 |
| Titel |
Flood alert system based on bayesian techniques |
| VerfasserIn |
Z. Gulliver, J. Herrero, C. Viesca, M. J. Polo |
| Konferenz |
EGU General Assembly 2012
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 14 (2012) |
| Datensatznummer |
250069414
|
|
|
|
|
|
| Zusammenfassung |
| The problem of floods in the Mediterranean regions is closely linked to the occurrence of
torrential storms in dry regions, where even the water supply relies on adequate water
management. Like other Mediterranean basins in Southern Spain, the Guadalhorce River
Basin is a medium sized watershed (3856 km2) where recurrent yearly floods occur , mainly
in autumn and spring periods, driven by cold front phenomena. The torrential character of the
precipitation in such small basins, with a concentration time of less than 12 hours, produces
flash flood events with catastrophic effects over the city of Malaga (600000 inhabitants).
From this fact arises the need for specific alert tools which can forecast these kinds of
phenomena.
Bayesian networks (BN) have been emerging in the last decade as a very useful and
reliable computational tool for water resources and for the decision making process. The joint
use of Artificial Neural Networks (ANN) and BN have served us to recognize and simulate
the two different types of hydrological behaviour in the basin: natural and regulated. This led
to the establishment of causal relationships between precipitation, discharge from upstream
reservoirs, and water levels at a gauging station. It was seen that a recurrent ANN model
working at an hourly scale, considering daily precipitation and the two previous
hourly values of reservoir discharge and water level, could provide R2 values of
0.86. BN’s results slightly improve this fit, but contribute with uncertainty to the
prediction.
In our current work to Design a Weather Warning Service based on Bayesian techniques
the first steps were carried out through an analysis of the correlations between the water level
and rainfall at certain representative points in the basin, along with the upstream reservoir
discharge. The lower correlation found between precipitation and water level emphasizes the
highly regulated condition of the stream. The autocorrelations of the variables were also
analyzed, where the water level, with time lags of 12 hours related to the concentration time,
was found to be most significant. In short, the fits to the different distribution functions
of extremes were unsatisfactory, as the data were of poor quality and scant. This
problem with data is not unusual in small and medium sized Mediterranean basins
and becomes the real challenge to any prediction system based only on statistical
methods.
The aim of the resulting tool is to develop and maintain a numerical short-range weather
forecasting system for operational use by the regional water management entities. The
development of this tool is also corroborated by recent survey results, which identify the need
to develop site specific models for water management in these Mediterranean regions, so
prone to flash flood events (NOVIWAM, 2011 Novel Integrated Water Management systems
for Southern European Regions, Seventh Framework Programme, EC, 2010-2013). |
|
|
|
|
|
|