|
Titel |
Applying clustering approach in predictive uncertainty estimation: a case study with the UNEEC method |
VerfasserIn |
Nilay Dogulu, Dimitri Solomatine, Durga Lal Shrestha |
Konferenz |
EGU General Assembly 2014
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 16 (2014) |
Datensatznummer |
250091385
|
Publikation (Nr.) |
EGU/EGU2014-5992.pdf |
|
|
|
Zusammenfassung |
Within the context of flood forecasting, assessment of predictive uncertainty has become a
necessity for most of the modelling studies in operational hydrology. There are several
uncertainty analysis and/or prediction methods available in the literature; however, most of
them rely on normality and homoscedasticity assumptions for model residuals occurring in
reproducing the observed data. This study focuses on a statistical method analyzing model
residuals without having any assumptions and based on a clustering approach: Uncertainty
Estimation based on local Errors and Clustering (UNEEC). The aim of this work is to
provide a comprehensive evaluation of the UNEEC method’s performance in view of
clustering approach employed within its methodology. This is done by analyzing normality
of model residuals and comparing uncertainty analysis results (for 50% and 90%
confidence level) with those obtained from uniform interval and quantile regression
methods. An important part of the basis by which the methods are compared is
analysis of data clusters representing different hydrometeorological conditions.
The validation measures used are PICP, MPI, ARIL and NUE where necessary. A
new validation measure linking prediction interval to the (hydrological) model
quality - weighted mean prediction interval (WMPI) - is also proposed for comparing
the methods more effectively. The case study is Brue catchment, located in the
South West of England. A different parametrization of the method than its previous
application in Shrestha and Solomatine (2008) is used, i.e. past error values in addition to
discharge and effective rainfall is considered. The results show that UNEEC’s notable
characteristic in its methodology, i.e. applying clustering to data of predictors upon which
catchment behaviour information is encapsulated, contributes increased accuracy of
the method’s results for varying flow conditions. Besides, classifying data so that
extreme flow events are individually represented helps to achieve normality in model
residuals’ distribution for each cluster identified. Reference: Shrestha, D. L., &
Solomatine, D. P. (2008). Data-driven approaches for estimating uncertainty in
rainfall-runoff modelling.International Journal of River Basin Management,6(2), 109-122. |
|
|
|
|
|