Identifying Non-Stationary Risk in an Era of Changing Environmental Perils

Extremes in meteorological, hydrological, and climatic events have increasing impacts on social, economic, and political components of modern society. To provide measures of the impacts of extreme events, many statistical analyses have been conducted to identify characteristics of physical quantities such as precipitation, wind speed, storm surges, and river discharge under extreme conditions. In particular, research has focused on estimating parameters of key distributions to characterize tails of distributions where data are sparse. Additionally, the uncertainty in parameters can become large for small probability events. An additional complication to the evaluation of extreme values of many environmental perils is that many events result from mixed forcing due to compound processes. Therefore, mixed distributions are required to appropriately capture the statistical character of the forcing and the resulting extreme values.

In this presentation, the nature of extreme value statistics is examined in relation to historical and future conditions of key physical factors related to precipitation and wind speeds. Emphasis is placed on addressing the non-stationary aspect of extreme events under climate change. A unique modeling framework is used to scale historical extreme events under future climate scenarios. Additionally, mixed-distribution models are defined to account for contributions from physical characteristics unique to extreme precipitation and wind speed events.

A key aspect of this work is measure of incremental changes in the statistics of extreme events under traditional extreme value analyses and those that account for mixed physical processes and the non-stationary aspect of the climate system. Therefore, estimates of changes in factors such as return periods and return levels of key precipitation and wind events are made and the changes are placed in the context of climate change scenario, future year, and contribution of incorporation of non-stationary and mixed distributions in the extreme value analysis.