The impact of errors in hydrological predictions on water resource system performance

This is peer-reviewed extended abstract presented at the Engineers Australia, HWRS2023: Hydrology & Water Resources Symposium 2023: Streamflow predictions from hydrological models are widely used in long-term water resource planning. These predictions have errors associated with inadequacies in model structure, input/output data, and parameter misspecification. While these “hydrological errors” are typically neglected when designing water resource systems, they can be accounted for by modelling “predictive errors”. This study evaluates the impact of neglecting and modelling hydrological errors on water resource system performance. A case study is used to explore the impact of hydrological errors on the yield and risk of water resource systems with multiple storage capacities and drought risk criteria. Commonly used tools are applied, including the GR4J hydrological model, a residual error model applied at annual time step, a stochastic rainfall model, and a simplified reservoir storage model to estimate yield and risk. The case study demonstrates that hydrological errors have potentially large impacts on yield; in one particular catchment, neglecting hydrological errors results in yield being over-estimated by as much as 35% for systems with annual design risk of 10% (failure 1 in 10 years), and up to 55% for annual design risk of 0.1% (failure 1 in 1,000 years). This over-estimation of yield can result in the actual risk of running out of water being up to around30 times greater than the design risk. These large errors in yield and risk are particularly startling given the predictions from GR4J appear to have reasonable performance in terms of standard streamflow metrics (e.g., NSE of around0.7). These outcomes highlight that the common practise of neglecting hydrological predictive errors in risk-based water resource design is likely to lead to erroneous design risk estimates and over-confident design.