Introduction to the analysis of spatial data using R

Dr. Ana I. Moreno-Monroy

15-18 November 2016

Chapter 2: Spatial data

Spatial data: definition and types

Spatial objects and projections
Merging statistical information into spatial objects
Additional operations with polygons

Spatial data

Spatial data is data that contains information about geographical locations and/or about the shapes of
the geographical features (geometry)
Spatial data can be stored as precise (point) coordinates (lat-lon/XY ) or topology (i.e., the geometric
boundaries of municipalities, countries, etc.)
A (postal) address or zip-code is not spatial data. It needs to be geolocated in order to obtain point
coordinates
In order to visually represent (i.e. map) an administrative unit, we need information about its geometry
(lines/polygons)

Shapefiles

Polygon boundaries are a collection of lines that form shapes. These are usually contained in shapefiles
The shapefile format is a geospatial vector data format for geographic information system (GIS) software.
Shapefile folders include at least four files with extensions: .shp (the feature geometry itself); .shx (a
positional index of the feature geometry); .dbf (columnar attributes). They can also include a .prj file
(with the coordinate system and projection information)

R and GIS

Spatial data is usually managed in a Geographic Information System (GIS) software (e.g. ArcGIS,
QGIS)
Some packages in R make it possible to handle and analyze spatial data. These include sp, rgdal,
matpools, rgeos and spatstat
In R it is possible to geolocate addresses, manipulate and visualize spatial data and perform spatial
statistical analyses

Spatial data: an example

Super Neighborhoods division polygon data is freely available at the City of Houston GIS open data
website
We can use the function readOGR from the package rgdal to import this data
library(rgdal)
neigh <- readOGR(dsn = "Super_Neighborhoods", layer = "Super_Neighborhoods")

1
## OGR data source with driver: ESRI Shapefile
## Source: "Super_Neighborhoods", layer: "Super_Neighborhoods"
## with 88 features
## It has 11 fields
Other options for importing shapefiles include the readSpatialPoly function of the maptools package
(without projection data)

Inspection

Neigh is a SpatialPolygonDataFrame with several slots, two of which are @data (containing variables
associated to each polygon) and @polygons
We can work with @data the same way we work with a regular data object
str(neigh@data)

## 'data.frame': 88 obs. of 11 variables:

## $ OBJECTID : int 1 2 3 4 5 6 7 8 10 11 ...
## $ PERIMETER : num 16572 43119 39256 59785 75759 ...
## $ POLYID : int 60 63 61 59 21 5 78 73 17 22 ...
## $ SNBNAME : Factor w/ 88 levels "ACRES HOME","ADDICKS PARK TEN",..: 26 68 16 14 36 33 32 27 22 82
## $ cohgisCOHG: int 0 0 0 0 0 0 0 0 0 0 ...
## $ cohgisCO_1: int 0 0 0 0 0 0 0 0 0 0 ...
## $ COUNCIL_AC: Factor w/ 2 levels "No","Yes": NA 2 NA 2 NA 2 NA 2 2 2 ...
## $ RECOGNITIO: Factor w/ 20 levels "2000-03-06T00:00:00.000Z",..: NA 14 NA 5 NA 2 NA 6 20 1 ...
## $ SnbrInfoUR: Factor w/ 88 levels "http://www.houstontx.gov/superneighborhoods/profiles/SN_1.htm",..
## $ WeCan : Factor w/ 1 level "Y": 1 1 NA 1 NA NA NA NA NA NA ...
## $ Top10 : Factor w/ 1 level "Y": NA 1 NA NA NA NA NA NA NA NA ...

NA values

The is.na() function is useful to handle NA values

table(is.na(neigh@data$COUNCIL_AC))

##
## FALSE TRUE
## 60 28
The complete.cases function reports TRUE for cases without any NA values (it can be used to subset
as well)
table(complete.cases(neigh@data))

##
## FALSE TRUE
## 80 8

Map projections

Every spatial data has its own Coordinate System, which can be either a lat-lon system (Geographic
Coordinate System) or a projected coordinate system (map projection)
A map projection is always necessary to create a map. Projections are a way to represent the 3-D
surface of the earth on a 2-D plane

2
 Different spatial operations require either lat-lon or projected coordinate systems. In any case, the
systems should be the same when working with super-imposed layers
Please take a moment to read these Projection Basics

Finding out the Coordinate System

To find out the CRS of a spatial object, open .prj file in shapefile folder, or in R type
proj4string(neigh)

