R and osmar : How to merge two adjacent osm files with overlapping nodes and ways?

Question

Due to a problem loading large osm files in osmar, I have decided to split my osm file in small parts, load them individually and the merge them.

I currently face two problems :

1. The code splitting the osm file tends to "lose" data around the bounding box. Therefore, I can't merge the objects afterwards because I have lost information in the process. This is done using osmconvert : osmconvert filename.osm -b=4,44,5,45 -o=filename_out.osm for your information.
2. To solve this, I now take a margin around each bounding box to make sure I don't lose nodes or ways. Consequently, I have to merge objects with overlapping nodes / ways in some cases. This is done with c() applied to objects of class osmar. And the result is not satisfying (see images below, with a map split in 3*3 tiles and then combined again) : and

The code used to display these maps is spplot(as_sp(osmar_object, "lines"), zcol=4) for your information.

How am I supposed to merge these files ? Should I find a way to merge osm files with overlapping nodes and ways or should I learn a better way to split the file than osmconvert ?

I'm opened to any approach, including switching to sp objects if needed.

Answer 1

I managed to merge osmar objects with 99% accuracy, and would like to share my code with you. The idea is to rewrite the c() function in order to manually group the dataframes inside of osmar objects.

Warning : In the code below, relations are completely ignored because I don't use them. Those of you willing to properly use relations should adapt the code accordingly.

# This is the equivalent of the c function for osmar objects
# l : list of objects of class "osmar"
osmar_c_list <- function(l, verbose = F) {

  # C is the result of the script. We initialise it as l[[1]], but will rewrite its components
  c <- first(l) 

  # Data.table does not like POSIXlt, so we have to convert timestamps
  l <- lapply(X = l, FUN = function(elem) {
    elem$nodes$attrs$timestamp     <- as.POSIXct(elem$nodes$attrs$timestamp)
    elem$ways$attrs$timestamp      <- as.POSIXct(elem$ways$attrs$timestamp)
    elem$relations$attrs$timestamp <- as.POSIXct(elem$relations$attrs$timestamp)
    elem
  })

  # The idea here is to rbind every dataframe in the osmar object
  # After that, we call the "unique" function to make sure we don't add the same object twice
  require(data.table)
  c$nodes$attrs      <- unique(rbindlist(lapply(X = l, FUN = function(elem) elem$nodes$attrs    )))
  c$nodes$tags       <- unique(rbindlist(lapply(X = l, FUN = function(elem) elem$nodes$tags     )))
  c$ways$attrs       <- unique(rbindlist(lapply(X = l, FUN = function(elem) elem$ways$attrs     )))
  c$ways$tags        <- unique(rbindlist(lapply(X = l, FUN = function(elem) elem$ways$tags      )))
  c$ways$refs        <- unique(rbindlist(lapply(X = l, FUN = function(elem) elem$ways$refs      )))
  c$relations$attrs  <- unique(rbindlist(lapply(X = l, FUN = function(elem) elem$relations$attrs)))
  c$relations$tags   <- unique(rbindlist(lapply(X = l, FUN = function(elem) elem$relations$tags )))
  c$relations$refs   <- unique(rbindlist(lapply(X = l, FUN = function(elem) elem$relations$refs )))

  # Calling the unique function was not without problem though. Indeed, in the case of a loop
  # i.e. a road from node A to node A, we have destroyed the last segment.
  # Let's detect duplicated nodes in ways and append them

  duplic_i    <- rbindlist(lapply(X = l, FUN = function(elem) elem$ways$refs[duplicated(elem$ways$refs),]))
  c$ways$refs <- rbind(c$ways$refs, unique(duplic_i))
  c$ways$refs <- c$ways$refs[order(c$ways$refs$id),]

  # Lets go back to POSIXlt
  c$nodes$attrs     <- data.frame(c$nodes$attrs)
  c$ways$attrs      <- data.frame(c$ways$attrs)
  c$relations$attrs <- data.frame(c$relations$attrs)
  c$nodes$attrs$timestamp     <- as.POSIXlt(c$nodes$attrs$timestamp)
  c$ways$attrs$timestamp      <- as.POSIXlt(c$ways$attrs$timestamp)
  c$relations$attrs$timestamp <- as.POSIXlt(c$relations$attrs$timestamp)

  return(c)
}

Warning : This code produces 99% accuracy for me, not 100%. I advise against using it as-is if your application requires high levels of accuracy.

Final note, I use this code to merge files with overlapping bounding boxes. I don't think it would work without some overlap.