Lascanopy error, or not?

[basb...@gmail.com]

Hello,

For my master research project I try to create raster outputs based on point counts and densities at several height layers from point cloud information of the AHN2 in the Netherlands.

Therefore, I perform 2 steps:

- Recalculating the z from 'height above NAP' (Dutch z reference) towards 'height above ground'. 

- calculating point counts and density for 8 height layers.

My input dataset is a las file with ground points already classified. 

Now I am experiencing that the tiffs of the lowest three layers contain no information at all, while the new las file with height information seems to go smooth.

It almost seems that lascanopy uses point information of the original input file..

Does anyone have an idea what is happening?

Script and input data can be found in the following link:

https://www.dropbox.com/s/i2pbylce6dt5xz5/LiDAR_Processsing.zip?dl=0

Unzipping it in your D-drive and running the script should do the trick..

Any help is highly appreciated. Many thanks in advance!

Bas Boek

PS - attached script:

:: Calculating point height above ground

lasheight -i "D:\LiDAR_Processsing\0_First_Input\*.laz"

-replace_z 

-odir "D:\LiDAR_Processsing\1_New_Heights" -olas

:: Create height rasters 5m resolution

lascanopy -i "D:\LiDAR_Processsing\1_New_Heights\*.las" 

-step 5 -d 0.05 0.2 0.5 1 2 5 10 20 80 -c 0.05 0.2 0.5 1 2 5 10 20 80 

-odir "D:\LiDAR_Processsing\2_Rasters" -otif

PAUSE

[Floris Groesz]

Hello Bas,

I haven't looked at your data but you could start defining a -height_cutoff. The default value is 1.37 meter so this could explain why your lower height classes are empty. If you put it to 0.05 then you can get data in all your classes.
Another question is whether your lowest classes contain information about the vegetation. 5 to 20 cm as lowest class is quite low and I think you will be seeing a lot of noise in this class. You should also make sure you exclude overlapping flight lines from this analysis. A slight misalignment between flight lines will will add extra noise to your lowest height classes.

Floris

[basb...@gmail.com]

Hello Floris,

Thanks a lot, -height_cutoff indeed solved my problem! I got a warning that the height_cutoff was below 1.37 before, but I wasn't aware that points underneath would be omitted.

For my final product I will make height density rasters of 25m spatial resolution. Do you think that the size of the area (so 625 m2) would be enough to compensate for the lack of precision? Dataset info: the systematic error of the dataset should be max. 5 cm and the standard deviation is max. 5 cm as well.

Bastiaen

Op woensdag 27 mei 2015 10:40:06 UTC+2 schreef Floris Groesz:

[Floris Groesz]

A systematic error of max 5 cm and a standard deviation of max 5 doesn't mean that all terrain points will be within 5 cm and all points above 5 cm are vegetation.
according to the specifications of the AHN:
A minimum of 68.2% of all lidar points will have a height deviation of maximal 5 + 1*5= 10 cm
A minimum of 95.4% of all lidar points will have a height deviation of maximal 5 + 2*5= 15 cm
A minimum of 99.7% of all lidar points will have a height deviation of maximal 5 + 3*5= 20 cm

see: http://www.ahn.nl/binaries/content/assets/hwh---ahn/nieuws/2010/10/kwaliteitsdocumentahn.pdf

Your lowest height classes will mainly contain information about noise and deviations. Even though the specifications are for the whole dataset there will be some local differences. dense low vegetation will influence the ground classification and therefor the accuracy of the terrain model. There will also be local differences because of the fact that the dataset was acquired with different systems.
I don't think that using fairly large cells of 25x25 meter will compensate for this.

Anyway, it doesn't harm to make the lowest height classes you propose and inspect them carefully.

Floris

[Terje Mathisen]

Floris Groesz wrote:
> A systematic error of max 5 cm and a standard deviation of max 5
> doesn't mean that all terrain points will be within 5 cm and all
> points above 5 cm are vegetation.
> according to the specifications of the AHN:
> A minimum of 68.2% of all lidar points will have a height deviation of
> maximal 5 + 1*5= 10 cm
> A minimum of 95.4% of all lidar points will have a height deviation of
> maximal 5 + 2*5= 15 cm
> A minimum of 99.7% of all lidar points will have a height deviation of
> maximal 5 + 3*5= 20 cm

This is broken math!

