Changes in Large Spatiotemporal Climatic Extreme Events Beyond the Mean Warming Signal
Abstract:Weather and climate extremes impose substantial impacts on human societies and ecosystems. In particular, events that are large in space (areal extent), time (duration) or both are likely to be associated with highly significant consequences. Hence, a better detection, characterization and understanding of such anomalous events is crucial.
There is widespread consensus on a global and continental-scale warming trend, which leads to increases in the number, magnitude and frequency of temperature extremes (Hansen et al., 2012). It is less clear, however, if this warming also coincides with a broadening of temperature distributions (Huntingford et al., 2013). Moreover, the question whether other climate variables, such as large-scale precipitation deficits, likewise change, remains largely unanswered (Sheffield et al., 2012; Seneviratne 2012).
In this study, we address this issue by investigating the characteristics of large extremes, using an algorithm that detects the n largest spatiotemporally connected climate extremes for any time period. The deployed algorithm detects, depending on the chosen time step and variable, major heat waves, cold spells or droughts. We find a robust increase in the magnitude of large hot temperature extremes on a global and European scale in observations and reanalysis products, whereas the duration and affected area of those extremes does not show any pronounced changes. These results reveal that there is a detectable signal in temperature distributions beyond the mean warming trend, which might imply a structural change in the making of large extreme events.
Furthermore, we use the CMIP5 ensemble of models and an ensemble of 100+ members of a regional climate model for Europe (HadRM3P within the weather@home framework) in order to conduct a global and continental-scale analysis of large extreme events in temperature and precipitation. The employment of those model ensembles allows to sample more reliably the tails of the distribution, which will result in a better characterization of the magnitude, frequency as well as spatiotemporal structure of these events.