Department of Civil, Surveying and Environmental Engineering. Bibliography: leaves 246-255

In recent years, increased demands have been placed on hydrologists to find the most effective methods of making predictions of hydrologic variables in ungauged basins. A huge part of the southern African region is ungauged and, in gauged basins, the extent to which observed flows represent natural flows is unknown, given unquantified upstream activities. The need to exploit water resources for social and economic development, considered in the light of water scarcity forecasts for the region, makes the reliable quantification of water resources a priority. Contemporary approaches to the problem of hydrological prediction in ungauged basins in the region have relied heavily on calibration against a limited gauged streamflow database and somewhat subjective parameter regionalizations using areas of assumed hydrological similarity. The reliance of these approaches on limited historical records, often of dubious quality, introduces uncertainty in water resources decisions. Thus, it is necessary to develop methods of estimating model parameters that are less reliant on calibration. This thesis addresses the question of whether physical basin properties and the role they play in runoff generation processes can be used directly in the estimation of parameter values of the Pitman monthly rainfall-runoff model. A physically-based approach to estimating the soil moisture accounting and runoff parameters of a conceptual, monthly time-step rainfall-runoff model is proposed. The study investigates the physical meaning of the model parameters, establishes linkages between parameter values and basin physical properties and develops relationships and equations for estimating the parameters taking into account the spatial and temporal scales used in typical model applications. The estimationmethods are then tested in selected gauged basins in southern Africa and the results of model simulations evaluated against historical observed flows. The results of 71 basins chosen from the southern African region suggest that it is possible to directly estimate hydrologically relevant parameters for the Pitman model from physical basin attributes. For South Africa, the statistical and visual fit of the simulations using the revised parameters were at least as good as the current regional sets, albeit the parameter sets being different. In the other countries where no regionalized parameter sets currently exist, simulations were equally good. The availability, within the southern African region, of the appropriate physical basin data and the disparities in the spatial scales and the levels of detail of the data currently available were identified as potential sources of uncertainty. GIS and remote sensing technologies and a widespread use of this revised approach are expected to facilitate access to these data.

This thesis presents a discussion on the study undertaken in the application of the monthly time step Pitman rainfall-runoff model to the Kafue River basin. The study constituted one of the initial steps in the capacity building and expansion of the application of hydrologic models in the southern African region for water resources assessment, one of the core areas of the Southern African FRIEND project (Flow Regimes from International Experimental Network Data). The research process was undertaken in four major stages, each stage working towards achieving the research objectives. The first stage was the preparation of spatial data which included the selection and delineation of sub-catchments and inclusion of spatial features required to run the Pitman model and transferring the spatial data into SPATSIM. The second stage was the preparation of input data, mainly rainfall, streamflow, evaporation, and water abstraction data. This information was then imported into SPATSIM, which was able to assist in the further preparation of data by assessment of the input data quality, linking of observed flows and spatial interpolation of point rainfall data to average catchment rainfall in readiness for running and calibration of the model. The third stage was the running and calibration of the Pitman model. Use was made of both the automatic calibration facility, as well as manual calibration by means of the time series graph display and analysis facility of SPATSIM. Model calibration was used to obtain the best fit and an acceptable correlation between the simulated and the observed flows and to obtain simulation parameter sets for sub-catchments and regions within the Kafue catchment. The fourth stage was the analysis and evaluation of the model results. This included verification of results over different time periods and validation and testing of parameter transfers to other catchments. This stage also included the evaluation of SPATSIM as a tool for applying the model and as a database for the processing and storage of water resources data. The study’s output includes: A comprehensive database of hydrometeorological, physical catchment characteristics, landuse and water abstraction information for the Kafue basin; calibrated Pitman model parameters for the sub-catchments within the Kafue basin; recommendations for future work and data collection programmes for the application of the model. The study has also built capacity by facilitating training and exposure to rainfall-runoff models (specifically the Pitman model) and associated software, SPATSIM. In addition, the dissemination of the results of this study will serve as an effective way of raising awareness on the application of the Pitman model and the use of the SPATSIM software within Zambia and the region. The overall Pitman model results were found to be satisfactory and the calibrated model is able to reproduce the observed spatial and temporal variations in streamflow characteristics in the Kafue River basin.

