DSBGT: Spatio-temporal Prediction of Atmospheric Optical Turbulence for LEO Satellite-to-Ground Laser Communication Networks

Atmospheric optical turbulence significantly degrades the stability of Low Earth Orbit (LEO) satellite-to-ground laser communication (SGLC) links. This study retrieves Fried parameter fields from high-resolution stratified meteorological data to characterize dynamic atmospheric states and proposes the Diffusion Schrödinger Bridge Graph Transformer (DSBGT) for minute-scale spatio-temporal forecasting. The DSBGT architecture utilizes a GWN-based Transformer (GWNformer) as its backbone, integrating multi-scale spatial features and dynamic graphs with a parallel attention mechanism to resolve complex dependencies. To capture the nonstationary nature of the atmosphere, a Diffusion Schrödinger Bridge (DSB) is incorporated to establish a precise nonlinear mapping from retrieved Fried parameter features to future Fried parameter distributions via probabilistic optimal transport. Results demonstrate that DSBGT achieves a Mean Absolute Error (MAE) of 0.22 cm and a Root Mean Square Error (RMSE) of 0.26 cm, significantly outperforming existing benchmarks. These findings underscore the efficacy of the proposed model for operational link scheduling in high-capacity SGLC networks.