The evaporation duct is a prevalent atmospheric structure in the marine lower troposphere with significant regional nonuniformity. This structure can trap radio waves inside its layer and enable them to propagate over the horizon with less loss. Therefore, accurately predicting the over-the-horizon propagation loss (OHPL) is important for optimizing the performance of radio-electronic systems. Considering the OHPL characteristics in nonuniform evaporation ducts, this article establishes an OHPL prediction model by incorporating prior information into the long short-term memory (LSTM)–transformer structure (IPILT–OHPL). The combination of the LSTM network and transformer is used to construct the LSTM–transformer, aimed at leveraging their respective strengths to extract important features of OHPL effectively. In addition, to improve the prediction accuracy, this article incorporates the evaporation duct height as an environmental prior information into the LSTM–transformer. Finally, this article comprehensively evaluates IPILT–OHPL model performance in different application scenarios. The evaluation results show that the established model not only has high prediction accuracy but also strong generalization ability, which provides a new method for efficiently predicting the OHPL in nonuniform evaporation ducts.

This study introduces RFCFormer, an innovative framework for retrieving evaporation duct refractivity from radar sea clutter, which synergistically integrates a dual-stream Transformer architecture with Gramian Angular Field (GAF) methodology. The framework first converts the input sea clutter into two distinct GAF matrices, effectively transforming the raw clutter into more manageable sparse matrix representations. This transformation enhances the model’s ability to identify underlying patterns within the sea clutter, improving the accuracy of the mapping between clutter characteristics and duct parameters. The dual-stream Transformer architecture then utilizes parallel processing pathways, where coordinated self-attention mechanisms process the GAF matrices in depth, facilitating the integration of multi-scale clutter features through cross-stream feature fusion. Experimental evaluations show that RFCFormer surpasses existing approaches in both inversion accuracy and generalization capability. Furthermore, the model achieves simultaneous retrieval of four duct parameters while maintaining adaptability to diverse sea states, which conclusively demonstrates its practical utility in real-world marine environments.

Given the significant spatial non-uniformity of marine evaporation ducts, accurately predicting the regional distribution of evaporation duct height (EDH) is crucial for ensuring the stable operation of radio systems. While machine-learning-based EDH prediction models have been extensively developed, they fail to provide the EDH distribution over large-scale regions in practical applications. To address this limitation, we have developed a novel spatiotemporal prediction model for EDH that integrates multiple environmental information sources, termed the EDH Spatiotemporal Network (EDH-STNet). This model is based on the Swin-Unet architecture, employing an Encoder–Decoder framework that utilizes consecutive Swin-Transformers. This design effectively captures complex spatial correlations and temporal characteristics. The EDH-STNet model also incorporates nonlinear relationships between various hydrometeorological parameters (HMPs) and EDH. In contrast to existing models, it introduces multiple HMPs to enhance these relationships. By adopting a data-driven approach that integrates these HMPs as prior information, the accuracy and reliability of spatiotemporal predictions are significantly improved. Comprehensive testing and evaluation demonstrate that the EDH-STNet model, which merges an advanced deep learning algorithm with multiple HMPs, yields accurate predictions of EDH for both immediate and future timeframes. This development offers a novel solution to ensure the stable operation of radio systems.

We develop ExPLNet, an explainable deep learning framework for predicting radar wave propagation loss (PL) in non-uniform evaporation ducts, which integrates a hybrid Long Short-Term Memory (LSTM)-Transformer model with SHapley Additive exPlanations (SHAP) technology. ExPLNet employs a data-driven methodology to generate PL predictions by leveraging the LSTM-Transformer framework to discern PL patterns from multi-source driving factors. Subsequently, the SHAP deep explainer module systematically quantifies feature contributions through coalitional game theory, revealing the relative importance of key influencing factors. This architecture simultaneously achieves accurate PL forecasting and physically explainable factor analysis, advancing radio propagation modeling for intelligent communication systems.

To accurately cognize the long-term variations in evaporation duct height (EDH), this article develops a novel environmental knowledge-driven prediction model based on the Informer for long-term EDH (EKDInformer-LTEDH). The model utilizes the Informer’s powerful time series processing abilities to capture long-term dependencies in EDH. Considering the influence of the marine environment on EDH, this article adopts a knowledge-driven method, which incorporates multiple MEPs as prior knowledge inputs into the model. The results of testing the performance show that the EKDInformer-LTEDH model has significant advantages over other models in long-term EDH prediction. Additionally, integrating MEPs into the model as environmental priori knowledge reduces prediction errors and significantly improves its long-term EDH prediction performance.

The study of surface-layer evaporation ducts (EDs) maintains persistent scientific significance due to their critical role in modulating tropospheric propagation. While multiple ED quantification methodologies exist, the inherent complexity of these atmospheric structures radically constrains the observational completeness achievable through singular technologies. To address this limitation, we develop a novel Bayesian fusion framework that systematically integrates multi-source knowledge through probabilistic reasoning to estimate ED from point-to-point propagation loss (PL) observations. This approach reformulates the inverse problem as a maximum likelihood estimation challenge, systematically synthesizing prior ED distributions (derived from diverse knowledge sources) with PL-constrained likelihood functions through a rigorous Bayesian approach. This probabilistic integration enables robust determination of posterior ED distributions while inherently quantifying retrieval uncertainties. Validation experiments using a dedicated dataset confirms the framework’s viability, demonstrating sufficient accuracy to support real-world marine applications.