Recent mapping technology advances such as direct digital imagery
acquisition and Light Detection and Ranging (LiDAR)-derived data
sets are quickly changing the remote sensing industry’s service
offerings. In fact, the two technologies have begun converging into
a single acquisition and georeferenced platform, enabling faster
data collection and processing.
In 2004, Sanborn began testing and implementing digital production
processes using Vexcel Corp.’s UltraCam-D digital camera. By
combining the camera’s multi-spectral capabilities with LiDAR sensor
technology, the company has dramatically improved its ability to
provide geographic information system (GIS) professionals with
robust, inexpensive data sets.
Applying the Technologies
In June 2004, the Wyoming Natural Resources Conservation Service (NRCS)
partnered with several local government groups and contracted
Sanborn to complete multiple LiDAR and imagery flights across the
central and north-central portions of the state. The project’s end
products included the development of filtered, classified and “bald
earth” LiDAR data sets; multispectral orthophotos; and 2-foot
contours.
Says Wyoming NRCS’ Randy Wiggins, “NRCS and its partners are
exploiting the data for the traditional uses we expected. But as we
continue to work with and understand the significance of the data
sets provided by Sanborn, we’re discovering many new applications
that will benefit our organization and others. The advantages of
direct digital imagery acquisition and LiDAR-derived datsets,
compared to analog photogrammetry, are becoming apparent to us.”
Direct Digital Imagery Acquisition
Although small- and medium-format digital cameras have been used for
several years, the new large-format sensors produced by companies
such as Intergraph Z/I Imaging, Leica Geosystems and Vexcel offer a
faster, better solution to conventional film-based aerial imaging.
Image
Clarity: Radiometric clarity is the most evident advantage that
digital sensors have over analog technology. For example, the Vexcel
UltraCam-D offers a base panchromatic (black and white) resolution
of 11,500 x 7,500 pixels, along with four color channels—red, green,
blue (RGB) and near-infrared. Forward-motion compensation can
support pixels as small as three centimeters on the ground. In
addition, digital images don’t contain analog anomalies such as
scratches, lint, dust or grain-noise. Finally, a digitally acquired
image simply looks better than an analog scanned image.
Multispectral Data: The second greatest advantage of using a
digital sensor instead of analog technology is the multispectral
content. The ability to provide clients panchromatic, color (RGB)
and color infrared (CIR) in a single flight is significant. With an
analog camera, three separate flights over the project area are
needed to provide equivalent image products. Typically, the
additional flights would be cost prohibitive, so many clients
haven’t recognized the benefits of multispectral data sets for
sophisticated remote sensing applications such as land use/land
classification identification, irrigated land delineation,
impervious surface identification, vegetation delineation and
wildfire hazard assessment. As the digital revolution continues to
expand, more GIS professionals will begin tapping into multispectral-derived
GIS data layers.
As part of the NRCS project, Wyoming’s Casper Mountain area was
flown with Sanborn’s UltraCam-D multispectral sensor. Stereo models
were used for breakline enhancements via stereo compilation. In
addition, the multispectral data will enable wild fire hazard
assessment professionals to derive land use/land classifications and
diseased trees calculations to help identify fire risk areas.
Dynamic Range: Digital sensor technology acquires data at a
12-bit sample rate. Traditional analog collection and subsequent
scanning result in a digital image less than 8 bits. The difference
is significant for automatic image processing techniques. For
example, the increased data range allows for greater accuracy and
redundancy when completing processes such as auto-match aerial
triangulation, auto-correlated digital elevation model (DEM)
extraction and auto-seaming for orthophoto mosaics. The superior
12-bit image quality and greater accuracy of automated techniques
lead to schedule and cost efficiencies.
In addition, the dynamic range allows greater detail in shadow
areas, such as those in rugged terrain or dense urban settings. As
with many western states, Wyoming exhibits challenges due to terrain
relief. NRCS has seen significant benefits due to shadow reduction
in mountainous and canyon areas.
