Entwicklung eines EDV-basierten Frühwarnsystems für die Blankaalabwanderung an der Mosel

The eel (Anguilla anguilla L.) is a fish that is mainly found in European waters. The River
Moselle is among the bodies of water inhabited by this specimen. During the downstream
migration into their Atlantic spawning ground, silver eels often experience severe to fatal
injuries while passing the turbines at the barrages. This annual migration takes place in a
relatively narrow timeframe. Therefore, knowing the trigger or beginning of said migration,
the mortality rate of the eels could be reduced by a fish-adapted turbine control or
comparable protective measures. This thesis introduces an early warning system, which
predicts the periods of silver eel emigration by means of certain abiotic factors.
On the basis of the information gleaned from different studies and the experience gained
from many years of professional fishing, those environmental factors were identified which
are connected with the migration of the silver eel. Extensive data analyses were used to
substantiate these findings. The water flow, the flow differences and the lunar phase were
particularly relevant. Furthermore, the season and the water temperature were taken into
account. In view of the different sources of information (experience and expert knowledge,
data sets and the findings derived from it), a hybrid structure of the early warning system
was realized. After examining different methods from the fields of soft computing and
mathematics or statistics, the fuzzy logic (knowledge-based), the case-based reasoning (casebased)
and the artificial neural networks (data-based) were selected. With each of these
methods, an independent prediction model was designed, tested and optimized. Special
characteristics were found during the data analysis and were taken into account by the use of
adequate modifiers.
The models were tested on the basis of the present data sets for the Moselle. It was shown
that it is possible to correctly predict most of the situations with increased catches
(suggesting a migration). Threshold values for a migration were defined based on the
catches. The same was done for the forecast values. Thus, for the 1963 to 1973 data record, a
total of 63% (artificial neural networks), 74% (fuzzy logic), and 83% (case-based reasoning) of
the events with increased catches could be detected. Since not every situation with a
favorable constellation of abiotic factors also led to a migration or higher catches, a lot of
"false" forecasts (up to 50%) were made as well. Good results have also been achieved when
using data from recent years and most events were identified.
A stand-alone program was developed for the practical application of the prognosis models.
This early warning system is a software which contains a user interface for reading data and
displaying prognosis values and into which the developed prognosis models are
implemented. In addition, recommendations for use were compiled and presented.