Sub-basins with no observed discharge data available for optimization were assigned parameter values of neighbouring sub-basins. The same applied to the downstream sections (e.g. Zambezi at Tete) with no reliable gauge data. The three optimized parameters that vary between (groups of) sub-basins

include: • Soil storage capacity. The first two parameters affect storage of rainfall in the soil for evapotranspiration and thereby control mean volume of flow. Further, they control how long it takes (up to several months) in the rainy season before the soils are sufficiently wet to enable runoff generation (see also Scipal et al., 2005 and Meier et al., 2011). The third parameter defines the fractions of runoff representing surface flow – which leaves the sub-basin within the same month – and base flow with a delayed response

controlling dry season discharge. Observed discharge data of the period 1961–1990 at 14 gauges were GSI-IX purchase used to automatically calibrate these three parameters of the water balance model with the Shuffled Complex Evolution search algorithm (Duan et al., 1992). As objective function we used a slightly modified version of the KGE-statistic ( Gupta et al., 2009; modified according to Kling et al., 2012): equation(1) KGE′=1−(r−1)2+(β−1)2+(γ−1)2 β=μsμo γ=CVsCVo=σs/μsσo/μowhere KGE′ is the modified version of the KGE-statistic (dimensionless), r is the correlation coefficient Adriamycin datasheet between simulated and observed discharge (dimensionless), β is the bias ratio (dimensionless), γ is the variability ratio (dimensionless), μ is the mean discharge in m3/s, CV is the coefficient of variation (dimensionless), σ is the standard deviation of discharge in m3/s, and the indices s and o represent simulated and observed discharge values, respectively. KGE′, r, β and γ have their optimum at unity. For a full discussion of the KGE-statistic and its advantages over the often used Nash–Sutcliffe Efficiency (NSE, Nash and Sutcliffe, 1970) or the related mean squared error see Gupta

et al. (2009). The KGE-statistic offers interesting diagnostic insights into the Isoconazole model performance because of the decomposition into correlation (r), bias term (β) and variability term (γ). In this paper we use this decomposition of the model performance to report on the evaluation of discharge simulations at five key locations within the Zambezi basin in the calibration period 1961–1990 as well as in the independent evaluation period 1931–1960. Because of the long observed discharge time-series these statistics were also computed at the gauge Kafue Hook Bridge, even though this gauge was not included in the original set-up of the model. In addition to the parameters of the water balance model, there were also a large number of parameters that had to be specified for the water allocation model. These parameters were not calibrated in a classical sense.