R - Finding closest neighboring point and number of neighbors within a given radius, coordinates lat-long

Question

I am trying to figure out how isolated certain points are within my data set. I am using two methods to determine isolation, the distance of the closest neighbor and the number of neighboring sites within a given radius. All my coordinates are in latitude and longitude

This is what my data looks like:

    pond            lat         long        area    canopy  avg.depth   neighbor    n.lat   n.long  n.distance  n.area  n.canopy    n.depth n.avg.depth radius1500
    A10             41.95928    -72.14605   1500    66      60.61538462                                 
    AA006           41.96431    -72.121     250     0       57.77777778                                 
    Blacksmith      41.95508    -72.123803  361     77      71.3125                                 
    Borrow.Pit.1    41.95601    -72.15419   0       0       41.44444444                                 
    Borrow.Pit.2    41.95571    -72.15413   0       0       37.7                                    
    Borrow.Pit.3    41.95546    -72.15375   0       0       29.22222222                                 
    Boulder         41.918223   -72.14978   1392    98      43.53333333                                 

I want to put the name of the nearest neighboring pond in the column neighbor, its lat and long in n.lat and n.long, the distance between the two ponds in n.distance, and the area, canopy and avg.depth in each of the appropriate columns.

Second, I want to put the number of ponds within 1500m of the target pond into radius1500.

Does anyone know of a function or package that will help me calculate the distances/numbers that I want? If it's an issue, it won't be hard to enter the other data I need, but the nearest neighbor's name and distance, plus the number of ponds within 1500m is what I really need help with.

Thank you.

Answer 1

Best option is to use libraries sp and rgeos, which enable you to construct spatial classes and perform geoprocessing.

library(sp)
library(rgeos)

Read the data and transform them to spatial objects:

mydata <- read.delim('d:/temp/testfile.txt', header=T)

sp.mydata <- mydata
coordinates(sp.mydata) <- ~long+lat

class(sp.mydata)
[1] "SpatialPointsDataFrame"
attr(,"package")
[1] "sp"

Now calculate pairwise distances between points

d <- gDistance(sp.mydata, byid=T)

Find second shortest distance (closest distance is of point to itself, therefore use second shortest)

min.d <- apply(d, 1, function(x) order(x, decreasing=F)[2])

Construct new data frame with desired variables

newdata <- cbind(mydata, mydata[min.d,], apply(d, 1, function(x) sort(x, decreasing=F)[2]))

colnames(newdata) <- c(colnames(mydata), 'neighbor', 'n.lat', 'n.long', 'n.area', 'n.canopy', 'n.avg.depth', 'distance')

newdata
            pond      lat      long area canopy avg.depth     neighbor    n.lat    n.long n.area n.canopy n.avg.depth
6            A10 41.95928 -72.14605 1500     66  60.61538 Borrow.Pit.3 41.95546 -72.15375      0        0    29.22222
3          AA006 41.96431 -72.12100  250      0  57.77778   Blacksmith 41.95508 -72.12380    361       77    71.31250
2     Blacksmith 41.95508 -72.12380  361     77  71.31250        AA006 41.96431 -72.12100    250        0    57.77778
5   Borrow.Pit.1 41.95601 -72.15419    0      0  41.44444 Borrow.Pit.2 41.95571 -72.15413      0        0    37.70000
4   Borrow.Pit.2 41.95571 -72.15413    0      0  37.70000 Borrow.Pit.1 41.95601 -72.15419      0        0    41.44444
5.1 Borrow.Pit.3 41.95546 -72.15375    0      0  29.22222 Borrow.Pit.2 41.95571 -72.15413      0        0    37.70000
6.1      Boulder 41.91822 -72.14978 1392     98  43.53333 Borrow.Pit.3 41.95546 -72.15375      0        0    29.22222
        distance
6   0.0085954872
3   0.0096462277
2   0.0096462277
5   0.0003059412
4   0.0003059412
5.1 0.0004548626
6.1 0.0374480316

Edit: if coordinates are in degrees and you would like to calculate distance in kilometers, use package geosphere

library(geosphere)

d <- distm(sp.mydata)

# rest is the same

This should provide better results, if the points are scattered across the globe and coordinates are in degrees

Answer 2

I add below an alternative solution using the newer sf package, for those interested and coming to this page now (as I did).

First, load the data and create the sf object.

