Exploring multiple model structures

Astagneau et al. (2022) hypothesize that the response of a catchment to high-intensity rainfall events is highly heterogeneous due to complex interactions among the hydrological processes at short temporal and spatial scales. The aim of their study is to improve the simulation of summer floods by using a lumped conceptual rainfall–runoff model. They modify the GR5H model and test three hypotheses: i) large rainfall intensities increase the volume of effective rainfall, ii) large rainfall intensities induce a faster routing of effective rainfall to the catchment outlet, and iii) a combination of these two hypotheses. A large database consisting of 10,652 flood events in 229 French catchments are used. The results show that when the storages and fluxes of a lumped conceptual model dynamically depend on rainfall intensities, the errors in flood volume are less (at least in simulations at hourly time step). It is noted that these conclusions specifically hold good for a particular model structure and further testing with other models and the data from other regions would be required to establish the wider applicability. Since intense rainfall events do not last long, the intensity-dependent functions are triggered for very small number of time steps. To address the calibration issues arising due to the above hypotheses, Astagneau et al. (2023) suggest regionalizing the parameters of the intensity-dependent function.
Saavedra et al. (2022) investigates if hydrological consistency in contrasting climate periods can be improved by sampling the model space with a simple pareto framework and if such a model selection procedure can reduce uncertainties in precipitation elasticities and temperature sensitivities. They use the Framework for Understanding Structural Error (FUSE) to produce 78 different hydrological model structures from four different conceptual parent models. To test the ability of models to predict impacts of climate change, they perform differential split-sample tests of the models by calibrating on dry periods and evaluating on wet periods and vice versa. The models are tested on three catchments in Peru with areas ranging from 3545 km2 to 9586 km2. The results show that it is possible to identify some model structures that robustly simulate catchment-scale hydrology under different climate conditions, and that these are not necessarily the structures that perform the best for traditional efficiency metrics. The results also show that the model selection procedure resulted in a significant reduction in the spread in precipitation elasticities and temperature sensitivities.
Sinha et al. (2022) perform an intercomparison test of the GR4J lumped conceptual model against the spatially distributed mHM model using data from 50 catchments in UK. The models are calibrated by optimizing the Nash-Sutcliffe efficiency. Subsequently, the model performances in validation periods are evaluated by four performance metrics as well as five hydrological signatures characterizing the ability of the models to reproduce different components of the flow. The results support previous findings that a lumped conceptual models can perform equally well, and in some cases slightly better, than a more complex model, when the modelling objective is limited to discharge simulation.