High-efficiency and high-resolution numerical modeling for two-dimensional infiltration processes, accelerated by a graphics processing unit

Determining the dynamic flow of soil water is an important part of water resource management and evaluation of agricultural production, and the high-efficiency and high-resolution simulation of soil-water dynamics has become the focus of numerical model research. In this study, a numerical model of groundwater flow, accelerated by a graphics processing unit (GPU) based on the compute unified device architecture (CUDA), is developed to investigate the efficiency and behavior of soil infiltration in three cases. Due to the advantages of the multithreaded operation of GPU programs, the model can reduce computation time 163-fold when the number of grids exceeds 250 × 250 under the same conditions, and acceleration is promoted with an increasing number of grids. To ensure the validity of the model, a reasonable maximum change in soil saturation should be used to control the change in time step to prevent model calculation instability. The developed model’s average absolute error and relative error do not exceed 0.94 and 0.31%, respectively, which are small compared with the results from HYDRUS. The GPU-accelerated infiltration model was found to simulate soil-water dynamics under hole irrigation accurately based on field-scale application (R² > 0.9). The mass conservation analysis of two types of soil with alternating distribution shows that the model’s relative error is 1.47%, with an average absolute error of 0.00014 m. The GPU-accelerated infiltration model is, therefore, considered to be an effective tool due to its high efficiency and high resolution when solving two-dimensional water infiltration processes.