Abstract

Estimation of return levels, based on extreme value distributions, is of importance in the earth and environmental sciences. To incorporate non-stationarity in the modelling, the statistical framework of generalised additive models for location, scale and shape is an option, providing flexibility and with a wide range of distributions implemented. With a large set of selections possible, model choice is an issue. As a case study, we investigate annual minimum temperatures from measurements at a location in northern Sweden. For practical work, it turns out that care must be taken in examining the obtained distributions, not solely relying on information criteria. A simulation study illustrates the findings.

Keywords

GEV distribution; GAMLSS; Non-stationary models; Extreme temperatures; Model selection

Published in

Acta Geophysica
2019, volume: 67, number: 6, pages: 1599-1604

Authors' information