Customizing global hydrological models for local applications

Special thanks to the organizers of the CEST 2023 conference (30/08 - 02/09 2023) for inviting me as Speaker to the Hydrology and Water Resources session. 

The focus of my presentation was on the scientific tools to customize global hydrological models for local applications. 

Abstract: Hydrological modelling at the continental and global scales is scientifically attractive since, for instance, the large samples generated, as model outputs, can lead to process understanding along the Earth's strong hydroclimatic gradients. Such large-scale modelling is also important for the society with the hydro-meteorological information being provided at poorly gauged and even ungauged locations addressing the various water-related challenges. Nevertheless, continental and global hydrological models are expected to provide reliable and useful information at the regional scale, and hence model setup and parameter identification requires application of state-of-the-art scientific practices. The information content in the "traditional" data used in catchment modelling is not sufficient to diagnostically guide the parameter identification and improve process understanding in large-scale applications. The large spatial scales require many "unknown" features, e.g., lake extents, snowpack cover, land cover changes, reservoir regulations, dam capacities, irrigation practices, etc., that directly influence the hydrological response, while fluxes in the catchment can be left unresolved. Due to the large physical heterogeneity, the complexity of hydrological models and the limited data traditionally available, the current model identification practices in catchment modelling cannot accurately represent the physical processes and lead to a robust model for large-scale applications. Even when more sophisticated identification practices are applied, large scale hydrological models are still over-tuned to capture the dominant processes and their spatiotemporal dynamics, and little attention is given to fluxes at the regional scale. To make large-scale models useful for local/regional purposes, effort is needed to customize the models using local data from various sources (e.g. in-situ and earth observations) and for various fluxes. Here, we showcase the Lake Hume catchment in Australia, where we used the data from the global World-Wide HYPE (WWH) hydrological model as a benchmark to quantify how much the model performance (and hence process understanding) can be improved by fine-tuning the parameters by conditioning them to local in-situ and earth observations (EO). We note that the WWH was set up using global datasets and with the objective to perform adequately over the entire globe. We show the information that WWH missed for application at this local scale, including mis-delineation and omission of dams and reservoirs. In the local customization, the model was refined to include local lakes and managed reservoirs and recalibrated on locally-available discharge measurements and the freely available global EO-based MODIS evapotranspiration data from NASA. The results show significant improvements in the model estimation of discharge and the actual and potential evapotranspiration. This underpins the importance of including as much local information as possible in modelling chains used for decision-making. Where data are not available, the results show the potential of EOs to fill in that gap. Overall, we highlight the success of local customization of large-scale hydrological models through information extraction and use in model identification.