R measuring distance from a coastline

问题:

I have a set of coordinates:

d1 <- data_frame(
title = c("base1", "base2", "base3", "base4"),
lat = c(57.3, 58.8, 47.2, 57.8, 65.4, 56.7, 53.3),
long = c(0.4, 3.4, 3.5, 1.2, 1.5, 2.6, 2.7))

I would like to know whether the coordinates fall on land, in the sea, or are 3 miles inside a coastline. The coordinates should fall somewhere within the UK, so I know that I need to draw a shape file of the UK and plot the points onto it.

I just don't know how to measure whether the points fall in the sea, land or 2 miles from the coast. Obviously I can tell from looking at a map where they fall, but what I would like is to add another column to my data set so it looks like this:

 d2 <- data_frame(
title = c("base1", "base2", "base3", "base4", "base5", "base6", "base7"),
lat = c(57.3, 58.8, 47.2, 57.8, 65.4, 56.7, 53.3),
long = c(0.4, 3.4, 3.5, 1.2, 1.5, 2.6, 2.7),
where = c("land", "land", "sea", "coast", "land", "sea", "coast"))

• note that the data in d2 column 'where' is illustrative, these lat/long points may all be on land, or something else

回答1:

Distance to the coastline can be calculated by downloading openstreetmap coastline data. You can then use geosphere::dist2Line to get the distance from your points to the coastline.

I noticed that one of your example points was in France so you may need to expand the coastline data beyond just the UK (can be done by playing with the extents of the bounding box).

library(tidyverse)
library(sf)
library(geosphere)
library(osmdata)

#get initial data frame
d1 <- data_frame(
  title = c("base1", "base2", "base3", "base4", 
"base5", "base6", "base7"),
  lat = c(57.3, 58.8, 47.2, 57.8, 65.4, 56.7, 53.3),
  long = c(0.4, 3.4, 3.5, 1.2, 1.5, 2.6, 2.7))

# convert to sf object
d1_sf <- d1 %>% st_as_sf(coords = c('long','lat')) %>% 
st_set_crs(4326)

# get bouding box for osm data download (England) and 
# download coastline data for this area
osm_box <- getbb (place_name = "England") %>%
  opq () %>% 
  add_osm_feature("natural", "coastline") %>% 
  osmdata_sf() 


# use dist2Line from geosphere - only works for WGS84 
#data
dist <- geosphere::dist2Line(p = st_coordinates(d1_sf), 
                         line = 
st_coordinates(osm_box$osm_lines)[,1:2])

#combine initial data with distance to coastline
df <- cbind( d1 %>% rename(y=lat,x=long),dist) %>%
  mutate(miles=distance/1609)


#  title    y   x   distance       lon      lat     miles
#1 base1 57.3 0.4  219066.40 -2.137847 55.91706 136.15065
#2 base2 58.8 3.4  462510.28 -2.137847 55.91706 287.45201
#3 base3 47.2 3.5  351622.34  1.193198 49.96737 218.53470
#4 base4 57.8 1.2  292210.46 -2.137847 55.91706 181.60998
#5 base5 65.4 1.5 1074644.00 -2.143168 55.91830 667.89559
#6 base6 56.7 2.6  287951.93 -1.621963 55.63143 178.96329
#7 base7 53.3 2.7   92480.24  1.651836 52.76027  57.47684



#plot
p <- ggplot() + 
  geom_sf(data=osm_box$osm_lines) +
  geom_sf(data=d1_sf) +
  geom_segment(data=df,aes(x=x,y=y,xend=lon,yend=lat))

That's just for the distance to the coastline. You also need to know how whether it is inland or at sea. For this, you would need a separate shapefile for the sea: http://openstreetmapdata.com/data/water-polygons and see if each point of your points sits in the sea or not.

#read in osm water polygon data
sea <- read_sf('water_polygons.shp')

#get get water polygons that intersect our points
in_sea <- st_intersects(d1_sf,sea) %>% as.data.frame() 

#join back onto original dataset
df %>% mutate(row = row_number()) %>%
  #join on in_sea data
  left_join(in_sea,by=c('row'='row.id')) %>%
  mutate(in_sea = if_else(is.na(col.id),F,T)) %>%
#categorise into 'sea', 'coast' or 'land'
  mutate(where = case_when(in_sea == T ~ 'Sea',
                           in_sea == F & miles <=3 ~ 'Coast',
                           in_sea == F ~ 'Land'))



