With the advancement of smart city development and precise navigation technologies, pedestrian path planning research has gradually shifted from a single efficiency-oriented paradigm to one driven by multidimensional and personalized demands. The primary objective is to develop path planning models that account for complex urban environments and individual user preferences, thereby providing efficient, flexible, and personalized route recommendations for pedestrians. However, current research still faces key challenges, including insufficient modeling of group heterogeneity, the absence of dynamic preference mechanisms, and limited adaptability to complex scenarios. To address these issues, This paper proposes a pedestrian path planning method based on multi-dimensional preference modeling and adversarial deep reinforcement learning. The proposed method first constructs a “context-aware and dynamically adaptive” multidimensional preference model, which provides dynamic preference weights for pedestrian route selection. These weights guide the reshaping of the reward function in the deep reinforcement learning framework, enabling a multi-objective collaborative optimization mechanism that balances efficiency, safety, and comfort. Subsequently, a preference-enhanced adversarial deep Q-network algorithm (PEA-DQN) is developed, incorporating a dual-experience replay pretraining strategy and an adaptive training mechanism to accelerate model convergence and reduce redundant computation. Experiments conducted in Wuhan under dynamic disturbances within a mixed urban road network validate the performance of the model trained by PEA-DQN. Compared with the DQN algorithm, PEA-DQN improves obstacle-avoidance success rates by more than 50% and reduces average path length by 40.40%. Ablation studies further demonstrate that, relative to Dueling DQN, the incorporation of a multi-objective reward function improves path quality by 100.4%, while the adaptive mechanism increases computational efficiency by 40% in dynamic obstacle scenarios. Overall, PEA-DQN significantly outperforms dynamic A^* algorithm and other comparable deep reinforcement learning approaches.

Traffic prediction is a core requirement for building intelligent transportation systems. In complex traffic scenarios, different prediction models exhibit significant performance variations across spatial regions and time periods, making it difficult for any single model to stably adapt to diverse prediction demands. Existing ensemble learning methods enhance prediction stability by leveraging the strengths of multiple models. However, they typically rely on globally fixed or locally optimal ensemble strategies, which overlook the synergistic constraints of global spatio-temporal correlations and spatio-temporal heterogeneity in the ensemble process, limiting the predictive performance and generalization ability. Therefore, this study proposes a spatiotemporal adaptive ensemble learning method with local-global awareness (LGA-EL) for traffic prediction tasks, which optimizes the performance of the base models under different traffic conditions by adaptively adjusting the ensemble parameters. The method first embeds road network topology and traffic state evolution characteristics to jointly capture spatio-temporal information of monitoring stations from both local and global perspectives, thereby collaboratively representing the spatio-temporal correlation and heterogeneity in the ensemble process. Based on the embedding vectors, the method adaptively solves the ensemble parameters for each spatio-temporal location and dynamically weights the output features of the base models to generate the final prediction. Experiments on short-term and long-term prediction tasks for traffic flow and speed demonstrate that the proposed method outperforms six mainstream ensemble prediction methods in terms of both prediction accuracy and computational efficiency. Further interpretability analysis shows that the method can accurately capture performance differences among models under different traffic states, leveraging the strengths of various models through adaptive ensemble weights, significantly enhancing the performance and robustness of ensemble learning in traffic prediction tasks.

With the acceleration of urbanization and digitalization, the importance of urban resource allocation, commercial facility layout, and emergency management has become increasingly prominent. While traditional methods have achieved positive results in static scenarios, they reveal significant limitations when dealing with high-dimensional and dynamic geospatial data. In recent years, artificial intelligence technologies, particularly deep reinforcement learning (DRL) methods, have offered novel approaches to optimizing urban facility allocation. By continuously learning through interaction with its environment, DRL can handle complex sequential decision-making problems. Supported by geographic big data, it demonstrates strong adaptability and intelligent advantages, effectively addressing the shortcomings of traditional methods. However, its application still faces challenges such as high model training costs and strong dependence on data quality. Future research should focus on optimizing DRL algorithm structures, enhancing model training efficiency, strengthening generalization capabilities across diverse scenarios, and exploring the integration of DRL with other intelligent optimization methods. This will further expand the depth and breadth of its application in urban facility allocation optimization.

