Abstract: To investigate the influence of microscopic spatial structures such as pores and fractures inside media on the diffusion law of transient electromagnetic fields, a 3D forward modeling method for transient electromagnetics applicable to rough media is proposed based on the spectral method and the analytical form of the Riemann-Liouville operator in the frequency domain. Taking the half-space model as an example, the correctness of the algorithm is verified by comparing the 1D analytical solution with the 3D numerical solution. Furthermore, three 3D models including layered, columnar and fault fracture zone models are used as cases to deeply explore the potential effects of roughness on the diffusion law of transient electromagnetic fields and water detection. Compared with the traditional model, the rough homogeneous half-space model presents a lower response value at the early stage of transient electromagnetic response and a smaller slope of the attenuation curve at the late stage, and the transient response value exceeds that of the traditional model after a certain time. Under specific parameters, the single-point transient electromagnetic response of the rough medium model attenuates almost linearly in the log-log coordinate system, which is similar to the diffusion law of the homogeneous half-space, making it difficult to identify low-resistivity bodies. In contrast, the transient electromagnetic response of the traditional model undergoes three stages: slow, fast and slow attenuation, with a more direct response to low-resistivity bodies; the greater the roughness, the more significant the impact on the transient response, which will increase the difficulty of anomaly identification or distort the inversion results in water-bearing area exploration. When applying the transient electromagnetic method to hydrogeological surveys, it is necessary to evaluate the underground roughness combined with the formation lithology revealed by drilling and its rock fractal dimension, and comprehensively consider its influence on the transient response. This can provide a reference for water prospectors to identify water-bearing bodies and help improve the accuracy and safety of water detection in coal mines and other scenarios.