The low-altitude economy aims to make the best use of low-altitude airspace and to shape three-dimensional transportation, and demonstrates distinctive characteristics of “air-ground” cooperation. It is therefore becoming obligatory to digitize the three-dimensional low-altitude airspace, to perform 3D analysis, and to conduct digital-intelligent controlling in the 3D space. Recently, China has promoted its national 3D mapping program and the resulting data product, the 3D realistic geospatial landscape model(3DRGLM), provides reliable fundamental 3D framework data for the low-altitude economy. This paper analyzed the requirements and challenges of the unitization of 3DRGLM in supporting the low-altitude economy. Several fundamental problems were examined and development strategies were proposed, including the digitalization of low-altitude elements, the establishment of low-altitude data spaces, and 3D spatial analysis for supporting the low-altitude economy. Furthermore, the near-future key tasks were identified and examined, such as the planning of the low-altitude skyway, the development of 3D navigation maps and systems, the construction of digital infrastructure for the low-altitude economy, as well as the specialized territorial spatial planning for low-altitude applications. Future efforts should be devoted to emphasizing coordinated planning, enhancing technological innovation, developing typical application scenarios, accelerating pilot demonstrations, and promoting cross-domain integration and cooperation.

[Objective] 3D realistic geospatial landscape model (3DRGLM) China, as an emerging spatiotemporal infrastructure, furnishes a unified three-dimensional spatiotemporal foundation crucial for the advancement of Digital China. Nonetheless, amidst its comprehensive development, substantial obstacles persist concerning the establishment of an accomplishment framework, key technological breakthroughs, and the expansion of application domains. Ningxia pioneered the initiation and completion of province-wide 3DRGLM construction in 2020. This study adopts Ningxia as a case study to methodically consolidate its practical experiences, thereby offering valuable perspectives for the national deployment and widespread adoption of real scene 3D technology.
[Methods] Grounded in the analysis of Ningxia's provincial-wide construction practices, this research unfolds through three dimensions:(1) Categorizing and characterizing the construction outcomes, aligning them with practical necessities to formulate a “three-category, four-tier” achievement system. This encompasses three model types, namely 3DRGLM terrain, rural 3D frameworks, and urban 3DRGLM models, alongside four tiers of model representation precision. It also encapsulates the multi-source nature, usability, timeliness, and sharing attributes of these achievements. (2) Tackling the hurdles and challenges encountered during construction by investigating automated modeling methodologies rooted in digital line graphics and point cloud data. This involves devising seamless integration strategies between terrain scenes and individual structures, and refining modeling techniques for intricate, irregular edifices, thereby augmenting the automation quotient of 3D modeling. (3) Assessing the efficacy of applications and services in vital areas such as major project site selection, urban planning and design, cultural tourism, and natural resource administration, guided by operational requirements and the inherent value of the achievements.
[Results] The results show that Ningxia's 3DRGLM initiative has established multi-tiered accomplishments spanning terrain, rural, and urban contexts, enabling the creation of 3D models with diverse levels of detail through the utilization of multifaceted data sources including aerial and satellite imagery. Key technological explorations have successfully addressed automatic modeling challenges in rural areas, the harmonious integration of terrain scenes with standalone models, and the nuanced modeling of complex architectural forms. Practical implementations validate that these accomplishments facilitate 3D visualization services, enhance operational efficiency, and expand service reach.
[Conclusion] Ningxia has cultivated a distinctive 3DRGLM construction paradigm through early experimentation and pilot projects, charting a pragmatic course of action in achievements systems, core technologies, and application services. This approach has not only been empirically validated but also extensively implemented. The insights garnered from this experience present a viable blueprint for the nationwide proliferation and enhancement of 3DRGLM construction endeavors.

[Objective] Population movement reflects the complex interplay between human activities and geographical dynamics, facilitating the spatial diffusion and concentration of resources, capital, and technology. To deepen the understanding of its characteristics, patterns, and development trends, this study analyzes the structural features and evolutionary dynamics of population flow networks.
[Method] This research utilizes Amap migration big data to examine the spatiotemporal patterns and structural evolution of the population flow network within the Chengdu-Chongqing economic circle over the past five years. By integrating complex network analysis and GIS methods, the study provides a comprehensive investigation.
[Result] The results show that population inflows and outflows in most cities within the Chengdu-Chongqing Economic Circle have generally increased over the past five years, though fluctuations with periodic patterns persist. Chengdu, Chongqing, and Leshan exhibit a net daily population inflow pattern, contrasting with the other thirteen cities. A clear weekly rhythm characterizes population flows: in Chengdu and Chongqing, inflow peaks occur on Sundays and outflow peaks on Saturdays, while in the remaining fourteen cities, inflow peaks fall on Saturdays and outflow peaks on Sundays. Rather than forming a typical “Chengdu-Chongqing” dual-center structure, the population flow evolves into a single-hub network centered on Chengdu. The flow network demonstrates distinct spatial proximity and hierarchy, with larger flows primarily occurring between Chengdu-Chongqing and their satellite cities. Over five years, inter-tier-one-city flows (Chengdu-Chongqing) exhibit an increasingly polarized trend, accounting for 48%, 45%, 49%, 51%, and 56% of the total annually. Smaller flows dominate interactions among non-core cities. Three cohesive subgroups emerge: a western cluster around Chengdu, an eastern cluster around Chongqing, and a southern Sichuan cluster, collectively representing around 70% of intercity mobility. Additionally, population flow networks derived from Amap data versus railway data reveal significant spatial discrepancies, with Amap-based networks highlighting stronger hub-and-spoke dynamics in core cities.
[Conclusion] Through analyzing five-year Amap migration datasets, this study systematically elucidated the spatiotemporal dynamics, distribution, network structure, and evolutionary trends of population flows in the Chengdu-Chongqing economic circle. The findings enhance regional population flow theory, offering insights for regional planning and governance. Additionally, the comparison of Amap and railway data challenges single-source research paradigms, providing a methodological reference for future studies.

[Objective] An increasing number of the public tend to share information via social media posts during disaster events. This is significant for supporting disaster risk reduction decision-making by capturing timely public response information. This study proposes a method for generating public disaster response maps through mining social media text data, demonstrating its application by constructing such maps for major earthquakes, typhoons, and cold waves in China using Weibo data.
[Method] Weibo text data related to earthquakes, typhoons, and cold waves (2018 - 2022) were collected via the Sina Weibo API and subjected to preprocessing steps, including deduplication, filtering, and word segmentation. Based on administrative division data, a comprehensive gazetteer containing national township-level (street) administrative names was developed. By integrating Python's requests and pandas libraries with jieba word segmentation technology, geographic entities in Weibo texts were precisely identified and extracted. For different disaster types, keyword frequency statistics were conducted, followed by spatial analysis.
[Result] The results reveal that: (1) Earthquake responses clustered in Sichuan province, reflecting its status as a seismically active region;(2) Typhoon-affected areas predominantly included coastal provinces (Guangdong, Zhejiang, Fujian, Shandong), with Guangdong exhibiting the highest impact proportion, consistent with typhoon track distributions; (3) Cold waves howed higher prevalence in central/southern China (Jan-Mar) and primarily affected northern regions (Inner Mongolia, Beijing) during Nov-Dec. These findings align well with statistical validation, confirming methodological reliability. The generated maps, visualize spatiotemporal distribution patterns and public response hotspots for different disaster types.
[Conclusion] This approach effectively reveals spatial-temporal distribution patterns of public responses to various disasters across large regions. It provides decision-making support for precise disaster prevention and mitigation, addressing spatial information gaps in traditional surveys. Despite challenges from data bias and quality inconsistencies in social media, the methodology demonstrates significant potential for disaster monitoring. Future work should focus on: (1) Refining geolocation extraction algorithms; (2) Enhancing emotional information analysis; (3) Integrating multi-source data to build a comprehensive disaster monitoring system.

