From the Ground Up: Horizontal Accuracy Assessment in LiDAR.
Project specifications typically spell out accuracy requirements for both the horizontal and vertical components in LiDAR collection.
But while a thorough assessment and reporting of the vertical accuracy of the elevation surface is almost always completed for LiDAR projects, a formal evaluation of the horizontal accuracy is less often required.
It is common to see accuracy requirements in the form of a root mean square error (RMSE) that is then used to approximate the accuracy of the elevation surface. Maximum vertical RMSEs of 15 to 18.5 centimeters (0.49 to 0.61 feet) are common for elevation models used to generate 2-foot contours. Similarly, horizontal accuracies (in terms of RMSEs) ranging from 0.5 to 1.0 meter are often specified for these same elevation models.
The collection of check points and the ensuing statistical assessment of the vertical surface are relatively easy and straightforward. Evaluating the horizontal accuracy is significantly more challenging, but with proper planning, it can be successfully accomplished.
Traditional vector-based mapping and digital orthophotos are often evaluated in terms of their horizontal accuracy, and reporting is provided as a project deliverable. What makes a LiDAR collection different? The answer lies in the nature of LiDAR collection and the relative coarseness of the elevation points collected. Features that are crisp in imagery are much harder to precisely define in coarse imagery generated from LiDAR returns.
All LiDAR projects are planned for some amount of side overlap between adjacent flight lines. This is similar to the requirement for side overlap in the collection of aerial imagery. While most specifications require 30-percent overlap for imagery collection (indicating that 30 percent of the ground coverage in one flight line is also found in the adjacent flight line), it is fairly common to have side overlap ranging from 20 to 40 percent or more with LiDAR collection. The amount of planned overlap depends on a number of variables, including required accuracies, the presence of development and the amount of terrain relief found at the project site.
Comparisons made in the side overlap area provide the first glimpse of the horizontal accuracy of LiDAR data. This analysis is particularly valuable when comparing adjacent flight lines that were flown in opposing directions. Buildings with pitched roofs or a sloped ground surface should be evaluated in the area of overlap. These features should fit almost perfectly on top of one another if the system is properly calibrated, the acquisition is carried out under strong GPS configurations, and the post processing is completed in an accurate fashion.
Some potential horizontal errors, however, will not be evident in a comparison of opposing flight lines in the side overlap area. It is necessary to take the analysis one step further to completely evaluate the horizontal accuracies gained in the LiDAR collection.
At least three strategies can be used to provide a reasonable test of horizontal accuracies from LiDAR. Intensity images are used in two of the three most common assessment methods. During LiDAR collection, units determine the 3D position of the laser as it is reflected off the ground surface. The sensors also measure the amount of energy in the reflected signal. This is known as the intensity of the return. Metal roofs have very different intensities compared to asphalt shingles. Concrete returns significantly more energy than asphalt. Deciduous trees have varying intensities when compared to closely mown grass. These intensities are perfectly registered to the 3D position of each laser pulse. We often generate images from these intensities to aid in the overall post processing or to be used in the collection of additional information, such as breaklines, to enhance the LiDAR surface.
For the first method of accuracy analysis, features must be carefully selected within the project area at points that will be visible in an intensity image so that their horizontal position can be determined with field GPS procedures and then compared to their location in the intensity image. This situation is analogous to the accuracy assessment of traditional mapping, but the difficulty of identifying the precise location of a feature in a coarse intensity image provides a higher degree of uncertainty when compared to traditional imagery.
The intensity image in Figure 1 on page 54 provides a good example. The red “x” marks the horizontal location of a photo-identifiable ground control point that was collected at the southeast corner of a parking lot. Notice the scan lines from the LiDAR following along a northeast-southwest pattern. During our accuracy assessment for this project, lines were constructed along the southern and eastern limits of the parking lot from the visual interpretation of the intensities, and we then refined these lines by looking at the elevation surface. The elevation surface was useful because the control point was collected at the base of the curb. The intersection of these two lines was then compared to the position gained from GPS. In this case, the LiDAR interpretation fit within 1.7 feet of the GPS position.
We have confidence that this comparison was accurate within 1.0 foot given the use of both intensities and the elevation surface to determine the corner of the parking lot. Other features also work well in the assessment of horizontal accuracy. Painted features like stop bars or striping in parking lots provide very good opportunities for control points. The paint shows up well in intensity images because it is a highly reflective surface and returns most of the energy from the laser. Transitions from concrete to asphalt surfaces, which are common where sidewalks meet driveways or roads, also serve as good control points.
With the second assessment method, it is possible to overlay the intensity image on digital orthophotos if they exist for the project area. However, to be valid in an accuracy assessment, the digital orthophotos must be of sufficiently higher accuracy than the project requirements for horizontal accuracy. The paint striping along roadways and in parking lots again provides excellent opportunities for the validation of the horizontal accuracy. Water features (where most of the laser returns are absorbed) and asphalt surfaces can also provide valid checks. Figure 2 illustrates how well the LiDAR intensities for this project fit the orthophotos. The higher intensities are rendered as bright white. Notice how well these higher intensities fit with the paint striping on the roadway and the stop bar painted at the intersection. You can also see how well the low-intensity returns (rendered as darker shades of gray) fit over surfaces that tend to reflect a small amount of energy like, in this case, asphalt roadways and driveways.
To be valid, roadways running in multiple directions should be carefully reviewed. A systematic east-west shift of the LiDAR data would not show up in a straight section of roadway running east-west. But confidence is gained if there is a close match in both east-west and north-south roads within a project area.
A third way to validate horizontal accuracy is to compare field run cross sections against sections generated from the LiDAR elevation model. The sections should be collected in areas with significant slope. Levees, roadway embankments or streams can provide ideal surfaces for this comparison. The argument for different orientations for the comparisons made above would apply here, as well. Ideally the sections will coincide with one another. Any horizontal shifts in the data will be evident in the cross sections as an offset between the two data sets and can easily be measured and compared against project requirements.
As technology changes, the challenges of a horizontal assessment are somewhat mitigated. A prime example is the hardware improvements that provide us with the ability to collect more dense elevation data sets. Increased point density makes it much easier to define photo-identifiable features in an intensity image and compare those features to field-determined GPS positions.
The horizontal integrity of a LiDAR surface is critically important to all projects, but a formal assessment is less frequently required as a project deliverable than the assessment of the vertical integrity. While major horizontal blunders would almost certainly show up in the vertical assessment, smaller horizontal errors could be missed if only elevation checks on relatively flat surfaces are performed. But with proper planning, the more challenging horizontal accuracy assessment can also be successfully accomplished and included as a valuable project deliverable.
1- In my July 2008 column, “From the Ground Up: Collecting Breaklines,” I compared the resolution of a typical LiDAR collection for a 2-foot elevation surface to that of digital imagery. In this example, the LiDAR resolution is 1 to 3 points per square meter compared to 43 points in 1/2-foot imagery and 172 points per square meter in 1/4-foot imagery.