Recently I worked with LiDAR from an Optech Gemini scanner in the Philippines and with LiDAR from a RIEGL Q680i scanner in Thailand. The two devices scan the terrain below with a very different pattern: the Optech uses an oscillating mirror producing zig-zag scan lines, whereas the RIEGL uses a rotating polygon producing parallel scan lines. In the following we investigate the point distribution in a small piece cut from each flightline.
First we compute an estimate for the average point density and the average point spacing with lasinfo:
D:\lastools\bin>lasinfo -i optech.laz ^ -nh -nv -nmm -cd number of last returns: 2330671 covered area in square units/kilounits: 1280956/1.28 point density: all returns 2.25 last only 1.82 (per unit^2) spacing: all returns 0.67 last only 0.74 (in units)
The option ‘-nh’ asks lasinfo not to print the header (aka ‘no header’), the option ‘-nv’ means not to print the variable length records (aka ‘no vlr’), the option ‘-nmm’ supresses the output of minimum and maximum point values (aka ‘no min max’), and the option ‘-cd’ requests that an average point density is computed (aka ‘compute density’) from which the average point spacing is derived.
D:\lastools\bin>lasinfo -i riegl.laz ^ -nh -nv -nmm -cd number of last returns: 3707217 covered area in square units/kilounits: 889816/0.89 point density: all returns 4.58 last only 4.17 (per unit^2) spacing: all returns 0.47 last only 0.49 (in units)
What we really want to know are pulse density and pulse spacing which we get by only counting one return from every pulse. Commonly one uses the last return, which is reported as ‘last only’ values by lasinfo. According to lasinfo the Optech and the RIEGL scan have an average pulse density of 1.82 and 4.17 [pulses per square meter] and an average pulse spacing of 0.74 and 0.49 [meters] respectively.
Single number averages do not capture anything about the actual distribution of the last returns in the swath of the scan. Let us compute density rasters with lasgrid. Because the density of the two scans is different we use 3 by 3 meter cells for the Optech, which should average 3 * 3 * 1.82 = 16.38 points per cell, and 2 by 2 meter cells for the RIEGL, which should average 2 * 2 * 4.17 = 16.68 points per cell.
lasgrid -i optech.laz -last_only ^ -density -step 3 ^ -odix _density3x3 -obil lasview -i optech_density3x3.bil lasgrid -i riegl.laz -last_only ^ -density -step 2 ^ -odix _density2x2 -obil lasview -i riegl_density2x2.bil
A visual comparison of the two resulting density rasters in BIL format with lasview shows first differences in pulse distribution for the two scanners.
It is noticable that the number of last returns per cell increases at the edges of the swath for the Optech whereas it decreases for the RIEGL. For the Optech it is higher due to the slowing down of the oscillating mirror and the reversing of scan direction when reaching either side of the scanline. This decreases the distances between pulses at the edges of the zig-zag scan and increases the number of last returns falling into cells there. For the rotating polygon scan of the RIEGL these pulse distances are more uniform. Here the numbers are lower because many cells along the edges of the swath are only partly covered by the scan and therefore receive fewer last returns.
lasgrid -i optech.laz -last_only ^ -density -step 3 ^ -false -set_min_max 0 20 ^ -odix _density3x3 -opng lasgrid -i riegl.laz -last_only ^ -density -step 2 ^ -false -set_min_max 0 20 ^ -odix _density2x2 -opng
Here another visualization of the pulse distribution by letting lasgrid false-color density rasters to a fixed range of 0 to 20. This range is based on the near identical expected values of around 16.38 and 16.68 last returns per cell based on the average densities that lasinfo reported and the cell sized used here.
