option to fill nodata in raster

[wiesehahn]

When processing e.g. CHMs /DTMs we might face areas/ pixels with nodata.

I am wondering if it would make sense to add a focal filtering stage which would allow e.g. to apply a median focal filter just on nodata areas.
There is pit_fill which does something similar, however there is no option run it only on nodata and hence it smoothes the entire raster.
We could also use a TIN but this also kind of smoothes the ratser and is probably not wanted.

library(lasR)
f <- system.file("extdata", "Topography.las", package="lasR")

read = reader_las(filter = "-keep_first -keep_class 2 -filter_and")
del1 = triangulate(filter = keep_ground())
norm = transform_with(del1, "-")
chm1 = rasterize(2, "max")
pit = pit_fill(chm1)
del2 = triangulate(filter = keep_first())
chm2 = rasterize(2, del2)
pipeline <- read + del1 + norm + chm1 + pit + del2 + chm2 
ans <- exec(pipeline, on = f, with = list(ncores = concurrent_points(half_cores())))

terra::plot(ans[[1]])

terra::plot(ans[[2]])

terra::plot(ans[[3]])

[Jean-Romain]

I think it is important to have no data. It shows users what they are currently doing such as requesting a too fine grain output with a too low density point cloud. Also lasR like lidR are point cloud processing software I'm not keen too reinvent the wheel too much with raster features.

Anyway I planned to add a focal function for raster to smooth the canopy and improve tree detection and segmentation. I could add a function to fill NAs.

[wiesehahn]

I think it is important to have no data. It shows users what they are currently doing such as requesting a too fine grain output with a too low density point cloud.

Maybe, but a focal filter with small kernel would not fill the water areas above but simply fill single NA pixels which are then assumed to be as high as average sourrounding.

Also lasR like lidR are point cloud processing software I'm not keen too reinvent the wheel too much with raster features.

Alright, I understand that, could be done with gdal as well.

[Jean-Romain]

Alright, I understand that, could be done with gdal as well.

Ideally yes post prod should be done with terra or another GIS software. But sometime post-prod is required to feed next stages and lasR will try to cover these cases when possible. I also plan to add a raster processing engine to work with raster without the point cloud

chm = reader_raster()
lmf =  local_maximum_raster(chm)
seg  = region_growning(chm, lmf)
pipeline = chm + lmf + seg

[Jean-Romain]

v0.10.0

library(lasR)
f <- system.file("extdata", "Topography.las", package = "lasR")

chm = rasterize(2, "zmax")
chm2 = lasR:::focal(chm, 8, fun = "mean")
chm3 = lasR:::focal(chm, 8, fun = "max")
pipeline <- reader_las() + chm + chm2 + chm3
u = exec(pipeline, on = f)

terra::plot(u[[1]])

terra::plot(u[[2]])

terra::plot(u[[3]])

^{Created on 2024-08-01 with reprex v2.1.0}

[wiesehahn]

For my future self:
Its simple and efficient as following:

library(lasR)
#> Warning: package 'lasR' was built under R version 4.4.2
f <- system.file("extdata", "Topography.las", package = "lasR")

chm = rasterize(2, "zmax")
pipeline <- reader_las() + chm
ans = exec(pipeline, on = f)
terra::plot(ans)

gdalraster::fillNodata(terra::sources(ans), max_dist = 3, band=1)
#> 0...10...20...30...40...50...60...70...80...90...100 - done.
terra::plot(ans)

^{Created on 2024-12-07 with reprex v2.1.0}