## [1] "+proj=longlat +datum=WGS84 +no_defs +ellps=WGS84 +towgs84=0,0,0"

A Geographic Coordinate System (Lat/Lon coordinates) does not have a projection. It is represented
in degrees, so distance is not measured in usual distance units (meters, kilometers)
The following map shows the axes values in degrees for the unprojected spatial data
plot(neigh, axes=TRUE)
29.5N29.6N29.7N29.8N29.9N 30N 30.1N

95.8W 95.6W 95.4W 95.2W 95W

Projecting coordenates

To find out the equivalent projection to a GCS, see this Spatial Reference website

3
 Transformations from unprojected to projected systems (and vice-versa) can be made in R using the
spTransform function of the sp package
library(sp)
neigh <- spTransform(neigh, CRS("+init=epsg:3673"))

The axes of the map are now in a different unit

plot(neigh, axes=TRUE)
4260000
4240000
4220000
4200000

920000 940000 960000 980000

Selecting a single area

Suppose we are only interested in the area DOWNTOWN. We can construct a TRUE/FALSE variable to
subset the spatial object
sel <- neigh@data$SNBNAME=="DOWNTOWN"
downtown<-neigh[sel, ]
plot(downtown, axes=TRUE)

4
 4219000
4217000

949000 951000 953000 955000

Merging external datasets

More information by Super Nieghborhood from the 2010 US Census is available at the City of Houston
GIS open data website in .csv format
We can import this csv into R using the read.csv function (see ?read.csv)
SN_data<-read.csv("Census_2010_By_SuperNeighborhood.csv", header=TRUE)

Alternatively, we can download and import the data directly using read.csv2
SN_data<-read.csv2("http://arcg.is/2dZ7naE", sep=",")

The variable POLYID in SN_data is also in neigh@data. It is a unique identifier of the spatial units
(Super Neighborhoods)
We can merge the SN_data into the neigh SpatialPointsDataFrame using this unique identifier
neigh<-merge(neigh, SN_data, by="POLYID")

We can subset to keep the variables we are interested in

keep<-c("POLYID", "SNBNAME", "SUM_TotPop")
neigh1<-neigh[,colnames(neigh@data) %in% keep]

The new shapefile can be exported using the writeOGR function

Handling merged data

We can make basic queries on the merged data the same way we would with a regular data object. For
instance, we could find out how many people in all neighborhoods, or how many people live Downtown

5
sum(neigh@data$SUM_TotPop)

## [1] 2066749
neigh@data$SUM_TotPop[neigh@data$SNBNAME=="DOWNTOWN"]

## [1] 16716
To add more criteria (e.g., more neighborhoods) use & for AND and | for OR
neigh@data$SUM_TotPop[neigh@data$SNBNAME=="DOWNTOWN" | neigh@data$SNBNAME=="MEMORIAL"]

## [1] 16716 45294

Simple cloropeth map

We can use the spplot to get a simple cloropeth map of any continuos variable
spplot(neigh, "SUM_TotPop", main = "Population distribution", col = "transparent")

Population distribution

1e+05

8e+04

6e+04

4e+04

2e+04

0e+00

Cloropeth map: more options

We can manipulate all the attributes of the cloropeth map by modifying different options. Unfortunately,
there is no friendly Map editor (as in QGIS or ArcMap) for these tasks

6
 To change the colors, we have to first choose a palette from the RColorsBrewer package options (type
display.brewer.all() to see the options) and then define the number of cuts. In this case the palette is
Orange Red and the number of cuts 7 (which becomes 6 when passed on the spplot function)
library(RColorBrewer)
my.palette <- brewer.pal(n = 7, name = "OrRd")
spplot(neigh, "SUM_TotPop", col.regions = my.palette, main = "Population distribution", cuts = 6, col =

Population distribution

1e+05

8e+04

6e+04

4e+04

2e+04

0e+00

Identify specific areas in a plot

We can subset polygons in the same way we subset data based on a condition
plot(neigh, axes=TRUE)
sel <- neigh@data$SUM_TotPop>mean(neigh@data$SUM_TotPop)
plot(neigh[sel, ], axes=TRUE, col="dark red", add=TRUE)

7
 4260000
4230000
4200000

900000 940000 980000

Find centroid of an area

The function gCentroid of the rgeos package allows us to find the geometric center (centroid) of a
map or a single polygon
library(rgeos)
plot(neigh, axes = TRUE)
center <- gCentroid(neigh[neigh$SNBNAME == "DOWNTOWN",])
plot(center, cex = 0.3, col="red", add=TRUE)

