Automatic calibration of a lumped Xinanjiang hydrological model by genetic algorithm

Genetic algorithm (GA) is a global search technique and useful in finding near optimal solutions to complex non-linear optimization problems. Therefore, it is applied in this study to calibrate a lumped Xinanjiang hydrological model having 16 parameters. The lumped model consists 4 components: runoff generation, evapotranspiration , runoff subdivision and slope runoff routing. It is “lumped”, for the spatial variability of the physical characteristics of the studied area is not considered. Three types of data were used for Xinanjiang model: rainfall, evapotranspiration and discharge. Two months' data were used for calibration and the following three months for testing. The criterion used to measure the fitness of the calculated against the observed discharges was the Deterministic Coefficient (DC). The calibration processes included first of all defining the feasible domains of the model parameters, and initialize the parameters in the feasible domains. Then the model parameters were iteratively evaluated and updated by applying 3 genetic operations: selection, crossover and mutation, until the maximum amount of generations was reached. The specific methods used for the 3 genetic operations are respectively: normalized geometric selection, arithmetic crossover and non-uniform mutation. The studied area was the Qingjiang river basin in China. The results show that both calibration and testing results are satisfactory: the DC values of which are respectively 0.84 and 0.81. Visual examinations on the simulated hydrographs indicates 2 potential ways of improving present Xinanjiang model: (1) Introducing varying lag in time parameter (L) which varied with the magnitudes of flood peaks; (2) improving the quality of the areally averaged rainfall data, by introducing more advanced spatial interpolation method, to avoid simulating fake basin-wide rainfall events.