Excerpt
Abstract
In many drainage basins around the world, no runoff data are available. This situation is more pronounced in developing countries, where many river basins lack runoff data and so are ungauged. In West Africa, the general situation of insufficient data is exacerbated by the decline of the measuring network observed since the late eighties. With the aim of predicting hydrological variables in ungauged basins, regionalization methods are usually used. The main objective of this study is to make prediction of streamflow hydrographs on the Bani basin to improve the knowledge of water resources availability. Firstly, the hydrological model SWAT was calibrated on many gauged catchments on the period 1983-1992 and validated on 1993-1997 using the Generalized Likelihood Uncertainty Estimation (GLUE) approach. Secondly, the studied catchments were categorized into clusters of similar physioclimatic characteristics by the means of a multivariate statistical analysis. And finally, in each case, the entire set of optimized model parameters was transferred from gauged to ungauged catchments based on physical similarity and spatial proximity approaches, and the discharge hydrograph was simulated on the target catchment for the period 1983-1997. Results indicated that the daily model performs as good as the monthly model at catchment and subcatchment scales, despite the limited data condition underlying the hydrological modeling. On a daily basis, a good performance of the SWAT model at the whole catchment scale has been obtained as depicted by a Nash-Sutcliffe Efficiency (NSE) of 0.76 and 0.84 and a coefficient of determination R2 of 0.79 and 0.87 for calibration and validation periods, respectively. In addition, the PBIAIS values were smaller than 25% in magnitude for both calibration and validation periods, reflecting a reasonable prediction of the water balance. Predictive uncertainties were acceptable despite being larger during high and low flows conditions. The 61% of observed data (P-factor = 0.61) were enclosed within a small uncertainty band (R-factor = 0.91). A better model performance and smaller predictive uncertainties have been achieved with monthly calibration compared to daily calibration, except for the water balance, where errors have slightly increased. A total of 12 model parameters were identified that best simulate the observed discharges. The test catchments principally aggregated into three groups: a group of northerly flat and semi-arid catchments, another group of southerly hilly and humid catchments, and a third group located in the center of the study basin, inside which, none of the descriptors seems to exert a strong control on the similarity. Overall, regionalization yielded v satisfactory to very good results at many target catchments. The best efficiencies have been recorded in the arid zone and at the whole catchment outlet with NSE values ranging between 0.56 and 0.83. However, predictive uncertainty showed an increase with aridity. A good mutual hydrological similarity was found in a set of catchments belonging to different physical regions, and between which, spatial proximity was found to be a better surrogate of this similarity. The knowledge of water resources availability where it is not measured is very useful for many applications such as water allocation for consumption and irrigation especially in West Africa frequently facing water deficit and food insecurity due to the impacts of a changing climate. Results also contribute to the advance in understanding of hydrological processes of a newly investigated area in the field of Prediction in Ungauged Basins (PUB), and constitute a first step toward further investigations on catchment functioning on which depends largely the success of any regionalization of hydrological information.