Research Doctorate - Doctor of Philosophy (PhD) / The purpose of rainfall-runoff modelling, like all environmental modelling is to generate simulations that accurately mimic those encountered in the system being modelled. Once this is achieved, the model may then be used to study the catchment response under conditions that have not previously been observed, such as the determination of extreme flood levels. The complex behaviour of the processes involved in the generation of streamflow mean that to achieve a usable model, simplifications must be made. This inevitably leads to the introduction of model error into the simulations, as these simplifications cannot reproduce the level of response variation encountered in a natural system. As a consequence, a model that performs well at some times may be inappropriate at other times. The MultiModel approach is an alternative method of rainfall-runoff modelling that uses numerous alternative process descriptions to generate a suite of unique rainfall runoff models. These models are calibrated and applied to allow for simulation responses that incorporate not only parameter variability but model structure variability. It is shown that the application of the MultiModel method to four test catchments produced simulated confidence limits that are much more likely to contain flood peaks that are beyond the range encountered during the calibration process than using a single model. This is due to the wider confidence limits generated as a result of the greater structure variability available to the MultiModel. The wider confidence limits are therefore a better reflection of our true understanding of the system being modelled. The prediction of ungauged basins presents an additional challenge to rainfallrunoff modelling. Most methods involve some form of regionalization of model parameters. These approaches are very limited in that they are restricted by model selection and application range. Two unique methods for the prediction of ungauged basins are presented that overcome these restrictions. The first attempts to condition a rainfall-runoff model using uncertain criteria, normally used as a supplement to more common calibration procedures. These criteria include estimates of flood peaks, baseflow, recession and saturated area. It is shown that combinations of these criteria provide a powerful means of constraining the parameter space and reducing the simulation uncertainty. The second approach to model conditioning for ungauged basins uses an alternative method of regionalization that focuses on the estimation of flow characteristics rather than model parameter values. Strong relationships between flow characteristics (such as runoff coefficients, flow duration curves and coefficient of variation) and catchment conditions (such as area, mean annual rainfall and evaporation) are identified for catchments across Australia. Using the estimated ranges of these flow characteristics as assessment criteria, a rainfall-runoff model is successfully conditioned to adequately reproduce the streamflow response of the four test catchments. In particular it is shown that the use of numerous characteristics in tandem further improves the conditioning for the test catchments.

Research Doctorate - Doctor of Philosophy (PhD) / The purpose of rainfall-runoff modelling, like all environmental modelling is to generate simulations that accurately mimic those encountered in the system being modelled. Once this is achieved, the model may then be used to study the catchment response under conditions that have not previously been observed, such as the determination of extreme flood levels. The complex behaviour of the processes involved in the generation of streamflow mean that to achieve a usable model, simplifications must be made. This inevitably leads to the introduction of model error into the simulations, as these simplifications cannot reproduce the level of response variation encountered in a natural system. As a consequence, a model that performs well at some times may be inappropriate at other times. The MultiModel approach is an alternative method of rainfall-runoff modelling that uses numerous alternative process descriptions to generate a suite of unique rainfall runoff models. These models are calibrated and applied to allow for simulation responses that incorporate not only parameter variability but model structure variability. It is shown that the application of the MultiModel method to four test catchments produced simulated confidence limits that are much more likely to contain flood peaks that are beyond the range encountered during the calibration process than using a single model. This is due to the wider confidence limits generated as a result of the greater structure variability available to the MultiModel. The wider confidence limits are therefore a better reflection of our true understanding of the system being modelled. The prediction of ungauged basins presents an additional challenge to rainfallrunoff modelling. Most methods involve some form of regionalization of model parameters. These approaches are very limited in that they are restricted by model selection and application range. Two unique methods for the prediction of ungauged basins are presented that overcome these restrictions. The first attempts to condition a rainfall-runoff model using uncertain criteria, normally used as a supplement to more common calibration procedures. These criteria include estimates of flood peaks, baseflow, recession and saturated area. It is shown that combinations of these criteria provide a powerful means of constraining the parameter space and reducing the simulation uncertainty. The second approach to model conditioning for ungauged basins uses an alternative method of regionalization that focuses on the estimation of flow characteristics rather than model parameter values. Strong relationships between flow characteristics (such as runoff coefficients, flow duration curves and coefficient of variation) and catchment conditions (such as area, mean annual rainfall and evaporation) are identified for catchments across Australia. Using the estimated ranges of these flow characteristics as assessment criteria, a rainfall-runoff model is successfully conditioned to adequately reproduce the streamflow response of the four test catchments. In particular it is shown that the use of numerous characteristics in tandem further improves the conditioning for the test catchments.

<p>Streamflow prediction in ungauged basins (PUB) is a process generating streamflow time series at ungauged reaches in a river network. PUB is essential for facilitating various engineering tasks such as managing stormwater, water resources, and water-related environmental impacts. Machine Learning (ML) has emerged as a powerful tool for PUB using its generalization process to capture the streamflow generation processes from hydrological datasets (observations). ML’s generalization process is impacted by two major components: data splitting process of observations and the architecture design. To unveil the potential limitations of ML’s generalization process, this dissertation explores its robustness and associated uncertainty. More precisely, this dissertation has three objectives: (1) analyzing the potential uncertainty caused by the data splitting process for ML modeling, (2) investigating the improvement of ML models’ performance by incorporating hydrological processes within their architectures, and (3) identifying the potential biases in ML’s generalization process regarding the trend and periodicity of streamflow simulations.</p><p>The first objective of this dissertation is to assess the sensitivity and uncertainty caused by the regular data splitting process for ML modeling. The regular data splitting process in ML was initially designed for homogeneous and stationary datasets, but it may not be suitable for hydrological datasets in the context of PUB studies. Hydrological datasets usually consist of data collected from diverse watersheds with distinct streamflow generation regimes influenced by varying meteorological forcing and watershed characteristics. To address the potential inconsistency in the data splitting process, multiple data splitting scenarios are generated using the Monte Carlo method. The scenario with random data splitting results accounts for frequent covariate shift and tends to add uncertainty and biases to ML’s generalization process. The findings in this objective suggest the importance of avoiding the covariate shift during the data splitting process when developing ML models for PUB to enhance the robustness and reliability of ML’s performance.</p><p>The second objective of this dissertation is to investigate the improvement of ML models’ performance brought by Physics-Guided Architecture (PGA), which incorporates ML with the rainfall abstraction process. PGA is a theory-guided machine learning framework integrating conceptual tutors (CTs) with ML models. In this study, CTs correspond to rainfall abstractions estimated by Green-Ampt (GA) and SCS-CN models. Integrating the GA model’s CTs, which involves information on dynamic soil properties, into PGA models leads to better performance than a regular ML model. On the contrary, PGA models integrating the SCS-CN model's CTs yield no significant improvement of ML model’s performance. The results of this objective demonstrate that the ML’s generalization process can be improved by incorporating CTs involving dynamic soil properties.</p><p>The third objective of this dissertation is to explore the limitations of ML’s generalization process in capturing trend and periodicity for streamflow simulations. Trend and periodicity are essential components of streamflow time series, representing the long-term correlations and periodic patterns, respectively. When the ML models generate streamflow simulations, they tend to have relatively strong long-term periodic components, such as yearly and multiyear periodic patterns. In addition, compared to the observed streamflow data, the ML models display relatively weak short-term periodic components, such as daily and weekly periodic patterns. As a result, the ML’s generalization process may struggle to capture the short-term periodic patterns in the streamflow simulations. The biases in ML’s generalization process emphasize the demands for external knowledge to improve the representation of the short-term periodic components in simulating streamflow.</p>