Explainable Evaporation Duct Prediction: A Hybrid Transformer-BiLSTM Model with SHAP Factor Analysis

Accurate prediction of Evaporation Duct Height (EDH) is vital for marine radar systems. While machine learning improves prediction efficiency, existing models often lack physical interpretability. This paper proposes a hybrid Transformer-BiLSTM deep learning model to predict EDH using five surface meteorological parameters. We incorporate SHapley Additive exPlanations (SHAP) to quantify the impact of each meteorological factor on the prediction. Evaluated on a comprehensive buoy dataset, the proposed model outperforms Transformer, GRU, and LSTM baselines. It achieves high predictive accuracy with a RMSE of 0.471 meter and an R² of 0.9477. SHAP analysis reveals that relative humidity and wind speed are the primary drivers of EDH, whereas atmospheric pressure has a negligible impact. The proposed framework successfully ensures both high-precision EDH estimation and reliable physical interpretability.