气候变化领域集成服务门户

Modelling seasonal temperature extremes in weather patterns allows for better forecasting and prediction. Analysis of extreme values over a given time period is usually done by fitting a generalized extreme value (GEV) distribution to the maximum values in the data; however, lack of sufficient weather recordings due to missing data or violation of independence assumptions (block maxima should be over a large time interval) may result in a poor fit for the GEV model. Modelling over larger, clustered regions may overcome some of these problems and can provide insight into macroscopic weather and climate changes. In this article, we analyse temperature measurements in July and August taken from stations across Texas and New Orleans, Louisiana. We introduce clustering techniques which group stations by temperature trends and mutual information before performing extreme value analysis on the clusters of time series. This obviates some of the problems commonly encountered in analysing single station weather data. Extreme analysis of the resulting clusters provides compelling evidence of nonstationarity of the distributional parameters in the GEV model and points to an increased likelihood from the period roughly 1980 to present of observing higher extreme temperatures for the months of July and August. We tabulate the probabilities of extreme temperatures in the clusters according to a nonstationary model. Our techniques can be easily adapted to a wide range of climatological problems.