Extreme Precipitation in the Eastern Mediterranean in ERA5

Datasets with precipitation indices from the coastal areas of Syria, Lebanon and Israel are defined from the ERA5-Land database (0.1° resolution). In each coastal area the grid point with the highest hourly precipitation is selected. The declustered datasets are modelled by generalised Pareto distribution. The parameters of the stationary models are estimated using the maximum likelihood (MLE) and Bayesian inference methods.

Non-stationary models with several different covariates, i.e., time and teleconnection indices are incorporated into the scale parameter. The parameters of the non-stationary models are estimated using the MLE. The goodness-of-fit of stationary models is assessed by the Anderson-Darling test. QQ-plots subjectively assess the goodness-of-fit for both stationary and non-stationary models. 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 show clear non-stationarity with the time covariates. Non-stationarity with teleconnection covariates is incoherent, except for the North Atlantic oscillation (NAO) in Syria. Return levels are estimated for stationary and non-stationary models which are obtained from different quantiles of the time-changing scale parameter vector according to -risk scenarios. The results show that return levels are highest in Syria and lowest in Israel.