[Objective] Ground substrate surveying, as an emerging field in the survey and monitoring of natural resources, has seen extensive exploratory research and demonstration efforts by numerous domestic experts and scholars in recent years. These endeavors have focused on the classification and surveying of ground substrates. Consequently, a preliminary technical system for ground substrate survey methods has begun to form. Simultaneously, advancements in information technology related to ground substrate surveys have progressed in tandem. Research and exploration by multiple teams have paved the way for the informatization of ground substrate surveys. However, there is currently a lack of an integrated information system to support the digitization and standardization of the entire investigation process. There is an urgent need to establish a digital software system to achieve the digitization of the entire business process and the full lifecycle management of business data. This would support the creation of a unified foundation and a single set of data for surface substrate investigations.
[Method] To address this issue, this paper designs and develops a digital surface substrate investigation system, which includes the following aspects: (1) Based on the analysis of data content and characteristics during the ground substrate survey process, and considering the features of current mature databases, an integrated hybrid database storage model was developed. This model combines “cloud storage + PostgreSQL relational database + MinIO object storage + MongoDB database.” (2) Centered on the concept of “one cloud, one database, one platform, three application terminals, two integrations, and two safeguards,” a five-layer architecture was adopted to construct the overall technical framework of the ground substrate survey system. (3) Drawing from the technical route of “indoor research, field survey, database modeling, and platform services” for ground substrate layer surveys, the digital workflow and system functional composition for ground substrate surveys were designed. (4) Key technologies were developed, such as online services for public base map data and big data storage based on hybrid databases. (5) The digital ground substrate survey system was developed. This system includes: A mobile subsystem for digital ground substrate survey field data collection; A desktop subsystem for digital ground substrate survey data editing and mapping; and A web subsystem for digital ground substrate survey data management and services.
[Result] The digital ground substrate survey system effectively supports the main process informatization of ground substrate surveying, including project management, data collection, data editing, data aggregation, mapping, database construction, and services. It enhances the efficiency of base map data services, field data collection, and data editing and mapping, achieving unified storage and management of data throughout the entire workflow.
[Conclusion] As compliance software for ground substrate survey standards, it lays the foundation for the standardization of data and the unified construction of a national ground substrate survey database.

[Objective] Landslides represent a significant geological hazard in mountainous areas of western China. Particularly, the Qinzhou district of Tianshui city, Gansu province—characterized by steep terrain, complex geological structures, and uneven precipitation distribution—experiences frequent landslides that pose severe risks to ecological stability and infrastructure safety. However, limited long-term, high-precision landslide monitoring data and insufficient analysis of multifactorial triggers hinder effective risk management. This study aims to identify the spatiotemporal evolution characteristics of large-scale landslides in this region and elucidate the dominant environmental drivers and their interactions.
[Method] A total of 50 Sentinel-1A descending orbit SAR images, acquired between June 2021 and June 2024, were processed using the Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) technique to derive surface deformation time series. Seventeen representative large-scale landslides were identified based on deformation features. Subsequently, the Geographical Detector model was employed to quantify the influence of eight environmental variables—elevation, slope, relief amplitude, aspect, precipitation, humidity, seismic activity, and anthropogenic disturbance—on landslide distribution. The spatial explanatory power (q-values) of each factor and their pairwise interactions were systematically analyzed.
[Result] Surface deformation rates in Qinzhou district exhibited significant spatial heterogeneity, ranging from -5.08 mm/a to 13.7 mm/a. Landslides were predominantly concentrated in zones with moderate elevation (1300-1750 m), moderate slope gradients (10°-15°), and annual precipitation between 535-550 mm. Landslides were further classified into three types: high-speed active, moderate-to-low-speed, and stable. Elevation (q = 0.374), precipitation (q = 0.252), and soil moisture (q = 0.216) emerged as the most influential single factors. Notably, multifactor interactions demonstrated strong nonlinear enhancement effects, such as precipitation interacting with topographic relief (q = 1) and elevation interacting with human activity (q = 0.91), substantially improving explanatory power for landslide distribution.
[Conclusion] This study reveals that landslide occurrence in Qinzhou district is governed by the interplay of multiple environmental factors, exhibiting distinct spatial clustering, threshold responses, and nonlinear coupling effects. The integrated approach combining SBAS-InSAR monitoring and the Geographical Detector framework provides a robust methodology for capturing spatiotemporal dynamics and driving mechanisms of landslides. These findings offer scientific guidance for early warning systems, spatial planning, and ecological risk mitigation in mountainous regions with similar geological and climatic conditions.

[Objective] Over the past three decades, the expansion of arable land and the increase in agricultural water consumption in the Xinjiang plain region have led to the over-exploitation of groundwater. This has consequently resulted in a range of issues, including but not limited to declining groundwater levels, reduced runoff, deteriorating water quality, land subsidence, and ecological degradation. The mechanism underlying changes in groundwater storage in the Xinjiang plain is complex. Both domestically and internationally, there have been relatively few studies on large-scale groundwater storage changes in this area. Therefore, it is essential to employ modern observational technology to objectively study the information on groundwater storage changes in this region.
[Method] Investigating the spatial and temporal evolution of groundwater reserves in the Xinjiang plain region is crucial for ensuring the security of regional water resources, food supply, and ecological health. By utilizing GRACE satellite and GLDAS model data, which are well-known for their high accuracy and reliability in hydrological research, we estimated the alteration in groundwater storage in the Xinjiang plain from 2003 to 2022. Subsequently, we analyzed the spatial and temporal evolution of groundwater storage using methods such as the Theil-Sen slope, Mann-Kendall trend test, and grading of groundwater storage changes. This comprehensive analysis provides an understanding of the groundwater dynamics in the region.
[Result] The results indicate that: (1) From 2003 to 2022, the overall groundwater reserves in the Xinjiang plain region exhibit a decreasing trend in four stages, with average monthly change trends of -0.58 mm/mon, -0.26 mm/mon, -0.11 mm/mon, and -1.69 mm/mon for the periods of 2003-2008, 2009-2013, 2014-2018, and 2019-2022, respectively. (2) There are irregular spatial changes; groundwater storage in the pre-mountain plain areas of the northern foothills of the Kunlun Mountain and parts of the southern foothills of the Altai Mountain remains stable or slightly increases (0-2 mm/mon). In contrast, the rest of the region experiences varying degrees of decline, with the Tianshan Mountains region's pre-mountain plain area showing the most significant decrease (-3 mm/mon to -1 mm/mon), and the area affected by this decline increasing annually. The decrease is most pronounced in the plain area in front of the Tianshan Mountains (-3 mm/mon to -1 mm/mon), with the affected area expanding each year. (3) The slowdown in the decline of groundwater reserves during certain periods is related to the implementation of the “Three Red Lines” policy and measures to control groundwater over-exploitation. However, the overall decrease in groundwater reserves is attributed to the combined effects of rising temperatures and increased groundwater extraction for irrigation.
[Conclusion] The research findings provide a scientific basis for monitoring, analyzing, and managing groundwater storage changes in large-scale regions.

[Objective] The rapid development of navigation positioning and IoT technology has generated a large amount of trajectory data, which plays an important role in the field of spatiotemporal data mining. In many application scenarios, it is usually necessary to efficiently query the k-nearest trajectories from large-scale trajectory data under given spatiotemporal constraints, known as trajectory spatiotemporal k-nearest neighbor query. The spatiotemporal features of trajectory data pose challenges to data query design, and existing work still has some shortcomings in handling spatiotemporal k-nearest neighbor queries that coexist with spatiotemporal constraints. Therefore, this paper studies large-scale spatiotemporal k-nearest neighbor distributed queries for trajectory data using the distributed column family database HBase. The aim is to improve query efficiency through advanced indexing strategies and optimized query mechanisms, and provide technical support for practical applications.
[Method] This paper first provides a formal definition of trajectories and their spatiotemporal k-nearest neighbor query, including point query mode and trajectory query mode. It also provides methods for calculating the distance between points and trajectories and the distance between trajectories. By combining XZ2 spatial encoding and XZT temporal encoding, two new spatiotemporal indexing strategies were designed: XZ2T+and TXZ2+. These two indexing strategies solve the problem of rough time partitioning in previous spatiotemporal indexing strategies. Based on this, this paper designs and implements a multi round distributed spatiotemporal k-nearest neighbor query. In each round, the query scope is encoded by a spatiotemporal indexing strategy and divided into different subqueries. Due to the combination of indexing and distributed structure proposed in this paper, a data shard counting system is introduced. After encoding the query range in time and space, we can optimize the effective number of query ranges based on data fragmentation. This solution addresses the issue of expanding query ranges in previous research and improves data scanning efficiency. Additionally, to reduce data processing, this paper also uses HBase's coprocessor mechanism on the storage side to implement pruning strategies that consider the spatiotemporal characteristics of trajectory data.
[Result] We conducted a comparative experiment with the existing index strategy XZ2+T, and the experimental results showed that the two indexing strategies proposed in this paper, XZ2T+ and TXZ2+, did not take longer to construct the dataset than XZ2+T. Most importantly, both of these indexing strategies exhibit better query performance compared to XZ2+T, and can effectively support trajectory spatiotemporal k-nearest neighbor queries. Without optimizing the query window, in the experiment when the parallelism is 24, the efficiency of XZ2T+ is increased by 25.7%, TXZ2+ by 18.9% in point mode, XZ2T+ by 36.4% and TXZ2+ by 40.8% in trajectory mode. This paper effectively accelerates the query process through distributed parallel queries, and optimizes query efficiency by adjusting the number of query scopes based on data sharding and setting spatiotemporal pruning strategies. Further experiments have shown that our scheme exhibits stable and good performance at different k values. When the time window is small, XZ2T+ queries are faster because they do not scan duplicate data. When the time window is large, the advantage of TXZ2+ gradually becomes apparent because it maintains aggregation in the time dimension.
[Conclusion] Overall, this paper has successfully implements distributed trajectory spatiotemporal k-nearest neighbor queries, addressing some of the shortcomings of previous research. It has been validated on large-scale datasets, providing strong technical support for processing large-scale trajectory data queries and laying a solid foundation for future research and application of trajectory data management systems.

