Federated Learning (FL) has been extensively applied in urban environmental prediction tasks of mobile edge computing by training a global machine learning model without data sharing. However, the training of FL faces the challenges such as the poor generalization capability of a single global model over heterogeneous data and hefty communication overhead caused by the frequent model exchange between massive edge servers and remote cloud servers. To address such issues, we propose HPFL-CN, a novel communication-efficient Hierarchical Personalized Federated edge Learning framework with Complex Network clustering. HPFL-CN introduces Privacy-preserving Feature Clustering (PFC) to extract privacy-preserving low-dimensional feature representations of each edge server via mapping the environmental data to different complex network domains for clustering similar edge servers accurately. Based on the clustering results of PFC, an edge-mediator-cloud hierarchical architecture is proposed to realize personalization at the cluster level by Effective Hierarchical Scheduling (EHS). Furthermore, to adapt to dynamic scenarios of new edge servers joining and streaming data generation, we further extend HPFL-CN to Adaptive personalized federated learning with dynamic grouping (Ada-HPFL-CN), which can flexibly re-group edge servers and adjust mixed model weights and the model aggregation frequency adaptively. Our extensive experiments on real-world datasets demonstrate the efficacy of our framework, which outperforms state-of-the-art FL methods regarding personalization and communication efficiency performance.

Satellite data often exhibit spatially discontinuities, and the unstructured nature of these data make it difficult for researchers to use structured spatial analysis methods such as convolutional neural networks (CNNs) in studies of sea surface parameters retrieval. To address these issues, we propose a dynamic graph physical residual network (DGPRN) for retrieving sea surface vector winds and sea surface temperatures via data from the microwave radiometer (MWRI) onboard the Fengyun-3 satellite. This model consists of a position encoder, a dynamic adjacency matrix generator, a graph convolution module, and a physical residual module. The results indicate that the DGPRN model performs better than traditional retrieval algorithms do under complex wind conditions. The DGPRN incorporates the radiative transfer equation as an inductive bias to convert relative wind direction information into brightness temperature residuals, thereby increasing the retrieval accuracy. Moreover, we capture the spatial distribution characteristics of sea surface wind speed and direction and retrieve continuous wind fields in areas with complex topography and cyclonic circulation. Extensive experiments indicate that the DGPRN model outperforms other competing methods in both retrieval accuracy and generalization performance. Specifically, DGPRN model achieves root mean square error (RMSE) values of 0.87 m/s for the wind speed, 14.80° for the wind direction, and 0.81 K for the sea surface temperature, with correlation coefficients exceeding 0.93.