ReNuMa model is designed to estimate nutrient fluxes at the scale of large watershed.The calibration module of ReNuMa uses Generalized Reduced Gradient (GRG) optimal algorithm. In practical application, the optimal module of ReNuMa has many shortcomings. For the purpose of improving the calibration efficiency of ReNuMa and the capacity of finding global optimal result, the paper put forward multi-start GRG, Genetic Algorithm (GA) and combination of both to improve the optimal results. The improved calibration module of ReNuMa shows a better performance on the validity of calibration result and the capacity of finding the global optimal result on the basis of the case in Lianjiang river basin.