[Objective] 3D realistic geospatial landscape model (3DRGLM) serves as a crucial infrastructure and foundational element in advancing the digital transformation of urban governance. It encounters challenges such as high modeling expenses, data integration difficulties, and inadequate spatial analysis capabilities. As the demand for nuanced urban management grows, there is a corresponding need to refine the granularity of management entities supporting this governance. The adoption of three-dimensional information models facilitates an elevation from two-dimensional to three-dimensional perspectives, and from static to dynamic representations, thereby imposing new demands on element coding and the integration of business information. Currently, the platform's spatial analysis and decision-making capacities are insufficient and necessitate further enhancement.
[Methods] This study addresses these challenges by exploring lightweight hierarchical household model reconstruction techniques to swiftly model detailed management units at various levels; establishing encoding rules grounded in planar mesh subdivision to optimize data encoding and correlation efficiency; investigating business information fusion methodologies that combine the nine-intersection model with nearest neighbor matching, thereby creating multi-source data spatial constraints associations; and developing a flexible analytical framework rooted in spatiotemporal knowledge graphs to offer adaptable computational support for grassroots services. A practical application in Xuhui district, Shanghai, involved constructing a live 3D data system encompassing both interior and exterior spaces, facilitating entity mapping and associating business management information pertinent to urban governance. Efficient visualization was attained through server cloud rendering and WebRTC technology, leading to the creation of diverse demonstration scenarios including grassroots community governance, panoramic digital business operations, and transparent firefighting simulations.
[Results] Employing lightweight hierarchical household model automation has enabled rapid construction of refined management units tailored to urban governance needs. The 3D encoding technique, predicated on planar mesh segmentation and prioritizing machine recognition while maintaining human readability, effectively controls encoding length to enhance data handling efficiency. Integrating business information with consideration for spatial constraints has successfully mapped entities and associated business management data within urban governance, accurately depicting residential and commercial structures. Computation powered by spatiotemporal knowledge graphs offers versatile and customizable capabilities for grassroots services, alleviating workload pressures.
[Conclusion] The development of a 3DRGLM platform for urban governance propels the progressive evolution towards visualization, intelligence, precision, and interactivity in city management. These research findings not only bolster urban governance capabilities but also furnish valuable insights and direction for the digital metamorphosis and advancement of urban governance underpinned by real-world 3D scenarios.

[Objective] Epipolar imagery is crucial in the 3D reconstruction process within photogrammetry. Traditional image analysis requires identifying feature points across an entire graphic range, whereas epipolar imagery simplifies this by focusing on corresponding points along the epipolar line of another image after a feature point is detected. This advantage extends to dense matching and 3D scene construction in computer vision. Unlike simple calibration used in computer vision, close-up photogrammetry offers precise absolute 3D coordinates and higher calibration accuracy. However, binocular orientation elements obtained by technicians typically rely on photogrammetric methods. Despite shared theories, differing coordinate system definitions between the two fields complicate direct integration.
[Method] Addressing the incompatibility of traditional epipolar image generation methods from photogrammetry with computer vision, this paper introduces a new method tailored for the Fusiello calibration model in computer vision, using a binocular camera setup for mobile robot navigation and hazard avoidance. The study begins by comparing geometric constraints and parameter expressions of the poles between the Fusiello and close-up photogrammetric models, deriving an adaptation formula based on rotational and translational parameters. Next, it adapts and processes parameters from both models for epipolar image generation. Finally, SIFT feature matching and RANSAC mismatch rejection evaluate calibration accuracy and the effectiveness of the epipolar correction adaptation through reprojection errors of matched feature points.
[Result] Experiments demonstrate that the proposed adaptation method achieves an average matching error below 0.9 pixels, a maximum error under 2 pixels, and a root-mean-square error around 1 pixel. It successfully verifies the accuracy of binocular epipolar correction and relative orientation calibration for the navigation and hazard avoidance camera, outperforming the epipolar rearranging method in speed and accuracy, and marginally surpassing the Bouguet calibration method.
[Conclusion] This methodology paves the way for subsequent stereo mapping tasks involving the navigation and hazard avoidance binocular camera. Future work will explore extended baseline scenarios and significant viewpoint discrepancies to further validate the approach presented herein, providing a robust foundation and reference for ongoing research in this domain.

[Objective] Mainstream semantic segmentation methods, primarily designed for small natural images, face significant challenges when applied to large-scale remote sensing imagery, e.g., 5000×5000 pixels. These challenges include spatial feature loss due to fragmented processing, block stitching artifacts from patch-based strategies, and prohibitive computational resource demands. To overcome these limitations, this study proposes large scale segment anything model (LS-SAM), an enhanced fine-tuning framework based on the segment anything model (SAM), specifically optimized for accurate and efficient building extraction from ultra-high-resolution remote sensing images. The primary objectives are to: Enable end-to-end processing of full-scale images while preserving spatial and contextual integrity. Balance computational efficiency with high segmentation accuracy for practical deployment. Address the limitations of existing methods in handling large-scale geospatial data.
[Method] The proposed LS-SAM framework addresses challenges in large-scale remote sensing image processing through four innovations: (1) A dynamic positional encoding generator (PEG) replaces SAM's fixed positional encoding, using depthwise convolutions (kernel size=3) to adaptively partition input images, e.g., H×W, into patches and project spatial coordinates into learnable embeddings. This enables arbitrary-sized input processing, e.g., 5000×5000 pixels, while preserving positional relationships. (2) A hybrid encoder integrates a CNN backbone with Transformer, where the CNN extracts hierarchical local features (edges, textures) and fuses them with SAM's global attention outputs via skip connections. (3) A SMS-AdaptFormer employs parallel convolutional branches with varying kernel sizes 1×1, 3×3, 5×5 and dilation rates, r=8, 14, 20, small kernels refine local details, while dilated convolutions expand receptive fields. Features are aggregated via weighted summation for precise segmentation of diverse buildings. (4) A dynamic training strategy is used: during training, the model takes full-resolution images and applies random crops, e.g., 512×512 pixels, while PEG generates adaptive positional encodings. At inference, PEG handles any input size, and the combined CNN-Transformer encoder processes large images, e.g., 5000×5000 pixels, end-to-end—no chunking or stitching required.
[Result] Experiments on four public datasets, IAILD, MBD, WBDS, WAID, demonstrate LS-SAM's superiority: Achieves 86.7% mIoU on IAILD, outperforming DeepLabV3+ 81.25% and SAM 76.98%. On WBDS and WAID datasets, LS-SAM attains 96.11% and 94.14% mIoU, respectively, demonstrating robust generalization. Reduces GPU memory usage to 12GB, vs. 24GB for vanilla SAM, during training. Attains 10.1 FPS inference speed on 5000×5000 pixels images, NVIDIA RTX 3090Ti. Visual results on Inria and WBDS datasets show LS-SAM effectively mitigates boundary ambiguities and block stitching errors, particularly in dense urban areas and complex terrains. Additionally, ablation experiments reveal that removing PEG reduces mIoU by 2.06%, while disabling SMS-AdaptFormer reduces accuracy by 1.02%, confirming the contribution of each component.
[Conclusion] LS-SAM provides an effective solution for large-scale geospatial analysis by harmonizing global context modeling with local detail preservation. The framework significantly mitigates block stitching errors and computational bottlenecks, achieving state-of-the-art performance in building extraction tasks. This work establishes a foundation for advancing large-scale remote sensing interpretation, with potential applications in urban planning, disaster response, and environmental monitoring. Future work will focus on scaling the architecture for ultra-large imagery, 10000×10000 pixels, and enhancing cross-modal adaptability for multi-sensor data fusion.

