Efficiency and Feasibility of an Integrated Algorithm for Distributed Hydrological M odel Calibration

Increasing complexity of distributed hydrological model(DHM)has lowered the efficiency of convergence.In this study,global sensitivity analysis(SA)was introduced by combining multiobjective(MO)optimization for DHM calibration.Latin Hypercube-once at a time(LH-OAT)was adopted in global parameter SA to obtain relative sensitivity of model parameter,which can be categorized into different sensitivity levels.Two comparative study cases were conducted to present the efficiency and feasibility by combining SA with MO(SA-MO).WetSpa model with non-dominated sorting genetic algorithm-Ⅱ(NSGA-Ⅱ)algorithm and EasyDHM model with multi-objective sequential complex evolutionary metropolis-uncertainty analysis(MOSCEM-UA)algorithm were adopted to demonstrate the general feasibility of combining SA in optimization.Results showed that the LH-OAT was globally effective in selecting high sensitivity parameters.It proves that using parameter from high sensitivity groups results in higher convergence efficiency.Study case I showed a better Pareto front distribution and convergence compared with model calibration without SA.Study case II indicated a more efficient convergence of parameters in sequential evolution of MOSCEM-UA under the same iteration.It indicates that SA-MO is feasible and efficient for high dimensional DHM calibration.

Increasing complexity of distributed hydrological model （DHM） has lowered the efficiency of convergence.In this study,global sensitivity analysis （SA） was introduced by combining multiobjective （MO） optimization for DHM calibration.Latin Hypercube-once at a time （LH-OAT） was adopted in global parameter SA to obtain relative sensitivity of model parameter,which can be categorized into different sensitivity levels.Two comparative study cases were conducted to present the efficiency and feasibility by combining SA with MO（SA-MO）.WetSpa model with non-dominated sorting genetic algorithm-Ⅱ （NSGA-Ⅱ） algorithm and EasyDHM model with multi-objective sequential complex evolutionary metropolis-uncertainty analysis （MOSCEM-UA）algorithm were adopted to demonstrate the general feasibility of combining SA in optimization.Results showed that the LH-OAT was globally effective in selecting high sensitivity parameters.It proves that using parameter from high sensitivity groups results in higher convergence efficiency.Study case Ⅰ showed a better Pareto front distribution and convergence compared with model calibration without SA.Study case Ⅱ indicated a more efficient convergence of parameters in sequential evolution of MOSCEM-UA under the same iteration.It indicates that SA-MO is feasible and efficient for high dimensional DHM calibration.