# Using sf
mydata <- structure(
  list(pond = c("A10", "AA006", "Blacksmith", "Borrow.Pit.1", 
                "Borrow.Pit.2", "Borrow.Pit.3", "Boulder"), 
       lat = c(41.95928, 41.96431, 41.95508, 41.95601, 41.95571, 41.95546, 
               41.918223), 
       long = c(-72.14605, -72.121, -72.123803, -72.15419, -72.15413, 
                -72.15375, -72.14978), 
       area = c(1500L, 250L, 361L, 0L, 0L, 0L, 1392L), 
       canopy = c(66L, 0L, 77L, 0L, 0L, 0L, 98L), 
       avg.depth = c(60.61538462, 57.77777778, 71.3125, 41.44444444, 
                     37.7, 29.22222222, 43.53333333)), 
  class = "data.frame", row.names = c(NA, -7L))


library(sf)
data_sf <- st_as_sf(mydata, coords = c("long", "lat"),
                    # Change to your CRS
                    crs = "+proj=longlat +ellps=WGS84 +datum=WGS84 +no_defs")
st_is_longlat(data_sf)

sf::st_distance calculates the distance matrix in meters using Great Circle distance when using lat/lon data.

dist.mat <- st_distance(data_sf) # Great Circle distance since in lat/lon
# Number within 1.5km: Subtract 1 to exclude the point itself
num.1500 <- apply(dist.mat, 1, function(x) {
  sum(x < 1500) - 1
})

# Calculate nearest distance
nn.dist <- apply(dist.mat, 1, function(x) {
  return(sort(x, partial = 2)[2])
})
# Get index for nearest distance
nn.index <- apply(dist.mat, 1, function(x) { order(x, decreasing=F)[2] })

n.data <- mydata
colnames(n.data)[1] <- "neighbor"
colnames(n.data)[2:ncol(n.data)] <- 
  paste0("n.", colnames(n.data)[2:ncol(n.data)])
mydata2 <- data.frame(mydata,
                      n.data[nn.index, ],
                      n.distance = nn.dist,
                      radius1500 = num.1500)
rownames(mydata2) <- seq(nrow(mydata2))

mydata2
          pond      lat      long area canopy avg.depth     neighbor    n.lat    n.long n.area n.canopy
1          A10 41.95928 -72.14605 1500     66  60.61538 Borrow.Pit.1 41.95601 -72.15419      0        0
2        AA006 41.96431 -72.12100  250      0  57.77778   Blacksmith 41.95508 -72.12380    361       77
3   Blacksmith 41.95508 -72.12380  361     77  71.31250        AA006 41.96431 -72.12100    250        0
4 Borrow.Pit.1 41.95601 -72.15419    0      0  41.44444 Borrow.Pit.2 41.95571 -72.15413      0        0
5 Borrow.Pit.2 41.95571 -72.15413    0      0  37.70000 Borrow.Pit.1 41.95601 -72.15419      0        0
6 Borrow.Pit.3 41.95546 -72.15375    0      0  29.22222 Borrow.Pit.2 41.95571 -72.15413      0        0
7      Boulder 41.91822 -72.14978 1392     98  43.53333 Borrow.Pit.3 41.95546 -72.15375      0        0
  n.avg.depth n.distance radius1500
1    41.44444  766.38426          3
2    71.31250 1051.20527          1
3    57.77778 1051.20527          1
4    37.70000   33.69099          3
5    41.44444   33.69099          3
6    37.70000   41.99576          3
7    29.22222 4149.07406          0

For obtaining the closest neighbor after calculating distance, you can use sort() with the partial = 2 argument. Depending on the amount of data, this could be much faster than using order as in the previous solution. The package Rfast is likely even faster but I avoid including additional packages here. See this related post for a discussion and benchmarking of various solutions: https://stackoverflow.com/a/53144760/12265198

Answer 3

In Rfast, there is a function called "dista" and computes the Euclidean or the Manhattan distances only (at the moment). It gives the option to compute the k-smallest distances. Alternatively it can return the indexes of the observations with the smallest distances. The cosinus distance is basically almost the same as the Eucledean distance (exluding a constant, 2 I think).

Answer 4

I add below a solution using the spatialrisk package. The key functions in this package are written in C++ (Rcpp), and are therefore very fast.

First, load the data:

df <- data.frame(pond = c("A10", "AA006", "Blacksmith", "Borrow.Pit.1", 
                          "Borrow.Pit.2", "Borrow.Pit.3", "Boulder"), 
                 lat = c(41.95928, 41.96431, 41.95508, 41.95601, 
                         41.95571, 41.95546, 41.918223), 
                 long = c(-72.14605, -72.121, -72.123803, -72.15419, 
                          -72.15413, -72.15375, -72.14978), 
                 area = c(1500, 250, 361, 0, 0, 0, 1392), 
                 canopy = c(66, 0, 77, 0, 0, 0, 98), 
                 avg.depth = c(60.61538462, 57.77777778, 71.3125, 41.44444444,
                               37.7, 29.22222222, 43.53333333))

The function spatialrisk::points_in_circle calculates the observations within radius from a center point. Note that distances are calculated using the Haversine formula. Since each element of the output is a data frame, purrr::map_dfr is used to row-bind them together:

ans1 <- purrr::map2_dfr(df$long, df$lat, 
                        ~spatialrisk::points_in_circle(df, .x, .y, 
                                                       lon = long, 
                                                       radius = 100000)[2,])

colnames(ans1) <- c("neighbor", "n.lat", "n.long", "n.area", 
                    "n.canopy", "n.avg.depth", "distance_m")

      neighbor    n.lat    n.long n.area n.canopy n.avg.depth distance_m
1 Borrow.Pit.1 41.95601 -72.15419      0        0    41.44444  765.87823
2   Blacksmith 41.95508 -72.12380    361       77    71.31250 1053.35200
3        AA006 41.96431 -72.12100    250        0    57.77778 1053.35200
4 Borrow.Pit.2 41.95571 -72.15413      0        0    37.70000   33.76321
5 Borrow.Pit.1 41.95601 -72.15419      0        0    41.44444   33.76321
6 Borrow.Pit.2 41.95571 -72.15413      0        0    37.70000   42.00128
7 Borrow.Pit.3 41.95546 -72.15375      0        0    29.22222 4158.21978

