Calibration of hydrological models for ecologically-relevant streamflow predictions: a trade-off between performance and consistency

Abstract

The ecological integrity of freshwater ecosystems is intimately linked to natural fluctuations in the river flow regime. Anthropogenic alterations in flow regimes threaten water security and freshwater biodiversity in many regions of the world. The impacts of climate change on the hydrological cycle change local flow regimes and thus impact on the ecological systems. In catchments with little human-induced hydro-morphological changes, existing hydrological models can be used to predict changes in local flow regime in order to assess whether its rivers remain a suitable living environment for endemic 5 species. However, hydrological models are traditionally calibrated to give a good general fit between observed and simulated hydrographs, e.g., using an optimisation with an objective function such as the Nash-Sutcliffe, or the Kling-Gupta efficiencies. Much ecological research has shown that aquatic species respond to very specific characteristics of the hydrograph, whether magnitude, frequency, duration, timing, and rate of change of flow events. Since each community in a river may be particularly sensitive to a few very specific streamflow characteristics, alternative hydrological model calibration strategies are needed, 10 focussing on good performance for those specific characteristics. This study investigates the performance of a set of specially developed, bespoke, objective functions made of combinations of specific streamflow characteristics relevant for fish and invertebrate communities. These are compared with the more traditional objective functions on a set of 33 Irish catchments with little human regulation. A split-sample test with a rolling-window procedure is applied to reduce the influence of variations between the calibration/evaluation periods on the conclusions. These bespoke objective functions are shown to be better suited to 15 predict the targetted streamflow characteristics in terms of performance in evaluation; however, traditional objective functions yield more consistent behavioural parameter sets, indicating a trade-off between model performance and model consistency when predicting streamflow characteristics, especially when the number of target streamflow characteristics are low.