A generalized extreme value approach for the analysis of stationary climatic covariate in a Mediterranean city

Abstract

Extreme value theory (EVT) is used as univariate extreme value analysis (EVA) in order to analyze and model the covariates temperature, relative humidity (RH) and the thermal comfort index (humidex) issued from a dataset of 38 years in Tunis. It is a South Mediterranean area known as a hotspot for climate change. The best approach is to reduce the data considerably by taking annual block maxima from mean monthly data. It will converge to a generalized extreme value distribution in order to estimate the return levels of the studied parameters. The stationarity of the series are checked by augmented Dickey-Fuller test. The modeling of the three parameters shows a Weibull distribution pattern. The extreme/maximum monthly means temperature of 30.2°C and humidex of 39.4 have a common return level between 300 and 350 years. The highest mean monthly RH of 86.0% is expected to be exceeded every 50 years. For the next 38 years, the maxima monthly mean temperatures are expected to be stable, and the maxima monthly mean RH values, as well as the humidex monthly mean maxima are expected to decrease. The percentile air temperature hot day (TX90p) and night (TN90p) indices show globally linear upward trends and the ones of cold days (TX10p) and cold nights (TN10p) have a downward trend. The diurnal yearly temperature range shows an almost flat trend for its evolution through the years of study.