In recent years, the propagation characteristics of Gaussian beams in anisotropic non-Kolmogorov maritime atmospheric turbulence have attracted significant attention. However, beam spreading is inevitable upon propagation, leading to the decrease of beam quality of Gaussian beams in far field. The analysis of beam spreading of Gaussian beams, using the power spectrum inversion and the multi-layer phase screen method, is quite time-consuming and challenging to efficiently address the rapid and accurate prediction of beam spreading in practical applications. Although many researches have been focused on the propagation characteristics of Gaussian beams in maritime atmospheric turbulence, the scaling law for beam spreading is still lacking. Especially, compared to terrestrial environments, the maritime environment exhibits distinct characteristics, including higher humidity levels and temperature variations. These environmental differences lead to variations in the propagation characteristics of Gaussian beams. Consequently, establishing scaling law is essential for predicting the beam spreading of Gaussian beams in maritime atmospheric turbulence. To effectively predict and evaluate the impact of maritime atmospheric turbulence on the beam spreading, this study investigates the scaling law for beam spreading of Gaussian beams in maritime atmospheric conditions.A comprehensive physical model has been developed to characterize maritime turbulence using the power spectrum inversion method to generate the turbulence phase screen. Subsequently, numerical simulations have been carried out by the use of the multi-layer phase screen method. Based on the Kolmogorov turbulence spectrum and taking into account the atmospheric coherence length for non-Kolmogorov scenarios, the scaling law for beam spreading of Gaussian beams in maritime atmospheric turbulence has been established. The effects of key parameters, including atmospheric anisotropy, spectral power index, initial beam quality, and wavelength, on the beam spreading of Gaussian beams propagating through maritime turbulence have been investigated. Simulation results, obtained under various parameter settings, yield curves that represent beam spreading and calibration factors under different parameters. The least squares method was then employed to fit these results, leading to the derivation of a beam spreading calibration formula. Furthermore, an error analysis has been performed on all numerical calculation data in relation to the derived calibration formula to validate the effectiveness and applicability of the proposed calibration approach.The results indicate that, within the wavelength range from 1 μm to 4 μm, the initial beam quality from 1 to 3, the refractive index structure from 5×10^-16 to 5×10^-13 m^3-α, the anisotropy factor from 1 to 4, and the spectral power index from 3.1 to 3.8, the error between the predicted beam spreading and the numerical calculation results remains within 15%. Upon these results, the calibration formula exhibits a maximum error of approximately 14.4%, with an average error of around 2.82%. Particularly, in cases where β₀>1 or ζ>1, the maximum error is reduced to below 10%. This reduction arises from the amplified impact of the initial beam quality on beam spreading as initial quality of the laser beam deteriorates. Furthermore, as anisotropy factor increases, the influence of turbulence on beam spreading decreases, augmenting the relative impact of the initial beam quality and consequently reducing the error of the scaling law.The scaling law was proposed to predict the beam spreading of Gaussian beam in anisotropic non-Kolmogorov maritime atmospheric turbulence rapidly and accurately. Firstly, the variations of beam spreading with laser and turbulence parameters, including atmospheric anisotropy, spectral power index, initial beam quality, and wavelength, were analyzed in detail. Then, the simulation results yield curves that represent beam spreading and calibration factors under different parameters. Subsequently, the least squares method was employed to fit these results, leading to the derivation of the scaling law. Furthermore, the errors between the results predicted by the scaling formula and the simulation results were analyzed. The results show that, within the specified parameter range, the errors are less than 14.4%, with the average error below 2.82%.