ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001
70
The discrepancy for spatial features can be mathematically
expressed by a multiple integral. However, it is known that
integrals may not have an exact solution. Moreover, there are no
standard procedures of finding the exact solution of the integral.
As a result, a traditional technique is unsatisfactory to solve this
problem in GIS.
Gaussian quadrature, a numerical integration technique,
approximates the integral of a function by integrating the linear
function that joins points of the function’s graph (Burden and
Faires 1993). In other words, the integral Jg(x) dx is
approximated by X Wi g(xj) where the nodes Xi, x 2 , .... x n and
coefficients wi, w 2 , ..., w n are chosen to minimize the expected
error between the integral and the approximation. This technique
can be modified in a straightforward manner for use in the
approximation of multiple integrals. This Gaussian quadrature is
implemented to calculate the multiple integral in Eq. 5, Eq. 7, Eq.
8 and Eq. 9.
4. RESULTS AND DISCUSSION
The analytical model for the uncertainty of spatial features is
applied to the example data of Shi and Cheung (1999). The two
expected nodes of the line segments are (0, 0, 0) and (1000, 0,
0). a,, ¿>1, Ci, a 2 , th and c 2 are 100ft, 196ft, 148ft, 30ft, 78ft and
90ft respectively. The covariance matrix, in the pdf of the
multivariate normal distribution f, is a 6x6 diagonal matrix with
nodes x1, yl, zl, x2, y2 and z2 and these nodal errors are
assumed to be independent. The confidence coefficient (1 -or) is
0.95. The expected discrepant area of the line segment is
62145.0ft 2 (as shown in Table 1).
Table 1. The expected discrepant area of spatial features
Spatial
feature
The expected discrepancy
Ratio
Analytical model
Simulation
model
Line
segment
62145.0ft 2
59344.1ft 2
1.047
Linear
85553.6ft 2
89036.1ft 2
0.961
Areal
15209871.4ft 3
18570274.0ft 3
0.819
Volumetric
37688986.2ft 3
44339983.8ft 3
0.849
* Ratio - ana| y tical resu ' t
simulated result
For the polyline, the three expected nodes are (0, 0, 0), (500,
500, 707.1) and (1500, 500, 707.1). a,, b u Ci, a 2 , tfc, c 2 , a 3 , ¿13
and c 3 are 100ft, 196ft, 148ft, 30ft, 78ft, 90ft, 100ft, 196ft and
148ft respectively. The covariance matrix in f is a 9x9 diagonal
matrix. The confidence coefficient (1 -a) is 0.95. The expected
discrepant area of the line feature is 85553.6ft 2 . Using the three
nodes of the polyline for the areal feature results in an expected
discrepant area equal to 5209871.4ft 3 .
For the example of a volumetric feature, an additional node (to
the existing three) is considered. This addition now specifies a
volumetric feature. This additional node is (500, 707.1, 500), and
its error ellipsoid has parameters a4 = 30ft, b 4 = 78ft and c 4 =
90ft. The expected discrepant area of the volumetric feature is
37688986.2ft 3 .
In Table 1, the expected discrepant areas of the spatial features
are recorded for both the numerical integration and the
simulation techniques. The ratio from the analytical model to that
from the simulation model is in the range of 0.8 to 1.1. In an
ideal situation, this ratio should be 1. A ratio varying from 1 is
due to the approximation of the expected discrepancy for both
techniques (numerical integration and simulation techniques).
In the simulation model, the accuracy of the result depends on
the number of simulation. The larger the number of simulation,
the higher the accuracy of the result. The accuracy of the
approximation in the analytical model is related to the number of
nodes chosen in the integral region. To a certain extent, the
oscillatory nature of the integral function affects the
approximation. In any case, the expected discrepant areas
calculated from the numerical integration and from the simulation
technique, are close to each other. Both techniques are valid.
The above four examples considered the discrepancy of the
spatial features when there is no correlation of nodal errors.
5. CONCLUSIONS
A newly developed analytical model to measure the uncertainty
of a spatial feature in 3D GIS was presented in this paper. The
uncertainty is determined by the ‘true’ location and the measured
location of the spatial feature. The discrepant area (or volume)
was used as a measure of the uncertainty. It was expressed as
a mathematical function of which the measured location and the
‘true’ location of the spatial feature were variables. Given the
measured location of the spatial feature, the discrepant area (or
volume) could be obtained and the derived result was the
discrepant area for this measured location. In general, the
measured location may be in the vicinity of the ‘true’ location.
Based on the assumption of error of a spatial feature, a number
of possible measured locations were considered in our proposed
model rather than one measured location of the spatial feature.
Therefore, the expected discrepant area was in the form of a
multiple integral. Since this multiple integral could be solved
analytically, the Gaussian quadrature, a numerical integration,
was implemented to provide an approximate solution for the
analytical model. The estimated expected discrepancy was
finally compared to the simulated solution.
In our previous uncertainty model for 3D spatial features, the
uncertainty model of 3D spatial features was studied using the
simulation technique. This simulation model was generated
some possible measured locations of a spatial feature based on
the same assumption of error of the spatial feature as stated in
this paper, and computed the average discrepant area (or
volume). In a mathematical point of view, there are a number of
infinite points inside a region. However, the simulation model
only sampled a certain number of the possible measured
location of the spatial feature instead of all possible measured
locations of the spatial feature. The accuracy of the expected
discrepant area (or volume) is question although more
simulations can provide a more precise result. Moreover, the
simulation model is quite time-consuming. Thus, we proposed
the analytical model by taking all possible measured locations of
the spatial feature into account, in order to provide the expected
discrepant area (or volume) with great accuracy in real time.
In this paper, an analytical model was provided to validate the
simulation model. The numerical results obtained from the
analytical model and the simulation results given in our previous
study can approximate a similar discrepant area (or volume).
However, the calculation for the numerical solution is much
faster than that for the simulated solution for the same problem.
Therefore, the numerical integration technique is considered the
preferred approach in studying the uncertainty of 3D spatial
features.
REFERENCE
Burden, R.L. and J.D. Faires. 1993. Numerical Analysis,
International Thomson Publishing, pp. 211-222.
Burrough, P.A. 1986. Principles of Geographical Information
Systems for Land Resources Assessment, Oxford: Clarendon
Press, pp. 103-135.
Burrough, P.A. and R.A. McDonnell. 1998. Principles of
Geographical Information Systems, Oxford University Press, pp.
220-240.
Caspary, W. and R. Scheuring. 1992. Error-band as measure of
geographic accuracy. Proceedings of EGIS’92, pp. 226-233.
Cassettari, S. 1993. Introduction to Integrated Geo-Information
Management, Chapman & Hall, pp. 2-3.