Urban air mobility offers new options for residents' travel. The location and spatial layout of key infrastructure such as vertiports directly affect the travel patterns and behavioral characteristics of future urban residents. Focusing on long-distance commuting scenarios of urban residents, this paper analyzes the optimal layout of vertiports based on real commuting demand data. A bi-level programming model is proposed to model the interaction mechanism between the location of vertiports and residents' travel choices, aiming to find the locations that can minimize the one-way commuting time of commuters and improve the operational efficiency of key ground transportation routes during peak hours. At the upper level, the location problem is formulated as a multi-objective optimization model, with the combination of candidate sites as the decision variable, and solved using the multi-objective genetic algorithm; at the lower level, the activities and travel chains of typical commuters are modeled through multi-agent traffic simulation to evaluate the comprehensive impact of the layout schemes on commuting efficiency. Taking the long-distance commuting scenarios in Nanjing as a case study, the experimental results show that the proposed method can effectively improve the efficiency of long-distance commuting, reducing the aggregate commuting time by approximately 5%. This paper provides theoretical basis and support for the planning and management of multi-modal transportation in future cities.

Stay point detection serves as a critical preliminary step in trajectory data mining, providing essential support for related research such as point-of-interest mining and movement pattern classification. However, traditional detection methods are typically designed for dense trajectories, such as GPS data, and struggle to address the challenges posed by sparse fixed-point trajectories like those from traffic checkpoints or mobile signaling data. These challenges include insufficient feature extraction and threshold estimation biases due to uneven data density and complex distribution patterns. To tackle these issues, we propose a dual-threshold stay point detection method based on adaptive expanded density peak clustering (AE-DPC) for sparse fixed-point trajectories. First, global thresholds are derived from overall data characteristics to preliminarily identify candidate stay points. Then, local thresholds are further refined based on the clustering results of AE-DPC, which constructs initial clusters by considering neighborhood relationships and improved density peaks, followed by cluster expansion and merging to enhance clustering performance. Finally, the integration of global and local thresholds enables precise stay point detection. Experiments were conducted on both open-source synthetic datasets and real-world sparse fixed-point trajectory datasets to evaluate AE-DPC and the dual-threshold method, respectively. The results demonstrate that AE-DPC significantly outperforms comparative algorithms (such as DBSCAN, HDBSCAN, and SNN-DPC) in terms of the adjusted rand index (ARI) and adjusted mutual information (AMI). Moreover, the dual-threshold method leveraging AE-DPC-based local thresholds shows superior performance in real-world stay point detection tasks, achieving improvements in precision of 14.10% and 9.88% compared to the local threshold method based on HDBSCAN and the dynamic threshold approach, respectively.

Jointly developed by the National Aeronautics and Space Administration (NASA) and the Indian Space Research Organization (ISRO), the NASA-ISRO synthetic aperture radar (NISAR) mission was successfully launched on 30 July 2025. With the advantage of wide-swath coverage, high spatial resolution, left-looking imaging geometry, and dual-frequency (L-and S-band), NISAR is expected to substantially enhance monitoring capabilities for the solid Earth, cryosphere, and ecosystems. Furthermore, the mission's open-access SAR/InSAR processing toolchain and data products will substantially lower the technical barriers for non-specialist users. In this paper, we present a comprehensive review of the characteristics of NISAR and its contributions to global surface deformation monitoring. First, this paper analyzes the design concept of the NISAR satellite system. Then, unique observational strengths of the mission are discussed in detail, especially in deformation monitoring. Finally, we analyze the differences in surface deformation characteristics associated with various types of geological hazards. And the potential improvements in monitoring deformation induced by different disasters using NISAR products are summarized.

To address the dependency of high-precision underwater positioning models on ray-tracing algorithms, this paper proposes a static acoustic delay compensation positioning model. Inspired by the GNSS atmospheric static delay model, a conventional sound speed (defined as the speed of sound in seawater) is established. A mapping function model between the zenith acoustic delay and the slant-range acoustic delay, dependent on the elevation angle, is constructed alongside its nonlinear least-squares parameter estimation algorithm. This approach transforms the sound speed profile into static acoustic delay and mapping function information products, enabling static acoustic delay compensation directly on the slant-range observations in the acoustic ranging positioning model. The proposed second-order continued fraction based on the sine of the elevation angle can accurately approximate the acoustic delay mapping function. When combined with dynamic acoustic delay compensation, it not only achieves centimeter-level precision for seafloor geodetic positioning but also improves computational efficiency by approximately 80% compared to traditional methods and by over 50% compared to existing fast ray-tracing algorithms. The static acoustic delay compensation positioning model proposed in this study provides a valuable reference for theoretical research in underwater acoustic ranging and positioning and offers a novel solution for constructing large-scale underwater acoustic navigation and positioning service information products.

