Skip to contents

Estimate HMD using tree crown polygons and normalized point cloud data

Source: R/trees_hmd.R
trees_hmd.Rd

trees_hmd() uses the input tree crown polygons (e.g. as exported by raster2trees()) with the columns treeID and tree_height_m to extracting the height of the maximum crown diameter (HMD) using height normalized point cloud data (e.g. as exported by cloud2raster()).

HMD is extracted directly from the height normalized point cloud by finding the height of the non-ground point farthest from the tree center (i.e. tree top).

An early version of this process was developed by Andrew Sanchez Meador.

There are likely to be trees for which there is insufficient data in the point cloud to successfully estimate HMD. The user can elect to estimate missing HMD values which is accomplished via:

  • Attempt to extract HMD from all trees

  • Successfully extracted HMD trees become training data used to estimate the height-HMD allometry relationship that is spatially informed using the relative tree location compared to the training data

  • The height and location predicting HMD model built from the point cloud training data is used to predict HMD for the non-training (i.e. missing HMD) data

Usage

trees_hmd(
  trees_poly,
  norm_las = NULL,
  tree_sample_n = NA,
  tree_sample_prop = NA,
  estimate_missing_hmd = TRUE,
  force_same_crs = F,
  outfolder = tempdir()
)

Arguments

trees_poly

must be one of the following that has required attributes treeID and tree_height_m:

  • sf class object with POLYGON geometry (see sf::st_geometry_type()). Recommended for smaller tree lists (e.g. <100k) that can fit in memory.

  • character vector with the path to a single or multiple spatial files that can be read by sf::st_read() and have POLYGON geometry. Recommended for large tree lists (e.g. 100k+) that might cause memory issues.

  • character with the path to a directory that has "final_detected_crowns*" files from cloud2trees() or raster2trees(). Recommended for large tree lists (e.g. 100k+) that might cause memory issues.

norm_las

character. a directory with nomalized las files, the path of a single .laz|.las file", -or- an object of class LAScatalog. It is your responsibility to ensure that the point cloud is projected the same as the trees_poly data

tree_sample_n, tree_sample_prop

numeric. Provide either tree_sample_n, the number of trees, or tree_sample_prop, the proportion of the trees to attempt to extract a HMD from the point cloud for. If neither are supplied, tree_sample_n = 777 will be used. If both are supplied, tree_sample_n will be used. Increasing tree_sample_prop toward one (1) will increase the processing time, perhaps significantly depending on the number of trees in the trees_poly data.

estimate_missing_hmd

logical. it is not likely that HMD will be extracted successfully from every tree (especially in low density clouds). Should the missing HMD values be estimated using the tree height and location information based on trees for which HMD is successfully extracted?

force_same_crs

logical. force the same crs between the point cloud and polygon if confident that data are in same projection. data created by a cloud2trees pipeline (e.g. cloud2raster()) will always have the same projection even if not recognized by lidR functions

outfolder

string. The path of a folder to write the model data to. Note, in the actual missing value estimation many RF models are estimated and model averaging is used. However, only the first estimated model is saved in this export which does not fully represent the process used to fill in missing values.

Value

Returns a spatial data frame of individual trees with the added columns: max_crown_diam_height_m, is_training_hmd

References

An early version of this process was developed by Andrew Sanchez Meador.

Examples

 if (FALSE) { # \dontrun{
  library(tidyverse)
  library(sf)
  # example tree crown polygons
  f <- paste0(system.file(package = "cloud2trees"),"/extdata/crowns_poly.gpkg")
  crowns <- sf::st_read(f, quiet = T)
  # example normalized las files are in this directory
  norm_d <- paste0(system.file(package = "cloud2trees"),"/extdata/norm_las")
  # now run the trees_hmd()
  trees_hmd_ans <- trees_hmd(
     trees_poly = crowns
     , norm_las = norm_d
     , force_same_crs = T
     , estimate_missing_hmd = T)
  # what?
  trees_hmd_ans %>% dplyr::glimpse()
  # spatial polygons
  trees_hmd_ans %>% ggplot2::ggplot() +
     ggplot2::geom_sf(ggplot2::aes(fill=max_crown_diam_height_m))
  # relationship between height and hmd
  trees_hmd_ans %>%
     ggplot2::ggplot(
       ggplot2::aes(
         x = tree_height_m, y = max_crown_diam_height_m, color=is_training_hmd
       )
     ) +
     ggplot2::geom_point()
  #### try a file list
  #### Recommended for large tree lists (e.g. 100k+) that might cause memory issues.
  # we'll split the crowns
  # as is done automatically for tree lists >250k by raster2trees() and cloud2trees()
  crowns <- crowns %>%
    dplyr::mutate(
      # makes 2 groups of data
      grp = ceiling(dplyr::row_number()/(dplyr::n()/2))
    )
  # make file names
  my_dir <- tempdir()
  fnm_1 <- file.path(my_dir, "crowns1.gpkg")
  fnm_2 <- file.path(my_dir, "crowns2.gpkg")
  fnm_1
  # write the data
  sf::st_write(crowns %>% dplyr::filter(grp==1), dsn = fnm_1, append = F) # grp 1
  sf::st_write(crowns %>% dplyr::filter(grp==2), dsn = fnm_2, append = F) # grp 2
  # try trees_cbh with our file list
  flist <- c(fnm_1,fnm_2)
  # now run the trees_hmd()
  trees_hmd_ans2 <- trees_hmd(
     trees_poly = flist
     , norm_las = norm_d
     , force_same_crs = T
     , estimate_missing_hmd = T)
  # tabulate training data
  trees_hmd_ans %>%
    sf::st_drop_geometry() %>%
    dplyr::count(is_training_hmd)
 } # }