With the recent awarding of large statewide LiDAR contracts and the exploding number of RFPs specifically requesting LiDAR mapping, it is clear that this technology is not going to go away. A brief survey of published accuracy specifications reveals that LiDAR results in accuracies of 0.5' vertical and 2' to 4' horizontal. However, there is growing concern among surveyors about the achievable accuracy of airborne LiDAR data and how this accuracy impacts products certified to various established mapping standards.
When discussing the accuracy of LiDAR data, it is important to keep in mind that the theoretical error based on a rigorous engineering analysis of the system is generally not achievable in the field. Operational considerations, such as variations in GPS quality, will significantly affect the final accuracy of the data. In addition, a surveyor will be faced with various interpretations—and misinterpretations—of what is meant by the accuracy of the LiDAR data. As many surveyors who have negotiated LiDAR survey contracts are finding, expectations and interpretations of what was negotiated at the beginning of a contract can become very different when presented with the actual LiDAR data from the vendor.
System Error BudgetThe total error for a LiDAR system is the contributing error budgets from each subsystem: the laser rangefinder, the GPS position and the IMU orientation error. Contributions include such diverse factors as the inherent pointing error of the laser, sensor mounting biases (which are the small angular misalignments between the laser reference frame and the IMU reference frame) and the error in recording the scanner angle at the moment of each laser pulse. An excellent review and detailed examples of how each system parameter contributes to overall LiDAR system accuracy can be found in Baltsavias (1999a).1
Unfortunately, the operational accuracy that can be achieved is generally worse than the theoretical error limit. As a result, there is a lack of clear definition of what is meant when stating accuracy for LiDAR data. The ASPRS LiDAR subcommittee is working hard to establish guidelines for calibrating LiDAR systems and providing commonly accepted definitions of LiDAR accuracy in an effort to standardize these specifications and address surveyors’ concerns about this new technology. Some notes for surveyors to keep in mind when discussing LiDAR accuracy include:
- Manufacturer accuracy specifications are derived from statistical sampling and are generally quoted as a 1 sigma specification, meaning ~68 percent of the data will fall within this limit; 2 sigma (95 percent) or 90 percent (1.6 sigma) specifications are generally not mentioned.
- Accuracy analysis tends to focus on vertical accuracy (Z), and details on how planimetric accuracy (XY) is verified are vague.
- Accuracies will vary under different conditions across a project, such as in areas of steep slope or from the maximum angle of the scan to the minimum.
- LiDAR does not sample infinitely small points on the ground, and the complex interaction of the transmitted pulse energy with the finite footprint on the target needs to be considered carefully. A bright target within the footprint can skew the return signal away from the geometrical center of the footprint.
- Certain artifacts can occur with LiDAR-derived products such as a lack of grade-break definition, planimetric accuracy issues, and artifacts from feature removal algorithms (bare earth extraction).
- Geoid height model errors will impact final accuracy. Empirical error estimates using current geoid height models produced by NGS show large differences using single-tie geoid modeling techniques (Smith 2000)2. Any vertical GPS error, such as geoid height modeling, will directly influence the accuracy of any LiDAR product.
When considering LiDAR for a project it is always prudent for a surveyor to request a certified system calibration and accuracy analysis from the LiDAR system manufacturer or the service provider. (For independently published analysis of LiDAR accuracy using commercial sensors, see Kraus & Pfeifer (1998)3, Shrestha et. al. (2000)4 and Gutierrez, R. et. al. (1998).5)
Range ErrorsUnder normal operating conditions the range error from a typical laser rangefinder that is properly calibrated can be expected to be between 2" to 3". However, the atmosphere acts to distort the path of the laser pulse as it travels to the target and back again, introducing a timing error that needs to be corrected. These corrections become critical at higher altitudes. These atmospheric affects are usually minimized but not eliminated by incorporating an appropriate atmospheric model (Marini and Murray, 19736) in the post-processing of the LiDAR data. A surveyor needs to pay close attention as specified operational altitudes increase. An excellent discussion on geolocation of laser altimetry data can be found in Hoften et. al. (2000)7.
Position ErrorsAirborne GPS systems are used in LiDAR to provide positioning information regarding the trajectory of the sensor. It is important for surveyors to have a good understanding of GPS-related errors that fall into several broad categories. Sources of error include satellite geometry (PDOP), orbital biases, multipath, antenna phase center modeling, integer resolution and atmospheric errors. Compounding some of these errors is the operational distance from the ground GPS stations. It is important to remember that for every GPS-related error source, a method can be employed to detect, eliminate or minimize that error. In general, when properly taken into account and with proper project planning, GPS error contributes on the order of 2" to 4" of position error to the final product.