The acquisition of high-precision attitude data is a crucial research aspect in the processing of gravity satellite payload raw data. Each GRACE-FO satellite is equipped with three star cameras and a gyroscope to measure the satellite's attitude data. The fusion of these two types of data is an important approach to obtain high-precision attitude data. Firstly, we establish a calibration algorithm for the star cameras installation matrix according to the anisotropic noise of star camera measurements. Secondly, a low-frequency error processing strategy for star cameras is developed, utilizing a moving average method to suppress low-frequency errors. Then, an indirect Kalman filtering algorithm is proposed, with the quaternion correction based on the Gibbs vector and gyroscope bias correction as state vectors. Finally, the GRACE-FO Level-1A data is used for validation and analysis. The computational results show that using different star cameras as references to calibrate the installation matrix of the other star cameras yields only minor differences in the derived satellite attitude and the low-frequency errors at the 1 CPR frequency band in the attitude data are significantly reduced. The established low-frequency error processing strategy for star cameras can effectively suppress low-frequency errors within the 30 CPR frequency band. For the fusion of star cameras and gyroscope data, compared to the JPL results, the attitude data calculated in this paper reduces noise in the 0.000 5~0.1 Hz frequency band, demonstrating that the fusion fully exploits the instruments'maximum performance. Regarding the impact on time-variable gravity field recovery, both the degree variances of the recovered gravity fields and the equivalent water height differences show that the accuracies of the time-variable gravity models derived from different attitude datasets are essentially identical. At the current level, further improvements in attitude data accuracy have only a minor effect on enhancing the precision of time-variable gravity field models.

Real-time kinematic (RTK) has become a common positioning technique for urban vehicle terminal integrity monitoring algorithms due to its high accuracy and fast convergence. However, due to the differences in application scenarios, terminal equipment, and positioning techniques, the integrity monitoring algorithms in the aviation domain cannot be directly applied to urban vehicle-mounted GNSS terminals. Therefore, this paper optimizes the RTK integrity monitoring algorithm based on the solution separation method, focusing on fault detection and protection level calculation, and verifies its application effectiveness. First, the similarity of the corresponding detection statistics across directions is used to construct a new fault detection statistic, enhancing the algorithm's ability to detect abnormal observations. Then, the nominal bias is extracted from the observation residuals using a two-step Gaussian overbounding method, and the impact of the bias term on the protection level is analyzed to enhance the overbounding effect on positioning errors. Finally, the effectiveness and applicability of the optimized algorithm are validated using urban vehicle datasets. For fault detection, the solution separation method with a two-direction combination is better than that of the single direction, with an average improvement of 21.7%. Compared to the residual detection method, the solution separation method improves the positioning accuracy in the east, north and up directions by 8.3%, 11.5% and 34.2%, respectively, and the correct fix rate of ambiguity is improved by 18.9%. In the protection level calculation, the protection levels of the floating and the fixed solution after considering the nominal bias are 7.273 m and 0.218 m, respectively. The float solution is more significantly affected by the nominal bias, and its missed rate is reduced from 8.8% to 0.3%, reflecting the need to consider the nominal bias term.

Global navigation satellite system reflectometry (GNSS-R) observations, such as surface reflectivity, have complex coupling effects with various parameters including soil moisture, vegetation, and terrain. There is strong spatial heterogeneity at the regional scale, making macroclimate-surface zoning method ineffective in depicting the responses under the varying influencing factors. This study quantitatively analyzes the driving factors of the spatial heterogeneity of GNSS-R surface reflectivity and proposes a zoning framework based on the quantification of the explanatory power (q value) of driving factors. Firstly, the geodetector model is used to quantitatively evaluate the q value of multiple surface driving factors (including topography, vegetation, soil moisture, and land cover) on the spatial differentiation of CYGNSS reflectivity. The results show that surface roughness, digital elevation model (DEM), and land cover type are the three dominant factors affecting the spatial heterogeneity of reflectivity. To examine the superimposition effect of multiple factors, based on the coupling relationship, the roughness factor with the strongest explanatory power is superimposed with soil moisture and vegetation water content respectively, to construct two types of composite zoning models: “roughness+soil moisture” and “roughness+vegetation water content”. Combined with the linear regression method, the fitting effects of different zoning strategies and surface parameter combinations on reflectivity are systematically compared (using weighted R² as the evaluation index). The research results show that: ①The model fitting goodness of the zoning superposition mode (especially the “roughness+vegetation water content” combination) is generally better than that of the single-factor zoning; ②The use of the PN value representing the scattering mechanism can significantly improve the model performance; ③The high weighted R² of the land cover zoning is due to the highprecision fitting in the large sample grassland area rather than the global optimum; ④The model performs well in low-altitude flat areas but is limited in accuracy in high-altitude complex terrain areas. Compared with macroclimate-surface zoning methods, the zoning model constructed based on the quantification of driving factors shows better fitting effects in the GNSS-R surface reflectivity linear regression model. This work provides important methodological support for the precise application of GNSS-R remote sensing data in the retrieval of surface parameters, environmental monitoring, and climate change research. It contributes to advancing the integration of multi-source remote sensing data and the coupled analysis of surface processes toward a more refined and mechanistic direction. Furthermore, it holds positive theoretical and practical significance for enhancing the simulation accuracy of global and regional-scale land surface hydrological, ecological, and climate models.