[Objective] Mathematical foundations are integral to maps, enabling users to precisely interpret spatial relationships, feature positions, and geometric configurations. Restoring these foundations is critical for maps lacking coordinate information, particularly historical or digitized scans. Map registration serves as the primary means to align map images with standard geographic coordinates, facilitating their integration into geospatial analyses. However, existing registration methods prioritize image feature extraction over leveraging inherently associated geographic coordinates, compromising accuracy and robustness. This limitation is pronounced for maps with complex projections, variable scales, divergent symbology, or significant imaging distortions. Consequently, there is a pressing need for an adaptable, automated approach that harnesses semantic feature points with embedded geographic coordinates to restore cartographic mathematical frameworks.
[Method] This study introduces a map registration framework combining deep learning-based semantic keypoint detection with a cubic transformation model. A YOLOv8-pose architecture is trained on annotated data to efficiently identify visually discernible semantic keypoints while preserving their geographic coordinates. These paired image-geographic coordinates are then input into a weighted least squares algorithm to derive cubic transformation parameters, effectively modeling the spatial-to-geographic transformation. This process automates the recovery of mathematical foundations for unreferenced maps, minimizing manual intervention and enhancing resilience across diverse cartographic conditions.
[Result]Experiments validated the method's performance on six maps with varying projections, scales, symbology, and imaging artifacts (rotation, perspective distortion, overexposure, texture interference). The approach achieved over 90% precision and recall in semantic keypoint matching, demonstrating strong adaptability to challenging scenarios. By reconciling recovered mathematical frameworks with standard geographic data, the method successfully integrated unreferenced maps into hybrid geospatial datasets, vector, and raster formats.
[Conclusion] By integrating semantic features with geographic coordinates within a deep learning paradigm, this study achieves efficient, accurate, and robust restoration of map mathematical foundations. The proposed method addresses limitations of traditional approaches, such as rigid transformation models, heavy manual reliance, and poor generalizability. This work advances applications in historical cartography, thematic mapping, geospatial data fusion, and semantic geographic space analysis.

[Objective] Extracting unlabeled urban roads is crucial for autonomous driving, urban planning, and emergency response. Traditional remote sensing-based extraction methods struggle with accuracy and efficiency, especially in areas with scarce labels, where deep learning models face challenges due to their reliance on large-scale labeled datasets. To address this limitation, we introduce a teacher-student framework using cross-domain transfer learning, leveraging D-LinkNet model distillation to extract urban roads from unlabeled remote sensing images. This framework enables adaptation from a labeled source domain to an unlabeled target domain, reducing the dependency on human-annotated data. A feedback mechanism further enhances pseudo-label quality, ensuring progressive improvement in segmentation accuracy.
[Method] The proposed approach employs a teacher-student learning strategy with knowledge distillation and cyclic refinement to adaptively improve road extraction performance across domains. Initially, a D-LinkNet-based teacher model is trained using labeled data from a source domain. The trained model generates pseudo-labels for the unlabeled target domain, serving as the primary supervisory signals for student model training. The student model iteratively learns from these pseudo-labels, refining its segmentation capability through a feedback mechanism that progressively enhances pseudo-label accuracy. To further reduce domain gaps, cyclic distillation is introduced, continuously updating both the teacher and student models. The method is evaluated using remote sensing datasets to validate its effectiveness in urban road extraction without requiring manual annotations.
[Result] Experimental evaluations on the Massachusetts and CHN6-CUG datasets demonstrate substantial improvements in remote sensing-based road extraction. Compared to the baseline D-LinkNet model, the proposed method achieves notable performance gains, with recall, F1 score, and IoU increasing by 16.291%, 10.191%, and 11.669% on the Massachusetts dataset, respectively. Similarly, on the CHN6-CUG dataset, recall, F1 score, and IoU improve by 26.305%, 23.453%, and 20.099%. These results confirm that the integration of the teacher-student framework and D-LinkNet model distillation significantly enhances segmentation accuracy in unlabeled target domains. Furthermore, the incorporation of cyclic distillation effectively refines pseudo-label quality, reducing false predictions and improving spatial continuity, ultimately enabling more accurate and reliable urban road extraction from remote sensing imagery.
[Conclusion] The proposed method effectively addresses the challenges of unlabeled urban road extraction in remote sensing imagery, providing a scalable solution for large-scale applications. By incorporating a teacher-student framework with cross-domain transfer learning, pseudo-label refinement, and knowledge distillation, the approach significantly enhances segmentation performance in label-scarce target domains. The ability to generalize across different environments without human-annotated labels makes this method highly suitable for urban road extraction in diverse geographic regions. Its successful application to the Massachusetts and CHN6-CUG datasets highlights its potential for broader deployment in remote sensing-based urban planning, intelligent transportation systems, and infrastructure monitoring.

[Objective] Geo-entity relationships (GERs) form the bedrock of contemporary surveying and mapping systems, elucidating multifaceted connections among geo-entities while reflecting the interplay of spatial, social, and technological elements inherent in geo graphical phenomena. Despite their importance, conventional GIS data models falter in effectively encapsulating intricate GERs. This study aims to bridge this void by introducing a graph-based GER model and erecting an exhaustive GER network, thereby augmenting the sophistication of geographic information services.
[Method] The research devises a comprehensive framework for GER modeling, harmonizing viewpoints from physical, social, and digital realms. Physical-world relationships underscore spatial topology, temporal progression, and semantic interactions; social-world ties delve into human-environment interactions, societal norms, and institutional regulations. Digital-world integration leverages digital twins to emulate virtual associations. A graph-based model G = (V, E, P) is introduced, where V (entities), E (relationships), and P (attributes) are managed through a hybrid spatial-graph database system. Methodologically, it adopts a multidimensional strategy to distill explicit relationships from diverse textual repositories, like administrative records, through advanced text mining algorithms. Spatial relationships, exemplified by adjacency, are computed using geometric methodologies that precisely capture entity positioning. Deep learning strategies, including graph convolutional neural network (GCNN) and graph autoencoders, uncover latent patterns within the dataset, enhancing comprehension of complex dynamics. These techniques collectively fortify the robustness and efficacy of the GER modeling framework, enabling a rich analysis of relationships across different domains.
[Result] The experimental validation results underscore the network’s impressive capacity to model highlights the network’s prowess in modeling hierarchical spatial constructs, such as administrative demarcations, alongside dynamic processes, notably resource flows. This prowess enables the framework to surpass traditional GIS models, particularly concerning scalability and interpretability. Case studies spanning urban planning and emergency response furnish tangible evidence of the framework’s real-world decision-making utility. Key enhancements include: a refined multimodal extraction technique amalgamating textual data, geometric depictions, and cutting-edge deep learning (especially GCNN and graph autoencoders); a hybrid storage solution optimizing data handling via spatial and graph database synergy; and demonstrated practicality in vital tasks like relationship deduction and centrality analysis, crucial for deciphering network behaviors.
[Conclusion] This investigation establishes a unified paradigm for GER network construction, merging physical, social, and digital lenses to transcend conventional GIS constraints. The graph-centric model enriches geographical interaction representation, substantiated by its application in areas like infrastructure planning and disaster mitigation. Future endeavors will center on perfecting heterogeneous data synthesis, advancing AI-driven analytics for predictive relational modeling, and global scalability. By integrating spatial data science with domain-specific decision support, this work paves the way for the next generation of intelligent geographic services.

[Objective] Surface water is a vital natural resource for numerous African countries along the Belt and Road, crucial for their survival and development. Efficient monitoring of these resources is essential for the scientific management and utilization of local water supplies. However, acquiring optical imagery in Africa’s cloudy and rainy regions is challenging, complicating efforts to accurately map surface water distribution. There is an urgent need to overcome limited sample sizes by leveraging synthetic aperture radar (SAR) imagery for this purpose. This research proposes an effective method for extracting surface water from SAR imagery in Kenya.
[Method] This study introduces a novel approach to surface water extraction using SAR satellite remote sensing, combining the random forest algorithm with a semi-supervised learning method. Sentinel-1 satellite imagery is used as the primary data source. Initially, the gray level co-occurrence matrix (GLCM) is employed to extract texture features from the SAR images. A multi-dimensional feature space is then constructed by integrating these texture features, polarization characteristics, water-related indices such as the Sentinel-1 dual-polarized water index (SDWI), and other relevant features. Next, the Boruta algorithm is utilized for efficient feature selection and optimization. Finally, a new method based on random forest and semi-supervised learning is developed to accurately extract surface water from the SAR images in the study area.
[Result] Experimental results demonstrate that the proposed method effectively distinguishes between water and non-water classes, even with extremely limited sample sizes. In the study area, the overall accuracy for surface water reached an impressive 91.54%, the recall reached an impressive 88.31%, the F1 score reached an impressive 92.08%, a significant achievement given the complexity of the data and challenging environments. Qualitative and quantitative comparisons with existing methods further verify the superiority of the proposed approach in handling limited sample sizes and difficult conditions.
[Conclusion] In conclusion, the proposed method successfully addresses long-standing challenges in surface water monitoring in regions lacking optical imagery. It offers a reliable, accurate, and efficient solution for surface water resource monitoring in African countries along the Belt and Road. This not only contributes significantly to the sustainable use and scientific management of water resources in these regions but also provides a valuable reference for similar studies in other areas facing comparable data limitations and monitoring challenges. Moreover, the successful application of this methodology in Kenya paves the way for its potential use in other African countries encountering similar water monitoring difficulties.