Orientation ErrorsKnowing the correct orientation or pointing direction of the LiDAR sensor is necessary for calculating an accurate spot location on the ground. In practice, the orientation of the platform is recorded by an on-board inertial measurement unit (IMU) that is hard mounted to the LiDAR sensor. While a variety of IMUs are available commercially, a typical specification for the price/performance levels common in most commercial LiDAR sensors would be 0.005Â° pitch/roll, 0.008Â° heading (POS/AV 510 from Applanix, Richmond Hill, Ontario, Canada post-processed solution) although some systems perform to a 0.0025Â° pitch/roll accuracy. A 0.005Â° angular error corresponds to a 0.87 foot error from 10,000 feet and a 1.70 foot error from 20,000 feet. However, there are additional contributions to the angular pointing error including contributions from the scanning subsystem. Many of these additional errors can be minimized but not eliminated by proper system calibration prior to data collection and proper system modeling during post-processing.
Field Project ResultsSurveyors need to exercise caution when reviewing LiDAR data specifications or products since proper project planning and execution along with skilled personnel also have a significant impact on data accuracy and quality. Surveyors should be wary when presented with reports that tout the superiority of one sensor over another in a single or limited number of field studies. An excellent sensor design is wasted in the hands of inexperienced or poorly trained operators while all commercial system designs on the market today, including the Optech ALTM (Optech Inc., Toronto, Canada) and Azimuth AeroScan (Azimuth Corporation, Westford, Mass.) product lines, are reliable instruments that will produce good data in the hands of competent personnel.
Project Example #1A volumetric survey conducted using an Optech ALTM 1225 was flown for ±0.5 ft RMS vertical accuracy. The survey was flown at an altitude 2,500 feet and 20,000,000 points were collected covering 5 sq. miles. The client established ground control based on ~400 static control points and kinematic GPS profiles. The LiDAR elevations were compared to the ground survey and the results are shown in Histogram 1 below, which shows the error distribution normalized to the number of control points. Each bin in the histogram is labeled with its uppermost value (the 0.00 bin includes all points with deviations between –0.25 and 0.00 feet). The minimum deviation was –0.58, the maximum was 0.40, the mean was –0.07, the standard deviation was 0.16 and the RMS was 0.17; and 99.5 percent of the measured deviations fell within ±0.5 feet.
Project Example #2A recent survey in California was flown for ±0.5 RMS vertical accuracy. Accuracy and rapid turn around were the prime drivers for the client. The aerial survey was completed in late December 2000 and the final data product delivered to the client 10 business days after completion of the fieldwork. The survey was flown at 3,000 feet and 150,000,000 points were collected covering 40 sq. miles. The client based ground control on ~4,500 control points and kinematic GPS profiles established prior to the aerial survey. The results are shown in Histogram 2 below. The normal distribution of the data is more evident due to the larger number of control points. The minimum deviation was –1.39, the maximum was 1.19, the mean was 0.04, the standard deviation was 0.40 and the RMS was 0.40; and 78.0 percent of the measured deviations fell within ±0.5 feet.
Project Example #3A survey for a major Hollywood studio was flown for high density, ±0.5 RMS vertical accuracy. The client’s desire was to rapidly obtain a high resolution DTM to support special effects modeling of a large outdoor film location. The survey was flown at 2,000 feet and over 150,000,000 points were collected covering 10 sq. miles. Ground control was based on ~90 points and kinematic GPS profiles. The results are shown in Histogram 3 below. The minimum deviation was –0.40, the maximum was 0.24, the mean was 0.01, the standard deviation was 0.10 and the RMS was 0.10; and 100 percent of the measured deviations fell within ±0.5 feet.
ConclusionAs LiDAR technology becomes more prevalent, surveyors will need to become more educated about the strengths and weaknesses of this new technology. Discussions about LiDAR accuracy need to start from a common understanding of the technology’s capabilities and an agreed definition of what LiDAR accuracy is and how it is to be measured.
Martin Flood is the chief technical officer for Airborne 1 Corporation, Los Angeles, Calif. His background is in the design and development of advanced LiDAR systems for mapping and remote sensing applications. He was a former project manager and terrestrial survey team leader at Optech Inc., Toronto, Canada.
Jay Satalich is the director of surveying for Airborne 1. He is a licensed land surveyor in California and several other western states and a former project surveyor at CALTRANS.