I'm working on an environmental epidemiology project where I have point exposures (~ 2,000 industrial hog operations - IHOs). These IHOs spray on nearby fields, but the feces water droplets and smell can travel miles. So these point exposures get 3mi buffers, and I want to know the number of IHO exposures (of various kinds - sum of amount of manure, number of hogs, whatever; most simplest, just the number of overlapping exposure buffers) per NC census blocks (~200,000). Exclusion census blocks (blue) are (1) anything in the top 5 most populous cities and (2) counties that do not border a county with an IHO in it (note: that was done with the gRelate function and DE-9IM codes - very slick!). See below image for a visual

enter image description here

The last step is to aggregate the buffered exposure representation to every census block. Here's where I'm stumped.

I've had good times with the %over% functions in the sp package so far, but understand from the over vignette that poly-poly and poly-line over are implemented in rgeos. The vignette only covers line-poly and self-referencing poly, and not with aggregation, so I'm a bit confused on what my options are for poly-poly with function aggregation, like sum or mean.

For a test case, consider the below, somewhat verbose snippet working with the world country borders file. This should be able to be copied out and run as is, since I'm using a random seed for the points and since I'm downloading and unzipping the world file in code.

First, we create 100 points, then use the over function with the fn argument to add up the element in the data frame. There are a lot of points here, but take a look at Australia: 3 points, number 3 as a label. So far, so good.

enter image description here

Now we transform geometries so we can create buffers, transform back, and map those buffers. (Included on previous map, since I'm limited to two links.) We want to know how many buffers overlap each country - in Australia's case, by eye, that's 4. I can't for the life of me figure what's going on though to get that with the over function. See my mess of an attempt in the final lines of code.

EDIT: Note that a commenter on r-sis-geo mentioned the aggregate function - also referenced on stack exchange question 63577 - so a work around / flow might be through that function, but I don't understand why I'd need to go to aggregate for polypoly when over seems to have that functionality for other spatial objects.


download.file("http://thematicmapping.org/downloads/TM_WORLD_BORDERS_SIMPL-0.3.zip", destfile="world.zip")
world.map = readOGR(dsn=".", "TM_WORLD_BORDERS_SIMPL-0.3", stringsAsFactors = F)
orig.world.map = world.map #hold the object, since I'm going to mess with it.

#Let's create 500 random lat/long points with a single value in the data frame: the number 1
lat.v = runif(n, -90, 90)
lon.v = runif(n, -180, 180)
coords.df = data.frame(lon.v, lat.v)
val.v = data.frame(rep(1,n))
names(val.v) = c("val")
names(coords.df) = c("lon", "lat")
points.spdf = SpatialPointsDataFrame(coords=coords.df, proj4string=CRS("+proj=longlat +datum=WGS84"), data=val.v)
points.spdf = spTransform(points.spdf, CRS(proj4string(world.map)))
plot(world.map, main="World map and points") #replot the map
plot(points.spdf, col="red", pch=20, cex=1, add=T) #...and add points.

#Let's use over with the point data
join.df = over(geometry(world.map), points.spdf,  fn=sum)
plot(world.map, main="World with sum of points, 750mi buffers") #Note - happens to be the count of points, but only b/c val=1.
plot(points.spdf, col="red", pch=20, cex=1, add=T) #...and add points.
world.map@data = data.frame(c(world.map@data, join.df))
#world.map@data = data.frame(c(world.map@data, over(world.map, points.spdf, fun="sum")))
invisible(text(getSpPPolygonsLabptSlots(world.map), labels=as.character(world.map$val), cex=1))
#Note I don't love making labels like above, and am open to better ways... plus I think it's deprecated/ing

#Now buffer...
pointbuff.spdf = gBuffer(spTransform(points.spdf, CRS("+init=EPSG:3358")), width=c(750*1609.344), byid=T)
pointbuff.spdf = spTransform(pointbuff.spdf, world.map@proj4string)
plot(pointbuff.spdf, col=NA, border="pink", add=T)

#Now over with the buffer (poly %over% poly).  How do I do this?
world.map = orig.world.map
join.df = data.frame(unname(over(geometry(world.map), pointbuff.spdf, fn=sum, returnList = F)) ) #Seems I need to unname this...?
names(join.df) = c("val")
world.map@data = data.frame(c(world.map@data, join.df)) #If I don't mess with the join.df, world.map's df is a mess..
plot(world.map, main="World map, points, buffers...and a mess of wrong counts") #replot the map
plot(points.spdf, col="red", pch=20, cex=1, add=T) #...and add points.
plot(pointbuff.spdf, col=NA, border="pink", add=T)
invisible(text(getSpPPolygonsLabptSlots(world.map), labels=as.character(world.map$val), cex=1)) 
#^ But if I do strip it of labels, it seems to be misassigning the results?
# Australia should now show 4 instead of 3.  I'm obviously super confused, probably about the structure of over poly-poly returns.  Help?
  • Appreciate the redirect - should I delete from here and repost over there? What's the best move? Thanks. Commented May 23, 2015 at 14:22

2 Answers 2


Thanks for the clear question and reproducible example.