[Objective] Geo-entities refer to natural features, artificial facilities, and geographic units that occupy a certain spatial range, and have similar attributes, or complete functions in the objective real world. With the promotion of new fundamental surveying and mapping and the construction of China’s National 3D Mapping Program (3DRGLM), geo-entities have become a research hotspot in recent years. However, several problems in the study of geo-entities remain unresolved: (1) The lack of effective logical organization methods for geo-entity data, resulting in low efficiency in retrieving and applying geo-entity data; (2) The SQL database alone is inadequate for storing and managing geo-entity data.
[Method] To address these issues, this paper has conducted research on relevant methods and technologies: (1) To overcome the lack of effective data organization methods for geo-entities, this paper breaks through the traditional method of organizing geographical spatial data using “vertical layering and horizontal partitioning”. By combining the characteristics of the geo-entity data model structure, we propose nested, hierarchical, networked, and linear multidimensional geo-entity data organization methods based on spatiotemporal domains, entity classes, relationship classes, and lifecycles. This approach achieves flexible organization and control of geo-entity data across different scales, spatiotemporal, and types; (2) In response to the inadequacy of the SQL database for handling geo-entity data storage and management, this paper proposes a distributed heterogeneous database hybrid architecture. This architecture integrates the advantages of SQL databases, file databases, graph databases, and column databases to manage different types of data, achieving multi-source heterogeneous data fusion, aggregation, and storage management with geo-entities as the core. The architecture provides geo-entity data services for various application scenarios through layers such as data storage, data indexing, data caching, data service, and data application.
[Result] To verify the feasibility of the proposed method, this paper constructed 3933 categories of geo-entity data (totaling 4.23 TB), and designed four sets of functional experiments for the proposed multidimensional data organization methods: (1) Cross scale fusion and seamless switching of geo-entities with different spatiotemporal domains and data types, including cosmic space, earth space, field space, indoor space, etc; (2) Classification display control of different types of components in BIM geo-entities; (3) Hierarchical and classified control of geographic social networks; (4) Full lifecycle display control of global submarine cable geo-entities between 1993 and 2024. The experimental results show that the proposed method can achieve efficient storage, flexible organization, and display control of geographic entity data across different spatiotemporal domains, entity classes, relationship classes, and lifecycles.
[Conclusion] The geo-entity data organization methods and distributed heterogeneous database hybrid architecture proposed in this paper can provide reference for the organization and management of massive geo-entity data in new fundamental surveying and mapping and the construction of 3D realistic geospatial scene of China.

[Objective] Raster geological maps constitute vital data resources in geological research and mineral exploration. However, these maps are often stored in non-standardized, unstructured raster formats, posing significant challenges for large-scale data retrieval, integration, and intelligent utilization. Traditional geological data management systems frequently struggle with fragmented storage, inefficient querying, weak inter-data relationships, and inadequate support for semantic-level searches. These limitations hinder the comprehensive exploitation of geological information. To address these challenges, this study explores large-scale textual element extraction from raster geological maps and proposes a novel map-text retrieval framework to enhance semantic accessibility and intelligent processing of unstructured geological data.
[Method] This study proposes a method for text extraction and map-text retrieval from raster geological maps. A distributed architecture leveraging HBase and the Hadoop distributed file system (HDFS) is constructed to efficiently manage large-scale unstructured raster geological maps and associated documents. For text extraction from geological maps, a deep learning-based optical character reader (OCR) pipeline is implemented, combining a differentiable binarization network (DBNet) for text region detection with a convolutional recurrent neural network (CRNN) for sequence-based text recognition. This approach substantially improves text detection and recognition accuracy under complex map backgrounds. In processing geological reports, the term frequency-inverse document frequency (TF-IDF) algorithm is employed for semantic similarity analysis, establishing meaningful associations between map elements and document content. Building on this, a sequence labeling model integrating bidirectional encoder representations from transformers (BERT) and conditional random field (CRF) is utilized to automatically extract geological entities and domain-specific keywords. Additionally, a full-text retrieval module based on the Apache Solr search engine is incorporated, enabling high-efficiency, semantic-aware retrieval of geological documents.
[Result] The experimental results indicate that the proposed method substantially enhances the usability of textual information and the efficiency of keyword extraction in real-world applications. The proportion of usable text increased from 51.7% to 82.3%, reflecting a marked improvement in the accuracy and completeness of text extraction. Furthermore, the efficiency of keyword extraction achieved a 426% improvement compared to the TextRank algorithm. The proposed framework demonstrates strong scalability and adaptability, enabling efficient processing of large-scale geological datasets, supporting real-time storage and rapid retrieval across datasets of various sizes, and significantly advancing the utilization efficiency of unstructured geological data.
[Conclusion] This study proposes a method for text extraction and map-text retrieval tailored to unstructured geological data, integrating distributed big data infrastructure, deep learning-based OCR technologies, and advanced semantic extraction models. The proposed method significantly strengthens the connection between raster geological maps and textual geological knowledge, thereby improving data utilization efficiency, research productivity, and geological decision-making. Looking ahead, future work will focus on incorporating cutting-edge Transformer-based architectures, constructing domain-specific geological knowledge graphs, and enhancing user interaction via intuitive interfaces. This integrated approach introduces a novel technological paradigm for geological data processing and knowledge discovery, with potential applications in scientific research, natural resource management, and digital geoscience services.

[Objective] The rational planning and allocation of medical facilities are crucial for enhancing urban public services and promoting the integrated development of urban and rural areas. This study proposes a comprehensive analysis framework and model parameter adaptation strategy at the prefecture-level city level to guide the hierarchical and accurate allocation of medical facilities.
[Method] This study focuses on the central area of Ganzhou City, using residential areas as assessment units. From the perspective of hierarchical diagnosis and treatment, it employs an improved potential model, service coverage and overlap rate, spatial autocorrelation analysis, minimum facility point model, and GIS spatial analysis. By analyzing the adaptability of relevant model parameters, the study obtains the optimal parameter combination to systematically analyze and evaluate accessibility, equality, and spatial optimization. Firstly, the distribution characteristics of medical facilities in the study area are examined based on location entropy. The improved potential model is then used to measure accessibility for hospitals of various levels. Subsequently, hierarchical evaluations from non-spatial and spatial equality perspectives are conducted based on accessibility results. Finally, using the minimum facility point model, the study performs spatial layout optimization analysis and proposes corresponding suggestions.
[Result] The research findings indicate that: (1) Localized calibration of limit travel time and friction coefficient in the improved potential model significantly enhances regional model adaptability within the “accessibility-equality-spatial optimization” analysis system. (2) The spatial allocation and distribution of medical facilities in the study area are imbalanced, with accessibility decreasing in a circular layered pattern. Tertiary, secondary, and primary hospitals show gradually decreasing accessibility levels, with notable differences. High accessibility communities are located in areas with dense medical facilities and convenient transportation. (3) Medical facilities generally exhibit inequality. The coverage and overlap of the 15-minute service range of primary hospitals and the 30-minute service range of secondary hospitals are relatively high, with significant spatial agglomeration among hospitals at all levels. (4) It is recommended to add new or renovate primary and secondary hospitals in under-resourced areas such as Shuidong Town and plan tertiary hospitals in peripheral areas like Meilin Town to achieve balanced medical facility distribution.
[Conclusion] This research enriches the systematic framework for studying medical resource allocation. The proposed analysis framework and model parameter adaptation methods support hierarchical and accurate allocation of medical facilities and provide methodological references for similar urban studies. The empirical results offer auxiliary decision-making support for optimizing medical facility layout in Ganzhou, enhancing overall efficiency and rationality, and ensuring balanced and coordinated development of regional medical resources.

[Objective] With the diversification and cost-effective development of remote sensing platforms, the availability of remote sensing data has significantly increased. The current challenge lies in effectively managing this multi-source, massive, and variable data. unmanned aerial vehicle (UAV), as one of the most convenient platforms for remote sensing data acquisition, have rapidly developed and gained widespread use in recent years. The data collected by UAV has proven invaluable for both civil and military applications. Geo-localization of UAV imagery is a key step in many applications. However, geo-localization becomes a significant challenge when the global navigation satellite system (GNSS) is unavailable or performs poorly due to external influences.
[Method] Based on the block matching of Transformer, we proposed a cross-view image-image retrieval method named TomGeo (Transformer oblong matching for geo-localization). This method can be used for geo-localization of UAV images and image-based navigation. TomGeo uses pyramid vision Transformer (PVT) as a feature encoder to extract multi-scale features from UAV and satellite images. Block classification and block matching are then performed based on these features to establish correspondence between the same regions in images from different viewpoints at the same location. Finally, salient area identification is used to enhance key instance category information in the cross-view images.
[Result] TomGeo has implemented multi-scale feature fusion based on PVT, which further improves the shortcomings of low utilization of location differences and contextual information of key features in cross view image retrieval through block classification, block matching, and salient region recognition. On the publicly available dataset University-1652, when retrieving satellite perspective images of corresponding locations based on UAV perspective images, TomGeo’s R@1 was 85.54% and AP was 87.62%; When retrieving UAV perspective images of corresponding locations based on satellite perspective images, R@1 is 91.43% and AP is 85.87%.
[Conclusion] Research has shown that TomGeo performs well in cross view image retrieval from UAV and satellite perspectives, offering better accuracy and higher stability. Additionally, the performance of TomGeo was comprehensively evaluated under different experimental conditions (ablation experiment), such as the backbone network, different network structures, and loss functions. TomGeo’s excellent performance in cross-view image retrieval between UAV and satellite views supports UAV image geo-localization and area navigation based on imagery. this makes it beneficial for UAV use in special situations and contributes to the development of the low-altitude economy.

[Objective] Conducting geological surveys in challenging and hazardous terrains, particularly in remote plateau regions, poses significant difficulties when integrating “remote sensing + geology” multi-modal data. Given these complexities, this article focuses on innovating the storage, representation, intelligent extraction of geological remote sensing information, and automated services for geological survey data, leveraging advanced technologies like cloud computing, big data, and artificial intelligence. These innovations collectively establish a collaborative cloud service technical framework for multimodal geological survey data, catering to the demand for intelligent high-resolution data services and integrated multi-source data in difficult and dangerous geological survey areas.
[Method] (1) To address the challenge of unified expression and storage for multimodal geological survey data, we propose a novel, multi-dimensional, and multidisciplinary “remote sensing + geology” data storage model. This model constructs a centralized database that harmonizes data across various levels and scales throughout its lifecycle, allowing dynamic updates. It facilitates comprehensive, area-wide, and element-wise data support tailored to the arduous conditions of geological surveys. (2) In response to inefficiencies in image data utilization and service distribution, we introduce an automatic extraction technique for linear features from high-resolution remote sensing images, combining deep learning with wavelet transform technology. Additionally, we present a service distribution strategy for geological survey image base maps optimized for multi-platform operation. This dual approach automates linear feature extraction from intricate images and enhances the distribution efficiency of large-scale, multi-type, high-precision image base maps.(3) Addressing shortcomings in the existing data storage model, our proposal entails developing a unified database encompassing multiple hierarchical levels, scales, lifecycle stages, with continuous updates. This ensures coherent, full-coverage data management crucial for geological surveys in demanding environments. (4) To overcome low automation levels and the inability to promptly meet field survey demands, we have designed a framework centered around field geological survey location sensing, knowledge discovery, and proactive knowledge services. This framework propels the dissemination of geological survey information in challenging territories, resolving key issues related to timeliness, precision, and comprehensiveness of such services. Collectively, these technological advancements constitute a collaborative cloud service ecosystem for multimodal geological survey data, forming an “Intelligent Spatial Platform for Geological Surveys in Difficult and Dangerous Areas” underpinned by a “cloud + terminal” service paradigm.
[Result] The research findings have been successfully implemented and validated across over 400 regional geological and mineral surveys, including Quaternary geology and ophiolite belt surveys in formidable regions like Qinghai-Tibet.
[Conclusion] The outcomes of this research effectively fulfill the operational needs of geological surveys in difficult terrains, bolstering the enhancement of geological surveying capabilities.

[Objective] Enhancing the efficiency and effectiveness of the land use approval process through digitalization and intelligent technologies is crucial for consolidating efforts in natural resource management, specifically achieving “dual unification”. This study tackles challenges posed by fragmented data management and the complexity of policy retrieval within the land use approval workflow. By leveraging the ontology of land use approval processes, we have developed a collaborative framework that integrates the construction of a knowledge graph with intelligent question-and-answer (Q&A) capabilities. This framework is designed to support and streamline land use approval activities, providing a robust decision-support tool that addresses issues related to weak business associations and difficult policy access.
[Method] The methodological approach involves systematically extracting and integrating information from various data sources relevant to land use policies and approval procedures. Utilizing advanced information extraction techniques and graph construction algorithms, we built a dynamic knowledge graph encapsulating the complex dependencies and regulations governing land use. Additionally, a knowledge retrieval-augmented generation model was developed to facilitate sophisticated Q&A interactions, allowing users to engage with the system via natural language queries and receive accurate, context-aware responses. This integrated framework was implemented within an intelligent service platform tailored for land use approval, and its effectiveness was assessed through a qualitative comparative analysis against traditional search engines, such as Baidu.
[Result] The implementation of the proposed framework led to the successful development of an intelligent service platform that significantly enhances the land use approval process. The constructed knowledge graph introduces a novel organizational structure for land use-related information, enabling seamless integration and retrieval of policy data. The intelligent Q&A system outperforms conventional search engines in delivering precise and relevant answers, demonstrating its ability to comprehend and process complex queries within the land use domain. The comparative analysis indicates that the platform substantially improves the accessibility and usability of policy information, thereby facilitating more informed and timely decision-making by approval personnel. Furthermore, the framework’s capability to systematically organize and leverage domain-specific knowledge highlights its potential to transform traditional land management practices into more streamlined and intelligent operations.
[Conclusion] In conclusion, this study presents an innovative approach to overcoming the inherent challenges in the land use approval process through the collaborative construction of a knowledge graph and the implementation of an intelligent Q&A system. The developed framework not only offers a new paradigm for knowledge organization within the land use sector but also provides a practical tool that enhances decision-making capabilities. Despite the significant advancements demonstrated, limitations remain in the automation of knowledge graph construction and the sophistication of Q&A interactions. Future research will focus on increasing the automation levels in knowledge graph development and expanding the applicability of the Q&A system to encompass a broader range of business scenarios. By further integrating these technologies with existing natural resource and land use planning systems, the framework aims to strengthen digital and intelligent governance capacities, ultimately contributing to more efficient and effective land management practices.

The 3D realistic geospatial landscape model (3DRGLM) China initiative constitutes a vital component of the nation’s emerging infrastructure, with its standard system serving as the cornerstone to guarantee the successful execution of 3DRGLM development. In response to the challenge posed by an incomplete standard framework, this paper undertakes a comprehensive review of pertinent 3DRGLM standards from both domestic and international contexts, meticulously identifying shortcomings within existing norms.
Guided by overarching principles of standard formulation—systematic approach, scientific rigor, progressiveness, scalability, and operability—this study is anchored in the 3DRGLM product ecosystem and aspires to propel advancements in technology and production organization methodologies. It proceeds to delineate the design philosophy and comprehensive structure of the 3DRGLM standard system, encompassing thirty core elements distributed across five key dimensions: holistic design, acquisition and processing, database administration, application services, and quality assurance.
Leveraging the full potential of established surveying and cartographic geographic information standards, the paper introduces eighteen novel standards along with their principal tenets, ensuring coverage throughout the entire lifecycle of 3DRGLM implementation. This work furnishes a technical compass for the research and development endeavors surrounding 3DRGLM standards, pivotal to realizing a national “one map” that is seamlessly interconnected horizontally and vertically integrated. Such an accomplishment holds profound implications for the realm of 3DRGLM construction.

3D realistic geospatial landscape model (3DRGLM) stereoscopic reconstruction technology plays a pivotal role in China’s digital transformation by leveraging the spatiotemporal complementarity and multi-view synergy of multi-source remote sensing data to achieve high-precision, multi-dimensional virtual space modeling. This article systematically reviews the technical framework of multi-source remote sensing data-driven 3D realistic geospatial landscape model reconstruction, covering data sources, geographic scene and entity modeling methods, technical challenges, and emerging trends.
Key data sources include optical imagery (satellite, aerial, and close-range), LiDAR point clouds (airborne, terrestrial, and mobile systems), and SAR data. Satellite optical imagery facilitates large-scale terrain monitoring, while aerial and close-range imagery improve urban and component-level modeling. LiDAR provides high-precision 3D spatial information, with mobile systems enhancing efficiency through colored point cloud acquisition. SAR data, when combined with InSAR-derived deformation point clouds, strengthens the reconstruction of complex terrain. Additionally, IoT-generated real-time data and historical geospatial data contribute to the dynamic maintenance of 3D models.
Geographic scene modeling primarily relies on mesh generation using multi-view stereo (MVS) and 3D Gaussian splatting (3DGS). Traditional MVS methods encounter difficulties in feature matching and environmental adaptability, while deep learning frameworks optimize pixel-level geometry. Transformer-based models enable joint camera calibration and 3D reconstruction from unconstrained images. 3DGS excels in visual fidelity and real-time rendering but faces challenges in maintaining multi-view geometric consistency. Large-scale reconstruction approaches balance detail preservation and computational efficiency through dynamic partitioning and distributed training, although cross-region fusion remains challenging.
Geographic entity modeling integrates model-driven (template-based) and data-driven (primitive segmentation) approaches. Model-driven methods excel in structured but low-detail reconstructions, while data-driven techniques provide flexibility at the cost of higher computational demands. Deep learning methods, such as Transformers and graph neural networks (GNN), facilitate large-scale urban reconstruction but require extensive training data. As a result, semi-automated workflows remain dominant, underscoring the need for a balance between efficiency and quality.
Critical challenges persist in advancing 3D realistic geospatial landscape model reconstruction: (1) Generative AI-based methods enable cross-modal 3D generation but encounter challenges related to data dependency and maintaining the plausibility of urban scenes. (2) Dynamic scene reconstruction faces difficulties in addressing long-interval changes (e.g., building demolition) and integrating rigid and non-rigid structures, with limited adaptability observed in methods such as Street Gaussians. (3) Multi-source data synergy remains constrained by spatiotemporal misalignment and complex preprocessing, necessitating the development of integrated platforms to enhance interoperability. (4) Application-driven product derivation demands standardized yet flexible models (e.g., photovoltaic assessment) to broaden the application of 3D models in smart cities and natural resource management.
Future developments focus on four promising areas: (1) Integrating generative AI with differentiable rendering to achieve lightweight and dynamic modeling. (2) Developing temporal reconstruction techniques that combine physical simulation with historical data-driven prediction to enhance long-term scene modeling. (3) Advancing intelligent multi-source registration and distributed computing to improve efficiency and scalability in large-scale reconstruction tasks. (4) Designing application-oriented model systems to enhance domain-specific services, such as digital twin platforms, by tailoring models to the needs of various industries. By addressing these challenges, 3D realistic geospatial landscape model reconstruction will strengthen its role as a spatiotemporal backbone supporting China’s digital economy and ecological civilization initiatives.

[Objective] The rapid proliferation of information technologies has engendered unprecedented volumes of heterogeneous data across disparate systems and sensors. A primary challenge in multi-source data integration lies in reconciling divergent data models and organizational frameworks. This issue is particularly acute in geological survey domains, where massive, spatiotemporally correlated datasets are managed across fragmented systems with heterogeneous models and standards. Such structural and semantic discrepancies hinder data management efficiency, exacerbating redundancy, inconsistency, and dispersion.
[Method] We propose a metadata-driven, semantic modeling approach to construct a unified data model for geological survey business management. The methodology comprises five sequential stages: (1) requirements analysis and data organization, (2) metadata extraction,(3) domain semantic analysis, (4) hierarchical model construction, and (5) iterative evaluation and refinement.
[Result] To address challenges associated with multi-source data, inconsistent standards, semantic ambiguity, and insufficient correlation in geological survey business management systems, the proposed methodology systematically organizes data sources, extracts and analyzes metadata from relevant systems, and performs semantic integration. This process yields a standardized data framework comprising unified entities, attributes, and relational structures. By leveraging geospatial coordinates and master reference data, a hierarchical organizational model is developed to harmonize heterogeneous datasets, enabling consistent data description and domain-specific abstraction. Bidirectional mapping protocols and synchronization mechanisms are established between the unified model and disparate sources. The resulting conceptual, logical, and physical data models have been validated through implementation in a geological survey business management data center.
[Conclusion] The proposed approach successfully constructs a unified data model for geological survey business management, integrating data from 20 heterogeneous sources. This model enables centralized data description and holistic, one-stop management services, resolving issues of standard inconsistency, semantic ambiguity, and multi-source data-induced redundancy. By facilitating interoperability, shared analytics, and advanced applications, the framework supports optimized workflows and evidence-based decision-making. This work provides a scalable technical solution to reconcile structural and semantic discrepancies in multi-source heterogeneous data, including spatiotemporal datasets.

[Objective] Accurately distinguishing between the wood (trunk) and leaf components of trees is crucial for obtaining structural parameters such as the leaf area index (LAI). Terrestrial laser scanning can capture point clouds at millimeter spatial resolution, allowing for detailed depiction of a tree's fine structure. The separation of wood and leaf point clouds from terrestrial laser scanning data serves as the foundation for deriving these structural parameters. However, current research methods exhibit significant misclassification in separating wood and leaf components.
[Method] To enhance the accuracy of wood-leaf separation, this paper introduces a refined methodology leveraging local features within terrestrial laser scanning point clouds. Initially, a graph-based leaf-wood separation (GBS) model is employed for preprocessing to segregate the initial point cloud into branches and leaves. Subsequently, local features are extracted, and both branch and leaf point clouds are further refined using curvature thresholds and circumcircle radius comparisons. Given that local features pertain to neighborhood relationships, the study examines the effect of varying the number of neighboring points on separation performance, settling on 100 neighbors to balance computational efficiency and accuracy. Additionally, the choice of surface variation threshold is vital; here, the ratio of the maximum standard value (SV) to a variable parameter α is used as the segmentation criterion. Optimum results are achieved when α is set to 1.45. For circumcircle radius comparisons, an empirical estimate γ of 5 cm effectively purges misclassified wood points from the leaf point cloud.
[Result] To validate the efficacy of our proposed method, we utilized both public datasets and self-collected data to acquire point clouds representing diverse tree structures. Comparisons against random forest, path tracking detection, and the GBS model reveal our approach yields the highest overall accuracy, Kappa coefficient, and F1 score, outperforming even the GBS model by 0.027 and 0.072 higher in average overall accuracy and Kappa values. Specifically, our method achieves an overall accuracy of 0.945 and Kappa value of 0.811.
[Conclusion] Our methodology offers a valuable enhancement to existing wood-leaf separation algorithms, eliminating the need for manual labeling while significantly boosting classification performance.

[Objective] Land surface temperature(LST) variations significantly affect surface water resource distribution, particularly in heavily dependent on these resources. Egypt, located in a tropical desert climate, is highly dependent on its surface water resources, primarily supplied by the Nile river. Therefore, accurate surface temperature inversion is crucial for formulating long-term water resource response strategies in the country. This study aims to: (1) investigate the impact of LST on surface water distribution in Egypt, (2) enhance single-channel algorithm accuracy for LST inversion, and (3) establish an inversely proportional functional model linking LST to surface water distribution.
[Method] Due to limited temperature measurement data in Egypt, this study employed a combination of variable normalized difference vegetation index (NDVI) thresholds and high-precision atmospheric water vapor content data to calculate a high-precision surface temperature dataset for Egypt using the Google Earth Engine (GEE) platform. By creating buffer zones at varying distances from the Nile River, quantitative characteristics of LST in riverine and surrounding areas were analyzed. A mathematical model was developed by assigning weighted coefficients to water resource distribution patterns, enabling analysis of the relationship between LST and surface water dynamics in Egypt.
[Result] LST inversion results, validated against Landsat 8 temperature products, achieved root mean square errors (RMSE) below 1.0 for two study areas—exceeding World Meteorological Organization (WMO) accuracy standards—and below 1.5 for the third. High linear fit accuracy was observed between inverted LST and meteorological station measurements (R2 = 0.828), closely matching historical long-term correlation (R2 = 0.849). Analysis revealed: (1) In water-abundant regions, LST reduction effects diminished rapidly with distance from water bodies; (2) In sparse, braided water distribution areas, LST effects decreased gradually. The inversely weighted proportional function model demonstrated strong correlation (R2 = 0.96) in regions with significant surface water resources.
[Conclusion] This research validates the efficacy of variable NDVI thresholds and atmospheric water vapor data for LST inversion in tropical desert regions. The inversely weighted proportional model provides a quantitative framework for understanding LST-water resource interactions in Egypt, offering critical insights for water management in arid regions. The model is broadly applicable to mid-low-latitude deserts facing water scarcity. Future work will refine NDVI threshold optimization (e.g., via genetic algorithms), integrate higher-resolution LST data, advance surface water distribution modeling, and employ machine learning to enhance model precision.