No Film, Film Processing or Scanning: Digital technology
eliminates film processing and scanning, allowing for a much faster
turnaround/delivery schedule. Moreover, direct digital acquisition
permits the collection of 90 percent forward lap between exposures
within a flight line. This is a significant benefit when considering
that the 90 percent forward lap allows up to nine different views of
the same feature. Also, additional overlap is needed to guarantee
appropriate coverage and reduced building lean when traditional
analog photography is flown in urban and mountainous terrain. Any
time more photography is captured, cost for film, film processing
and scanning increases. The cost to increase overlap is minimal if
direct digital acquisition is completed.
LiDAR Surface Mapping
LiDAR systems measure the time of flight of a laser pulse to
construct a surface elevation model. The two most prevalent
commercial systems are produced by Leica Geosystems (the ALS series)
and Optech (the ALTM series). Both systems operate on the principle
of reflecting a pulsed laser off an oscillating mirror and measuring
the time of flight to determine the distance traveled by the laser
pulse. An integrated 6-DOF geopositioning system (both companies use
the Applanix POS AV) determines the sensor’s precise position (x, y,
z) and attitude. This information, combined with the angle of the
mirror, is used to calculate 3-D positions of terrain points. Such
capabilities offered NRCS several advantages
Efficiency Improvements: Most mapping companies currently use
LiDAR for contour and orthophoto mapping, as LiDAR is more efficient
than traditional stereo compilation. A stereo compiler typically
collects 1,400 points an hour, but LiDAR can collect 100,000 points
per second! Thus, clients receive mapping products faster and
cheaper. In addition, LiDAR data provide a greater amount of
information than traditional stereo-compiled products. As a result,
NRCS received greater accuracy and more data sets at a lower cost
when compared with analog photogrammetry.
Speed: LiDAR technology offers fast, near-real-time
collection of 3-D points that are typically used to generate digital
terrain models (DTMs). LiDAR doesn’t discriminate, collecting
features such as trees, cars and buildings, as well as the ground.
The entire data set is known as a “point cloud.” Automatic
algorithms can remove surface classes such as vegetation, power
lines, buildings and automobiles. Once removed, the processed LiDAR
data set is known as a “bald earth” DEM, which is used for
orthorectification and contour creation.
3-D Analysis: Analyzing data in 3-D helps users visualize
spatial relationships for sound decision-making, and facilitates
effective and clear communication of planning ideas. The NRCS
project included LiDAR collections for several canyon corridors, the
city of Kaycee and Casper Mountain. LiDAR-derived DEMs, surface
classifications and intensity image data sets were created.
Furthermore, canopy density and tree outlines were derived from the
LiDAR data and delivered in vector format. Compared with
public-domain U.S. Geological Survey DEMs, the more accurate LiDAR-derived
DEMs made it much easier for NRCS to complete hydrographic and
hydraulic modeling and dam breach analysis.
Contour Mapping: As part of the NRCS project, Sanborn
colleted LiDAR and Vexcel UltraCam-D imagery to produce 2-foot
contours meeting Federal Emergency Management Agency accuracy
specifications. NRCS used elevation data sets and a 3-D
visualization tool from the Orton Family Foundation’s CommunityViz
software to re-create a major 2002 flood in Kaycee.
For many applications, LiDAR is an ideal tool. However, certain
applications, such as floodplain mapping or 2-foot contour mapping,
require guaranteed accuracy. For these applications, raw or filtered
LiDAR data often are unsuitable because automatic filtering to
remove buildings and vegetation is imperfect. Overly aggressive
filtering will remove important topographic features, such as
hilltops. Conversely, some vegetation and large buildings aren’t
removed if the filtering isn’t aggressive enough. Other problems
arise when the terrain surface is under-sampled. In other words,
LiDAR has finite point spacing, and it doesn’t capture sharp terrain
breaks precisely. LiDAR doesn’t penetrate thick vegetation, which
typically occurs around stream channels. Furthermore, certain types
of land cover, such as thick grass in a field, can cause a
systematic bias.
Processed LiDAR contains residual errors,
while individual LiDAR points are accurate (15 centi-meters RMSE).
