Prediction of Tropospheric Scatter Transmission Loss Based on Meteorological Factors

Troposcatter communication is an important beyond-line-of-sight transmission technology for long-distance terrestrial links in infrastructure-limited environments. However, its transmission loss is strongly affected by meteorological conditions, and the resulting short-term fluctuations make accurate link-state prediction difficult. To address this issue, this paper investigates the prediction of troposcatter transmission loss based on meteorological factors and proposes an attention-enhanced prediction model for short-term forecasting. First, the physical relationship between atmospheric conditions and troposcatter transmission loss is analyzed. On this basis, an hourly dataset is constructed by combining meteorological parameters extracted from the ERA5 reanalysis dataset with the corresponding transmission-loss observations of a troposcatter link. The meteorological field is represented through spatial sampling along the link direction and vertical reconstruction across multiple altitude levels, so that both spatial structure and temporal evolution can be incorporated into the model input. The proposed model utilizes a convolutional neural network (CNN) for the extraction of spatial features from meteorological profiles and then uses a Transformer encoder to capture long-range temporal dependencies within the historical observation window. Simulation results suggest that the proposed model can track the time-varying pattern of transmission loss more accurately than benchmark models, including CNN and CNN-LSTM. The results demonstrate that the attention mechanism can effectively enhance the short-term prediction capability for troposcatter transmission loss and provide useful support for dynamic link assessment and operational decision-making.