Premium
Comparing Australian heat waves in the CMIP5 models through cluster analysis
Author(s) -
Gibson Peter B.,
PerkinsKirkpatrick Sarah E.,
Alexander Lisa V.,
Fischer Erich M.
Publication year - 2017
Publication title -
journal of geophysical research: atmospheres
Language(s) - English
Resource type - Journals
eISSN - 2169-8996
pISSN - 2169-897X
DOI - 10.1002/2016jd025878
Subject(s) - heat wave , robustness (evolution) , climatology , climate model , uncorrelated , spatial ecology , cluster (spacecraft) , contrast (vision) , environmental science , climate change , scale (ratio) , meteorology , computer science , geology , geography , statistics , mathematics , ecology , biochemistry , oceanography , chemistry , cartography , artificial intelligence , biology , gene , programming language
Quantitative projections of climate extremes on local to regional scales are highly valuable for planners and decision makers and necessary for effective local climate change adaptation. However, in contrast to the model robustness of simulated extremes at the global scale, the robustness in simulating past and future extremes often diminishes over finer spatial scales. In this study we analyze heat waves simulated by state‐of‐the‐art global climate models over the Australian region. For the first time we present results explicitly detailing the model spread in simulated heat wave trends and climatology for this region for the recent past (1958–2005). As expected, large intermodel spread is observed at the local to regional scale for both heat wave trends and climatology. By analyzing multiple initial condition runs from individual models, we show that model internal variability strongly influences the spatial patterns of heat wave trends, while intermodel differences in heat wave climatology appear more influenced by model uncertainty. From a model evaluation perspective, cluster analysis is shown to be useful in characterizing robust spatial features of heat waves simulated by the models. In contrast to the multimodel mean, where uncorrelated spatial features tend to be averaged out, cluster composites preserve these features. Since previous examinations have tended to focus on the multimodel mean the extent of model spread may have been overlooked. Further examination of the processes that lead to model differences and biases is needed.