8
 4260000
4230000
4200000

900000 940000 980000

Use the option byid=TRUE to obtain the centroids of each polygon

center_all <- gCentroid(neigh, byid=TRUE)

Select areas touching a buffer

Buffers around a point are often used to define areas of interest. The function gBuffer of the rgeos package
allows us to calculate buffers of different width (in meters) around a given spatial geometry
plot(neigh, axes=T)
buffer_2k <- gBuffer(center, width = 2000)
plot(neigh[buffer_2k,], col = "lightblue", add = T)
plot(buffer_2k, add = T, border = "red", lwd = 2)

9
 4260000
4230000
4200000

880000 920000 960000 1000000

Calculating a distance matrix between centroids

We can calculate the euclidian distance between every pair of centroids (as an approximation for the distance
between areas) using the distm function of the geosphere package. As it requires points in degrees (lon/lat),
we have to transform back our spatial object to the original CRS
library(geosphere)
neigh <- spTransform(neigh, CRS("+proj=longlat +datum=WGS84 +no_defs +ellps=WGS84 +towgs84=0,0,0"))
center_all <- gCentroid(neigh, byid=TRUE)
dist_matrix <- distm(center_all)
dist_matrix[1:5,1:5]

## [,1] [,2] [,3] [,4] [,5]

## [1,] 0.000 5151.180 2094.728 11450.322 9166.35
## [2,] 5151.180 0.000 3077.924 6310.274 14285.42
## [3,] 2094.728 3077.924 0.000 9360.479 11260.36
## [4,] 11450.322 6310.274 9360.479 0.000 20595.69
## [5,] 9166.350 14285.421 11260.359 20595.694 0.00

Testing for spatial autocorrelation

Variables may have a spatial pattern, meaning that they seem to be more clustered or dispersed than would
be expected by random location. To check for this, we first plot the spatial distribution of the variable of
interest (in this case Black population)
spplot(neigh, "SUM_NH_Black", col.regions = my.palette, main = "Number of Blacks", cuts = 6, col = "tran

10
 Number of Blacks

25000

20000

15000

10000

5000

Constructing a neighbors list

To find out the extent of spatial autocorrelation in our variable, we need to define some criteria for
what we consider being a neighbor. The first step in this direction is to construct a neighbors list
using the poly2nb function of the spdep package.
library(spdep)
neigh_nb <- poly2nb(neigh)
head(neigh_nb, n=3)

## [[1]]
## [1] 3 10 78 85
##
## [[2]]
## [1] 3 11 12 73 77 82
##
## [[3]]
## [1] 1 2 10 11 19 75 77 78
Here the default contiguity condition is Queen, meaning that areas sharing any boundary point are
considered neighbors
The object neigh_nb contains a list with the ids of neighbors of each polygon

Plot neighbors

We can visualize the nb object in a plot, together with the coordinates of the centroids of each polygon

11
coords<-coordinates(neigh)
plot(neigh)
plot(neigh_nb, coords, add=T)

Construct weights between neighbors

The next step is to construct weights between neighbors. There are many coding schemes that can be chosen
to represent the spatial relationship between areas. In this simple example, we assume row-standardized
weights between neighbors, so that areas with more neighbors weight more than those with less. We do this
using the nb2listw function
neigh_lw_W <- nb2listw(neigh_nb, zero.policy = T)
print(neigh_lw_W, zero.policy=T)

## Characteristics of weights list object:

## Neighbour list object:

12
## Number of regions: 88
## Number of nonzero links: 430
## Percentage nonzero weights: 5.552686
## Average number of links: 4.886364
## 2 regions with no links:
## 36 38
##
## Weights style: W
## Weights constants summary:
## n nn S0 S1 S2
## W 86 7396 86 40.11216 353.6972
We have to introduce the option zero.policy = TRUE because we have 2 areas (green areas/parks) with no
neighbors in the original shapefile

Calculating a measure of spatial autocorrelation

Finally we are ready to calculate measures of spatial autocorrelation. In this example we calculate the
Morans I statistic under normality
moran_u <- moran.test(neigh@data$SUM_NH_Black, listw = neigh_lw_W, zero.policy = T, randomisation=FALSE)
moran_u

##
## Moran I test under normality
##
## data: neigh@data$SUM_NH_Black
## weights: neigh_lw_W
##
## Moran I statistic standard deviate = 5.1102, p-value = 1.609e-07
## alternative hypothesis: greater
## sample estimates:
## Moran I statistic Expectation Variance
## 0.354440899 -0.011764706 0.005135344
According to these results, we can reject the hypothesis of no spatial autocorrelation in the distribution of
blacks in Houston

13