Non-Stationary Modelling of Extreme Temperatures in a Mountainous Area of Greece

Abstract

The generalised extreme value (GEV) distribution is often fitted to environmental time series of extreme values such as annual maxima and minima of temperatures. It is often necessary to allow the distribution’s parameters to depend on time or other covariates (non-stationary GEV). Increasingly, model fitting within the GAMLSS framework is being used as an alternative approach. A case study is presented of temperature extremes in a mountainous area of Greece divided into nine zones by altitude. Model fitting supported non-stationary GEV models for temperature with the location parameter depending linearly on year and zone, showing the expected dependence on altitude along with an increasing trend in annual maxima and declining trend in annual minima. The scale parameter for maxima depended on zone, with greater variability at higher altitudes. The scale parameter for minima increased over time. Fitting non-stationary Inverse Gaussian, Lognormal and Gamma distributions within the GAMLSS framework identified the same dependence on zone and year. There was little difference in goodness of fit of the various distributions.