Laboratoire d'Océanographie et du Climat - Expérimentation et Analyse Numérique
Meeting room nr. 417, corridor 45-55, 4th floor
4 place de Jussieu
75252 PARIS CEDEX 05
Seminar: Trends in Mean and Extreme Values of River Discharge Time Series
by Malaak KALLACHE , Potsdam Institute for Climate Impact Research (PIK)
Statistical characteristica of river discharge are valuable indicators for
water management authorities. Here data analysis methods are presented to
assess trends in mean and extreme river run-off. This includes the
consideration of the auto-correlation structure of the data. Such an approach
is very useful to, e.g., assess the anticipated intensification of the
hydrological cycle due to anthropogenic climate change. The costs related to
more frequent or more severe floods and droughts are enormous. Therefore an
adequate estimation of these hazards and the related uncertainties is of
major concern. A study of discharge of basins of the Danube River and Neckar
River in Southern Germany is presented.
To evaluate trends of average discharge data a trend test is used, where the auto-correlation structure of the data is explicitly modelled using stochastic FARIMA processes. This is a crucial task, because auto-correlations are capable of producing spurious trends. It is assumed that the data consists of an additive combination of natural variability, which is represented by the stochastic process, and potentially a deterministic trend component. The trend is estimated using wavelets and represents the fluctuation of the data on large time scales. In case the trend adds more variability to the data than the natural variability is likely to generate the trend is considered as significant. When analysing about 90 discharge records in the Neckar and Danube River basin a spatially heterogeneous auto-correlation structure incorporating short- and long-term correlations and heterogeneous trend patterns are found.
To model trends in the extremes of river discharge data point processes are used. Thereby the exceedances over a threshold are used as extremes and they are assumed to be distributed according to a generalized Pareto distribution. In order to eliminate auto-correlation, the data are thinned out. Contrary to ordinary extreme value statistics, potential non-stationarity is included by allowing the model parameters to vary with time. By this, changes in frequency and magnitude of the extremes can be tracked. The model which best suits the data is selected out of a set of models which comprises the stationary model and models with a variety of polynomial and exponential trend assumptions. Common assessment measures, such as 100-year return levels, can be calculated from this model. Analysing winter discharge data of about 50 gauges within the Danube River basin, we find trends in the extremes in about one third of the gauges examined. The proposed approach allows to quantify the uncertainty of assessment measures such as return levels.
For more information: M. KALLACHE ()
Link to presentation: