Hydrological modelling of sergoit basin for the estimation of catchment yield: a comparative study of MIKE 11-NAM and SWAT models

Abstract

Hydrological models provide a way to conceptualize and investigate the relationships between climatic variables and management practices with water resources. The decision on which modelling approach to employ poses a challenge to water resource managers and researchers. Due to their different structures and varying data requirements, models should be tested prior to application. The challenge of lack of current data that hinders quantification of a catchment’s water yield. The objective of this research was to set up, calibrate and validate the lumped conceptual model (MIKE 11-NAM)and the semi-distributed physically based model (SWAT), evaluate both models using statistical and graphical techniques and thereafter apply the models to estimate the current catchment yield of the Sergoit basin and compare the results. Meteorological data was sourced from the Kenya Meteorological Department. Both models were setup, calibrated and validated.Data for the periods 1975 to 1977 and 1982 to 1984 was used for calibration and validation of the NAM model, while 1975 to 1979 and 1981 to 1984 data inclusive of a one year warmup period was used for calibration and validation of the SWAT model respectively. Goodness of fit statistics and graphical methods were used to evaluate model performance. The models were then used to estimate thecatchment yield for the period 2005 to 2009. The overall results from the goodness-of-fit statistics shows differences in performance and overall behaviour of the two models. NAM performed better than SWAT during the calibration period with an NSE, R2, IA, and PBIAS of 0.81, 0.81, 0.94 and 1.80% and 0.69, 0.70, 0.90, and 15.11% respectively. The validation period marked a slight performance drop with NAM and SWAT attaining an NSE, R2, IA, and PBIAS of 0.78, 0.80, 0.95 and 0.65% and 0.65, 0.65, 0.89 and -11.82% respectively. There is a general tendency to underestimate the peak values in both models. On the basis of extreme value analysis, the NAM model performed better than the SWAT model. The general underestimation increases for larger values, indicating poor extrapolation capabilities. The semi-distributed nature of the SWAT model and the large number of model parameters makes it difficult to calibrate and is vulnerable to the quality of data, whereas the lumped nature of the NAM model and low number of model parameters makes it easier to calibrate and gives a better overall performance as most values are averaged throughout the basin. The NAM model estimates the mean annual basin yield for the period 2005 to 2009 at 94.8 MCM/year while the SWAT model gives a lower estimate of 69.6 MCM/year. The study recommends the installation of well distributed weather stations within the Sergoit Basin to improve on the representativeness and the data captured and further study is recommended to incorporate effects of land use and climate change as these have an impact on the catchment’s water yield