University of Idaho - From Here You Can Go Anywhere  
 UI Home Contact UI Search

measurements_banner_2.jpg (65612 bytes) Idaho Fire Initiative for Research and Education (IFIRE)

College of Natural Resources








Dr Alistair Smith

alistair [at]


Introduction to Discrete Return LiDAR Web Tutorial v 2.01

Compiled by:
Jeffrey S. Evans & Andrew T. Hudak (USDA Forest Service) &  Alistair M.S. Smith (University of Idaho)

Part I - LiDAR Basics:

Light detection and Ranging (LiDAR), sometimes referred to as scanning laser altimetry, is a widespread technology that allows the relative height of surfaces to be measured from an aircraft or satellite sensor.The concept of LiDAR is simple, as it uses the simple relation that:

distance to an object = speed x time.

Essentially, the time for the light to travel to and from the target is then used to determine the distance. The distance to the target and the position of the airplane is then used to determine the elevation and location.



The most important property of LiDAR is that unlike other remote sensing methods, which rely on inference and correlation (e.g. NDVI and LAI), LiDAR provides DIRECT measurements of relative height (or elevation). 

In LiDAR the footprint size decreases with increasing post-spacing and importantly the last return from a discrete return system is not always the ground. LiDAR sensor systems vary in the number of returns from a surface. Below is an example of how many of the 1st, 2nd, and 3rd returns were obtained from within a pulse:



The LiDAR Intensity Image: This is a commonly unused bi-product of a LiDAR acquisition and is the intensity of object that the laser pulse is striking. This is an uncalibrated 8-bit (0-255) image that is orthorectified as therefore can be used as an orthophoto. These images are not typically used in quantitative analysis as the image gains are always set to 'the adaptive gain' setting when the images are acquired.



Getting Heights from LiDAR Data:

Raw LiDAR data contains millions of XYZ points. The area highlighted in Yellow is of Priest River Experimental Forest (ID) and is approximately 1 by 0.75 miles and contains over 400,000 returns. The widespread use of LiDAR data has been in the production of relative elevation maps from which absolute heights of objects such as trees or buildings can be measured. A common product from a LiDAR acquisition is a Digital Elevation Model (DEM) and a 2-m example from Priest River Experimental Forest is shown:

To produce a DEM we must filter the raw LiDAR XYZ points to remove all returns that are not from the ground (i.e. that are not due to
topography). Several filtering methods exist an indeed one of the best examples is the 'Progressive Curvature Filter'. Detailed information of the progressive curvature filter algorithm and comparisons with other methods can be found at:

Other methods are also being used throughout the wider Lidar literature. Each method is typically iterative and essentially in each pass removes non-ground objects.

XYZ points above a pre-defined slope gradient are removed.

BLOCK MINIMUM: Minimum Z value in bin is assumed to be a ground value.


Left is an example of block minimum applied to a heavily forested area. In this example the block minimum bin size was set to 6 raster pixels. You can clearly see that this method fails to address areas where no ground returns exist.

Essentially, the ground is raised by the vegetation. Other methods, such as the  'Progressive Curvature Filter' are clearly needed in such high biomass areas.
In a similar fashion, the image on the right is the result of 'slope threshold' applied to an urban area [From Vosselman 2000]. Shadows, or footprints, of buildings area clearly visible and the 'mess' in the bottom right is a high biomass area.

As such it is clear that in such urban environments that other methods are necessary. One such method, is a morphological filter, such as that developed by Zhang et al (IEEE TGRS: 2003 ,41,4)

Once a bare earth model is produced, relative heights of objects can be obtained by subtracting this surface from each Z value. The figure below shows both the Ellipsoidal Elevations (i.e. the raw XYZ values) and the calculated Heights (i.e. the Ellipsoidal Elevations minus the filtered ground surface elevations). Following calculation of the Heights it is common for this to be converted into a RASTER image for use in traditional image processing or GIS (e.g., ARC) software packages. However, techniques do exist that allow the direct analysis of this Height point cloud.

University of Idaho, Moscow, ID, 83844