Considering the growing population of the earth and the decreasing water resources, the need for reliable and accurate estimation and prediction of streamflow time series is increasing. Due to the climate change and anthropogenic impacts on hydrologic systems, the estimation and prediction of streamflow time series remains a challenge and it is even more difficult for regions where watersheds are ungauged in terms of streamflow. The research presented in this dissertation, was scoped to develop a reliable and accurate methodology for daily streamflow prediction/estimation in ungauged watersheds. The study area in this research encompasses Ontario natural watersheds with various areas spread in different regions. In this research work nonlinear data-driven methods such as Artificial Neural Networks (ANN) and conventional methods such as Inverse Distance Weighted (IDW) as well as their combination are investigated for different steps in streamflow regionalization. As such, Watershed classification prior to regionalization is investigated as an independent step in regionalization. Nonlinear classification techniques such as Nonlinear Principal Component Analysis (NLPCA) and Self-Organizing Maps (SOMs) are investigated for watershed classification and finally a methodology which combines watershed classification, streamflow regionalization and hydrologic model optimization is presented for reliable streamflow prediction in ungauged basins. The results of this research demonstrated that a multi-model approach which combines the results of proposed individual models based on their performance for the gauged similar and close watersheds to the ungauged ones can be a reliable streamflow regionalization model for all watersheds in Ontario. Physical similarity and spatial proximity of watersheds was found to play an important role in similarity between the streamflow time series, hence, it was incorporated in all individual models. It was also shown that watershed classification can significantly improve the results of streamflow regionalization. Investigated nonlinear watershed classification techniques applicable to ungauged watersheds can capture the nonlinearity in watersheds physical and hydrological attributes and classify watersheds homogeneously. It was also found that the combination of watershed classification techniques, regionalization techniques and hydrologic models can impact the results of streamflow regionalization substantially. Furthermore, to evaluate the uncertainty associated with the predictions in ungauged watersheds, an ensemble modelling framework is proposed to generate ensemble predictions based on the proposed regionalization model. / Dissertation / Doctor of Philosophy (PhD)

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.

>Magister Scientiae - MSc / The Kars has a well-defined channel along the 62 km stretch from its sources in the Bredasdorp Mountains. After entering the Agulhas plain which has a very low gradient, this river changes into a triangular shaped wetland. This wetland is 7 km in length with no defined channel running through it. The wetland then discharges into another 7 km long channel that joins the Heuningnes River with its mouth at the Indian Ocean. The presence of the wetland causes frequent flooding which affects cultivated lands and a major highway linking towns on the coastal Cape Agulhas area with the rest of the country. Before this study, there was no monitoring of flows along the Kars River including water levels within the wetland. Consequently, the conditions leading to flooding of the wetlands were unknown. This study is aimed at understanding how the combination of local rainfall, Kars River inflow into the wetland, soil characteristics, and the morphology of the wetland influence flooding/inundation. The study monitored river inflows into and outflows from the wetland. A soil survey was conducted within the wetland using the augering method and an infiltrometer to determine soil type and infiltration rates. This was done to assess the hydrological characteristics of the wetland. Using the collected climate data and river flow data, a conceptual model was developed for predicting downstream outflows and possible flood events on a daily timescale. The results indicated that the Kars wetland comprises soil with high silt and clay content, and low infiltration capacity. The wetland causes flood attenuation and diffuse surface flows. Low infiltration rates result in ponding of local rainfall which can contribute to flooding.

