Presenter
Mr. Murat Yegin, Middle East Technical University (META)
Co-author(s)
Mr. Ahmet Korpinar, Middle East Technical University
Mrs. Gulsah Karakaya, Middle East Technical University
Mrs. Elcin Kentel, Middle East Technical University
Sub-theme
3. Building Resilience for Disaster Prevention and Mitigation
Topic
3-7. Management of water risks induced by extreme weather and climate events
Body
Although non-stationarity is common in nature, the time series in flood modeling are mostly assumed to be stationary for simplicity. However, this assumption may underestimate the severity of extreme events. In this study, stationarity analysis by fitting a Generalized Extreme Value distribution is carried out to investigate the effects of climate change on the distribution of maximum precipitation. The precipitation data ranging from 1976 to 2005 of 53 meteorological stations from the south of Turkey is used in the analysis. To represent non-stationarity, parameters of Generalized Extreme Value distribution are modeled as functions of the following commonly used covariates: maximum temperature of the day in which the block maxima event occurred, time in year, and the North Atlantic Oscillation index. In addition, two new covariates are introduced: the first one is the number of days in a year whose maximum temperature exceeds the long-term average of daily maximum temperature and the second one is the value obtained from a linear regression line fitted to historic daily maximum temperatures. Three types of non-stationary models are formulated based on the parameters of a Generalized Extreme Value distribution, that is, models with i) non-stationary location parameter, ii) non-stationary scale parameter, and iii) non-stationary location and scale parameters. A total of 30 non-stationary models are developed and an Akaike Information Criterion-based scoring system is introduced to assess the performances of these models; hence, the best non-stationary model for each meteorological station is determined. Afterward, the best non-stationary model is compared against the stationary model by the Likelihood Ratio test. The results show that non-stationary models perform better than stationary models for almost half of the stations. Hence, stationarity analysis of the hydrometeorological variables (such as annual maximum precipitation, annual maximum discharge) is crucial in the management of water risks by extreme weather and climate events.