Using the Generalized Pareto Distribution as a Tool for Identification and Attribution of Extreme Temperatures

Extreme temperature events are a topic of considerable importance due to the risks they pose to human and animal health. This is exacerbated by the increasing rate at which extreme temperatures are more frequently occurring globally, with new high temperature records being set and broken seemingly year after year. There are also concerns over the impacts that high heat has on local flora, fauna, hydrology, etc., as well as entire environmental and sociological systems. In spite of this, there is still much debate on the exact way to measure these events Frequent questions include how to define the start and end points of an event, the event locality, its extent, and its severity. Indeed, the question regarding the length scale of a given event, especially about separating out discrete heat events from seasonality and climate variability, proves quite difficult. In order to address these questions, a method of identifying extreme events and describing their characteristics based on the tail behavior of a fitted generalized Pareto distribution is proposed. We make use of NCEI’s nClimGrid Daily dataset averaged over each of the climate divisions to generate tail distributions of maximum and minimum temperature based on monthly data exceeding the 90th quantile along with warm-season exceedance thresholds. This produces a highly parseable daily dataset of quantiles that can be used to identify climate divisions currently experiencing abnormally high temperature events, the degree to which they are extreme, and the duration they have persisted. We make use of this index to identify changes to quantiles in temperatures between the mid-1900s and the most recent few decades and the subsequent changes in the return period for extreme temperatures exceeding certain thresholds for a given time scale. We use the notable extreme heat in the U.S. in the past several years to demonstrate the utility of this approach.