In reality, the systematic error and std dev are (almost by definition)
uncorrelated, so that the 10 cm limit will cover 68% plus that half of
the remaining points where the error went in the opposite direction from
the systematic error, i.e. 68 + 32/2 = 84%.

However, since 84% is >= 68% the AHN specification is still correct. :-)

>
> see:
> http://www.ahn.nl/binaries/content/assets/hwh---ahn/nieuws/2010/10/kwaliteitsdocumentahn.pdf
>
> Your lowest height classes will mainly contain information about noise
> and deviations. Even though the specifications are for the whole
> dataset there will be some local differences. dense low vegetation
> will influence the ground classification and therefor the accuracy of
> the terrain model. There will also be local differences because of the
> fact that the dataset was acquired with different systems.
> I don't think that using fairly large cells of 25x25 meter will
> compensate for this.

Since my background is in making orienteering maps, this was pretty
obvious, so I selected 30 cm as the lower cutoff point:

Anything below that can be either a small hillock, a rocky outcrop, or
just a patch of very dense brush:

There is no way to know a priori exactly how fractal the true ground
surface is, and in wilderness it can be quite hard to determine exactly
anyway. I.e. is the lowest layer of slowly rotting leaves, pine needles
and grass, moss and other ground cover really a part of the ground
surface itself or vegetation?

Terje

>
> Anyway, it doesn't harm to make the lowest height classes you propose
> and inspect them carefully.
>
> Floris

> --
> Download LAStools at
> http://lastools.org
> http://rapidlasso.com
> Be social with LAStools at
> http://facebook.com/LAStools
> http://twitter.com/LAStools
> http://linkedin.com/groups/LAStools-4408378
> Manage your settings at
> http://groups.google.com/group/lastools/subscribe


--
- <Terje.M...@tmsw.no>
"almost all programming can be viewed as an exercise in caching"

[Floris Groesz]

@ Terje

This is broken math!

It would be if the sentence hadn't started with "a minimum of". Otherwise I agree with you. This specification is very focused on the maximal errors and not so much on how accurate most of the lidar points are.

Since my background is in making orienteering maps, this was pretty
obvious, so I selected 30 cm as the lower cutoff point:

Anything below that can be either a small hillock, a rocky outcrop, or
just a patch of very dense brush: There is no way to know a priori exactly how fractal the true ground
surface is, and in wilderness it can be quite hard to determine exactly
anyway. I.e. is the lowest layer of slowly rotting leaves, pine needles
and grass, moss and other ground cover really a part of the ground
surface itself or vegetation?

This AHN dataset is slightly denser and has a higher accuracy than most datasets in Norway so maybe the lower cutoff could be at 15 to 20 cm.

There is no chance to meet a small hillock or a rocky outcrop in this dataset either. The same for wilderness. Dense vegetation could be a problem even though the data has been acquired before the growing season.

Floris

[basb...@gmail.com]

Hi,

The AHN2 is already split and downloadable in nonground-only points and ground-only points (though all point classification information is not embedded in the dataset). Therefore, for my research I classified the points in the ground datasets again as ground points. @Floris: do you perhaps know the cutoff that the AHN2 organisation has used for their initial split? Perhaps good to know..

In any case I will check the quality of the lowest (5-20cm) layer for noise, thanks for feedback.

Bastiaen

Op vrijdag 29 mei 2015 12:34:46 UTC+2 schreef Floris Groesz:

[Floris Groesz]

No cut-off has been used to split the data into ground and no-ground.

The data has been classified into ground/no-ground and afterwards it has been manually checked/edited. There can always be points on the ground surface that have been classified as no-ground. This depends on the on the type of classification algorithm (and the settings) and on the lidar data. Some algorithms/settings are quite "including", resulting in many points in the ground class. Others can be more "strict", resulting in a smoother more generalized ground mode. One of the requirements in the AHN was that as many real ground points as possible were classified as ground, without compromising the accuracy of the ground model. However, there is always a trade-off here.

Floris