>Magister Scientiae - MSc / Improved knowledge is required on the quantity and source of water resources, particularly evident during periods of drought currently being faced in South Africa. There is inadequate knowledge with regards to the flood attenuating properties of wetlands, particularly evident in the ungauged catchments of Southern Africa. This study aims to improve the knowledge on the contribution of flow from tributaries with headwaters in mountainous regions to low lying areas and the effects of wetlands on river flow patterns. Several river flow monitoring sites were established along the major upper tributaries of the Nuwejaars River at which daily water levels were recorded and bi-weekly discharge measurements were conducted. Weather data was collected using four automatic weather stations and three automatic rain gauges’ setup throughout the catchment. Rainfall data coupled with rating curves and daily discharges were used to assess the flow responses of these tributaries to rainfall events. Additionally, stable isotope analysis and basic water quality analysis was used to determine the major sources of flow within the major tributaries. The rainfall and river flow data collected, coupled with the characterization of the wetland was used to determine the flood attenuation capabilities of the wetland. Lastly, a conceptual model based on a basic water balance was developed to further explain the role of the wetland and its effects on river flows. The results showed a 27-hour lag time in peak flows from the upper tributaries at the inflows of the wetland to the outflow. Two of the upper tributaries had flow throughout the year and were fed by springs in the upper mountainous regions of the catchment and all tributaries were largely reliant on rainfall for peak flows. The temporary storage of flows within the wetland occurred as a result of the Nuwejaars River bursting its banks, filling of pools, or ponds and the Voëlvlei Lake. It was concluded that the wetland increased the travel time and decreased the magnitude of flows of the Nuwejaars River. However, due to the fact that wetlands are interlinked on a catchment scale and have a collective effect on flood attenuation this study may be improved by looking at the wetlands within the catchment holistically.

Reliable hydrological model simulations are the result of numerous complex interactions among hydrological inputs, landscape properties, and initial conditions. Determination of the effects of these factors is one of the main challenges in hydrological modelling. This situation becomes even more difficult in cold regions due to the ungauged nature of subarctic and arctic environments.<p> This research work is an attempt to apply a new approach for modelling snowcover ablation and snowmelt runoff in complex subarctic environments with limited data while retaining integrity in the process representations. The modelling strategy is based on the incorporation of both detailed process understanding and inputs along with information gained from observations of basin-wide streamflow phenomenon; essentially a combination of deductive and inductive approaches. The study was conducted in the Wolf Creek Research Basin, Yukon Territory, using three models, a small-scale physically based hydrological model, a land surface scheme, and a land surface hydrological model. The spatial representation was based on previous research studies and observations, and was accomplished by incorporating landscape units, defined according to topography and vegetation, as the spatial model elements.<p> Comparisons between distributed and aggregated modelling approaches showed that simulations incorporating distributed initial snowcover and corrected solar radiation were able to properly simulate snowcover ablation and snowmelt runoff whereas the aggregated modelling approaches were unable to represent the differential snowmelt rates and complex snowmelt runoff dynamics. Similarly, the inclusion of spatially distributed information in a land surface scheme clearly improved simulations of snowcover ablation. Application of the same modelling approach at a larger scale using the same landscape based parameterisation showed satisfactory results in simulating snowcover ablation and snowmelt runoff with minimal calibration. Verification of this approach in an arctic basin illustrated that landscape based parameters are a feasible regionalisation framework for distributed and physically based models. In summary, the proposed modelling philosophy, based on the combination of an inductive and deductive reasoning, is a suitable strategy for reliable predictions of snowcover ablation and snowmelt runoff in cold regions and complex environments.

Reliable hydrological model simulations are the result of numerous complex interactions among hydrological inputs, landscape properties, and initial conditions. Determination of the effects of these factors is one of the main challenges in hydrological modelling. This situation becomes even more difficult in cold regions due to the ungauged nature of subarctic and arctic environments.<p> This research work is an attempt to apply a new approach for modelling snowcover ablation and snowmelt runoff in complex subarctic environments with limited data while retaining integrity in the process representations. The modelling strategy is based on the incorporation of both detailed process understanding and inputs along with information gained from observations of basin-wide streamflow phenomenon; essentially a combination of deductive and inductive approaches. The study was conducted in the Wolf Creek Research Basin, Yukon Territory, using three models, a small-scale physically based hydrological model, a land surface scheme, and a land surface hydrological model. The spatial representation was based on previous research studies and observations, and was accomplished by incorporating landscape units, defined according to topography and vegetation, as the spatial model elements.<p> Comparisons between distributed and aggregated modelling approaches showed that simulations incorporating distributed initial snowcover and corrected solar radiation were able to properly simulate snowcover ablation and snowmelt runoff whereas the aggregated modelling approaches were unable to represent the differential snowmelt rates and complex snowmelt runoff dynamics. Similarly, the inclusion of spatially distributed information in a land surface scheme clearly improved simulations of snowcover ablation. Application of the same modelling approach at a larger scale using the same landscape based parameterisation showed satisfactory results in simulating snowcover ablation and snowmelt runoff with minimal calibration. Verification of this approach in an arctic basin illustrated that landscape based parameters are a feasible regionalisation framework for distributed and physically based models. In summary, the proposed modelling philosophy, based on the combination of an inductive and deductive reasoning, is a suitable strategy for reliable predictions of snowcover ablation and snowmelt runoff in cold regions and complex environments.

