Modelling seasonality in extreme precipitation: a UK case study

We examine a statistical model for the description of the seasonal variation of extreme daily precipitation at 689 stations across the UK. The probability distribution for monthly maximum precipitation intensity is modelled with a generalised extreme-value distribution (GEV). Instead of modelling the distribution of precipitation maxima separately for every month, we propose an overall model with seasonally-varying location and scale parameters and a constant shape parameter. This model is tested against an augmented version with the shape parameter allowed to vary as well. Furthermore, we compare model adequacy for block length of one and two month and found no major improvements for the longer block-length. Based on this model, the 10 and 100-year return levels are calculated conditioned on the month of the year. The interpolation of return levels to a complete coverage of the UK allows for an identification of spatial patterns and their temporal evolution. These patterns suggest that different mechanisms for extreme precipitation are dominant in different regions of the UK.