Assessing nonstationarity impacts for historical and projected extreme precipitation in Turkey

The temporal variability in yearly and seasonal extreme precipitation across Turkey is investigated using stationary and nonstationary frequency approach. Four frequency distributions namely, generalized extreme value (GEV), gumbel, normal, and lognormal distributions are used for the historical period (1971-2016) as well as the projection period (2051-2100). The nonstationarity impacts are determined by calculating the percentage difference of return levels (30 years) between stationary and nonstationary cases for both periods. The results are presented as nonstationarity impact maps across Turkey, providing information about the spatial variability of the magnitude of impacts as well as the impact types, i.e., the increase or decrease in return levels of extreme precipitation. For nonstationarity analysis during the projection period, the projected precipitation data is obtained from a 12-member ensemble of the Coordinated Regional Downscaling Experiment (CORDEX) regional climate models (RCM) based on the worst emission scenario (RCP8.5). In addition to this, the effects of bias-corrections on stationarities are also investigated for selected RCMs. Performance evaluation of CORDEX ensemble members suggested significant intramodel and intraregion variability in the simulation of historical precipitation. Overall, GEV provided the best fit while normal distribution provided the worst fit to precipitation extremes. However, with a few exceptions, all the distributions exhibited a similar pattern for the historical impacts of nonstationarities across the country. The yearly nonstationarity impacts for the 30-year return level reached 35, 30, 25, and 20% for the Mediterranean, Aegean, Marmara, and Black sea regions, respectively, while a mixed type of nonstationarity impacts was found in Central, Eastern, and South-Eastern Anatolia regions. Moreover, the magnitude and impact type of nonstationarity showed important variations between the seasons. The ensemble analyses of CORDEX-based projected precipitation demonstrated that during the projection period, the nonstationarity impacts are found to be different in magnitude and type as compared to the impacts during the historical period in many regions (particularly in the eastern part of Turkey). The 30-year return levels of extreme precipitations are expected to increase (up to 30%) across Turkey in the yearly, winter, spring, and autumn seasons during the projection period. However, during the same period, the summer extremes in Eastern Anatolia and Mediterranean regions decreased ( up to 30%) with nonstationarity. Projected increases in the precipitation extremes lead to more floods and winter storms, while decreases in the summer precipitations will further dwindle water availability