A more quantitative check of how well distributed the pulse are is to generate histograms. You can use lasinfo with the ‘-histo z 1’ option to print a histogram of the z values of the BIL rasters and import these statistics into your favorite software to create a chart.
lasinfo -i optech_density3x3.bil ^ -nh -nv -nmm -histo z 1 lasinfo -i riegl_density2x2.bil ^ -nh -nv -nmm -histo z 1
Both distributions have their peaks at the expected 16 to 17 last returns per cell although the Optech distribution has a significantly wider spead. The odd two-peak distribution of the Optech scan seems puzzling at first but looking at the density image shows that this is an aliasing artifact of putting a fixed 3 by 3 meter grid over zig-zagging scanlines.
The next experiments uses the new ‘-edge_shortest’ and ‘-edge_longest’ options available in las2dem that compute for each point the length of its shortest or longest edge in a Delaunay triangulation and then rasters these lengths. Below you see the command lines to do this and generate histograms of the rastered edge lengths.
las2dem -i optech.laz -last_only ^ -step 1.5 ^ -edge_longest ^ -odix _edge_longest -obil lasinfo -i optech_edge_longest.bil ^ -nh -nv -nmm -histo z 0.1 ^ -o optech_edge_shortest_0_10.txt las2dem -i optech.laz -last_only ^ -step 1.5 ^ -edge_shortest ^ -odix _edge_shortest -obil lasinfo -i optech_edge_shortest.bil ^ -nh -nv -nmm -histo z 0.05 ^ -o optech_edge_shortest_0_05.txt las2dem -i riegl.laz -last_only ^ -edge_longest ^ -odix _edge_longest -obil lasinfo -i riegl_edge_longest.bil ^ -nh -nv -nmm -histo z 0.05 ^ -o riegl_edge_longest_0_05.txt las2dem -i riegl.laz -last_only ^ -edge_shortest ^ -odix _edge_shortest -obil lasinfo -i riegl_edge_shortest.bil ^ -nh -nv -nmm -histo z 0.05 ^ -o riegl_edge_shortest_0_05.txt
These resulting histograms do look quite different.
The more important histograms are those of longest edge lengths. They illustrate the observed spacing between pulses showing us the maximal distance of each pulse from its surrounding pulses. A perfect pulse distribution that samples the ground evenly in all directions would have one narrow peak. The Optech scan is far from this ideal with a wide and flat peak that tells us that pulses are spaced apart anywhere from 1.2 to 2.2 meters. Visualizing the scan pattern shows that the 2.2 meter spacings happen at the tips of zig-zag and the 1.2 meter spacings at nadir. The RIEGL scan has much narrower peak with most pulses spaced apart at most 60 to 80 centimeters.
The histograms of shortest edge lengths illustrate how close pulses are spaced. Again, it helps to visualize the zig-zag scan pattern to explain the odd distribution of shortest edge lengths in the Optech scan: as we get closer to the edges of the flightline the pulse spacing along the scanline becomes increasingly dense. This is represented on the left side in the histogram that is slowly leveling off to zero. Again, the RIEGL scan has a narrower peak with most pulses spaced no closer than 35 to 55 centimeters.
To confirm our findings we illustrate where the scanners produce longest or shortest edges using ranges that we find in the histograms above and create false-colored rasters:
las2dem -i optech.laz -last_only ^ -step 1.5 -edge_longest ^ -false -set_min_max 1 2.5 ^ -odix _edge_longest -opng las2dem -i optech.laz -last_only ^ -step 1.5 -edge_shortest ^ -false -set_min_max 0 0.7 ^ -odix _edge_shortest -opng las2dem -i riegl.laz -last_only ^ -edge_longest ^ -false -set_min_max 0.5 1.0 ^ -odix _edge_longest -opng las2dem -i riegl.laz -last_only ^ -edge_shortest ^ -false -set_min_max 0.1 0.7 ^ -odix _edge_shortest -opng
The color-codings clearly illustrate that the Optech has a much wider pulse spacing on both edges of the flightline but also a much narrower pulse spacing. That reads contradictory but one are the “within-zig-zag” spacings and the other are the “between-zig-zag” spacings. The RIEGL has an overall much more even distribution in pulse spacings.
As a final confirmation to what causes the pulse spacings to be both widest and narrowest at the edges of the flightline for the Optech scan we scrutinizing the distribution of last returns and their triangulation visually.
