In view of the frequent occurrence of floods due to climate change,and the fact that a large calculation domain,with complex land types,is required for solving the problem of the flood simulations,this paper proposes an optimized non-uniform grid model combined with a high-resolution model based on the graphics processing unit (GPU) acceleration to simulate the surface water flow process.For the grid division,the topographic gradient change is taken as the control variable and different optimization criteria are designed according to different land types.In the numerical model,the Godunov-type method is adopted for the spatial discretization,the TVD-MUSUL and Runge-Kutta methods are used to improve the model\'s spatial and temporal calculation accuracies,and the simulation time is reduced by leveraging the GPU acceleration.The model is applied to ideal and actual case studies.The results show that the numerical model based on a non-uniform grid enjoys a good stability.In the simulation of the urban inundation,approximately 40％-50％ of the urban average topographic gradient change to be covered is taken as the threshold for the non-uniform grid division,and the calculation efficiency and accuracy can be optimized.In this case,the calculation efficiency of the non-uniform grid based on the optimized parameters is 2-3 times of that of the uniform grid,and the approach can be adopted for the actual flood simulation in large-scale areas.