Trajectory clustering algorithms analyze the movement trajectory of the target objects to mine the potential movement trend, regularity, and behavioral patterns of the object. Therefore, the trajectory clustering algorithm has a wide range of applications in the fields of traffic flow analysis, logistics and transportation management, and crime analysis. Existing algorithms do not make good use of the temporal attributes of trajectory data, resulting in a long clustering time and low clustering accuracy of spatial-temporal trajectory data. Meanwhile, the density-based clustering algorithms represented by DBSCAN are very sensitive to the clustering parameters. The radius value Eps and the minimal points number MinPts within Eps radius, defined by the user, have a significant impact on the clustering results, and tuning these parameters is difficult. In this paper, we present STRP-DBSCAN, a parallel DBSCAN algorithm based on spatial-temporal random partitioning for clustering trajectory data. It adopts spatial-temporal random partitioning to distribute balanced computation among different computing nodes and reduce the communication overhead of the parallel clustering algorithm, thus improving the execution efficiency of the DBSCAN algorithm. We also present the PER-SAC algorithm, which uses deep reinforcement learning to combine the prioritized experience replay (PER) and the soft actor-critic (SAC) algorithm for autotuning the optimal parameters of DBSCAN. The experimental results show that STRP-DBSCAN effectively reduces the clustering time of spatial-temporal trajectory data by up to 96.2% and 31.2% compared to parallel DBSCAN and the state-of-the-art RP-DBSCAN. The PER-SAC algorithm also outperforms the state-of-the-art DBSCAN parameter tuning algorithms and improves the clustering accuracy by up to 8.8%. At the same time, the proposed algorithm obtains a higher stability of clustering accuracy.

Tropical cyclones (TCs) can cause significant economic damage and loss of life in coastal areas. Therefore, TC prediction has become a crucial topic in current research. In recent years, TC track prediction has progressed considerably, and intensity prediction remains a challenge due to the complex mechanism of TC structure. In this study, we propose a model for short-term intensity prediction based on adaptive weight learning (AWL-Net) for the evolution of the TC’s structure as well as intensity changes, exploring the multidimensional fusion of features including TC morphology, structure, and scale. Furthermore, in addition to using satellite imageries, we construct a dataset that can more comprehensively explore the degree of TC cloud organization and structure evolution. Considering the information difference between multi-source data, a multi-branch structure is constructed and adaptive weight learning (AWL) is designed. In addition, according to the three-dimensional dynamic features of TC, 3D Convolutional Gated Recurrent (3D ConvGRU) is used to achieve feature enhancement, and then 3D Convolutional Neural Network (CNN) is used to capture and learn TC temporal and spatial features. Experiments on a sample of northwest Pacific TCs and official agency TC intensity prediction records are used to validate the effectiveness of our proposed model, and the results show that our model is able to focus well on the spatial and temporal features associated with TC intensity changes, with a root mean square error (RMSE) of 10.62 kt for the TC 24 h intensity forecast.

Arctic sea ice has a significant effect on global climate change, ship navigation, Arctic ecosystems, and human activities. Therefore, it is essential to produce high-resolution sea ice maps that accurately represent the geographical distribution of various sea ice types. Based on deep learning technology, many automatic sea ice classification algorithms have been developed using synthetic aperture radar (SAR) imagery over the last decade. However, sea ice classification faces two vital challenges: (1) it is difficult to distinguish sea ice types with close developmental stages solely from SAR images and (2) an imbalanced sea ice dataset has a significantly negative effect on ice classification model performance. In this article, a spatio-temporal deep learning model—the Dynamic Multi-Layer Perceptron (MLP)—is utilized to classify 10 sea ice types automatically. It consists of a SAR image branch and a spatio-temporal branch, which extracts SAR image features and spatio-temporal distribution characteristics of sea ice, respectively. By projecting similar image features to different positions in the spatio-temporal feature space dynamically, the Dynamic MLP model effectively distinguishes between similar sea ice types. Furthermore, to reduce the impact of data imbalance on model performance, the dynamic curriculum learning (DCL) method is used to train the Dynamic MLP model. Experimental results demonstrate that our proposed method outperforms the long short-term memory (LSTM) network approach in distinguishing between sea ice types with similar developmental stages. Moreover, the DCL training method can also effectively improve model performance in identifying minority ice types.

Trajectory data may suffer from missing locations due to environmental and equipment constraints, which impact the performance of downstream tasks. Trajectory recovery aims to enhance the utility of mobility data by reconstructing high-quality trajectories. However, existing methods face two significant limitations. First, most existing methods rely on individual historical trajectories, making it difficult to exploit global spatial dependencies across the population. Second, current approaches ignore the role of local multi-factor semantics in personalized trajectory recovery. To address the above challenges, we propose a joint trajectory recovery method named GLTrajRec, which integrates global spatial dependencies and local multi-factor semantics. Specifically, global trajectory flow graph modeling is proposed to capture shared mobility patterns from all users’ trajectories, which can provide effective spatial transition constraints even when individual historical data is insufficient. Then, multi-factor local semantics embedding is designed to comprehensively encode multi-dimensional personalized mobility information from multiple aspects, enabling more accurate personalized recovery results. In addition, a Transformer-based encoder–decoder framework is developed to bidirectionally encode the embedded global dependencies and local semantics and to decode multi-class information. Finally, we construct a transition matrix based on the global trajectory flow graph to improve the accuracy of the recovery output. Extensive experiments conducted on two real-world datasets demonstrate that the proposed method achieves mean average precision (MAP) scores of 0.7181 and 0.7345, respectively, representing an improvement of approximately 5% to 7% over traditional trajectory recovery methods.

Fine-scale identification of urban building functions is essential for understanding urban spatial structure, socioeconomic organization, and sustainable urban development. However, large-scale building function mapping remains constrained by reliance on proprietary data, insufficient representation of geographical context, and limited cross-city generalization. To address these challenges, this study proposes a multi-source geographical knowledge fusion framework for fine-scale building function classification using exclusively open-source data. In addition to conventional morphological, POI-based, and spectral features, the framework systematically integrates open-source geographical environmental features to characterize accessibility, infrastructure relationships, ecological surroundings, and environmental conditions at the building level. Beijing is selected as the training area, while Tianjin is used for independent cross-city validation. Ten representative models, including deep learning, ensemble learning, and traditional machine learning methods, are systematically evaluated for classifying six building function types. Results show that the systematic integration of geographical environmental features improves classification performance and interpretability. Deep learning and ensemble models outperform traditional methods, with CNN achieving the highest accuracy of 87.07% in Beijing and 69.83% in Tianjin. Feature contribution analysis further indicates that geographical environmental features play a dominant role in functional discrimination, while POI features provide important socio-semantic information. Overall, this study provides a reproducible open-data framework for urban building function mapping, supporting sustainable urban planning, land-use optimization, infrastructure allocation, and smart city governance.