A discussion about the LiDAR production process.

The LiDAR pulse bounces off whatever it encounters--the real art in LiDAR processing is knowing, with confidence, what the pulse bounces off. Image courtesy of Spectrum Mapping.

Many agree that aerial LiDAR is accepted as the most efficient and cost-effective means to create accurate digital elevation and terrain data. It has become the standard for flood mapping and many other applications requiring fast, accurate, inexpensive Digital Terrain Models (DTMs), Digital Elevation Models (DEMs) and other geospatial features. But exactly how those final deliverables are derived has been a mystery to many LiDAR customers and data users.

This lack of understanding creates a situation where LiDAR customers do not have enough information to specify how they want their data processed. Therefore, the opportunity exists for some LiDAR vendors to potentially cut corners in their production process. And, because many LiDAR customers do not verify the accuracy and quality of their data, they may never discover the discrepancies.

It is important to keep in mind that when it comes to remote sensing data, there are no panaceas--no silver bullets. Results can vary widely among different LiDAR providers depending on their production processes and the skill and talent of their technicians. Remote sensing technology such as LiDAR is a tool. Accordingly, no single tool provides the capability to entirely solve a particular problem; the solution invariably requires a combination of tools and the skills and talents of a trained tradesman to use the tools and materials to create the desired finished product. Remote sensing and LiDAR are no different.

To understand the issues related to the processing of LiDAR data and the production of various derived products, it is vital to understand the LiDAR mapping procedure.

GPS and IMU technology accurately position a plane to allow for precise mapping of the Earth’s surface. Image courtesy of Spectrum Mapping.

Collecting LiDAR Data

In an airborne LiDAR operation, a laser produces a light pulse projected to a mirror and is reflected out the bottom of a plane or helicopter. The light pulse travels down until it reaches an object and is then reflected back to the system. Since the speed of light is a constant, a standard mathematical equation can be used to determine the distance the light traveled in the recorded time interval. This measurement provides the height, or Z data point, of an object. In addition to knowing the height of the object, global positioning systems are used to verify longitude and latitude, providing the X and Y data points of a three-dimensional data set. Finally, an inertial measurement unit (IMU) provides the pitch, yaw and roll information of the aircraft.

The laser beam distends over distance from a micron wide at its source to a couple of feet wide by the time it gets to the ground. Therefore, it is possible for the beam to hit a variety of objects in succession, such as a branch, the side of a building, and then continue down to the ground. On some LiDAR equipment, more than five different reflections from a single pulse can be collected.

The laser collects more than 3,000,000 3D points per minute with ground point density of one point every square meter to 40 or more points per meter, depending on flying height and speed. This results in data accuracy of 15 cm absolute and relative accuracy from point-to-point within the data set of 2–4 centimeters.

LiDAR data can be merged with digital ortho imagery to create realistic and accurate 3D models for visualization. Image courtesy of Digimapas Chile. Inset: LiDAR data combined with true digital ortho imagery are key building blocks in the creation of 3D city models. Image courtesy of TopoSys GmbH.

Viewing LiDAR Data

LiDAR data is collected from everything the laser hits: the ground, buildings, trees, power poles and so on. These 3D data points of light, as imagined floating in space as a 3D model, are referred to as a “point cloud” in LiDAR vernacular. When the 3D model is rotated and viewed from different angles, it soon becomes obvious that a large number of the dots are on a “lowest plane,” and the rest seem to float above it.

When one views a LiDAR point cloud it is quite obvious that the discrimination of exactly what the LiDAR beam bounces off of is not very obvious. Spatial analysis alone is often inconclusive when attempting to determine whether a LiDAR point has hit a small bush, fire hydrant, boulder or a ground surface anomaly. Conventional surface classification filters can only go so far when removing aboveground phenomena, as natural and artificial objects may be spatially interpreted similarly and subjectively removed, inadvertently eliminating valid surface detail without differentiation.

A) Unfiltered LiDAR returns

It is critical to the success of a LiDAR project for the customer to know how filtering is done to get an accurate bare earth surface. There is no one single “magic” filter for all terrains, vegetation, buildings or other manmade structures. After automated filtering, human editing needs to be done to ensure that all the data anomalies have been removed or that those left in are correct.