In many parts of the world, developed as well as developing, rivers are not gauged for continuous monitoring. Streamflow prediction at such "ungauged" river catchments requires information transfer from gauged catchments that are perceived to be hydrologically similar to them. Achieving good predictability at ungauged catchments requires an in-depth understanding of the physical and climatic controls on hydrologic similarity among catchments. This dissertation attempts to gain a better understanding of these controls through three independent research studies that use data from catchments across the continental United States. In the first study, I explore whether streamflow similarity among nearby catchments is preserved across flow conditions. Catchments located across four river basins in the northeast United States are analyzed to quantify the spatio-temporal variability in streamflows across flow percentiles. Results show that similarity in catchment stream response is dynamic and highly dependent on flow conditions. Specifically, the coefficient of variation is high at low flow percentiles and gradually reduces for higher flow percentiles. This study concludes that high variability at low flows is controlled by the dominance of high evaporative demand, whereas low variability at high flows is controlled by the dominance of precipitation input relative to evapotranspiration. In the second study, I examine whether streamflow similarity among catchments exists across a wide range of climatic and geographic regions. Data from 756 catchments across the United States is used and daily streamflow at each catchment is simulated using distance-based streamflow interpolation from neighboring catchments. With this approach, high predictability at a catchment indicates that catchments in its vicinity have similar streamflows. Results show that high predictability catchments are mainly confined to the Appalachian Mountains, the Rocky Mountains, and Cascade Mountains in the Pacific Northwest. Low predictability catchments are located mostly in the drier regions of US to the west of Mississippi river. Results suggest that streamflow similarity among nearby catchments is more likely in humid runoff-dominated regions than in dry evapotranspiration-dominated regions. In the third study, my goal is to identify what constitutes the essential information that must be transferred from gauged to ungauged catchments in order to achieve good model predictability. A simple daily time-step rainfall-runoff model is developed and implemented over 756 catchments located across the United States. Results show that the rainfall-runoff model simulates well at catchments in humid low-energy environments, most of which are located in the eastern part of the US, the Rocky Mountains, and to the west of Cascade Mountains. Within these regions, transfer of the parameter characterizing hydrograph recession provides reliable streamflow predictions at ungauged catchments, with a loss in prediction efficiency of less than 10% in most catchments. The results presented in this dissertation show that climate exerts a strong control on hydrologic similarity among catchments. The results further suggest that an understanding of the interaction between climate and topography is essential for quantifying the spatial variability in catchment hydrologic behavior at a regional scale.

In most regions of the world, and particularly in developing countries, the possibility and reliability of hydrologic predictions is severely limited, because conventional measurement networks (e.g. rain and stream gauges) are either nonexistent or sparsely located. This study, therefore, investigates various systems methods and newly available data acquisition techniques to evaluate their potential for improving hydrologic predictions in poorly gaged and ungaged watersheds.Part One of this study explores the utility of satellite-remote-sensing-based rainfall estimates for watershed-scale hydrologic modeling at watersheds in the Southeastern U.S. The results indicate that satellite-based rainfall estimates may contain significant bias which varies with watershed size and location. This bias, of course, then propagates into the hydrologic model simulations. However, model performance in large basins can be significantly improved if short-term streamflow observations are available for model calibration.Part Two of this study deals with the fact that hydrologic predictions in poorly gauged/ungauged watersheds rely strongly on a priori estimates of the model parameters derived from observable watershed characteristics. Two different investigations of the reliability of a priori parameter estimates for the distributed HL-DHMS model were conducted. First, a multi-criteria penalty function framework was formulated to assess the degree of agreement between the information content (about model parameters) contained in the precipitation-streamflow observational data set and that given by the a priori parameter estimates. The calibration includes a novel approach to handling spatially distributed parameters and streamflow measurement errors. The results indicated the existence of a significant trade-off between the ability to maintain reasonable model performance while maintaining the parameters close to their a priori values. The analysis indicates those parameters responsible for this discrepancy so that corrective measures can be devised. Second, a diagnostic approach to model performance assessment was developed based on a hierarchical conceptualization of the major functions of any watershed system. "Signature measures" are proposed that effectively extract the information about various watershed functions contained in the streamflow observations. Manual and automated approaches to the diagnostic model evaluation were explored and were found to be valuable in constraining the range of parameter sets while maintaining conceptual consistency of the model.

