RSSeNEWS to be determined...

The Environmental Statistics Section and the East Midlands local group joined forces in April to organise a meeting on Extremes in the environment. The three talks all focused on weather-related extremes in the UK and generated plenty of discussion.

Simon Brown from the Met Office Hadley Centre used a point process characterisation of threshold exceedances to model extreme temperatures, as simulated by regional climate models. To understand the effect of global warming, the statistical model parameters were allowed to depend on global mean temperature. The results suggested that in many areas of the world global warming is associated with changes in both the location and scale of extreme temperature distributions.

For the UK, Simon compared models fitted to observations and to climate model simulations under 20th century conditions. In most parts of the country the models differed significantly. However, parameters representing changes in response to increases in global mean temperature were in much better agreement. This suggests that climate model simulations can be used to assess potential future changes in UK temperature extremes, although issues remain about how best to adjust for the biases in the simulations.

Spatial risk assessment

The second talk on spatial risk assessment was from Caroline Keef of JBA Consulting. She briefly reviewed asymptotic theory that can be applied to the simultaneous modelling of extremes at multiple sites. She then described a relatively tractable model for multivariate extremes.

In the first of two applications, this model was used to study storm surges in the North Sea. Caroline illustrated how the model can be used to visualise the spatial pattern of surge levels given that an extreme event is occurring at a specified location and also showed how it could be used to simulate spatial fields of extreme surges. In her second application, the model was used to simulate spatially realistic flood events in north-east England. Coupled with a model for the consequent flood damage to individual properties, this provided a tool for estimating economic losses from rare events.

An alternative model for multi-site extremes

To conclude the meeting, David Walshaw from Newcastle University described an alternative model for multi-site extremes with an application to daily rainfall in the UK. At individual sites, classical asymptotic theory motivates the use of generalised extreme value (GEV) distributions to model annual rainfall maxima. For multiple sites, however, asymptotic theory leads to a limited range of spatial dependence structures and, in practice, annual maxima do not appear sufficiently extreme for these dependence results to hold.

David therefore suggested a pragmatic solution in which the annual maxima at each site are transformed so that their marginal distribution is no longer GEV but standard normal such that the transformed maxima are then modelled using a multivariate normal distribution with spatial correlation structure.

This approach can be used to describe extreme rainfall within homogeneous regions. For application to the entire UK, he used a hierarchical structure in which the country was split into nine such regions with model parameters treated as region-specific random effects. Bayesian fitting of this hierarchical model suggested that it could be simplified by constraining some parameters to be the same for all regions.