As a whole, LiDAR data are “noisy.” For example, a cross-section of
LiDAR data in the area of a frozen lake would reveal a horizontal
band of points 5-10 centimeters thick. Consequently, contours
generated from raw LiDAR data have a ragged appearance and show many
tops and isolations that are due more to noise in the LiDAR data
than to actual ground features. This problem is also evident in
stream beds, because hydrology isn’t enforced. Due to systematic
errors and failure to penetrate vegetation near streams, contours
generated from LiDAR data can show water flowing uphill. In
addition, automatic filtering typically doesn’t cut the terrain to
allow water to flow through culverts or low-lying bridges. Again,
these are important issues for hydraulic modeling applications.
For the NRCS project, Sanborn used its proprietary Filtering and
Surface Estimation (FASE) process to edit the LiDAR data and produce
a triangulated irregular network (TIN) surface. Using a
sophisticated least-squares surface estimation algorithm, LiDAR
points are combined with manually captured breaklines and masspoints,
as well as surface classes that identify buildings, water and
vegetation
The process is performed iteratively. Stereocompilers examine
surface contours using stereo photogrammetric workstations. They
modify the surface by adding breaklines and masspoints, as well as
defining surface types. The surface is then re-estimated and new
contours are generated. This process is repeated until the surface
is accurate and complete.
In the image set below, Figures 1a and 2a show a shaded relief model
and contours derived from automatically filtered LiDAR data. Figures
1b and 2b show the same area computed and processed with FASE. In
Figure 1a, note the contour puddles along the drainage and the
“uncut” bridge in the upper part of the image. These examples show
the lack of hydro-enforcement and have been corrected with FASE in
Figure 1b. Note in Figure 2a how overpasses have been inconsistently
handled. Also note the stray contours that extend from the
riverbanks into the river, as well as the noise-related contours
that are scattered all over the river. FASE corrects these problems,
too, as shown in Figure 2b.
FASE requires stereo photography in most cases. Although this
lengthens the process, it is still less effort than using a
traditional photogrammetric approach. In addition, a surface
estimated with FASE addresses all the aforementioned deficiencies.
Ground-Based LiDAR: A ground-based
LiDAR instrument provides an alternate means of terrain and
digital image acquisition. The Trimble GS 200 3-D laser scanner
will enhance the end products that have been traditionally
delivered to LiDAR and surveying technology users. Civil
engineering, architectural and surveying companies rely on the
topographic information for various projects such as highway
design and bridge repair. A ground-based laser scanner will allow
for profiling and measuring of structures that may not be visible
from an aerial acquisition method. Furthermore, a ground-based
LiDAR sensor provides significantly more detail when compared with
traditional ground-survey methods. The Trimble GS200 3-D laser
scanner also provides imagery that is co-registered with the LiDAR
data for a ground-based orthophoto. The sensor’s accuracy depends
on how far it is from the terrain/structure (at 100 meters, the
manufacturer claims accuracy is 4mm). Future solutions of the GS
200 3-D LiDAR unit will include the integration of an Inertial
Measurement Unit and Global Positioning System (GPS) technologies,
enabling the procedures to move from static and stop-and-go
acquisition methods to mobile acquisition similar to aerial
flight.
Unlimited Utility
For years, photogrammetry professionals have incorporated LiDAR into a
variety of applications, including volumetric studies, DEM development,
hydrologic and hydraulic modeling, and topographic and digital
orthophoto mapping. Now integrators such as NRCS have shown how LiDAR
can be used with a digital multispectral sensor for increasingly diverse
applications, including 3-D modeling, dam breach analysis, wildfire
hazard assessment, soil type delineation and erosion, tree canopy
extraction and density calculations, as well as other surface class
extraction such as buildings and power lines. Moreover, habitat mapping
will be completed using the vegetation canopy heights and multispectral
vegetation classifications.
Such innovative uses of LiDAR and multi-spectral imagery demonstrate how
remote sensing and GIS professionals can receive more information faster
and at lower cost. The integration of these two technologies and the
fusion of their complementary data sets are vital for accurate GIS
applications.
