tree.detection.multi.scan: Tree-Level Variables Estimation

In FORTLS: Automatic Processing of Terrestrial-Based Technologies Point Cloud Data for Forestry Purposes

Tree-Level Variables Estimation

Description

Detects trees from point clouds corresponding to TLS multi-scan approaches and SLAM devices. For each tree detected, the function calculates the central coordinates and estimates the diameter at 1.3 m above ground level (which is known as dbh, diameter at breast height) and classifies it as fully visible or partially occluded. Finally, the function obtains the number of points belonging to normal sections of trees (those corresponding to dbh +/- 5 cm) and estimates them for both original and reduced (with random selection process) point clouds.

Usage

tree.detection.multi.scan(data, single.tree = NULL,
                          dbh.min = 4, dbh.max = 200, h.min = 1.3,
                          ncr.threshold = 0.1,
                          tls.precision = NULL,
                          stem.section = NULL, stem.range = NULL, breaks = NULL,
                          slice = 0.1, understory = NULL, bark.roughness = 1,
                          den.type = 1, d.top = NULL,
                          plot.attributes = NULL,
                          save.result = TRUE, dir.result = NULL)

Arguments

data	Data frame with same description and format as indicated for normalize ‘Value’.
single.tree	Optional argument to indicate if there is only one tree.
dbh.min	Optional minimum dbh (cm) considered for detecting trees. By default it will be set at 4 cm.
dbh.max	Optional maximum dbh (cm) considered for detecting trees. By default it will be set at 200 cm.
h.min	Optional minimum h (m) considered for detecting trees. By default it will be set at 1.3 m.
ncr.threshold	Local surface variation (also known as normal change rate, NCR). By default it will be set as 0.1. For better understanding of this argument see ‘Details’.
tls.precision	Average point cloud precision in cm.
stem.section	Section free of noise (shurb, branches, etc.) considered to detect trees. If not specified, an automatic internal algorithm will be applied (see ‘Details’).
stem.range	Section considered to estimate straightness tree attributes.
breaks	Height above ground level (m) of slices considered for detecting trees. By default it will be considered all possible sections from 0.4 m to maximum height by 0.3 m intervals (+/- 5 cm).
slice	Slice width considered for detecting trees. By default it will be considered as 0.1 m.
understory	Optional argument to indicate if there is dense understory vegetation.
bark.roughness	Bark roughness established in 3 degrees (1 < 2 < 3). By default it will be considered as 1.
den.type	Numeric argument indicating the dendrometic type used to estimate volumen when there are not sections enough to fit a taper equation. Dendrometrics types available are the following: cylinder = 0, paraboloid = 1 (by default), cone = 2 and neiloid = 3.
d.top	Top stem diameter (cm) considered to estimate commercial timber volume.
plot.attributes	Data frame with attributes at plot level. It must contain a column named id (character string or numeric value) with encoding coinciding with that used in id argument of normalize for identifying plots. If there are strata, another column named ‘stratum’ (numeric) will be required for other functionalities of FORTLS (see, for instance, estimation.plot.size or metrics.variables). If this argument is not specified by the user, it will be set to NULL by default and, as a consequence, the function will not add these possible plot attributes.
save.result	Optional logical which indicates whether or not the output files described in ‘Output Files’ section should be saved in dir.result. If this argument is not specified by the user, it will be set to TRUE by default and, as a consequence, the output files will be saved.
dir.result	Optional character string naming the absolute path of an existing directory where the files described in ‘Output Files’ section will be saved. .Platform$file.sep must be used as the path separator in dir.result. If this argument is not specified by the user, and the save.result is TRUE, it will be set to NULL by default and, as a consequence, the current working directory of the R process will be assigned to dir.result during the execution.

Details

Slices determined by breaks argument are clustered using the DBSCAN algorithm (Ester et al., 1996) on the horizontal plane according to Cartesian coordinates (x, y). Before and after this process, several algorithms are used to remove noisy points and apply classification criteria to select the clusters of trees.

dbh is directly estimated for the section of 1.3 m above ground level, and estimated from other sections using dbh~breaks linear regression. Finally, the mean value of all estimates is provided in ‘Value’ as the dbh of the tree section.

Volume is estimated modelling stem profile as a paraboloid and calculating the volumes of revolution; where trees dbh are estimated in tree.detection.single.scan, and total heights are estimated as percentile 99 of z coordinate of points delimited by Voronoi polygons.

The number of points corresponding to a normal section (+/- 5 cm) is estimated in proportion to dbh, using the average number of points per radius unit as reference. In this respect, only tree sections fully visible at 1.3 m above ground level will be considered for estimating the average number of points.

Local surface variation (also known as normal change rate ,NCR), is a quantitative measure of curvature feature (Pauly et al., 2002). This is useful for distinguishing points belonging to fine branches and foliage (e.g. leaves, shrubs) and stem points (e.g. Jin et al., 2016; Zhang et al., 2019). Just as we considered 5 cm as suitable for calculating local surface variation for the stem separation in forests, according to other authors (Ma et al., 2015; Xia et al., 2015), we also established the NCR threshold as 0.1, according to Zhang et al. (2019). However, this argument (ncr.threshold) may be modified in order to use more appropriate values.

Value

Data frame with the following columns for every tree detected (each row corresponds to one tree detected):

id	Optional plot identification encoded as a character string or numeric. If this argument is not specified by the user, it will be set to NULL by default and, as a consequence, the plot will be encoded as 1.
file	Optional file name identification encoded as character string or numeric. If it is null, the file will be encoded as id by default.
tree	tree numbering
Coordinates	Cartesian (according to https://en.wikipedia.org/wiki/Cartesian_coordinate_system notation):

• x: distance on x axis (m) of tree centre.

