[R-sig-Geo] remove nearby points

[Tim Howard]

Julian, 
Thanks for the reply. That seems like an interesting approach. I
suppose another 'GIS' way would be to buffer all the points 100m and
then find those where the buffers overlap. My dataset isn't too big
(1000's records), and what I came up with seems to do the trick, so I'll
stick with it for now.  
 
For the archives, I was a bit hasty in the solution I pasted below - it
caught the second of the pairs, where I wanted the first of the pairs
(after sorting descending by date). These are the lines that do it
correctly. 
 
b <- a[0,]
for(i in 1:nrow(a)){
    if(is.na(a[match(a$neigh[i],rownames(a)[1:i]),]$ID)){ 
        b <- rbind(b,a[i,])
    }
} 
 
Best to all,
Tim


>>> Julian Burgos <julian at hafro.is> 2/21/2013 3:24 AM >>>
Hi Tim,

Perhaps you should use clustering to identify groups of points that are
separated 100m or more from other points.  You could:
a) Calculate distances among points
b) Do some type of hierarchical clustering (e.g. the function agnes in
the cluster package).
c) Identify as clusters everything with a dissimilarity less than
100m.
d) Randomly select a single point from each cluster.

Julian

-- 
Julian Mariano Burgos, PhD
Hafrannsóknastofnunin/Marine Research Institute
Skúlagata 4, 121 Reykjavík, Iceland
Sími/Telephone : +354-5752037
Bréfsími/Telefax:  +354-5752001
Netfang/Email: julian at hafro.is

On 02/20/2013 06:39 PM, Tim Howard wrote:


I've found a very inelegant solution that continues on the path I was
going. Using the dummy dataset below, this code will strip neighbors as
I desire. 
 
b <- a[0,]
for(i in 2:nrow(a)){
    if(!is.na(a[match(a$neigh[i],rownames(a)[1:i]),]$ID)){ 
    b <- rbind(b,a[i,])
    }
    } 
> b
     ID dist neigh
3 three  5.1     2
4  four  2.2     1
I'd still be curious if anyone knows a cleaner solution. 
Best, 
Tim

>>> Tim Howard 2/20/2013 10:19 AM >>>
I am trying to remove spatial 'duplicates' from a point dataset. The
coordinates won't be exactly the same and so I can't use the normal
methods for removing the second instance of the points. This generalizes
to a question about removing points nearby others, either randomly or
based on other criteria (in my case, I want to keep the one with a more
recent date attribute). 
 
Although my research and fiddling has got me close, I wonder if there
already is a solution I'm missing within the various spatial packages so
I'm starting with sig-geo, even though I'm stuck at a spot that would
use regular R syntax. 
 
My approach (code at bottom of email):
 
1. Move the full point data set over to SpatialPoints as decimal
degrees longlat  [package=sp]
2. Reproject to utm, using spTransform
3. convert to ppp
4. find the distance from each point to its nearest using nndist() 
[package = spatstat]
5. identify that nearest using nnwhich()  [package = spatstat]
6. extract those with neighbors closer than 100m
 
** this is where I'm stuck **  I now have a list of neighbors, for
which I'd like to keep the first case of each neighbor but remove the
second (and sometimes third). Similar to unique(). Here's a dummy
example
 
#set up dummy data frame, the dist and neigh columns are from nndist()
and nnwhich(), respectively
a <-
data.frame(ID=c("one","two","three","four"),dist=c(2.2,5.1,5.1,2.2),neigh=c(4,3,2,1))
 
#here's as far as I've got, I can remove the neighbor to row one with
the following line.
#a looping solution seems problematic as the size of the dataframe
changes with each loop
 
b <- a[-match(a$neigh[1], rownames(a)),]
 
Questions:
- Is there already a function in a spatial package that offers a way to
remove points within a certain distance of others?
- if not, does anyone have any hints for taking the next step from what
I've done?
 
### code for what I've got so far.
### dat.wind.tall is the input DF of lat long decimal degree
coordinates
 
library(sp)
library(spatstat)
library(maptools)
library(rgdal)
llCRS <-  CRS("+proj=longlat +datum=NAD83")
wind.sp <- SpatialPoints(dat.wind.tall[,c(66,65)], proj4string=llCRS)
prjNew <- CRS("+proj=utm +zone=18 +datum=NAD83")
wind.utm <- spTransform(wind.sp, prjNew)
wind.ppp <- as(as(wind.utm, "SpatialPoints"), "ppp")
turb.dist <- nndist(wind.ppp)
turb.nearest <- nnwhich(wind.ppp)
dat.wind.tall.nbr <- cbind(dat.wind.tall, nneigh=turb.nearest,
dist=turb.dist)
closeNeighbors <- dat.wind.tall.nbr[turb.dist<100,]
#code for removing neighbor to first row. 
y <- closeNeighbors[-match(closeNeighbors$nneigh[1],
rownames(closeNeighbors)),]
 
Thanks in advance for any help. 
Tim


_______________________________________________
R-sig-Geo mailing
listR-sig-Geo at r-project.orghttps://stat.ethz.ch/mailman/listinfo/r-sig-geo