Jude Musuuza

Study region: Lake Hume in Australia and Harsha Lake in USA. Study focus: Large-scale hydrological models (LSHMs), though important for both scientific and societal reasons, require the representation of many unknown features that influence river system response. However, current model identification practices in catchment modelling cannot lead to robust LSHMs for local decision-making. To address this, it is necessary to customise the models by integrating local data and knowledge from various sources (e.g. in-situ and earth observations) and fluxes. We present a framework to customize LSHMs for impactful local applications and showcase this using the global WWH hydrological model as the reference LSHM. New hydrological insights: We present significant improvements in modelling streamflow and actual and potential evapotranspiration, following WWH refinements to include local lakes and reservoir management. Local streamflow measurements and earth observations from NASA MODIS evapotranspiration products were used to re-calibrate the locally adapted model, leading to spatial consistency in performance. Combining multiple variables and metrics during model identification improved streamflow performance and robustness, with combination sets reducing variability and enhancing representation of diverse hydrological processes, highlighting the need for tailored metric and variable selection. This underpins the importance of including informative data in customized multi-objective modelling chains. Finally, incorporating reservoir management improved simulation of a regulated system, with local insights informing reservoir parameterization in LSHMs and bridging the gap to global-scale applications.

The effect of model calibration on the projection of climate change impact on hydrological indicators was assessed by employing variants of a pan-European hydrological model driven by forcing data from an ensemble of climate models. The hydrological model was calibrated using three approaches: calibration at the outlets of major river basins, regionalization through calibration of smaller scale catchments with unique catchment characteristics, and building a model ensemble by sampling model parameters from the regionalized model. The large-scale patterns of the change signals projected by all model variants were found to be similar for the different indicators. Catchment scale differences were observed between the projections of the model calibrated for the major river basins and the other two model variants. The distributions of the median change signals projected by the ensemble model were found to be similar to the distributions of the change signals projected by the regionalized model for all hydrological indicators. The study highlights that the spatial detail to which model calibration is performed can highly influence the catchment scale detail in the projection of climate change impact on hydrological indicators, with an absolute difference in the projections of the locally calibrated model and the model calibrated for the major river basins ranging between 0 and 55% for mean annual discharge, while it has little effect on the large-scale pattern of the projection.