Abstract: To meet the demand for accurate high-resolution evaporation duct height forecasting over large areas in tropospheric over-the-horizon (OTH) communication, this study proposes a prediction model that integrates the Gannet Optimization Algorithm (GOA) with a Back-Propagation (BP) neural network, termed the GOA-BP model. First, regional environmental meteorological parameters are obtained using the WRF mesoscale numerical model. Then, the evaporation duct height is predicted by combining the NPS model from the U.S. Naval Postgraduate School, creating a dataset containing rich environmental information and forecasted duct height values. Next, the GOA algorithm is introduced to optimize the initial parameters of the BP neural network, significantly enhancing the model's global search ability and convergence speed, thus avoiding the drawback of traditional BP neural networks, which are prone to getting trapped in local optima. Finally, the GOA-BP model is obtained through training. The results show that the coefficient of determination (R2) reaches 97.21%, with an average verification error (RMSE) of 2.24 meters, demonstrating that the GOA-BP model can more accurately and effectively predict the evaporation duct height. This can provide valuable reference for planning and application of ultra-shortwave/microwave beyond-line-of-sight radar and radio communication systems.