Distributed Rainfall-Runoff models are gaining widespread acceptance; yet, a fundamental issue that must be addressed by all users of these models is definition of an acceptable level of watershed discretization (geometric model complexity). The level of geometric model complexity is a function of basin and climatic scales as well as the availability of input and verification data. Equilibrium discharge storage is employed to develop a quantitative methodology to define a level of geometric model complexity commensurate with a specified level of model performance. Equilibrium storage ratios are used to define the transition from overland to channel -dominated flow response. The methodology is tested on four subcatchments in the USDA -ARS Walnut Gulch Experimental Watershed in Southeastern Arizona. The catchments cover a range of basins scales of over three orders of magnitude. This enabled a unique assessment of watershed response behavior as a function of basin scale. High quality, distributed, rainfall -runoff data was used to verify the model (KINEROSR). Excellent calibration and verification results provided confidence in subsequent model interpretations regarding watershed response behavior. An average elementary channel support area of roughly 15% of the total basin area is shown to provide a watershed discretization level that maintains model performance for basins ranging in size from 1.5 to 631 hectares. Detailed examination of infiltration, including the role and impacts of incorporating small scale infiltration variability in a distribution sense, into KINEROSR, over a range of soils and climatic scales was also addressed. The impacts of infiltration and channel losses on runoff response increase with increasing watershed scale as the relative influence of storms is diminished in a semiarid environment such as Walnut Gulch. In this semiarid environment, characterized by ephemeral streams, watershed runoff response does not become more linear with increasing watershed scale but appears to become more nonlinear.

Climate change and a growing demand for freshwater resources due to population increases and socio-economic changes will make water a limiting factor (in terms of both quantity and quality) in development. The need for reliable quantitative estimates of water availability cannot be over-emphasised. However, there is frequently a paucity of the data required for this quantification as many basins, especially in the developing world, are inadequately equipped with monitoring networks. Existing networks are also shrinking due mainly to shortages in human and financial resources. Over the past few decades mathematical models have been used to bridge the data gap by generating datasets for use in management and policy making. In southern Africa, the Pitman monthly rainfall-runoff model has enjoyed relatively popular use as a water resources estimation tool. However, it is acknowledged that models are abstractions of reality and the data used to drive them is imperfect, making the model outputs uncertain. While there is acknowledgement of the limitations of modelled data in the southern African region among water practitioners, there has been little effort to explicitly quantify and account for this uncertainty in water resources estimation tools and explore how it affects the decision making process. Uncertainty manifests itself in three major areas of the modelling chain; the input data used to force the model, the parameter estimation process and the model structural errors. A previous study concluded that the parameter estimation process for the Pitman model contributed more to the global uncertainty of the model than other sources. While the literature abounds with uncertainty estimation techniques, many of these are dependent on observations and are therefore unlikely to be easily applicable to the southern African region where there is an acute shortage of such data. This study focuses on two aspects of making hydrologic predictions in ungauged basins. Firstly, the study advocates the development of an a priori parameter estimation process for the Pitman model and secondly, uses indices of hydrological functional behaviour to condition and reduce predictive uncertainty in both gauged and ungauged basins. In this approach all the basins are treated as ungauged, while the historical records in the gauged basins are used to develop regional indices of expected hydrological behaviour and assess the applicability of these methods. Incorporating uncertainty into the hydrologic estimation tools used in southern Africa entails rethinking the way the uncertain results can be used in further analysis and how they will be interpreted by stakeholders. An uncertainty framework is proposed. The framework is made up of a number of components related to the estimation of the prior distribution of the parameters, used to generate output ensembles which are then assessed and constrained using regionalised indices of basin behavioural responses. This is premised on such indices being based on the best available knowledge covering different regions. This framework is flexible enough to be used with any model structure to ensure consistent and comparable results. While the aim is to eventually apply the uncertainty framework in the southern African region, this study reports on the preliminary work on the development and testing of the framework components based on South African basins. This is necessitated by the variations in the availability and quality of the data across the region. Uncertainty in the parameter estimation process was incorporated by assuming uncertainty in the physical and hydro-meteorological data used to directly quantify the parameter. This uncertainty was represented by the range of variability of these basin characteristics and probability distribution functions were developed to account for this uncertainty and propagate it through the estimation process to generate posterior distributions for the parameters. The results show that the framework has a great deal of potential but can still be improved. In general, the estimated uncertain parameters managed to produce hydrologically realistic model outputs capturing the expected regimes across the different hydro-climatic and geo-physical gradients examined. The regional relationships for the three indices developed and tested in this study were in general agreement with existing knowledge and managed to successfully provide a multi-criteria conditioning of the model output ensembles. The feedback loop included in the framework enabled a systematic re-examination of the estimation procedures for both the parameters and the indices when inconsistencies in the results were identified. This improved results. However, there is need to carefully examine the issues and problems that may arise within other basins outside South Africa and develop guidelines for the use of the framework. / iText 1.4.6 (by lowagie.com)

