Initial look at EVT Return Levels

Continuing my study of Classical EVT

Main text

Coles, S. (2001). An introduction to statistical modeling of extreme values / Stuart Coles, London ; New York : Springer, [2001].

Also http://www.ral.ucar.edu/~ericg/extRemes/extRemes2.pdf which has been submitted to Journal of Statistical Software

fit1 <- fevd(TMX1, PORTw, units = "deg C")

plot(fit1) 

What the heck am I looking at? Visual ways to evaluate the fit of the model, which I don't have a feel for. The 2 quantiles plots are not convincing. 

Look at the Gumbel model

fit0 <- fevd(TMX1, PORTw, type = "Gumbel", units = "deg C")

plot(fit0) 

What do I see, Gumbel fit is worse than fit1

• Quantiles differ from regression lines more at higher quantiles ( I am not certain if that matters)
• The shape of the empirical and modelled density are less similar
• The Gumbel return levels exceed the CI and appears to be bias to lower temperatures

Bayesian

The most obvious difference is in the Return Period and Level plot. The Baysian return levels are more within the 95% CI. Not certain what this implies.