Plan for retirement of rgdal, rgeos

[Jean-Romain]

Tagging interested persons

i.e. developers of packages that depend on lidR

@ptompalski @bi0m3trics @tgoodbody @carlos-alberto-silva @Blecigne @oliviermartin7 @DRAAlmeida @Molina-Valero @jmmonnet

aRchi FORTLS leafR lidaRtRee weblidar-treetop viewshed3d rGEDI spanner lidRmetrics

Context

rgdal and rgeos will be retired on Jan 1st 2024.

• https://twitter.com/RogerBivand/status/1407705212538822656
• https://www.youtube.com/watch?v=cK08bxUJn5A

raster and sp are based on rgdal/rgeos. lidR is based on raster and sp because it was created before sf, terra and stars. This means that sooner or later lidR will run into trouble (actually it is more or less already the case). So we must fully move to sf, terra/stars without depending on sp and raster. Do do so we need to:

• design new functions that return sf/terra object in replacement of function that return sp/raster objects.
• redesign the LAS* classes such as they no longer inherit sp::Spatial. This point being by far the most complex task and it will necessarily break backward compatibility

In this thread, I'm putting down my plan to fully remove dependency to sp/raster in 2024. The plan is to do it with minimum backward incompatibilities and a lot of time to update the workflow.

Warning: I'm putting here draft of ideas because it helps me to make a plan. It is in no way a definitive plan and anybody reading this is welcome to contribute to this thread.

Replacement functions

Former functions that return sp/raster will be supported until it causes issues on CRAN i.e. maybe Jan 1st 2024 maybe later

Function	Replacement
grid_metrics	pixel_metrics
grid_terrain	rasterize_terrain
grid_canopy	rasterize_canopy
grid_density	rasterize_density
tree_metrics	crown_metrics
delineate_tree	crown_metrics
find_trees	locate_trees
track_sensor	track_sensor

LAS class

The current LAS definition is:

setClass(
  Class = "LAS", contains = "Spatial",
  representation(data = "data.table", header = "LASheader", index = "list")
)

Which implies it has @proj4string and @bbox slots. To minimize backward incompatibilities we need something like:

setClass(
  Class = "LAS",
  representation(data = "data.table", header = "LASheader", index = "list", crs = "crs")
)

This way all user code that uses las@data$Z instead of las$Z will still work. What will fail however is las@proj4string or las@bbox which is unlikely to be used. Notice that bbox is not useful because it is redundant with the header, it was there only by inheritance. Also function projection and crs will still return sp::CRS object even if they are outdated in the R spatial ecosystem for backward compatibility.

LAScatalog class

The LAScatalog definition is

setClass(
  Class = "LAScatalog",
  contains = "SpatialPolygonsDataFrame",
  representation(
    chunk_options = "list",
    processing_options = "list",
    output_options = "list",
    input_options = "list",
    index = "list"
  )
)

The best option would be to become an sf object instead of an S4 classes but this is inconsistent with the LAS class. One well backward compatible option is

setClass(
  Class = "LAScatalog",
  representation(
    data = "sf"
    chunk_options = "list",
    processing_options = "list",
    output_options = "list",
    input_options = "list",
    index = "list"
  )
)

This will still provide access to e.g. ctg@data[["Min.Z"]] or ctg$GUID without modifying the code. Other operation being unlikely to be performed by user. Only by internal code

LAScluster class

The LAScluster class definition is

setClass(
  Class = "LAScluster", contains = "Spatial",
  representation(
    center = "list",
    bbbox  = "matrix",
    width  = "numeric",
    height = "numeric",
    buffer = "numeric",
    shape  = "numeric",
    select = "character",
    filter = "character",
    wkt    = "character",
    files  = "character",
    name   = "character",
    save   = "character",
    alt_dir = "character",
    index = "list"
  )
)

But being an internal class it is easier to deal with it

New functions and functions that need an update

• New: st_crs ; st_crs<- for LAS and LAScatalog
• Change: projection ; crs ; projection<- ; crs<-
• Remove: spTransform

[DRAAlmeida]

Dear Jean, thanks a lot for your email. Cheers,

[Jean-Romain]

Update.

I created a version of lidR v4 that no longer uses sp at all. The code is 100% sf based using sf::as_Spatial() to return outputs preserving backward compatibility. Inheritances to spatial objects has been removed with only minor backward incompatibilities (few call that are not expected to be used). Tables comparing before/after compatibility are reported below. All CRAN packages depending on lidR were tested without problem. sp is now suggested only.

I will now try to remove raster which will be another game but that should not affect users. The major problem was the inheritance to sp objects.

---

Backward compatibility using:

library(lidR)

Note: some calls that do not work are likely to work loading e.g. sf. For example st_crs() did not work in v3 but worked with library(sf) because it becomes equivalent to sf::st_crs().

• Call: is the called function
• LAS: applied to a LAS object
• LAScatalog: applied to a LAScatalog object
• LASheader: applied to a LASheader object
• LAScluster: applied to a LAScluster object
• Regression: is there a regression
• ❌/✔️ compares v3/v4

CRS

Call	LAS	LAScatalog	LASheader	LAScluster	Regression
st_crs	❌/✔️	❌/✔️	❌/✔️	❌/✔️
sf::st_crs	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️
st_crs<-	❌/✔️	❌/✔️	❌/✔️	❌/❌
sf::st_crs<-	❌/✔️	❌/✔️	❌/✔️	❌/❌
projection	✔️/✔️	✔️/✔️	✔️/✔️	✔️/✔️
raster::projection	✔️/❌	✔️/❌	❌/❌	✔️/❌	🚩
projection<-	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️
raster::projection<-	✔️/❌	✔️/❌	❌/❌	✔️/❌	🚩
crs	✔️/✔️	✔️/✔️	✔️/✔️	✔️/✔️
raster::crs	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️
sf::st_is_longlat	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️

Area

Call	LAS	LAScatalog	LASheader	LAScluster	Regression
st_bbox	❌/✔️	❌/✔️	❌/✔️	❌/✔️
sf::st_bbox	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️
extent	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️
raster::extent	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️
bbox	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️
sp::bbox	✔️/✔️	✔️/✔️	❌/✔️	✔️/✔️
st_area	❌/✔️	❌/✔️	❌/✔️	❌/❌
sf::st_area	❌/✔️	❌/✔️	❌/✔️	❌/❌
area	✔️/✔️	✔️/✔️	✔️/✔️	❌/❌
raster::area	✔️/✔️	✔️/✔️	✔️/✔️	❌/❌

Accessor

Call	LAS	LAScatalog	LASheader	LAScluster	Regression
[	✔️/✔️	✔️/✔️	❌/❌	❌/❌
$	✔️/✔️	✔️/✔️	❌/✔️	❌/❌
[[	✔️/✔️	✔️/✔️	❌/✔️	❌/❌
@data$	✔️/✔️	✔️/✔️	❌/❌	❌/❌
@proj4string	✔️/❌	✔️/❌	❌/❌	✔️/❌	🚩

R base

Call	LAS	LAScatalog	LASheader	LAScluster	Regression
nrow/ncol	❌/✔️	❌/✔️	❌/❌	❌/❌
dim	❌/✔️	❌/✔️	❌/❌	❌/❌
rbind	✔️/✔️	✔️/ ✔️	❌/❌	❌/❌
names	❌ ✔️	❌/✔️	❌/❌	❌/❌

Others

Call	LAS	LAScatalog	LASheader	LAScluster	Regression
st_convex_hull	❌/✔️	❌/❌	❌/❌	❌/❌
sf::st_convex_hull	❌/✔️	❌/❌	❌/❌	❌/❌