Dans les régions méditerranéennes françaises, des épisodes pluvieux diluviens se produisent régulièrement et provoquent des crues très rapides et volumineuses que l'on appelle crues éclair. Elles font fréquemment de nombreuses victimes et peuvent, sur un seul évènement, coûter plus d'un milliard d'euros. Face à cette problématique, les pouvoirs publics mettent en place des parades parmi lesquelles la prévision hydrologique tient une place essentielle.C'est dans ce contexte que le projet BVNE (Bassin Versant Numérique Expérimental) a été initié par le SCHAPI (Service Central d'Hydrométéorologie et d'Appui à la Prévision des Inondations) dans le but d'améliorer la prévision des crues rapides. Ces travaux s'inscrivent dans le cadre de ce projet et ont trois objectifs principaux : réaliser des prévisions sur des bassins capables de ces réactions qu'ils soient correctement jaugés, mal jaugés ou non jaugés.La zone d'étude choisie, le massif des Cévennes, concentre la majorité de ces épisodes hydrométéorologiques intenses en France. Ce mémoire la présente en détails, mettant en avant ses caractéristiques les plus influentes sur l'hydrologie de surface. Au regard de la complexité de la relation entre pluie et débit dans les bassins concernés et de la difficulté éprouvée par les modèles à base physique à fournir des informations précises en mode prédictif sans prévision de pluie, l'utilisation de l'apprentissage statistique par réseaux de neurones s'est imposée dans la recherche d'une solution opérationnelle.C'est ainsi que des modèles à réseaux de neurones ont été synthétisés et appliqués à un bassin de la zone cévenole, dans des contextes bien et mal jaugés. Les bons résultats obtenus ont été le point de départ de la généralisation à 15 bassins de la zone d'étude. A cette fin, une méthode de généralisation est développée à partir du modèle élaboré sur le bassin jaugé et de corrections estimées en fonction des caractéristiques physiques des bassins. Les résultats de l'application de cette méthode sont de bonne qualité et ouvrent la porte à de nombreux axes de recherche pour l'avenir, tout en démontrant encore que l'utilisation de l'apprentissage statistique pour l'hydrologie peut constituer une solution pertinente. / In the French Mediterranean regions, heavy rainfall episodes regularly occur and induce very rapid and voluminous floods called flash floods. They frequently cause fatalities and can cost more than one billion euros during only one event. In order to cope with this issue, the public authorities' implemented countermeasures in which hydrological forecasting plays an essential role.In this contexte, the French Flood Forecasting Service (called SCHAPI for Service Central d'Hydrométéorologie et d'Appui à la Prévision des Inondations) initiated the BVNE (Digital Experimental Basin, for Bassin Versant Numérique Expérimental) project in order to enhance flash flood forecasts. The present work is a part of this project and aim at three main purposes: providing flash flood forecasts on well-gauged basins, poorly gauged basins and ungauged basins.The study area chosen, the Cévennes range, concentrates the major part of these intense hydrometeorological events in France. This dissertation presents it precisely, highlighting its most hydrological-influent characteristics.With regard to the complexity of the rainfall-discharge relation in the focused basins and the difficulty experienced by the physically based models to provide precise information in forecast mode without rainfall forecasts, the use of neural networks statistical learning imposed itself in the research of operational solutions.Thus, the neural networks models were designed and applied to a basin of the Cévennes range, in the well-gauged and poorly gauged contexts. The good results obtained have been the start point of a generalization to 15 basins of the study area.For this purpose, a generalization method was developed from the model created on the gauged basin and from corrections estimated as a function of basin characteristics.The results of this method application are of good quality and open the door to numerous pats of inquiry for the future, while demonstrating again that the use of statistical learning for hydrology can be a relevant solution.