In order to give an idea of the estimated ARMA parameters
and their variations within the regions, table 3 lists all
estimated process parameters. The parameters of a region
show a similar behaviour although the most profiles have
different directions in the terrain. Hence the mean values
of the parameters within a region are also given.
The standard deviations of the prediction errors and the
process parameters may be considered to be typical for a
region. However, the dependence on the sampling interval -
Ackermann /1/ assumes a linear dependence - must be
regarded. The relationship between the properties of ARIMA
processes and the sampling intervals remains a subject of
further investigations.
As a selection the first profiles of the regions Bohuslaen,
Uppland, Disko Island and Drivdalen are ploted in the
figures 1-4. The prediction errors are ploted too below the
profiles refered. The different scales of the plots should
be noticed.
7 Remarks
7.1 Breaklines
Different types of terrain require different processes. A
particular ARIMA process is only capable of modelling a
homogeneous terrain without breaklines.
On the other hand ARIMA processes are suitable for detec-
ting breaklines: suddenly arising great prediction errors
may indicate them. An example is given by a profile from
Drivdalen in figure 4.
For our investigations the measured terrain profiles were
separated according to the breaklines. Hence the profiles
examined in chapter 6 are from homogeneous terrain.
7.2 Observation Errors
The observation of the time series x(t) represented by an
ARMA process is usually contaminated with an observation
error n(t). It can be reasonably assumed that these errors
are additive white Gaussian noise with
n(t) = N (0, v 25.,
Therefore only the observed process y(t) is available
y(C) * x(t) 4 n(t).
In order to cope with the problem of observational noise,
two methods became known to the author.
- 434 -