# title    y   x   distance       lon      lat     miles row col.id in_sea where
#1 base1 57.3 0.4  219066.40 -2.137847 55.91706 136.15065   1  24193   TRUE   Sea
#2 base2 58.8 3.4  462510.28 -2.137847 55.91706 287.45201   2  24194   TRUE   Sea
#3 base3 47.2 3.5  351622.34  1.193198 49.96737 218.53470   3     NA  FALSE  Land
#4 base4 57.8 1.2  292210.46 -2.137847 55.91706 181.60998   4  24193   TRUE   Sea
#5 base5 65.4 1.5 1074644.00 -2.143168 55.91830 667.89559   5  25417   TRUE   Sea
#6 base6 56.7 2.6  287951.93 -1.621963 55.63143 178.96329   6  24193   TRUE   Sea
#7 base7 53.3 2.7   92480.24  1.651836 52.76027  57.47684   7  24143   TRUE   Sea


ggplot() + 
  geom_sf(data=osm_box$osm_lines) +
  geom_sf(data=d1_sf) +
  geom_segment(data=df,aes(x=x,y=y,xend=lon,yend=lat)) +
  ggrepel::geom_text_repel(data=df, 
aes(x=x,y=y,label=paste0(where,'\n',round(miles,0),'miles')),size=2)

Update 16/08/2018

Since you asked for an approach specifically using a shapefile I have downloaded this one here: openstreetmapdata.com/data/coastlines which I will use to carry out the same approach as above.

clines <- read_sf('lines.shp') #path to shapefile

Next I created a custom bounding box so that we can cut down the size of the shapefile to only include coastlines reasonably close to the points.

# create bounding box surrounding points 
bbox <- st_bbox(d1_sf) 

# write a function that takes the bbox around our points
# and expands it by a given amount of metres.
expand_bbox <- function(bbox,metres_x,metres_y){

  box_centre <- bbox %>% st_as_sfc() %>% 
    st_transform(crs = 32630) %>%
    st_centroid() %>%
    st_transform(crs = 4326) %>%
    st_coordinates()


  bbox['xmin'] <-  bbox['xmin'] - (metres_x / 6370000) * (180 / pi) / cos(bbox['xmin'] * pi/180)
  bbox['xmax'] <-  bbox['xmax'] + (metres_x / 6370000) * (180 / pi) / cos(bbox['xmax'] * pi/180)
  bbox['ymin'] <-  bbox['ymin'] - (metres_y / 6370000) * (180 / pi)
  bbox['ymax'] <- bbox['ymax'] + (metres_y / 6370000) * (180 / pi)


  bbox['xmin'] <- ifelse(bbox['xmin'] < -180, bbox['xmin'] + 360, bbox['xmin'])
  bbox['xmax'] <- ifelse(bbox['xmax'] > 180, bbox['xmax'] - 360, bbox['xmax'])
  bbox['ymin'] <- ifelse(bbox['ymin'] < -90, (bbox['ymin'] + 180)*-1, bbox['ymin'])
  bbox['ymax'] <- ifelse(bbox['ymax'] > 90, (bbox['ymax'] + 180)*-1, bbox['ymax'])
  return(bbox)
}


# expand the bounding box around our points by 300 miles in x and 100 #miles in y direction to make nice shaped box.
bbox <- expand_bbox(bbox,metres_x=1609*200, metres_y=1609*200) %>% st_as_sfc

# get only the parts of the coastline that are within our bounding box
clines2 <- st_intersection(clines,bbox) 

Now I used the dist2Line function here because it is accurate and it gives you the points on the coastline it is measuring to which makes it good for checking errors. The downside is, it's very slow for our rather large coastline file.

Running this took me 8 minutes:

dist <- geosphere::dist2Line(p = st_coordinates(d1_sf), 
                                 line = as(clines2,'Spatial'))

#combine initial data with distance to coastline
df <- cbind( d1 %>% rename(y=lat,x=long),dist) %>%
  mutate(miles=distance/1609)

df

 # title    y   x  distance        lon      lat    ID     miles
#1 base1 57.3 0.4 131936.70 -1.7711149 57.46995  4585  81.99919
#2 base2 58.8 3.4  98886.42  4.8461433 59.28235   179  61.45831
#3 base3 47.2 3.5 340563.02  0.3641618 49.43811  4199 211.66129
#4 base4 57.8 1.2 180110.10 -1.7670712 57.50691  4584 111.93915
#5 base5 65.4 1.5 369550.43  6.2494627 62.81381  9424 229.67709
#6 base6 56.7 2.6 274230.37  5.8635346 58.42913 24152 170.43528
#7 base7 53.3 2.7  92480.24  1.6518358 52.76027  4639  57.47684

plot:

ggplot() + 
  geom_sf(data=clines2) +
  geom_sf(data=bbox,fill=NA)+
  geom_sf(data=d1_sf) +
  geom_segment(data=df,aes(x=x,y=y,xend=lon,yend=lat))

If you don't mind about the slight loss of accuracy (results differ by around 0.3% on your data), and are not fussed about knowing where exactly on the coastline it is measuring to, you can measure the distance to the polygon:

