Hydrological and water resources modelling under uncertainty and climate change : an application to the Limpopo basin, Botswana

The application of water resources planning models to semi-arid or arid areas is expected to be particularly challenging because of the high variability of rainfall and streamflow, highly limited historic observations, sparse rain gauge and flow networks with significant periods of missing rainfall and potential data quality issues. These lead to high uncertainty in rainfall and hydrological models which need to be explicitly represented in model predictions. These uncertainties are expected to increase when considering future predictions associated with the effects of climate change. This has presented an opportunity for this thesis to develop a framework of uncertainty analysis in hydrological and water resources modelling. The framework consists of multi-site continuous time stochastic rainfall model to (1) identify suitable rainfall predictors, (2) infill the missing values in the historic rainfall data, (3) extend the limited rainfall observations, and (4) generate rainfall under climate change scenarios by downscaling global climate models outputs. The stochastically infilled rainfall data allows calibration of a hydrological model under input uncertainty. The rainfall model together with the uncertain hydrological model are then used to generate multiple realisations of reservoir inflows over a 100-year period, first assuming a stationary climate and secondly under a changed climate. This framework is applied to the upper Limpopo basin in Botswana, using 25 years of observed daily rainfall and flow data for model calibration. A Generalised Linear Model was used for the rainfall and a semi-distributed version of the IHACRES model was used for the hydrology. A proposed 382 Mm3 reservoir at the outlet of this basin, which is part of Botswana‘s national water resource strategy, is re-evaluated in light of the extended inflow data and the estimated uncertainty. Analysis within this thesis revealed that the effects of data and model parameter uncertainty on water resources planning models can be high, and thus should not be ignored. The thesis advocates a shift from deterministic to stochastic ways of infilling missing rainfall values, and for consideration of hydrological model uncertainty, climate model and climate scenario uncertainties. Given the high uncertainty in the semi arid case study, priority areas can be identified, which may include acquiring and expanding the gauge networks, building efficient and robust data collection processing and achieving to improve the existing database so as to support and enable quality research.EThOS - Electronic Theses Online ServiceCommonwealth CommissionGBUnited Kingdo