• y: distance on y axis (m) of tree centre.

Azimuthal angles:

• phi: angular coordinate (rad) of tree centre.

h.dist	horizontal distance (m) from plot centre to tree centre.
dbh	estimated tree diameter (cm) at breast height (1.3 m).
h	estimated tree total height (m).
h.com	estimated commercial tree height (m) according to the top diameter defined in the argument d.top.
v	estimated tree stem volume (m^3).
v.com	estimated commercial tree stem volume (m^3) according to the top diameter defined in the argument d.top.
n.pts	number of points corresponding to a normal section (+/- 5 cm) in the original point cloud.
n.pts.red	number of points corresponding to a normal section (+/- 5 cm) in the point cloud reduced by the point cropping process.
n.pts.est	number of points estimated for a normal section (+/- 5 cm) in the original point cloud.
n.pts.red.est	number of points estimated for a normal section (+/- 5 cm) in the point cloud reduced by the point cropping process.
partial.occlusion	yes (1) or no (0)

Output Files

At the end of the tree detection process, if the save.result argument is TRUE, the function will save the data frame described in ‘Value’ as a CSV file named ‘tree.tls.csv’. The data frame will be written without row names in the dir.result directory by using write.csv function from the utils package.

Note

Although tree.detection.multi.scan also works with reduced point clouds, thus reducing the computing time, use of the original point cloud is recommended in order to detect more trees. This will also depend on forest conditions, especially those related to visibility. The more distant the trees are, the lower the density of points will be, and using reduced point clouds will therefore complicate detection of the most distant trees.

Note that dbh.min and dbh.max are important for avoiding outlier values when inventory data are used for reference purposes. Otherwise, knowledge about the autoecology of species could be used for filtering anomalous values of dbh.

The argument breaks = 1.3 could be sufficient for detecting trees visible at dbh, involving lower computational cost. However, those trees not detected at dbh, may be estimated from lower and/or higher sections. Considering the three default sections in the argument breaks = c(1.0, 1.3, 1.6) maintains a good balance in the case study of this package.

Author(s)

Juan Alberto Molina-Valero and Adela Martínez-Calvo.

References

Ester, M., Kriegel, H. P., Sander, J., & Xu, X. (1996). A density-based algorithm for discovering clusters in large spatial databases with noise. In Kdd (Vol. 96, No. 34, pp. 226-231).

Jin, S., Tamura, M., & Susaki, J. (2016). A new approach to retrieve leaf normal distribution using terrestrial laser scanners. J. Journal of Forestry Research, 27(3), 631-638. \Sexpr[results=rd]{tools:::Rd_expr_doi("https://doi.org/10.1007/s11676-015-0204-z")}

Ma, L., Zheng, G., Eitel, J. U., Moskal, L. M., He, W., & Huang, H. (2015). Improved salient feature-based approach for automatically separating photosynthetic and nonphotosynthetic components within terrestrial lidar point cloud data of forest canopies. IEEE Transactions Geoscience Remote Sensing, 54(2), 679-696. \Sexpr[results=rd]{tools:::Rd_expr_doi("https://doi.org/10.1109/TGRS.2015.2459716")}

Pauly, M., Gross, M., & Kobbelt, L. P., (2002). Efficient simplification of point-sampled surfaces. In IEEE Conference on Visualization. (pp. 163-170). Boston, USA. \Sexpr[results=rd]{tools:::Rd_expr_doi("https://doi.org/10.1109/VISUAL.2002.1183771")}

Xia, S., Wang, C., Pan, F., Xi, X., Zeng, H., & Liu, H. (2015). Detecting stems in dense and homogeneous forest using single-scan TLS. Forests. 6(11), 3923-3945. \Sexpr[results=rd]{tools:::Rd_expr_doi("https://doi.org/10.3390/f6113923")}

Zhang, W., Wan, P., Wang, T., Cai, S., Chen, Y., Jin, X., & Yan, G. (2019). A novel approach for the detection of standing tree stems from plot-level terrestrial laser scanning data. Remote Sens. 11(2), 211. \Sexpr[results=rd]{tools:::Rd_expr_doi("https://doi.org/10.3390/rs11020211")}

See Also

normalize, tree.detection.single.scan, tree.detection.several.plots, distance.sampling, estimation.plot.size, simulations, metrics.variables

Examples

# Establishment of working directories (optional)
# By default here we propose the current working directory of the R process

dir.data <- getwd()
dir.result <- getwd()


# Loading example data of TLS multi-scan approach point cloud (LAZ file) to dir.data

download.file(
"www.dropbox.com/scl/fi/es5pfj87wj0g6y8414dpo/PiceaAbies.laz?rlkey=ayt21mbndc6i6fyiz2e7z6oap&dl=1",
              destfile = file.path(dir.data, "PiceaAbies.laz"),
              mode = "wb")

# Normalizing the whole point cloud data without considering arguments

pcd <- normalize(las = "PiceaAbies.laz",

                 id = "PiceaAbies",

                 scan.approach = "multi",

                 dir.data = dir.data, dir.result = dir.result)


# Tree detection without considering arguments

tree.tls <- tree.detection.multi.scan(data = pcd[pcd$prob.selec == 1, ],

                                      stem.section = c(1, 5),

                                      dir.result = dir.result)

  

FORTLS documentation built on Sept. 11, 2023, 5:09 p.m.