Extreme events in climate and weather an interdisciplinary workshop (10w5016)

Arriving in Banff, Alberta Sunday, August 22 and departing Friday August 27, 2010


Montserrat Fuentes (North Carolina State University)

(Norwegian Computing Center)


There is a perceived intensification of the Atlantic hurricanes over the last decade. The climate models predict higher sea surface temperatures, which in turn leads to more intense hurricanes. There has been several analyses of trends in hurricanes, and the relation to sea surface temperature, North Atlantic Oscillation, El Nio-Southern Oscillation, etc. Yet these analyses have not, from a statistical point of view, been fully convincing. There are, e.g., issues of temporal dependence and of ecological correlation. Hurricanes are of course strongly related to precipitation extremes. In fact, recent tropical systems (Fay and Hanna) brought much needed rain to the U.S. Southeastern states, and contributed to the easing of the prolonged drought condition in the region.

Another situation of practical interest is the case of high winds, temperatures around 0�C, and heavy snowfall. Because of the weight of wet snow on branches, trees are more vulnerable to the wind. This is a multivariate situation in which not all the variables of interest are extreme.

Climate models also predict many other observable features, such as increased sea levels, increased frequency of heat spells, increased rain intensity (but decreased frequency) over mid-latitudes, increased frequency of droughts, decreasing temperature range (but increasing minimum and maximum temperatures), and so on. These are all extreme events in a nonstationary climate, and in order to make sure that the predicted trends in these extremes actually are statistically significant, new tools for analyzing extremes are needed.

There has been substantial amounts of recent work on nonstationary temporal extreme values, on spatial extreme values for Gaussian processes, and on multivariate extreme values when not all components are extreme. However, there remains much to be done on extreme values for non-Gaussian spatial processes, on space-time extremes, on exceedances of levels that vary according to trends in space and time, and on multivariate approaches to each of these problems.

The format of the workshop will be as follows. Every day we will have two lectures, with prepared discussants and then a general discussion. In addition we will have working groups who summarize the current state of affairs, develop research agendas for and work towards solutions of particular types of problems, such as space-time extremes for nonstationary processes, multivariate limiting distributions when only some components are extreme, comparison of climate model extremes to observed extremes, Rice�s formula for mixtures of Gaussian random fields, etc. We will also have poster sessions, to which all participants are encouraged to contribute. During the final day we will draft a comprehensive statement about what needs to be done in order to be able to answer scientifically and politically crucial questions relating to climate change.