catalog_apply: LAScatalog processing engine

View source: R/catalog_apply.R

catalog_applyR Documentation

LAScatalog processing engine

Description

This function gives users access to the LAScatalog processing engine. It allows the application of a user-defined routine over a collection of LAS/LAZ files. The LAScatalog processing engine is explained in the LAScatalog class

catalog_apply() is the core of the lidR package. It drives every single function that can process a LAScatalog. It is flexible and powerful but also complex. catalog_map() is a simplified version of catalog_apply() that suits for 90% of use cases.

catalog_sapply() is a previous attempt to provide simplified version of catalog_apply(). Use catalog_map() instead.

Usage

catalog_apply(ctg, FUN, ..., .options = NULL)

catalog_sapply(ctg, FUN, ..., .options = NULL)

catalog_map(ctg, FUN, ..., .options = NULL)

Arguments

ctg

A LAScatalog object.

FUN

A user-defined function that respects a given template (see section function template)

...

Optional arguments to FUN.

.options

See dedicated section and examples.

Edge artifacts

It is important to take precautions to avoid 'edge artifacts' when processing wall-to-wall tiles. If the points from neighbouring tiles are not included during certain processes, this could create 'edge artifacts' at the tile boundaries. For example, empty or incomplete pixels in a rasterization process, or dummy elevations in a ground interpolation. The LAScatalog processing engine provides internal tools to load buffered data 'on-the-fly'. catalog_map() takes care of removing automatically the results computed in the buffered area to avoid unexpected output with duplicated entries or conflict between values computed twice. It does that in predefined way that may not suit all cases. catalog_apply() does not remove the buffer and leave users free to handle this in a custom way. This is why catalog_apply() is more complex but gives more freedom to build new applications.

Buffered data

The LAS objects loaded in theses functions have a special attribute called 'buffer' that indicates, for each point, if it comes from a buffered area or not. Points from non-buffered areas have a 'buffer' value of 0, while points from buffered areas have a 'buffer' value of 1, 2, 3 or 4, where 1 is the bottom buffer and 2, 3 and 4 are the left, top and right buffers, respectively. This allows for filtering of buffer points if required.

Function template

The parameter FUN of catalog_apply expects a function with a first argument that will be supplied automatically by the LAScatalog processing engine. This first argument is a LAScluster. A LAScluster is an internal undocumented class but the user needs to know only three things about this class:

  • It represents a chunk of the file collection

  • The function readLAS can be used with a LAScluster

  • The function extent or st_bbox can be used with a LAScluster and they return the bounding box of the chunk without the buffer. It must be used to clip the output and remove the buffered region (see examples).

A user-defined function must be templated like this:

myfun <- function(chunk, ...) {
   # Load the chunk + buffer
   las <- readLAS(chunk)
   if (is.empty(las)) return(NULL)

   # do something
   output <- do_something(las, ...)

   # remove the buffer of the output
   bbox <- bbox(chunk)
   output <- remove_buffer(output, bbox)
   return(output)
}

The line if (is.empty(las)) return(NULL) is important because some clusters (chunks) may contain 0 points (we can't know this before reading the file). In this case an empty point cloud with 0 points is returned by readLAS() and this may fail in subsequent code. Thus, exiting early from the user-defined function by returning NULL indicates to the internal engine that the chunk was empty.

catalog_map is much simpler (but less versatile). It automatically takes care of reading the chunk and checks if the point cloud is empty. It also automatically crop the buffer. The way it crops the buffer suits for most cases but for some special cases it may be advised to handle this in a more specific way i.e. using catalog_apply(). For catalog_map() the first argument is a LAS and the template is:

myfun <- function(las, ...) {
   # do something
   output <- do_something(las, ...)
   return(output)
}

.options

Users may have noticed that some lidR functions throw an error when the processing options are inappropriate. For example, some functions need a buffer and thus buffer = 0 is forbidden. Users can add the same constraints to protect against inappropriate options. The .options argument is a list that allows users to tune the behaviour of the processing engine.

  • drop_null = FALSE: not intended to be used by regular users. The engine does not remove NULL outputs

  • need_buffer = TRUE: the function complains if the buffer is 0.

  • need_output_file = TRUE the function complains if no output file template is provided.

  • raster_alignment = ... the function checks the alignment of the chunks. This option is important if the output is a raster. See below for more details.

  • automerge = TRUE by default the engine returns a list with one item per chunk. If automerge = TRUE, it tries to merge the outputs into a single object: a ⁠Raster*⁠, a ⁠Spatial*⁠, a ⁠LAS*⁠, an sf, a stars similarly to other functions of the package. This is a fail-safe option so in the worst case, if the merging fails, the list is returned. This is activated by catalog_map making it simpler.

  • autoread = TRUE. Introduced in v3.0.0 this option enables to get rid of the first steps of the function i.e readLAS() and ⁠if (is.empty())⁠. In this case the function must take two objects as input, first a LAS object and second a bbox from sf. This is activated by catalog_map making it simpler.

  • autocrop = TRUE. Introduced in v4.0.0 this option enables to get rid of the buffer crop steps of the function i.e something <- remove_buffer(something, bbox). In this case the function must take one LAS object as input. This is activated by catalog_map making it simpler.

