ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
mean heights of certain areas. Furthermore, it was not clearly
formulated.
The former and the new error description scheme will be
described in the following paragraphs, together with methods
to quantify the differently scaled error components. The
chosen thresholds for every error component, derived from
real data, will be presented. Finally, the benefit of the error
description scheme for the AHN user will be illustrated by
propagating the error components to the height precision of
derived products.
Thus, this paper deals with assessing the height precision of
laser altimetry DEMs and quantifying the effects of the
different error components on the measured heights. This
paper does however not aspire to give methods for
eliminating or minimizing these errors, or for quantifying the
errors themselves, e.g. roll, pitch and heading errors of the
INS, such as Schenk [2001] did.
2. FORMER WAY OF DESCRIBING LASER DEM
HEIGHT PRECISION
The demanded height precision for the AHN is strict: 5 cm
systematic error and 15 cm standard deviation. It turned out
that these requirements were not always achievable. In
addition, a fundamental drawback of this formulation is its
ambiguity: several interpretations are possible. The region
size, for which the thresholds for bias and standard deviation
are valid, is missing. Does the 5 cm bias apply for, for
example, 100 m? areas or for 1 km? or for 10 km? ? Or is this
maximal bias valid for all these areas? And what about
controlling this, as such large ground control fields can
hardly be measured?
Another disadvantage of this height error description is that
not all the occurring error types of current laser data are taken
into account. The error behaviour of laser altimetry data,
acquired by a complex system of different sensors, cannot be
expressed by only two parameters: a bias and a standard
deviation. These two parameters do not suffice for describing
the height precision of a laser altimetry DEM.
A more sophisticated approach for comprehensively
describing the height quality is required. This new approach
must take into account the specific scale of each error type.
Some errors are stochastic for a single laser point. Others are
systematic for a small area or for an entire laser strip, but
stochastic as we focus on a large number of these small areas
or strips. In the following paragraph, these different error
components will be described including their technical
causes.
3. NEW HEIGHT ERROR DESCRIPTION MODEL
In our opinion the total error budget of laser altimetry data
can be divided into four components with different
amplitudes and with different spatial resolution [Crombaghs
et al. 2000]. These errors, which are illustrated in figure 1,
are:
1: Error per point. Due to the measuring uncertainty of
the laser scanner each laser point is affected with a
random error. This error is also called ‘point noise’.
2. Error per GPS observation. Every GPS observation as
well is affected with a random error. This error,
however, is constant (systematic) for all laser points
measured during this second. Usually, these points are
lying in a strip-wide area of about 100 m in length. This
depends on flying speed and GPS observation interval.
3. Error per strip. GPS and INS sensors are needed to
measure the position and orientation of the aircraft
along the flight path. The GPS/INS-system introduces
systematic errors in strips, like vertical offsets, tilts in
along- and across-track direction and periodic effects
with a period of several kilometres.
4. Error per block. Terrestrial reference measurements
(ground control ‘points’) are used to transform blocks
of laser measurements into the national height system.
Errors in these control ‘points’ result in height
deviations which affect entire blocks of laser strips.
This influence depends on the block configuration
(position and number of strips, cross strips and control
‘points’) and on the correction procedure (strip
adjustment).
Error per GPS observation — Error per strip (GPS/INS)
/
Ni A |
e
Error per |laser point
Error per block
Figure 1. Different scaled error components.
At the Survey Department strip adjustment techniques are
developed to minimize error components 3 and 4
[Crombaghs et al. 2000] whereas for error components 1 and
2 it is impossible to correct for. The amplitudes of the four
error components differ per project and depend on hardware,
software, measurement setup (block confi-guration) and
measurement procedures (e.g. calibration). The next
paragraph describes how the amplitude of the error
components can be determined.
4. DETERMINATION OF ERROR COMPONENTS
In order to quantify the different error amplitudes, various
methods are used, such as cross correlation techniques,
analysis of empirical covariance functions and 1d strip
adjustment. In the following paragraphs these techniques will
be described.
4.1 Error per point
The amplitude of this error cannot be determined by simply
taking the standard deviation of the laser data because this
standard deviation covers the total error budget of a single
laser point. To obtain the pure point noise, cross correlation
techniques are used. Flat areas of 50m x 50m without
vegetation and buildings are selected. The height of each