Abstract: Forest dynamic simulation plays an important role in studying the long-term development of forest structure and composition. Seeds dispersal, the initial step of forest dynamic process, forms the vital component of forest dynamic simulation. However, it is time-consuming to perform the realistic and spatially-explicit seeds dispersal process. We propose a CUDA-accelerated parallel algorithm, by partitioning the seeds dispersal problem into fine sub-problems and using CUDA programming model to map these sub-problems to parallel processing threads. Meanwhile, we implement the parallel algorithm with CUDA optimization strategies to make full use of the GPU computing capacity. The Experimental results show a good speedup compared to the naive implementation on CPU.