When the function FUN returns a raster it is important to ensure that the chunks are aligned with the raster to avoid edge artifacts. Indeed, if the edge of a chunk does not correspond to the edge of the pixels, the output will not be strictly continuous and will have edge artifacts (that might not be visible). Users can check this with the options raster_alignment, that can take the resolution of the raster as input, as well as the starting point if needed. The following are accepted:

# check if chunks are aligned with a raster of resolution 20
raster_alignment = 20
raster_alignment = list(res = 20)

# check if chunks are aligned with a raster of resolution 20
# that starts at (0,10)
raster_alignment = list(res = 20, start = c(0,10))

Examples

# More examples might be avaible in the official lidR vignettes or
# on the github book <https://jean-romain.github.io/lidRbook/>

## =========================================================================
## Example 1: detect all the tree tops over an entire catalog
## (this is basically a reproduction of the existing function 'locate_trees')
## =========================================================================

# 1. Build the user-defined function that analyzes each chunk of the catalog.
# The function's first argument is a LAScluster object. The other arguments can be freely
# chosen by the users.
my_tree_detection_method <- function(chunk, ws)
{
  # The chunk argument is a LAScluster object. The users do not need to know how it works.
  # readLAS will load the region of interest (chunk) with a buffer around it, taking advantage of
  # point cloud indexation if possible. The filter and select options are propagated automatically
  las <- readLAS(chunk)
  if (is.empty(las)) return(NULL)

  # Find the tree tops using a user-developed method
  # (here simply a LMF for the example).
  ttops <- locate_trees(las, lmf(ws))

  # ttops is an sf object that contains the tree tops in our region of interest
  # plus the trees tops in the buffered area. We need to remove the buffer otherwise we will get
  # some trees more than once.
  bbox  <- st_bbox(chunk)
  ttops <- sf::st_crop(ttops, bbox)
  return(ttops)
}

# 2. Build a collection of file
# (here, a single file LAScatalog for the purposes of this simple example).
LASfile <- system.file("extdata", "MixedConifer.laz", package="lidR")
ctg <- readLAScatalog(LASfile)
plot(ctg)

# 3. Set some processing options.
# For this small single file example, the chunk size is 100 m + 10 m of buffer
opt_chunk_buffer(ctg) <- 10
opt_chunk_size(ctg)   <- 100            # Small because this is a dummy example.
opt_chunk_alignment(ctg) <- c(-50, -35) # Align such as it creates 2 chunks only.
opt_select(ctg)       <- "xyz"          # Read only the coordinates.
opt_filter(ctg)       <- "-keep_first"  # Read only first returns.

# 4. Apply a user-defined function to take advantage of the internal engine
opt    <- list(need_buffer = TRUE,   # catalog_apply will throw an error if buffer = 0
               automerge   = TRUE)   # catalog_apply will merge the outputs into a single object
output <- catalog_apply(ctg, my_tree_detection_method, ws = 5, .options = opt)

plot(output)


## =========================================================================
## Example 1: simplified. There is nothing that requires special data
## manipulation in the previous example. Everything can be handled automatically
##=========================================================================

# 1. Build the user-defined function that analyzes a point cloud.
my_tree_detection_method <- function(las, ws)
{
  # Find the tree tops using a user-developed method
  # (here simply a LMF for the example).
  ttops <- locate_trees(las, lmf(ws))
  return(ttops)
}

# 2. Build a project
LASfile <- system.file("extdata", "MixedConifer.laz", package="lidR")
ctg <- readLAScatalog(LASfile)
plot(ctg)

# 3. Set some processing options.
# For this dummy example, the chunk size is 100 m and the buffer is 10 m
opt_chunk_buffer(ctg) <- 10
opt_chunk_size(ctg)   <- 100            # small because this is a dummy example.
opt_chunk_alignment(ctg) <- c(-50, -35) # Align such as it creates 2 chunks only.
opt_select(ctg)       <- "xyz"          # Read only the coordinates.
opt_filter(ctg)       <- "-keep_first"  # Read only first returns.

# 4. Apply a user-defined function to take advantage of the internal engine
opt    <- list(need_buffer = TRUE)   # catalog_apply will throw an error if buffer = 0
output <- catalog_map(ctg, my_tree_detection_method, ws = 5, .options = opt)


## Not run: 
## ===================================================
## Example 2: compute a rumple index on surface points
## ===================================================

rumple_index_surface = function(las, res)
{
  las    <- filter_surfacepoints(las, 1)
  rumple <- pixel_metrics(las, ~rumple_index(X,Y,Z), res)
  return(rumple)
}

LASfile <- system.file("extdata", "Megaplot.laz", package="lidR")
ctg <- readLAScatalog(LASfile)

opt_chunk_buffer(ctg) <- 1
opt_chunk_size(ctg)   <- 140     # small because this is a dummy example.
opt_select(ctg)       <- "xyz"   # read only the coordinates.

opt    <- list(raster_alignment = 20) # catalog_apply will adjust the chunks if required
output <- catalog_map(ctg, rumple_index_surface, res = 20, .options = opt)

plot(output, col = height.colors(25))

## End(Not run)

lidR documentation built on Sept. 11, 2024, 5:21 p.m.