Strategies to address structural issues in hydrologic models

This thesis explores approaches for improving predictions of conceptual hydrologic models by addressing their structural issues, which typically arise from simplifications that occur when modelling complex hydrologic processes. The thesis documents the research undertaken to develop a generic and defensible modelling strategy. Hydrologic modelling is particularly challenging due to non-stationarity and limitations of data and in the methods used to analyse uncertainties. The thesis proposes two broad categories of methods to address these issues: (1) state uncertainty estimation and (2) model structure modification. A new Bayesian framework for estimating state uncertainty in hydrologic models was developed in Chapter 2, which was demonstrated in a synthetic study and a real-world case in the Bates catchment, Western Australia. This provided the theoretical underpinnings for Chapter 3, which presented a more practical approach named State and Parameter Uncertainty Estimation (SPUE). SPUE was applied to 46 catchments in Australia using the rainfall runoff model, GR4J. SPUE outperformed the classical approach of solely estimating uncertainty in hydrologic parameters based on validation fit to observed streamflow, reliability, and precision metrics. Chapter 4 explored a different approach, where direct modifications to the model structure were made to a river reach model, creating the River Bed/Bank Storage (RBS) model. This had the added advantage of more accurately representing the actual hydrological systems and their dynamic response to changing environmental conditions. In Chapter 5, the RBS model was further improved by combining it with SPUE, which resulted in better reliability and improved probability distributions. Overall, the thesis demonstrated that a comprehensive uncertainty formulation is essential for more accurate predictions of hydrologic models, and the use of both state uncertainty estimation and model structure modification methods can significantly enhance model performance