On Low-Frequency Rainfall Variability Bias in Climate Model Simulations

Water resource managers need to assess changes in low-frequency rainfall variability for future climates which can only be obtained through climate model simulations. These models excel at simulating selected atmospheric variables, yet struggle to represent various attributes of rainfall. The above is especially true for rainfall low-frequency variability. To address these deficiencies, a range of bias correction strategies have been developed which are commonly used in water resources assessments worldwide. While most bias correction alternatives focus on correcting differences in mean, variance or distributional shape, Nested Bias Correction (NBC) stands out as an optimal method for correcting low-frequency variability bias in simulations. This thesis investigates the ability of NBC to improve low-frequency rainfall variability features in both global and regional climate models in four ways: (1) By measuring global climate model (GCM) representation of low-frequency rainfall variability through the Aggregated Persistence Score before and after applying correction; (2) By applying multiple bias correction techniques, including NBC, on regional climate model (RCM) inputs and examining any improvement in the downscaled rainfall; (3) By exploring limiting factors within the RCM on improving rainfall attributes, particularly the lateral boundaries; and (4) By evaluating the lack of dynamical consistency in atmospheric bias correction approaches using potential vorticity. The results presented show that GCMs struggle to simulate low-frequency rainfall variability compared with observations and that NBC produces a substantial improvement. Also, RCMs similarly struggle with this attribute, but this can be improved through bias correcting input conditions. This work identifies limiting factors in improving RCM simulations which revolve around the lateral boundaries. Finally this work shows that potential vorticity can be used to evaluate the lack of dynamical consistency created when bias correcting multiple atmospheric variables individually versus correcting them whilst preserving inter-variable relationships. Using the contributions developed in this thesis, future climate modelling should aim to improve low-frequency rainfall variability within models and related bias correction techniques. Future work using these contributions will lead to better simulations of flood and drought and suitable policies for securing long-term water supply under a changing climate