Research papersValidation of a national hydrological model

•Modelled daily river flows across New Zealand were compared with observed flows.•Verification procedures were applied to quantify spatial and temporal performance.•Hypothesis regarding performance for signatures across catchments were tested.•Performance was better in larger and medium-wet catchments.•The procedures provide an objective basis for model verification and assessment.

Nationwide predictions of flow time-series are valuable for development of policies relating to environmental flows, calculating reliability of supply to water users, or assessing risk of floods or droughts. This breadth of model utility is possible because various hydrological signatures can be derived from simulated flow time-series. However, producing national hydrological simulations can be challenging due to strong environmental diversity across catchments and a lack of data available to aid model parameterisation. A comprehensive and consistent suite of test procedures to quantify spatial and temporal patterns in performance across various parts of the hydrograph is described and applied to quantify the performance of an uncalibrated national rainfall-runoff model of New Zealand. Flow time-series observed at 485 gauging stations were used to calculate Nash-Sutcliffe efficiency and percent bias when simulating between-site differences in daily series, between-year differences in annual series, and between-site differences in hydrological signatures. The procedures were used to assess the benefit of applying a correction to the modelled flow duration curve based on an independent statistical analysis. They were used to aid understanding of climatological, hydrological and model-based causes of differences in predictive performance by assessing multiple hypotheses that describe where and when the model was expected to perform best. As the procedures produce quantitative measures of performance, they provide an objective basis for model assessment that could be applied when comparing observed daily flow series with competing simulated flow series from any region-wide or nationwide hydrological model. Model performance varied in space and time with better scores in larger and medium-wet catchments, and in catchments with smaller seasonal variations. Surprisingly, model performance was not sensitive to aquifer fraction or rain gauge density.