Your understanding is correct, and this boils down to a bug in rgeos::over, which was fixed a month ago but has not made it into a CRAN release yet. The following is a work-around if you're only interested in the number of intersections:

world.map$val = sapply(over(geometry(world.map), pointbuff.spdf, returnList = TRUE), NROW)

I'm using NROW here instead of length so that it works with the wrong rgeos (0.3-8, from CRAN) as well as the corrected (0.3-10, from r-forge). The earlier suggestion of using

a = aggregate(pointbuff.spdf, world.map, sum)

also counts the number of intersections, but only with the fixed rgeos version installed. Its advantage, besides a more intuitive name, is that it directly returns a Spatial object, with the geometry of world.map.

To get rgeos 0.3-8 working, add

    signature(x = "SpatialPolygons", y = "SpatialPolygonsDataFrame"),

to your script, before you use over.

  • Very helpful, thank you. I particularly want to celebrate your offering a solution that works pre- and post-fix. Would you mind elaborating on: (1) What the bug IS that I'm hitting here-rgeos::over is returning a spatial polygon geography, not a spatial poly data frame? Don't some of the functions just return data frames...? (2) How this is generally supposed to work with aggregate and over? I'm a bit confused as to their intended differences and use cases. Really appreciate your weighing in, thank you. And sidenote: any suggestions for understanding the CRAN release cycle? Commented May 24, 2015 at 13:56
  • Also, as to the original question: I need to count the number of exposures, but I also really need to sum them - things like number of hogs in each exposure. Counting overlaps is a start... but it sounds like the solution I need is to pull in the newest rgeos, yes? No way to do that functional aggregation (not just counting) without it? Commented May 24, 2015 at 14:02
  • (1) rgeos::over for signature SpatialPolygons,SpatialPolygonsDataFrame should return a data.frame, but returns an index vector identical to when y would have been SpatialPolygons. sp::aggregate does what you do with over in a more user-friendly way, returning the Spatial object instead of the data.frame. CRAN packages are maintained by volunteers. Commented May 24, 2015 at 16:04
  • OK, thanks Edzer. It sounds like aggregate relies on rgeos over, so in order to get this functionality ahead of the CRAN release cycle (whenver that is), I'll need to find out how to download the newest rgeos and work off of that. Thank you. And thanks for all your work on the package!! Commented May 24, 2015 at 17:13
  • Also, Edzer, thanks so much for the note on R-sis-geo. Wasn't sure where the better place to post was, so I'm glad that thread now points here. Commented May 25, 2015 at 14:26

I whipped up a quick (and poorly coded) over-replacer in the meantime that creates the data frame I need, since my question isn't quite answered by the above counting-only solution or "work off the new rgeos", which I'm not quite skilled enough to understand how to do.

This function is clearly (1) incomplete (notice how I ignore the fn argument) and (2) inefficient, since I'm coming at it without R's powerful array manipulations / sapply... (clearly I'm coming from other languages without that power) but honestly, I'm still confused what the structure of the over function returns (list of lists...? And blank lists if NA?). For what it's worth (edits welcome), this function does the work I need done, successfully, and mimicks the action of the other over functions.

Edits welcome:

overhelper <- function(pol, pol.df, fn=sum, verbose=F){
   if(verbose) {cat("Building over geometry...\n"); t=Sys.time(); t}
   geolist = over(geometry(pol), pol.df, returnList = T)
   if(verbose) {cat("Geometry done. Aggregating df. \n"); Sys.time()-t;t=Sys.time();t;}
   results = data.frame(matrix(0,nrow=length(pol), ncol=ncol(pol.df)))
   names(results) = names(pol.df)
   end = length(geolist)

   for (i in 1:end){
     if(verbose) cat(i, "...")
     results[i,] = sapply(pol.df@data[unlist(geolist[i]),], fn)
   if(verbose) cat("Aggregation done! (", Sys.time()-t, ") \n Returning result vector.")
   return (results)
  • 1
    I added an alternative to get rgeos 0.3-8 fixed, to my answer. Commented May 25, 2015 at 19:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.