Semi-intelligent processing algorithms make assumptions that are right perhaps 95 percent of the time and with the proper analysis and interpretation by the processing technician, this rate can range between 98 and 99.9 percent. But if the LiDAR hits a large object, such as a rock, hidden by a tree, it will most likely be accidentally removed. That would be the wrong thing to do when creating a bare earth digital terrain model.

B) A single filter applied to LiDAR returns does not remove all artifacts

However, to be fair, no technology is without its limitations. Even interpretation in photogrammetry is subject to individual experience and intuition. And if that rock really was under the tree, the photogrammetrist wouldn’t see it either.

C) A full set of filters applied to LiDAR returns. Notice the eroded terrain indicated by red lines. This could be due to heavy vegetation (no ground returns) or over-filtering.

A common question among LiDAR customers is: “How do we know that the lowest point is the ground and not a lump of grass?” The answer is that we don’t, no more than a photogrammetrist who looks at a lump of grass would know that he’s seeing grass or a lump of dirt.

The filter removes about 95 percent of all objects. The remaining fragments are removed by manual editing of the data set. By using the DSM and DTM data sets, the laser point cloud will be classified into three classes: ground, non-ground and intermediate. Image courtesy of TopoSys GmbH.

Combining LiDAR and Imagery

Digital imagery taken at the same time as the LiDAR survey provides the best of both worlds (for more on this, read “From the Ground Up” on page 56). For quality control measures, many LiDAR operators collect digital imagery along with LiDAR simultaneously. In this case, the processing technician can view the imagery beneath the LiDAR at any point. This determination can usually help sort out problems where interpretation is not as obvious as using the LiDAR data file alone.

Collecting imagery and LiDAR simultaneously provides elevation and terrain data along with a color-coded 3D view and RGB true-ortho image data of a power line project. Image courtesy of TopoSys GmbH.

Some LiDAR mapping companies have found that the integration of LiDAR data along with high-resolution color or multispectral imagery are vital to the production of accurate bare earth surfaces and the classification of laser returns to support a wide range of mapping applications.

These firms find that, depending on the terrain and vegetation coverage, laser data cannot be “blindly” filtered. In their experience, for laser data certification and quality assurance, imagery is required to review the surface to determine the effectiveness of data filtering for bare earth and/or canopy laser returns in most areas. For projects such as FEMA floodplain mapping that requires breaklines, there is no other current solution than to use imagery because LiDAR alone will not allow the creation of accurate breaklines.

Additional benefits of simultaneous acquisition of LiDAR and digital imagery are:

• Higher accuracy, greater confidence in bare earth DTMs

• High-quality, accurate digital orthos acquired at the same time as LiDAR data

• Digital orthos provided for a minimal or nominal additional cost of LiDAR data

• Superior feature extraction, such as planimetrics; building footprints and height; land cover/land use classification; contours; enhanced change detection/classification;

• Creation of a myriad of geospatial information by merging LiDAR and image data.

By classifying the data, an analyst can determine whether the data belongs with the bare earth or artifacts classification, and either remove, replace or edit the data.

Realizing LiDAR's Potential

Airborne LiDAR technology is being used increasingly for a variety of engineering studies where detailed topography of the ground is required and many customers still request contours as a deliverable. It is important to keep in mind that these are only a visual aid. Engineers generally convert them to a DTM, which is what LiDAR data provides directly. LiDAR provides a density of points that gives most engineers far more information than they normally require.

Many LiDAR users are developers or engineers who have to move dirt and need to know volume. In general, LiDAR does a better job of providing base data than does any other technology, especially in vegetated areas. It’s also faster. If a project has a short time frame, LiDAR is the quickest way for getting data to a client.

Ultimately, the potential LiDAR data user needs to decide: “What are the costs of working with good or bad data?” It is absolutely critical to a project’s success to not only specify that the right tools be used, but also that the LiDAR technicians are sufficiently skilled and use the appropriate methods. A universal truth is that the cost of doing it right is less than the costs of doing it over or the potential pitfalls in the field that result from bad data.