Abstract
The latest report from the Intergovernmental Panel on Climate Change (IPCC) states that it is unequivocal that climate change is occurring. One of the largest impacts of climate change is anticipated to be an increase in the severity of extreme events, such as extreme precipitation. Floods caused by extreme precipitation pose a threat to human life and cause high economic losses for society. Thus, strategies to adapt to changes in extreme precipitation are currently being developed and established worldwide. Information on the expected changes in extreme precipitation is required for the development of adaptation strategies, but these changes are subject to uncertainties. The focus of this PhD thesis is the quantification of uncertainties in changes in extreme precipitation. It addresses two of the main sources of uncertainty in climate change impact studies: regional climate models (RCMs) and statistical downscaling methods (SDMs). RCMs provide information on climate change at the regional scale. SDMs are used to bias-correct and downscale the outputs of the RCMs to the local scale of interest in adaptation strategies. In the first part of the study, a multi-model ensemble of RCMs from the European ENSEMBLES project was used to quantify the uncertainty in RCM projections over Denmark. Three aspects of the RCMs relevant for the uncertainty quantification were first identified and investigated. These are: the interdependency of the RCMs; the performance in current climate; and the change in the performance of the RCMs from current to future climate. The interdependency of the RCMs was estimated using two different methods. These led to slightly different results but to the same conclusion; that the RCMs cannot be considered independent. The performance of the RCMs under current climate conditions was assessed using a range of precipitation indices, metrics, and observational data sets. It was found that these factors have a large influence on the performance estimated for the RCMs. This highlights the fact that it is not possible to identify a single best or worst RCM. The possible change in the performance of the RCMs under future climate conditions was explored using the relation between the bias of the RCMs and the observed precipitation intensity. For all the RCMs, the magnitude of the bias depends on the precipitation intensity. Hence, changes in bias can be expected to occur with changes in extreme precipitation. These findings were taken into account in the development of a Bayesian approach, which quantifies the statistical uncertainty in the change in extreme precipitation. In general, extreme precipitation intensity is expected to increase by the end of the century, but this change is associated with large uncertainties, especially in summer. With a probability of 95%, extreme precipitation is estimated to increase in winter, but in summer the values range from a decrease of 40% to an increase of 40%. A set of tests were carried out to assess the influence of accounting for the interdependency and change in bias of the RCMs in the quantification of uncertainty. The results highlight the importance of taking these two aspects into account. If they are not accounted for there is a risk of underestimating the uncertainty and reaching overconfident results. The second part of the study addressed the uncertainty arising from SDMs for two applications: river flooding in eleven European catchments; and urban flooding in Denmark. A range of SDMs were applied at daily and hourly resolution to the RCMs in the ensemble. The results for Denmark from both applications showed that in general the SDMs agree on an expected increase in extreme precipitation intensity. The uncertainty was explored by analysing the differences in the results of the SDMs and by comparing them with the differences within the RCM outputs. It was found that even though the variability within the SDMs is smaller than within the RCMs, it is not negligible. For example, in the river flooding application it represents approximately 30% of the total variance. This study contributes to the understanding of the uncertainties in climate change impact studies arising from RCMs and SDMs. The Bayesian approach suggested is a step forward towards a more comprehensive quantification of the uncertainties in a multi-model ensemble of RCMs. This approach could potentially be extended to include the uncertainty arising from other sources, such as SDMs. Further research is suggested in this direction. The findings of this study point out that there are large uncertainties in changes in extreme precipitation under climate change conditions. These uncertainties should not be seen as a reason for postponing action on climate adaptation. We have enough knowledge to carry on with the development of adaptation strategies, but their robustness must be ensured by including information on the uncertainties in climate change impact studies.