Comparison of the DPAT network and a scaling law for the evaluation of Gaussian beam propagation in the atmosphere

We present and compare two methods for evaluating the far-field mean intensity of Gaussian beam propagation through the atmosphere. Using simulation data encompassing multiple scenarios, including turbulence and thermal blooming, we constructed the dual path attention transformer (DPAT) model and compared it with an established scaling law model. We evaluated these models in terms of accuracy, influence of sample size, generalization capability, and computation time, in conditions of turbulence-only and combined effects of turbulence and thermal blooming. The results demonstrate that the DPAT model maintains consistent accuracy in both conditions, with its precision exhibiting enhancement with increased training sample size. Furthermore, the computation time of the scaling law model is the shortest, whereas the DPAT model demonstrates superior generalization capability.