We investigate how to make statistical predictions of extreme weather such that events predicted to occur with a probability of 1 % will occur 1 % of the time. We apply the methods we describe to a standard extreme weather attribution example from the recent climate literature. We find that the methods we describe imply that extremes are roughly twice as likely as when estimated using maximum likelihood. We have developed a software package to make it easy to apply these methods.

We present a novel stochastic weather generator conditioned on circulation patterns and regional temperature, accounting for dynamic and thermodynamic atmospheric changes. We extensively evaluate the model for the central European region.  It statistically downscales precipitation for future periods, generating long, spatially and temporally consistent series. Results suggest an increase in extreme precipitation over the region, offering key benefits for hydrological impact studies.

Global warming is a big issue: it is necessary to know more details to make a forecast model and plan adaptation measures. Warming varies in space and time and models often average it over large areas. However, it shows great variations between months of the year. It also varies between regions of the world and between lowland and highland regions. This paper uses statistical and machine learning techniques to quantify such differences between Italy and the UK at different altitudes.