 |
| Titel |
A similarity based approach to identify homogeneous regions for seasonal forecasting |
| VerfasserIn |
Simon Schick, Ole Rössler, Rolf Weingartner |
| Konferenz |
EGU General Assembly 2015
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 17 (2015) |
| Datensatznummer |
250107979
|
| Publikation (Nr.) |
 EGU/EGU2015-7707.pdf |
|
|
|
|
|
| Zusammenfassung |
| Seasonal runoff forecasting using statistical models is challenged by a large number of
candidate predictors and a general weak predictor-predictand relationship. As the area of the
target basin increases, often also the available data sets do, thus reinforcing the predictor
selection challenge. We propose an approach which follows the idea of “divide and conquer”
as developed in computational sciences and machine learning: First, the macroscale target
basin is partitioned into homogeneous regions using all its gauged mesoscale subbasins.
Second, one representative subbasin per homogeneous region is identified, for which models
are fitted and applied. Third, the resulting forecasts are combined at the scale of the
macroscale target basin.
This approach requires a suitable method to identify homogeneous regions and
representative subbasins. We suggest a way based on hydrological similarity, as catchment
similarity estimated with respect to physiographic-climatic descriptors does not necessarily
imply similar runoff response. Each descriptor is derived from daily runoff series and aimed
to reflect a specific catchment characteristic:
autocorrelation coefficient, parameters of fitted Gamma distribution and
low/high flow indices (based on daily runoff values)
fluctuation of the standard deviation within the yearly cycle (based on weekly
runoff values)
dominant harmonics obtained from the discrete Fourier transform (based on
monthly runoff values)
long term trend (based on yearly runoff values)
Where necessary, the runoff series first need to be standardized, aggregated, detrended or
deseasonalized.
As a preliminary study we present the results of a cluster analysis for the Swiss Rhine
River as macroscale target basin, which leads to about 40 mesoscale subbasins with runoff
series for the period 1991-2010. Problems we have to address include the choice
of a clustering algorithm, the identification of an appropriate number of regions
and the selection of representative subbasins per region. The results are finally
discussed with respect to the runoff regimes as defined in the Hydrological Atlas of
Switzerland. |
|
|
|
|
|
|