Robust hydrological evaluation of stochastic rainfall

• Evaluating rainfall model on streamflow metrics offer a more discerning assessment than rainfall evaluations alone. • Hydrological evaluation maintains robustness regardless of the hydrological model selected. • Rainfall model performance is sensitive to the features of the catchment and streamflow evaluation metrics. As climate conditions evolve, stochastic rainfall models are crucial instruments for assessing hydrological risks, by providing the capability to generate random yet plausible rainfall timeseries. Conventional hydrological risk analysis involves the use of stochastically generated rainfall timeseries as inputs to hydrological models to simulate end-of-system variables that might inform flood-control strategies, water resource management or infrastructure development. To ensure the reliability of these rainfall inputs and subsequent generated streamflow, stochastic rainfall models should be evaluated in terms of their hydrological performance and not merely in terms of rainfall metrics. However, differences within hydrological models could generate different outcomes of a hydrological evaluation despite their similar intent. A hydrological model may either dampen or amplify any potential discrepancies in stochastically generated rainfall when compared to observed rainfall, which poses a challenge for the robustness of interpretation of a hydrological evaluation. Therefore, this paper evaluates the robustness of hydrological assessments by analysing the impact of different hydrological models, catchment characteristics, and evaluation metrics on evaluation outcomes. Stochastically generated timeseries from two rainfall models (a Markov-based model and a latent variable model), as well as observed rainfall timeseries, are inputted to four conceptual rainfall-runoff models (AWBM, IHACRES, GR4J, and Sacramento) to derive daily streamflow timeseries that form the hydrological evaluation. The evaluation is conducted on 25 catchments in Tasmania, Australia. The results show that while the choice of hydrological model has minimal effect on robustness, the evaluation is influenced by variation in catchment features and the selection of evaluation metrics. The results support the use of hydrological models for robustly assessing the performance of stochastically generated rainfall.