# make data into polygons
clines3 <- st_intersection(clines,bbox) %>%
  st_cast('POLYGON')

#use rgeos::gDistance to calculate distance to nearest polygon
#need to change projection (I used UTM30N) to use gDistance
dist2 <- apply(rgeos::gDistance(as(st_transform(d1_sf,32630), 'Spatial'),
                               as(st_transform(clines3,32630),'Spatial'),
                               byid=TRUE),2,min)

df2 <- cbind( d1 %>% rename(y=lat,x=long),dist2) %>%
  mutate(miles=dist2/1609)

df2

#  title    y   x     dist2     miles
#1 base1 57.3 0.4 131917.62  81.98733
#2 base2 58.8 3.4  99049.22  61.55949
#3 base3 47.2 3.5 341015.26 211.94236
#4 base4 57.8 1.2 180101.47 111.93379
#5 base5 65.4 1.5 369950.32 229.92562
#6 base6 56.7 2.6 274750.17 170.75834
#7 base7 53.3 2.7  92580.16  57.53894

By contrast this took just 8 seconds to run!

The rest is as in the previous answer.

回答2:

First you need a file containing coastlines of UK.

You can use the method provided in this question to know whether a point falls inside the UK coastlines or outside.

Then, for points that fall into the UK, you can compute the Haversine distance between them and the coast line points, to determine if they are in your 3 miles from the coast.

回答3:

For a faster implementation of geosphere:::dist2Line that uses purrr for efficient looping and progress for a progress bar, thus retaining the accuracy of Chris' first answer, see below:

library(geosphere)
library(purr)
library(progress)

spDistPoint2Line <- function (p, line, distfun)
{ 
  ## rewrite of internal function from geosphere
  test <- !sp::is.projected(line)
  if (!isTRUE(test)) {
    if (is.na(test)) {
      warning("Coordinate reference system of SpatialPolygons object is not set. Assuming it is degrees (longitude/latitude)!")
    }
    else {
      stop("Points are projected. They should be in degrees (longitude/latitude)")
    }
  }

  x <- line@lines
  n <- length(x)
  res <- matrix(nrow = nrow(p), ncol = 3)
  colnames(res) <- c("distance", "lon", "lat")

  line_coords <- map(x, ~(map(.@Lines, ~(.@coords)))) #basically an unlist
  pb <- progress_bar$new(
    total = length(line_coords),
    format = "(:spin) :current of :total, :percent, eta: :eta"
  )

  res[] <- Inf
  result <- reduce(
    .x = line_coords,
    .init = res,
    .f = function(res, crd){
      pb$tick()
      crd <- crd[[1]]
      r <- dist2Line(p, crd, distfun) # have to live without ID
      k <- r[, 1] < res[, 1]
      res[k, ] <- r[k, ]
      return(res)
    }
  )
  return(result)
}

dist2Line <- function (p, line, distfun = distGeo) 
{
  p <- geosphere:::.pointsToMatrix(p)
  if (inherits(line, "SpatialPolygons")) {
    line <- methods::as(line, "SpatialLines")
  }
  if (inherits(line, "SpatialLines")) {
    return(spDistPoint2Line(p, line, distfun))
  }

  line <- geosphere:::.pointsToMatrix(line)
  line1 <- line[-nrow(line), , drop = FALSE]
  line2 <- line[-1, , drop = FALSE]
  seglength <- distfun(line1, line2)

  res <-
    p %>%
      array_branch(1) %>%
      map(
        function(xy){
          crossdist <- abs(dist2gc(line1, line2, xy))
          trackdist1 <- alongTrackDistance(line1, line2, xy)
          trackdist2 <- alongTrackDistance(line2, line1, xy)
          mintrackdist <- pmin(trackdist1, trackdist2)
          maxtrackdist <- pmax(trackdist1, trackdist2)
          crossdist[maxtrackdist >= seglength] <- NA
          nodedist <- distfun(xy, line)
          warnopt = getOption("warn")
          options(warn = -1)
          distmin1 <- min(nodedist, na.rm = TRUE)
          distmin2 <- min(crossdist, na.rm = TRUE)
          options(warn = warnopt)
          if (distmin1 <= distmin2) {
            j <- which.min(nodedist)
            return(c(distmin1, line[j, ]))
          }
          else {
            j <- which.min(crossdist)
            if (trackdist1[j] < trackdist2[j]) {
              bear <- bearing(line1[j, ], line2[j, ])
              pt <- destPoint(line1[j, ], bear, mintrackdist[j])
              return(c(crossdist[j], pt))
            }
            else {
              bear <- bearing(line2[j, ], line1[j, ])
              pt <- destPoint(line2[j, ], bear, mintrackdist[j])
              return(c(crossdist[j], pt))
            }
          }
        }
      ) %>%
      simplify %>%
      matrix(ncol = 3, byrow = TRUE)

  colnames(res) <- c("distance", "lon", "lat")
  return(res)
}