Representation of U.S. Warm Temperature Extremes in Global Climate Model Ensembles
Extreme temperature events can have considerable negative impacts on sectors such as health, agriculture, and transportation. Observational evidence indicates the severity and frequency of warm extremes are increasing over much of the United States, but there are sizeable challenges in both estimating extreme temperature changes and in quantifying the relevant associated uncertainties. This study provides a simple statistical framework using a block-maxima approach to analyze the representation of warm temperature extremes in several recent global climate model ensembles sampling different models and initial conditions uncertainties. We analyze mean and extreme temperature statistics at regional-to-local scales and evaluate model/ensemble skill based on the recent historical record (50 years).
Daily average and extreme temperatures vary widely between different models and ensembles, which influences skill in both bulk and extreme temperature distributions. Uncertainties due to structural model differences, grid resolution, and internal variability are also considerable, and the effects vary between ensembles. Taken together, these effects considerably influence historical and future projections of extreme temperature distributions, leading to large biases and wide ensemble spread. Results can help inform regional analysis that is particularly sensitive to extreme temperature events (e.g. agriculture).
This study provides a simple statistical framework using a block-maxima approach to analyze the representation of warm temperature extremes in several recent global climate model ensembles. Uncertainties due to structural model differences, grid resolution, and internal variability are characterized and discussed. Results show that models and ensembles differ greatly in the representation of extreme temperature over the United States, and variability in tail events is dependent on time and anthropogenic warming, which can influence estimates of return periods and distribution parameter estimates using Generalized Extreme Value (GEV) distributions. These effects can considerably influence the uncertainty of model hindcasts and projections of extremes. Several idealized regional applications are highlighted for evaluating ensemble skill and trends, based on quantile analysis and root mean square errors in the overall sample and the upper tail. The results are relevant to regional climate assessments that use global model outputs and that are sensitive to extreme warm temperature.