| Popis: |
Probabilistic event attribution (PEA) presents a method of quantifying the change in risk of extreme weather events, both in magnitude and probability of occurrence, due to an- thropogenic climate change. Studies have so far covered numerous different extreme events in different regions (Stott et al., 2016). One method of PEA relies on computing large ensembles of climate models for two different climate scenarios, one which represents the observed climate forcings, and another the climate forcings with anthropogenic greenhouse gas contributions removed. Other PEA studies using different methodologies have also been conducted. Until now there has been no formal validation of the climate models used in PEA studies. This thesis presents two ways of validating the models and applies these methods to a very large ensemble of a climate model simulating the observed climate for the period 1960 to 2010. The premise is, that for an attribution statement to be meaningful, the model should be able to accurately represent relevant weather statistics over a range of climate scenarios. In order to compare to observations, the period 1960-2010 is used as the range of scenarios. Conceived as an e-Science project, this thesis presents technical development of method- ologies in the detection and attribution of extreme weather events to climate change. The first method identifies and tracks storm-like features in meteorological data. A novel set of algorithms transforms the meteorological data to a hierarchical equal-area triangular grid, identifies storm-like feature points in the data and grows the points into objects. These objects are tracked as they evolve over time by a hybrid prediction-optimisation routine, which minimises a cost function to find a locally optimal set of feature tracks. Applying the algorithms to the large ensemble of climate models, and also to the ERA-Interim data, shows that the model can successfully capture the track length, persistence and position of low-pressure systems over Europe. However, the depression depth of the systems is not as well represented. The second method applies a forecast verification technique to the distributions of climate variables. Comparing the temperature, precipitation, mean sea level pressure (MSLP) and winds between the large ensemble and ERA-40 and ERA-Interim shows that the model can accurately represent temperature, precipitation and windspeed variables over Europe, after a bias correction has been applied. However, for low MSLP, there are irreparable biases in the low tails of the distribution. Overall, it is shown that using a large ensemble of climate models is a valid method of investigating the change in risk of extreme weather events due to climate change, as long as the variable to be attributed is carefully chosen. |
