Multisite Nonparametric Stochastic Streamflow Generation for the Eastern Nile Basin

Water resource planning in large river basins requires large sets of hydrological sequences to evaluate system performance under conditions that extend beyond the historical record, in turn requiring techniques capable of generating inflows at multiple locations that may be correlated with each other. This study introduces a multisite nonparametric streamflow generation technique that accurately reproduces temporal dependence of hydrological sequences at each location on a range of timescales, including long-memory persistence characterized by the Hurst coefficient. It also reproduces the spatial dependence between each location. The algorithm resamples the observed data at each location to generate randomized sequences, and then rearranges the elements of the sequences using simulated annealing to optimize a fit with statistical moments and temporal and spatial dependence statistics. The method is applied to 18 inflow locations in the Eastern Nile Basin. The simulated annealing method is compared with a widely used multistep procedure using a nearest neighbor resampling (k-NN) followed by spatial and temporal disaggregation. The two methods showed a similar ability to maintain spatial correlations among multiple sites when evaluating annual statistics; however, our proposed method replicates the correlations of monthly flows between sites significantly better than the k-NN method. Furthermore, our method to replicate long-term persistence as evaluated by the Hurst coefficient demonstrates a distinct advantage compared with the k-NN technique. The spatially and temporally flexible method can be used to generate large numbers of flow series for risk-based analyses of water management strategies.