Abstract: In this paper, GPU-based FDTD algorithm is applied to study the electromagnetic(EM) scattering from two-dimensional(2-D) target floating on one-dimensional(1-D) rough sea surface with Pierson-Moskowitz(PM) spectrum. The FDTD lattices are truncated by uniaxial perfectly matched layer (UPML), and the finite-difference equations are employed in the whole computation domain for the parallelization convenient to carry out. Also, the parallelism design is limited to the iteration of the near field that is extremely time consuming. To improve the performance, asynchronous transfers is implemented to mask the memory transfers time and the shared memory is used to achieve high memory bandwidth. Using compute unified device architecture(CUDA) technology, significant speedup ratios are achieved, which demonstrates the efficiency of GPU accelerated the FDTD method. The validation of our method is verified by comparing the numerical results with these obtained by sequential FDTD executing on CPU as well as method of moments (MOM), which shows favorable agreements. Furthermore, our parallel implementation is employed to study the influences of the incident angle, the wind speed, the depth of the target on the EM scattering from the target and a sea surface composite model.