To address the limitations in existing end-to-end heterogeneous change detection methods, which often neglect modality-specific feature differences and struggle to balance local details with global semantics, this paper introduces a multi-scale heterogeneous change detection network (MHCDNet) featuring cross-modal fusion for heterogeneous remote sensing imagery, which is built upon an encoder-decoder architecture. In the encoding part, a remote sensing foundation model is utilized to construct multi-scale feature representations for multi-modal images. To enhance the textural and structural information, a feature enhancement module (FEM) is introduced, which employs a bottleneck structure with multi-scale convolution design to effectively enhance detail information in different modal features while suppressing noise interference. Furthermore, to effectively account for the differences in multimodal features and achieve efficient fusion of shallow heterogeneous features, a selective cross-modal fusion module (SCFM) is introduced, which learns dynamic weights to enable adaptive fusion of multi-modal features, effectively capturing complementary information between modalities, thereby enhancing the robustness and representational capacity of fused features. Additionally, to effectively model the spatiotemporal context of deep heterogeneous features, a cross-modal cross-attention fusion module (CCFM) is introduced, which leverages both spatial and channel attention mechanisms to capture inter-modal spatiotemporal correlations, significantly enhancing the robustness and reliability of fused features. Finally, an adaptive up-sampling module (AUM) is proposed to achieve alignment and fusion of encoder-decoder features, effectively compensating for the loss of detail information during the decoding process, accumulating the change information, and generating change maps through a change head composed of three convolutional layers and up-sampling modules. To verify the effectiveness of the proposed method, experiments are conducted on two large-scale flood change detection datasets, CAU-Flood and Ombria. The results demonstrate that compared with other methods, MHCDNet achieves the best accuracy metrics on both datasets, while significantly reducing the false alarms and missed detections in change detection, yielding optimal visual results. Furthermore, ablation studies further verify the effectiveness of each module in MHCDNet. Model complexity analysis demonstrates that MHCDNet possesses low computational complexity, achieving the best balance between accuracy and efficiency.

Visual sensors are currently the most common perception sensors for crowdsourced road change detection. However, visual SLAM often suffers from limited accuracy and robustness in such applications. To address it, this study proposes a road pole-like object change detection technology framework supported by visual point cloud quality optimization. First, a visual point cloud optimization method is constructed by fusing semantic constraints, LiDAR point cloud depth, and GNSS global correction, significantly improving trajectory accuracy and point cloud quality. Second, based on the optimized visual point cloud, accurate extraction and localization of road pole-like objects are achieved. Then, a fast-matching strategy based on hash mapping is introduced to achieve robust change detection of pole-like objects across different periods. Finally, the effectiveness of the overall process is verified using two phases of experimental data collected around Tongji University. Experiments show that the proposed method reduces the mean absolute error (MAE) and root mean square error (RMSE) of the trajectory by 68.39% and 65.65%, respectively, while increasing the point cloud density by an average of 57.97%. In the tasks of element localization and change detection, the mean localization error of pole-like objects in a single direction is 2~3 m, and the mean plane error is around 3.5 m. The matching accuracy of pole-like objects reaches 94.8%, and the accuracy and recall rates of change detection for added and removed pole-like objects both reach 100%. The results demonstrate the effectiveness and stability of the proposed method, confirm the application value of multi-source data fusion and optimization technology in road pole-like object change detection, and provide a reliable technical path for high-definition map change detection and intelligent management of road facilities.

In recent years, large language models (LLMs) have made remarkable progress in semantic understanding and task reasoning, providing new technological opportunities for the intelligent design of maps. This study focuses on the color design of administrative maps and proposes a framework, MapColor-Agent, that integrates large language models with a multi-agent collaboration mechanism. The framework employs the LLMs as the semantic reasoning core and enables task decomposition and process coordination through multiple agents. It also combines natural language interaction with a graphical user interface, allowing users to intuitively generate semantically consistent color schemes for maps. The system performance was evaluated using the system usability scale and semi-structured interviews. The results show that MapColor-Agent achieved an overall usability score of 77.9, indicating a good level of usability. Participants generally found the system easy to learn, clear in operation, and natural in interaction, with high levels of interpretability and controllability of results. Difference analysis revealed that participants familiar with map color design scored higher in learning efficiency and perceived complexity, suggesting that background knowledge influences user experience. The interview results further indicated that the framework performs well in semantic understanding and task guidance, though improvements are needed in complex semantic parsing and generation stability. Overall, the findings demonstrate the feasibility of integrating large language models with multi-agent collaboration for map color design and provide a reference for future research on semantic reasoning and multimodal interaction in intelligent cartographic design.