DTM outputs have a shift between tiles about whitebox-tools

[tombe-nm]

I tried merging the laz into one file and then running lidar tile gridding.

I get a constant shift north west throughout the raster compared to lastools output

[vsydorov]

Hi,

As you have correctly suspected it's a matter of whitebox-tools performing sampling in the corner of the pixel, instead of the center like lastools does. In my personal opinion, this is a bug, and the behaviour of lastools is to be preferred. When displaying the DEM in QGIS we can see that generated raster differs from the position of the original las points.

This is especially visible on the raster corners (Picture 1):

Or when one interpolates at high enough resolution to see the mismatch between the TIN points and the underlying pointclouds (Picture 2):

I've managed to fix for this for my own purposes by sidestepping this behaviour (or maybe we can call it a bug). I forcibly shift the points by half-a-resolution-step before processing them with the LidarTINGridding tool. This gives me results identical to las2dem outputs, bar numerical instabilities. Here is the code that works on my data which uses EPSG:32648 CRS. I use lastools to perform this shift and set a right bounding box, but I hope to find a solution using other opensource tools. I don't think WBT itself has the functionality needed to set the bounding box directly.

# *** How to create DTM at 0.5 that match between lastools and White Box Tools ***

# Put the .LAS into tight borders that we want in the resulting GTIFF
las2las -i "C387014760.las" -o "C387014760_tight.las" -auto_reoffset
lasinfo -i "C387014760_tight.las" -set_bounding_box 387000.00 1476000.00 12.00 387500.00 1476500.00 36.00

# = Creating C387014760_LT.tif in lastools  =
# Apply las2dem on this data, forcing high precision of elevation outputs to match WBT precision
las2dem -i "C387014760_tight.las" -keep_class 2 -step 0.5 -cpu64 -o "C387014760_LT.tif" -force_precision -float_precision 0.000001

# = Creating C387014760_WBT.tif in White Box Tools =
# Shift las points by half a resolution step
las2las -i "C387014760_tight.las" -o "C387014760_shifted.las" -translate_x -0.25 -translate_y +0.25
# Force the bounding box to match the previous tight box
lasinfo -i "C387014760_shifted.las" -set_bounding_box 387000.00 1476000.00 12.00 387500.00 1476500.00 36.00
# Apply WBT on shifted, tight data
whitebox_tools -v=true --compress_rasters=True --max_procs=32 --run=LidarTINGridding --input "C387014760_shifted.las" --parameter="elevation" --returns="all" --resolution=0.5 --exclude_cls="1" --max_triangle_edge_length=100.0 -o "C387014760_WBT.tif"

For the moment being I will use this trick to get proper TIN gridding with WBT

[tombe-nm]

Thanks a lot for this. This works for me also.

However there is a very slight shift in the result on my end still due to the fact I have used a 0.5 step in lastools. (This ensures the edges between tiles are seemless.)

I am doing the transformation on the x and y in each as tile beforehand using laspy.

As my data is in feet and my required resolution is 3 so 1.5m shift. I am shifting:

                las.x += -4.92126 # Shift on the x-axis 
                las.y += 4.92126 # Shift on the y-axis 

To fix the step issue in the wb version. I round the XLLCORNER AND YLLCORNER. Eg:

NCOLS 2255
NROWS 844
XLLCORNER 804729.1
YLLCORNER 444923.72
CELLSIZE 3
NODATA_VALUE -32768.00

NCOLS 2255
NROWS 844
XLLCORNER 804729.0 #technically now XLLCENTER
YLLCORNER 444924.0 #technically now YLLCENTER
CELLSIZE 3
NODATA_VALUE -32768.00

Then I get perfect alignment:

Cheers Vlad