Pixelated DSM using LAS Dataset to Raster?

I am using ArcGIS 10.2.2, and I am trying to create a DSM out of some LIDAR data. After checking the LAS Dataset properties, I see that I have an average point spacing of approx 0.3.

I tried creating a dataset with all the basic settings, just setting the sampling value to times three the point spacing, so I set 1.

The image seems ok, but when I add hillshading, I feel it seems kind of clunky. I am wondering if I use the dataset to raster tool right? Is there some approach that is "best practice"?

• It looks to me like you are using all the points (not just those classified as ground). Is that what you are seeing? You will need to exclude the points that do not represent the ground! – jbchurchill May 23 '16 at 15:31
• The DSM is correct, im using all points on purpose. I just think it looks too pixelated, but I guess this is the way it is? – DOMINUS MIHI ADIUTOR May 23 '16 at 15:44

My general opinion is that this is a lousy tool for creating a DSM however, @Andre Silva hit the nail on the head in using the maximum argument and you should mark his answer as correct because it directly addresses your question.

Ideally, for a DSM you should use the first or last returns with an interpolation algorithm, and not merely a binning approach. To not sound like I am just ESRI bashing, my favorite tool for this is the "Topo to Raster" in the Spatial Analyst > Interpolation tools. This will take point input and uses an adaptive Thin Plate Spline interpolation, based on Hutchinson's ANUSPLINE algorithm. It is also quite fast with large data.

• Using first or last return really depends on what your want to emphases in the DSM. If you want to capture the top of canopy then first would be more appropriate. However, if you are in a urban environment with mostly buildings then I believe that last would work better. I think that the quality of the resulting DSM inevitability dictates it and both should be tried. Although, one common mistake that I see is assuming that last return is ground and suitable for creating a DEM. I edited my answer to reflect one or the other. – Jeffrey Evans May 24 '16 at 17:28

An average point spacing of 0.3 is approximately equal to a point density of 11 returns per unit of area. In theory, it should be sufficient to generate a DSM with pixel of area 1.

As explained here, my understanding of a LiDAR DSM is:

DSM as a raster. This represents the first echo the laser received for each laser pulse sent out, and represents the tops of buildings, trees, and other objects, or the ground, if unobstructed.

Therefore, you should change the parameter Binning/Cell Assignment Type from AVERAGE to MAXIMUM, considering you are using all points for processing and have no elevation outliers in the las dataset. See the ESRI's description for the 'maximum' argument:

Maximum — Assigns the maximum value found in the points within the cell.

An alternative would be to use AVERAGE, but with the las dataset filtered by first returns.