Geosphere/dplyr: create matrix of distance between coordinates

Question

I want to create a "matrix" of the distance between multiple coordinates with each other. Preferably using dplyr/geosphere. I already saw that the geosphere package offers this. I managed to create the distance between two vectors but I have difficulties creating the full matrix.

This is the sample table with multiple coordinates.

df <- data.frame(latitude = c(49.48609,-8.14671,11.28625),
                 longitude = c(8.463678,143.05793,-11.18285))

  latitude  longitude
1 49.48609   8.463678
2 -8.14671 143.057930
3 11.28625 -11.182850

And this is the output I am looking for:

  latitude    longitude    distance-latlon1    distance-latlon2   distance-latlon3                 
1 49.48609     8.463678    NA                  *latlon2><latlon1  *latlon3><latlon1
2 -8.14671   143.057930    *latlon1><latlon2   NA                 *latlon3><latlon2
3 11.28625   -11.182850    *latlon1><latlon3   *latlon2><latlon3  NA

I tried out using geosphere but I only found a way to calculate the distance between two columns (which in this snippet results in a 0).

library(geosphere) 
df$distance <- distVincentyEllipsoid(df[,c('longitude','latitude')],
                                     df[,c('longitude','latitude')])

Answer 1

You need the distm function of the geosphere -package. With:

# create a distance matrix
m <- distm(df[2:1], df[2:1], fun = distVincentyEllipsoid)

# replace the diagonal with NA
diag(m) <- NA

# make column names for the distance matrix
colnames(m) <- paste0('r',1:nrow(df))

# bind the distance matrix to the dataframe
cbind.data.frame(df, m)

you get:

  latitude longitude r1 r2 r3 1 49.48609 8.463678 NA 13792423 4606658 2 -8.14671 143.057930 13792423 NA 17189185 3 11.28625 -11.182850 4606658 17189185 NA 

Answer 2

We can use the st_distance function from the sf package, which uses functions from geosphere to calculate the distance if the sf object is in lon-lat projection (EPSG 4326). df2 is the example output.

# Load packages
library(dplyr)
library(sf)

# Create example data frame
df <- data.frame(latitude = c(49.48609,-8.14671,11.28625),
                 longitude = c(8.463678,143.05793,-11.18285))

# COnvert to sf object
df_sf <- st_as_sf(df, coords = c("longitude", "latitude"))

# Set the projection as ESPG 4326 (long_lat)
st_crs(df_sf) <- 4326

# Apply the st_distance function
dist_m <- st_distance(df_sf)

# Combine with df
df2 <- df %>%
  mutate(`distance-latlon1` = as.numeric(dist_m[, 1]), 
         `distance-latlon2` = as.numeric(dist_m[, 2]),
         `distance-latlon3` = as.numeric(dist_m[, 3])) 

# Replace 0 with NA
df2[df2 == 0] <- NA

df2
  latitude  longitude distance-latlon1 distance-latlon2 distance-latlon3
1 49.48609   8.463678               NA         13792423          4606658
2 -8.14671 143.057930         13792423               NA         17189185
3 11.28625 -11.182850          4606658         17189185               NA

Here is an alternate way to combine dist_m with df .

library(tidyr)

# Convert dist_m to data frame
dist_df <- dist_m %>%
  as.table() %>%
  as_data_frame() %>%
  spread(Var2, n) %>%
  select(-Var1) %>%
  mutate_all(as.numeric) %>%
  setNames(paste0("distance-latlon", 1:nrow(df)))

# Combine with df
df2 <- df %>%
  bind_cols(dist_df)

# Replace 0 with NA
df2[df2 == 0] <- NA

Answer 3

Instead of distVincentyEllipsoid I would use the faster and more precise distGeo

df = df[,2:1] # the order should be longitude, latitude!
distm(df, df, distGeo)

or

library(raster)
d <- pointDistance(df)

perhaps followed by

as.matrix(as.dist(d))