Now it should be clear. At the edge of the flightline the Optech scanner has increasingly close-spaced pulses within each zig-zagging scan line but also increasingly distant-spaced pulses between subsequent pairs of zig-zagging scanlines. At the very edge of the flightline the scanner is almost sampling a “linear area” twice but leaves unsampled gaps that are twice as wide as in the center of the flightline. I believe this is one of the reasons why people “cut off” the edges of the flightlines based on the scan angle when using oscillating mirror systems. Now let’s took at the RIEGL.
There is no obvious difference in pulse distribution at the edges and the center of the flightline. The entire width of the swath seems more or less evenly sampled.
The above results show that “average point density” and “average point spacing” do not capture the whole picture. A quality check that properly verifies that a scan has the desired point / pulse density and spacing should measure the actual spacings – especially when operating a scanner with oscillating mirrors. We have done it here by computing for each last return its longest surrounding edge in the Delaunay triangulation, rasterizing these edge length, and then generating a histogram of the raster values.
Good follow-up reads are Dr. Ullrich’s “Impact of point distribution on information content of point clouds of airborne LiDAR” presented at ELMF 2013 and “Assessing the Information Content of LiDAR Point Clouds” from PhoWo 2013.
this is really important. We experienced the problem several times. Up to now we do the same with gawk – we count the number of points in a 1x1m cell (by the way in ASCII). But the idea with lasgrid and lasinfo –histo is very interesting and will be much faster. I will try that out.
Prof. Phisan from Chulalongkorn University pointed out a related article from Ty Naus that uses the average edge lengths of the Delaunay triangulation of last returns to compute the pulse spacing and the area of the dual Voronoi cells as a density measurement.
I think our longest and shortest edge histograms above suggest that using edge length averages is not such a good idea. This will “average away” the different spacings we observe along scanline and between scanlines for oscillating scans (when considering the entire scanline). However, Ty goes on to suggest to only test small rectangles in the center of the swath. If the edges are generously cut-off due to a large overlap then testing only the swath center may be fine. But especially in easy and flat terrain survey companies will – unless explicitely specified – try to use as little overlap as possible to reduce total flying time. I think all areas of the swath from where points will later be used in the LiDAR delivery should be checked using “longest edge length” and not “average edge length” histograms.
very interesting article! In operations, in flat and rugged terrain , lidar survey companies uses from 20 to 40 % overlap, normally. The cut off of the edge of the sawth – 2 to 5° is made, for the Optech scanners, to limit not only the effect of increasing the point density and redundancy of information but also to limit the foot print “spead” of the laser pulse hitting the ground, hance somehow “diluting” the accuracy of the measurement which happen to be on a wider area.
But I belive you take this in account by cutting off the data at 35° FOV? Correct?
Interesting noticing that the histo z1 results of Optech scanner, having a bimodal distribution is due to an artifact..
Cheers and thank you again
I did not cut anything off in these expetiments. One purpose of the experiements with the Optech Gemini was to determine what true pulse spacing is on the edges of the swath. The other was to determine the minimal cut-off that definitely should be done. So there are 3 arguments now: (1) cut-off needs to happen because the returns where the mirror flips are are less accurate (see discussion here), (2) cut-off needs to happen because the pulses along the swath edges are poorly spaced and may not meet the terrain sampling targeted for based on calculations at nadir, and (3) cut-off needs to happen because the angle between pulse and terrain is more likely to enlargen and smear out the illuminated footprint, with the latter depending on the actual shape of the terrain (on sloped terrain this will be compensated on one but exaggerated on the other side of the plane) …
thanks for the clarification, I was referring to a caption in one of the Graphs ” The distribution of the number of last returns per 3 by 3 meter cell for the Optech scan (2720 cells with values above 35 are ignored).” which was not clear to me (the 35 is now clear referred to the nr of last return per cell).
Do you think that the original different point spacing might also affect the lenght (and spreading on the histogram) of the Delaunay triangles?