Difference in M3C2 range in CC vs Arc

[Linzoid3000]

Hi Daniel,

I have performed M3C2 on several clouds against my reference.
The resulting data contains 6 bands.
In CC they are listed as
1. none
2. SD of compared cloud
3. SD of reference cloud
4. Change significance
5. Distance uncertainty
6. M3C2 distance

I recorded the range of values in each band as listed in CC.
I then exported the data (and all the scalar bands) as rasters, and looked at them in Arc.
I then recorded the ranges listed for each band as they appear in Arc.
The ranges do not match. Additionally, band 1 in Arc has data, unlike in CC (as band 1 is listed as "none").

Can you help me understand why this may be the case?
The results come from CC, but are different in Arc....
Which should I use?

Thanks, Lindsay

[daniel]

How different are the ranges? And what do you mean by

band 1 in Arc has data, unlike in CC (as band 1 is listed as "none")

? Are you referring to the 'None' entry of the 'Active' drop-down list in the cloud properties ('Scalar field' section)?

If yes then this 'None' entry doesn't correspond to an actual band, but simply means that no band is currently active / displayed. The first band is actually the second entry in this list.

[Linzoid3000]

Hi Daniel,

Yes, you're right.
I'm talking about the "active" dropdown list.
In CC, there are 5 scalar fields. (6 if you include "none").
When I export as rasters, I choose the option to export all scalars.
When opened in Arc, these rasters have 6 bands.

Example of range in data:

CloudCompare (read from cloud properties)
None: n/a
STD cloud 1: 0.00001538 to 2.50038052
STD cloud 2: 0.00230297 to 2.93335223
significant change: 0 to 1
distance uncertainty: 0.00104423 to 0.30315742
M3C2 distance: -5.56904411 to 4.11246014

ArcGIS (read from cell statistics in properties)
Band 1: -0.95372 to 10.39015
Band 2: 0.0000154 to 2.500381
Band 3: 0.002303 to 2.933352
Band 4: 0 to 1
Band 5: 0.001044 to 0.303157
Band 6: -3.66435 to 2.09861

So, as you can see, most bands match in range between the platforms, with two important exceptions.
Band 1 and 6, or "none" and "m3c2 distance".

Because the M3C2 distance is of course what I'm most interested in, I'm worried about the difference in statistics from the same results....
Why would the same file have such different ranges when viewed in different software?
And what is represented in band 1? The range of elevation in the original cloud?

Thanks for your time and thoughts.
Lindsay

[daniel]

The first band in the raster is the 'height' band (generated by the Rasterize tool). This is why you can't find it in the cloud scalar fields.

Then the bands of the raster are obtained by projecting the points of the cloud inside the regular grid. In each cell, the height is either the minimum, average or maximum height of all the points falling inside. And it's the same for scalar values. Therefore the range of the raster band depends on:
- the 'interpolate SF(s)' method (min, max or average)
- how many points are falling inside each cell (which generally depends on how big the cells are)

If your raster resolution is low, then it may explain the big differences in the M3C2 distance bands. Also, if the largest distances in the original cloud are in fact only corresponding to a few points (i.e. the tail of the distribution), they will probably be 'smoothed' out by the rasterizing process. You should check the histogram of the M3C2 scalar field: make the 'M3C2 distance' active and then call 'Edit > SF > Show histogram'. Check the raster boundaries against the histogram.

Don't hesitate to post the snapshot of the histogram here. You can also send me the original cloud and the parameters you used for the Rasterize tool.

[daniel]

By the way you can also export the raster grid as a cloud. You'll see then that the scalar fields / bands are (should be) the same.