Hydrological models are widely used tools in water resources estimation, but they are simple representations of reality and are frequently based on inadequate input data and uncertainties in parameter values. Data observation networks are expensive to establish and maintain and often beyond the resources of most developing countries. Consequently, measurements are difficult to obtain and observation networks in many countries are shrinking, hence obtaining representative observations in space and time remains a challenge. This study presents some guidelines on the identification, quantification and reduction of sources of uncertainty in water resources estimation in southern Africa, a data scarce region. The analyses are based on example sub-basins drawn from South Africa and the application of the Pitman hydrological model. While it has always been recognised that estimates of water resources availability for the region are subject to possible errors, the quantification of these uncertainties has never been explicitly incorporated into the methods used in the region. The motivation for this study was therefore to contribute to the future development of a revised framework for water resources estimation that does include uncertainty. The focus was on uncertainties associated with climate input data, parameter estimation (and recognizing the uncertainty due model structure deficiencies) methods and water use data. In addition to variance based measures of uncertainty, this study also used a reservoir yield based statistic to evaluate model output uncertainty, which represents an integrated measure of flow regime variations and one that can be more easily understood by water resources managers. Through a sensitivity analysis approach, the results of the individual contribution of each source of uncertainty suggest regional differences and that clear statements about which source of uncertainty is likely to dominate are not generally possible. Parameter sensitivity analysis was used in identifying parameters which are important withinspecific sub-basins and therefore those to focus on in uncertainty analysis. The study used a simple framework for evaluating the combined contribution of uncertainty sources to model outputs that is consistent with the model limitations and data available, and that allows direct quantitative comparison between model outputs obtained by using different sources of information and methods within Spatial and Time Series Information Modelling (SPATSIM) software. The results from combining the sources of uncertainties showed that parameter uncertainty dominates the contribution to model output uncertainty. However, in some parts of the country especially those with complex topography, which tend to experience high rainfall spatial variability, rainfall uncertainty is equally dominant, while the contributions of evaporation and water use data uncertainty are relatively small. While the results of this study are encouraging, the weaknesses of the methods used to quantify uncertainty (especially subjectivity involved in evaluating parameter uncertainty) should not be neglected and require further evaluations. An effort to reduce data and parameter uncertainty shows that this can only be achieved if data access at appropriate scale and quality improves. Perhaps the focus should be on maintaining existing networks and concentrating research efforts on making the most out of the emerging data products derived from remote sensing platforms. While this study presents some initial guidelines for evaluating uncertainty in South Africa, there is need to overcome several constraints which are related to data availability and accuracy, the models used and the capacity or willingness to adopt new methods that incorporate uncertainty. The study has provided a starting point for the development of new approaches to modelling water resources in the region that include uncertain estimates.

The world is undergoing rapid changes and the future is uncertain. The changes are related to modification of the landscape due to human activities, such as large and small scale irrigation, afforestation and changes to the climate system. Understanding and predicting hydrologic change is one of the challenges facing hydrologists today. Part of this understanding can be developed from observed data, however, there often too few observations and those that are available are frequently affected by uncertainties. Hydrological models have become essential tools for understanding historical variations of catchment hydrology and for predicting future possible trends. However, most developing countries are faced with poor spatial distributions of rainfall and evaporation stations that provide the data used to force models, as well as stream flow gauging stations to provide the data for establishing models and for evaluating their success. Hydrological models are faced with a number of challenges which include poor input data (data quality and poorly quantified human activities on observed stream flow data), uncertainties associated with model complexity and structure, the methods used to quantify model parameters, together with the difficulties of understanding hydrological processes at the catchment or subbasin. Within hydrological modelling, there is currently a trend of dealing with equifinality through the evaluation of parameter identifiability and the quantification of uncertainty bands associated with the predictions of the model. Hydrological models should not only focus on reproducing the past behaviour of a basin, but also on evaluating the representativeness of the surface and subsurface model components and their ability to simulate reality for the correct reasons. Part of this modelling process therefore involves quantifying and including all the possible sources of uncertainty. Uncertainty analysis has become the standard approach to most hydrological modelling studies, but has yet to be effectively used in practical water resources assessment. This study applied a hydrological modelling approach for understanding the hydrology of a large Tanzanian drainage basin, the Great Ruaha River that has many areas that are ungauged and where the available data (climate, stream flow and existing water use) are subject to varying degrees of uncertainty. The Great Ruaha River (GRR) is an upstream tributary of the Rufiji River Basin within Tanzania and covers an area of 86 000 km2. The basin is drained by four main tributaries; the Upper Great Ruaha, the Kisigo, the Little Ruaha and the Lukosi. The majority of the runoff is generated from the Chunya escarpment, the Kipengere ranges and the Poroto Mountains. The runoff generated feeds the alluvial and seasonally flooded Usangu plains (including the Ihefu perennial swamp). The majority of the irrigation water use in the basin is located where headwater sub‐basins drain towards the Usangu plains. The overall objective was to establish uncertain but behavioural hydrological models that could be useful for future water resources assessments that are likely to include issues of land use change, changes in patterns of abstraction and water use, as well the possibility of change in future climates.