Semi-distributed parameter optimization and uncertainty assessment for large-scale streamflow simulation using global optimization

Value of semi-distributed calibration parameters for large-scale streamflow simulation using the spatially distributed LISFLOOD model is evaluated. In the semi-distributed approach, the spatial detail is defined by the resolution of the discharge stations. All the calibration parameters of the LISFLOOD model are well identifiable using 2 years of measured daily discharges. For all strategies, probabilistic streamflow predictions are obtained using 7500 parameter combinations sampled from the respective posterior parameter distributions after convergence the Shuffled Complex Evolution Metropolis (SCEM-UA) algorithm. The increase in model uncertainty when going from a semi-distributed to a lumped approach can be attributed to the less correct representation of the spatially varying hydrological characteristics of the system.