Now calculate the number of ponds within 1500m of the target pond. The function spatialrisk::concentration sums the number of observations within a radius from center points. 1 is substracted from the number of ponds to exclude the pond itself.

df$npond <- 1  
radius1500 <- spatialrisk::concentration(df, df, npond, lon_sub = long, 
                                         lon_full = long, radius = 1500, 
                                         display_progress = FALSE)$concentration - 1

Column-bind the data frames together:

cbind(df, ans1, radius1500)

          pond      lat      long area canopy avg.depth     neighbor    n.lat    n.long n.area n.canopy n.avg.depth distance_m radius1500
1          A10 41.95928 -72.14605 1500     66  60.61538 Borrow.Pit.1 41.95601 -72.15419      0        0    41.44444  765.87823          3
2        AA006 41.96431 -72.12100  250      0  57.77778   Blacksmith 41.95508 -72.12380    361       77    71.31250 1053.35200          1
3   Blacksmith 41.95508 -72.12380  361     77  71.31250        AA006 41.96431 -72.12100    250        0    57.77778 1053.35200          1
4 Borrow.Pit.1 41.95601 -72.15419    0      0  41.44444 Borrow.Pit.2 41.95571 -72.15413      0        0    37.70000   33.76321          3
5 Borrow.Pit.2 41.95571 -72.15413    0      0  37.70000 Borrow.Pit.1 41.95601 -72.15419      0        0    41.44444   33.76321          3
6 Borrow.Pit.3 41.95546 -72.15375    0      0  29.22222 Borrow.Pit.2 41.95571 -72.15413      0        0    37.70000   42.00128          3
7      Boulder 41.91822 -72.14978 1392     98  43.53333 Borrow.Pit.3 41.95546 -72.15375      0        0    29.22222 4158.21978          0

Answer 5

The Solution propose by @Zbynek is quite nice but if you are looking for a distance between two neighboor in km like I am , I am proposing this solution.

   earth.dist<-function(lat1,long1,lat2,long2){

           rad <- pi/180
           a1 <- lat1 * rad
           a2 <- long1 * rad
           b1 <- lat2 * rad
           b2 <- long2 * rad
           dlat <- b1-a1
           dlon<- b2-a2
           a <- (sin(dlat/2))^2 +cos(a1)*cos(b1)*(sin(dlon/2))^2
           c <- 2*atan2(sqrt(a),sqrt(1-a))
           R <- 6378.145
           dist <- R *c
           return(dist)
           }


    Dist <- matrix(0,ncol=length(mydata),nrow=length(mydata.sp))

  for (i in 1:length(mydata)){
      for(j in 1:length(mydata.sp)){
          Dist[i,j] <- earth.dist(mydata$lat[i],mydata$long[i],mydata.sp$lat[j],mydata.sp$long[j])
 }}



     DDD <- matrix(0, ncol=5,nrow=ncol(Dist))   ### RECTIFY the nb of col by the number of variable you want

   for(i in 1:ncol(Dist)){
       sub<- sort(Dist[,i])[2]
       DDD[i,1] <- names(sub) 
       DDD[i,2] <- sub
       DDD[i,3] <- rownames(Dist)[i]
       sub_neig_atr <- Coord[Coord$ID==names(sub),]
       DDD[i,4] <- sub_neig_atr$area
       DDD[i,5] <- sub_neig_atr$canopy
       ### Your can add any variable you want here 

   }

    DDD <- as.data.frame(DDD)

    names(DDD)<-c("neigboor_ID","distance","pond","n.area","n.canopy")
   data <- merge(mydata,DDD, by="pond")

You end up getting a distance in km if your coordinates are long and lat.

Any suggestions to make it better ?

Answer 6

Another answer which while probably slower might have an intuitive appeal for dplyr addicts.

You create a mega-grid of every single possible combination of lat/lons, then you can find the one that has the smallest distance using geosphere.

The example is where you have two datasets with different points to compare - but you can adjust it easily by duplicating the first dataset.

    library(tidyverse)
    library(geosphere)
    library(data.table)

    #This function creates a big dataframe with every possible combination
    expand.grid.df <- function(...) Reduce(function(...) merge(..., by=NULL), list(...))

shortest_distance <- expand.grid.df(df1,df2) %>%
      mutate(distance = distHaversine(p1 = cbind(lon_2,lat_2),
                                      p2 = cbind(lon,lat))) %>% 
      group_by(ACCIDENT_NO) %>% 
      slice(which.min(distance))