map <- add_colourbar (map, cols=terrain.colors (100), zlims=range (dat$z))
print_osm_map (map)
Note that the colours submitted to add_colourbar need not be the same as those used to plot the surface. (Although using different colours is rarely likely to be useful.) As for add_axes, and explained in the first vignette, Basic maps, the transparency of the boxes surrounding the elements of the colourbar may be controlled by specifying the value of alpha. Both alignment and position may also be adjusted, as illustrated in this example.
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_surface (map, dat_B, dat=dat, cols=heat.colors (30))
cols_adj <- adjust_colours (heat.colors (30), -0.2)
map <- add_osm_surface (map, dat_HP, dat=dat, cols=cols_adj, size=1.5)
map <- add_colourbar (map, cols=heat.colors (100), zlims=range (dat$z),
alpha=0.9, vertical=FALSE,
barwidth=c(0.1,0.12), barlength=c(0.5,0.9),
text_col="blue", fontsize=5, fontface=3,
fontfamily="Times")
map <- add_axes (map, colour="blue", fontsize=5, fontface=3,
fontfamily="Times")
print_osm_map (map)
Both barwidth and barlength can be specified in terms of one or two numbers. A single value for barwidth determines its relative width (0-1) from the border (the right side if vertical=TRUE or the top if vertical=FALSE), while two values determine the relative start and end positions of the sides of the bar. A single value for barlength produces a bar of the given length centred in the middle of the map, while two values determine its respective upper and lower points (for vertical=TRUE) or left and right points (for vertical=FALSE).
This example also demonstrates how colours, sizes, and other font characteristics of text labels can be specified (with text_col determining the colour of all elements of the colourbar other than the gradient itself). Finally, as for add_axes, the text labels of colourbars are not currently able to be rotated because ggplot2 does not permit rotation for the geom_label function used to produce these labels.
2.1 Surfaces and data perimeters
It may often be that user-provided data only extend across a portion of a map, leaving a perimeter beyond the data boundary for which interpolation should not be applied. add_osm_surface has a bg parameter specifying a background colour for objects beyond the perimeter of the data surface. Passing this parameter to add_osm_surface causes objects beyond the data perimeter to be coloured within this ‘background’ colour.
To illustrate, trim the above data to within a circular range of the centre of the map.
d <- sqrt ((dat$x - mean (dat$x)) ^ 2 + (dat$y - mean (dat$y)) ^ 2)
range (d)
## [1] 0.00000000 0.01118034
Remove from dat all rows translating to d>0.01:
dat <- dat [which (d < 0.01),]
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_surface (map, dat_B, dat=dat, cols=heat.colors (30),
bg="gray40")
cols_adj <- adjust_colours (heat.colors (30), -0.2)
map <- add_osm_surface (map, dat_HP, dat=dat, cols=cols_adj, size=c (1.5, 0.5),
bg="gray70")
print_osm_map (map)
(The perimeter is irregular because of the positions of the points in dat.)
2.3 Further control of surface appearance
The final add_osm_surface call in the above code (for dat_HP) illustrates additional parameters that may be passed for further control of map appearance. In this case, the two size parameters control the size of the lines within the data surface and beyond its perimeter. Single values may also be passed, in which case they determine the width of lines in both cases. One or two shape parameters may also be passed, with these also determining the shapes of SpatialPoints, as illustrated in the next example, which overlays trees on the map.
Both lines and points use the same ggplot2 colour gradient, and so adding the second of these again generates an error and means that the actual colour scheme will be determined by the final call to add either lines or points.
dat_T <- extract_osm_objects (key='tree', bbox=bbox)
map <- osm_basemap (bbox=bbox, bg='gray20')
map <- add_osm_surface (map, dat_HP, dat=dat, cols=terrain.colors (30),
size=c (1.5, 0.5), bg="gray70")
map <- add_osm_surface (map, dat_H, dat=dat, cols=terrain.colors (30),
size=c (1, 0.5), bg="gray70")
map <- add_osm_surface (map, dat_T, dat=dat, cols=heat.colors (30),
bg="lawngreen", size=c(3, 2), shape=c(8, 1))
print_osm_map (map)
## Scale for 'colour' is already present. Adding another scale for
## 'colour', which will replace the existing scale.
## Scale for 'colour' is already present. Adding another scale for
## 'colour', which will replace the existing scale.
The first two colour specificiations (terrain.colors) have been ignored, and all added items are coloured according to the final value of heat.colors (30). Other aspects such as line sizes and point shapes are nevertheless respected.
