Group seminar on 5. January, 14:15 CET
A Probabilistic Assessment of Extreme Precipitation in the Eastern Mediterranean
Precipitation indices of the Syrian, Lebanese and Israeli coastal areas are modelled using the generalised Pareto distribution. All datasets are declustered after a threshold has been selected. The parameters of the stationary models are estimated using the maximum likelihood, L-moments and Bayesian inference methods. The goodness-of-fit of stationary models is assessed quantitatively and qualitatively with the Anderson-Darling (AD) test and QQ-plots respectively. Trend tests using Spearman’s rho, the Mann-Kendall test and a modified Mann-Kendall test are applied to declustered peak-over-threshold (POT) datasets. Non-stationarity models with several different covariates, i.e. time and teleconnection indices, are applied to models which have passed the AD test. The goodness-of-fit of non-stationary models is assessed in comparison to the stationary models with the likelihood ratio test (LRT) and with the differences in the Akaike information criterion (AIC). The results of the non-stationary models do not show clear influences from teleconnection patterns on extreme precipitation indices, except for the North Atlantic oscillation (NAO) which is shown to have an influence on localised extreme precipitation in the Syrian coast. Non-stationary models indicate that localised extreme precipitation will increase with time in all coastal regions. Return levels are computed for the models that passed the AD test and those that fared better in the LRT and AIC. Two different return level approximation methods are reviewed for non-stationary models. The sensitivity of the return level with respect to the NAO and the sensitivity of the return level with respect to the return time is conducted on the Syrian coastal area dataset, in which higher sensitivities are yielded with negative NAO.