Abstract/Summary

Earth system models have become viable alternatives to traditional hydrological models in supporting global hydrological forecasting of river flow during the last decades. Hydrological forecasting systems rely on a climatology of flood hazard to derive flood thresholds, which are used to generate early flood warnings. However, the derivation of these climatologies is not straightforward and limitations, errors and uncertainties may play a major role and can significantly influence the quality of the flood warnings. This thesis evaluates some of the crucial characteristics of the reanalysis data sets used to produce the flood hazard climatologies, such as the land data assimilation and snow scheme complexity. Limitations of these data sets are identified, with suggestions presented to further improve the hydrological modelling and threshold generation methodologies. This in turn will lead to improved climatologies as crucial elements in delivering higher quality flood warnings. It was found that increments produced by the land data assimilation of snow and soil moisture can lead to systematic water budget errors and subsequently contribute to significant errors in river discharge simulations. Results have also shown that a more complex snow scheme with multiple layers can generally improve river discharge unless there is permafrost, where improvements required further adjustments of the snow and soil freezing parametrisations. In addition, the linear trend analysis of a state-of-the-art hydrological reanalysis data set revealed widespread, dominantly negative trends globally, that can adversely impact on the use of thresholds in flood warnings, derived from these reanalyses. In order to improve the quality of the flood hazard climatologies, an alternative threshold generation method, using ensemble reforecasts, has been developed and shown to deliver vastly improved forecast reliability and skill. This thesis contributes to better understanding of the global flood hazard climatologies in Earth system models and implements a more sophisticated method for producing these climatologies which will deliver better flood warnings. Additional research avenues are also recommended to further improve the hydrological representation of the Earth system models and the generation of the flood hazard climatologies in order to achieve the best possible hydrological forecast quality.