delineate_lakes: Individual waterbodies segmentation

In Jean-Romain/lidRplugins: Extra functions and algorithms for lidR package

Individual waterbodies segmentation

Description

Individual surfacic waterbodies segmentation to find the contour of the waterbodies (lakes, rivers). This function is designed to be used on raw point clouds with 1 point per square meter. Point-clouds must be loaded with option filter = "-thin_with_grid 1".

Usage

delineate_lakes(
  las,
  tol = 1/1000,
  tol2 = c(1/30 * tol, 2 * tol),
  trim = 1000,
  p = 0.5,
  res = 5,
  th1 = 25,
  th2 = 6,
  k = 10
)

Arguments

las	A LAS or LAScatalog object.
tol	numeric. Positive value. The method tracks the flat regions. This parameter gives the tolerance to the deviation to 'perfectly flat' region. 0 means 0 tolerance so meighboring points must be perfectly aligned exactly on a. A value ranging between 1/10000 and 1/100 might be good. This parameter is of major importance and greatly affect the output.
tol2	numeric. Positive values. To speed-up the computation a raster-based pre-processing is made. This parameter has the same meaning than tol but applies to the raster step. It can contains up to 2 values because the preprocessing step allows for two passes (see details and references). This parameter is of major importance and greatly affect the output. A very small value like 1/1000 is expected.
trim	numeric. Bodies with an area smaller than this value are removed. This param reduce oversegmentation of small bodies than may be mistaken with a waterbody
p	numeric. Value between 0 and 1. During the pre-processing raster-based step, cells that have a probability below 'p' to be a lake are not considered in subsequent processing (see references). This parameter is not of major importance
res	numeric. Resolution of the raster use in the pre-processing steps. This parameter is not of major importance
th1, th2, k	Param of the function shp_hplane that is used internally. Notice that th3 is set to 1-tol.

Details

See the reference for more details. The methods relies on two steps. A raster-based pre-processing step that segments roughtly the waterbodies and a vector-based step that fine accuraly the contours of the lakes. The raster-based step is made of two steps: (a) a very conservative estimation in which the expected output is a raster that contains all the waterbodies without false positive but the shape of the waterbodies might be very poorly estimated and (b) an permissive step in which the shapes of the waterbodies are almost correct but with potential false positive. The two steps are then merged internally in a single accurate raster. Using this raster to speed-up the computation of the vector-based step, the method computes a very accurate contour of the waterbodies.

Value

A SpatialPolygons.

Tips for finding adequante parameters

User can apply the function with tol = NULL to return only the raster-based step output. Also the parameter tol2 can be only 1 numbers to return only one estimation. tol2 is very important and the users must ensure than the conservative step only retains true positive (using a very conservative value such as 1/10000) and the permissive step have a good estimation of the shape even at the cost of many false positives that are anyway removed later using the conservative step. #'

# Run only the raster-based conservative step (output is a RasterLayer)
r <- delineate_lakes(las, tol = NULL, tol2 = 1/30000)
# Run only the raster-based permissive step (output is a RasterLayer)
r <- delineate_lakes(las, tol = NULL, tol2 = 2/1000)
# Run both raster-based steps (output is a RasterStack)
r <- delineate_lakes(las, tol = NULL, tol2 = c(1/30000, 2/1000))

Supported processing options

Supported processing options for a LAScatalog. For more details see the LAScatalog engine documentation:

• chunk size: How much data is loaded at once.

• chunk buffer: The buffer size is estimated internally

• chunk alignment: Align the processed clusters.

• progress: Displays a progress estimate.

• output_files: Return the output in R or write each cluster's output in a file. Supported templates are XLEFT, XRIGHT, YBOTTOM, YTOP, XCENTER, YCENTER ID and, if chunk size is equal to 0 (processing by file), ORIGINALFILENAME.

• select: 'xyz' are loaded by default.

• filter: "-thin_with_grid 1" is the default.

References

Roussel Jean-Romain, Segmentation of water bodies from ALS data without spectral information (in prep.)

Examples

LASfile <- system.file("extdata", "Topography.laz", package="lidR")
las <- readLAS(LASfile, filter = "-thin_with_grid 1")

lake <- delineate_lakes(las, trim = 700)
plot(header(las))
plot(lake, add = TRUE, col = "cornflowerblue")

## Not run: 
mapview::mapview(lake)

## End(Not run)

Jean-Romain/lidRplugins documentation built on Feb. 8, 2023, 5:39 a.m.