[Objective] Urban subsidence poses a significant threat to infrastructure and environmental stability, particularly in rapidly growing cities. Traditional monitoring methods struggle to accurately capture complex, nonlinear subsidence patterns. This study introduces the geographically weighted random forest (GWRF) approach, integrating multi-period subsidence monitoring data to enhance pattern recognition and prediction. The research analyzes subsidence trends at 74 monitoring points within a construction site from April 16 to July 29, 2023.
[Method] To overcome the limitations of conventional techniques, this study applies the GWRF method, which incorporates spatially adaptive weighting to account for regional subsidence variations. Unlike Kriging interpolation and traditional random forest methods, the GWRF dynamically adjusts prediction weights based on local spatial features, leading to more precise subsidence forecasting. The performance of GWRF, Kriging, and random forest is compared using root mean squared error (RMSE) and mean absolute error (MAE) as evaluation metrics.
[Result] In contrast to traditional random forest, which fails to account for spatial variation, the GWRF reduces prediction bias and enhances the accuracy of subsidence forecasts.The results suggest that during the monitoring period, the most significant subsidence occurred in the road surface and northwest building areas, influenced by construction activities. The findings show that GWRF outperforms both Kriging and random forest in identifying and predicting subsidence trends. Compared to the other methods, the GWRF reduces RMSE by 25% and MAE by 30%, particularly during periods of severe subsidence. The model also provides reliable forecasts for future subsidence trends, highlighting areas in the eastern and northern construction zones that require reinforcement to mitigate risks.Between April 16 and June 11, 2023, maximum subsidence reached -12 mm in the road surface and northwest building areas, primarily due to road construction and underground pipeline installation. From June 11 to July 29, 2023, subsidence decreased to below -3 mm in the southeast following construction scope adjustments. These modifications effectively controlled the subsidence level, demonstrating the model’s utility in guiding construction decisions and reinforcement strategies. The GWRF method successfully integrates spatial heterogeneity into subsidence predictions, offering a significant improvement over traditional Kriging interpolation and random forest methods. Compared to Kriging, which assumes stationarity and often misidentifies non-subsidence zones as areas with slight subsidence, the GWRF offers more reliable predictions. By adjusting construction strategies, the subsidence rate was effectively controlled in later periods.
[Conclusion] This research confirms the effectiveness of the GWRF method for subsidence pattern identification and prediction in construction sites. The GWRF method provides valuable insights for construction management, emphasizing the importance of localized reinforcement measures. This study lays the foundation for further optimizing of the GWRF model for extreme point prediction and the development of real-time monitoring and early warning systems. Future research can integrate dynamic construction data to enhance prediction accuracy, providing comprehensive technical support for urban planning, construction management, and disaster prevention.

This thesis focuses on industry applications and proposes a framework for semantic modeling of geo-entity relationships 3D realistic geospatial landscape model, compatible with multiple domains.
[Objective] Semanticizing of geo-entities in 3D realistic geospatial landscape model is crucial for constructing a unified 3D spatiotemporal substrate for Digital China. This process facilitates efficient information circulation and sharing, promoting high-quality industry development. However, current semantic modeling of relationships faces challenges such as insufficient standardization, poor scalability, and difficulty in cross-domain application. Establishing and enhancing this framework along with optimizing the relational semantic construction method, are essential for advancing the entire chain from high-quality data production to multi-domain applications. Therefore, we need to create a robust and practical semantic modeling framework.
[Method] By analyzing the multi-dimensional characteristics of geo-entities in 3D realistic geospatial landscape model domains, this paper delves into the content of the geo-entity semantic model system, and underscores the necessity of developing a semantic modeling framework. We proposes a framework that considers multiple domains formed by defining relationship types and description rules. Standardized and generalized methods ensure the accurate and consistent expression and storage of relationship semantics among geo-entities. Building on this basic framework, we add a new relationship domain index or extend the relationship triggering feature conditions to accommodate different business applications.
[Result] Using the emergency disposal scenario of community gas pipeline leakage as an example, this paper validates and elaborates on the application of the semantic modeling framework in detail. The researcher integrates proprietary relationship types and corresponding description rules from extended domains like real estate rights, and emergency security, to construct a multi-domain compatible semantic modeling framework. Practical results demonstrate that the framework effectively supports cross-domain decision-making and application requirements.
[Conclusion] The proposed framework for semantic modeling of relationships achieves compatibility and extensibility across multiple industries, fostering interconnectivity and interoperability of geo-entity semantics information. It holds significant value and prospects for supporting the informatization construction and services of related domains, offering strong support for their advancement.

[Objective] Surface water resources in arid African regions are scarce and unevenly distributed, presenting significant challenges for water access and management. Egypt, emblematic of such regions, endures a hot, dry climate with minimal rainfall, rendering its water resources heavily reliant on the Nile River and an extensive network of artificial canals. These canals are crucial for agriculture, population sustainability, and driving economic activities. However, their complex and variable spatial configurations, compounded by the presence of adjacent landforms like deserts and agricultural areas, render the precise delineation of surface water boundaries and small water bodies exceedingly difficult for conventional remote sensing methodologies. These methods often fall short due to incomplete extractions, ambiguous boundaries, and misclassifications, particularly within narrow canals. Addressing these issues is crucial for achieving accurate monitoring of surface water and facilitating the sustainable management of Egypt's water resources.
[Method] To address these challenges, this study proposes an improved U-Net deep learning model, namely GLF-MFUNet, designed to enhance the precision and robustness of surface water extraction from remote sensing imagery. The model features a dual-path encoder that seamlessly integrates Vision Transformer (ViT) and Manhattan self-attention (MVT), effectively capturing global contextual cues to ensure comprehensive extraction of artificial canals and enhance water body classification in intricate environments. Moreover, a multi-scale depthwise convolution mechanism is embedded within the spatial attention module, empowering the model to proficiently merge water features across diverse scales, thereby refining the definition of fine details and water boundaries. The implementation of channel attention mechanisms further serves to suppress noise and minimize misclassifications. The model's training utilized Sentinel-2 multispectral imagery of Egypt and was rigorously validated against ZY-3 satellite data, ensuring its robustness across varying environmental conditions.
[Result] Experimental outcomes underscore that the proposed GLF-MFUNet markedly elevates water body extraction accuracy in comparison to existing models. It outshines prevalent semantic segmentation models, including ViT, SwinTransformer, and DeepLabV3+, demonstrating superior performance across a majority of evaluation metrics. Specifically, relative to the pre-improvement baseline U-Net model, GLF-MFUNet achieves an impressive increase of 4.97% in IoU, 3.02% in F1-score, and a substantial 10% enhancement in precision. The synergistic fusion of global and local feature extraction through the MVT and SPCAI modules endows the model with heightened spatial continuity and a reduced incidence of false detections.
[Conclusion] The GLF-MFUNet model adeptly confronts the pivotal challenges associated with surface water extraction in arid African locales such as Egypt, yielding substantial improvements in detection accuracy, spatial coherence, and classification consistency. Through the integration of global-local feature synthesis, attentive mechanisms, and multi-scale information processing, it emerges as a fitting solution for surveillance of artificial water networks in arid environments. Evidently, GLF-MFUNet exhibits unparalleled performance in large-scale, automated surface water mapping, furnishing dependable data to inform water resource governance, irrigation scheming, and ecological preservation initiatives. Its successful application in Egypt highlights its potential for wider adaptation in arid and semi-arid areas, providing valuable support for zones globally, thereby contributing valuably to the sustainable water resource management in Africa.

[Objective] Lake Victoria, the largest freshwater lake in Africa, plays a critical role in regional water resource management, flood preparedness, and ecological conservation across the Nile River Basin. However, conventional methods relying on optical remote sensing are severely limited by persistent cloud cover and heavy rainfall in the tropical climate, resulting in fragmented temporal observations long-term monitoring consistency. These atmospheric conditions often lead to discontinuities in temporal observations, limiting the ability to conduct consistent and reliable long-term monitoring. Therefore, there is an urgent need for a more stable, weather-independent, and temporally consistent method for large-scale inland water body mapping.
[Method] We advanced water body extraction techniques by exploiting the all-weather imaging capability of Sentinel-1 SAR datasets. A rigorous comparative analysis identified an optimal SAR-based water index, which was coupled with an improved Edge-Otsu algorithm to replace conventional methods that rely on manually set initial thresholds. This innovation enables adaptive, automated water body segmentation, minimizing subjectivity and enhancing methodological reproducibility. The framework was applied to generate a 10-meter resolution monthly water surface area dataset for lake Victoria from 2017 to 2023. This dataset facilitated a comprehensive analysis of the spatiotemporal dynamics of lake Victoria.
[Result] The proposed method demonstrated exceptional stability and automation across multi-temporal SAR imagery, achieving an overall water extraction accuracy of 98.9%. Compared to global products such as the JRC Global Surface Water (GSW) dataset (R = 0.1) and Dynamic World (R = 0.29), our dataset exhibited a significantly higher correlation of 0.76 with in situ water level records, reflecting superior temporal consistency and reliability. Notably, the algorithm captured dynamic water boundary fluctuations without manual intervention, outperforming traditional threshold-dependent approaches.
[Conclusion] The high-resolution dataset revealed distinct temporal and spatial patterns in lake Victoria's surface area. From 2017 to 2022, the lake exhibited a gradual expansion trend, followed by a minor contraction in 2023. More specifically, water surface area increased rapidly during the long rainy season (March to May), typically peaking in June, then gradually receding through the dry season (July to September). Spatial heterogeneity was most evident in the northeastern and southern basins, particularly in Kenya's Winam Gulf and Tanzania's Mwanza Gulf. This study provides comprehensive and long-term monitoring of the dynamic changes in lake Victoria's water surface area, offering a solid scientific basis for transboundary water resource management. It plays a vital role in supporting ecological conservation and promoting regional sustainable development. Furthermore, the SAR-based methodology pioneered here is transferable to other tropical regions, supporting hydrological modeling, climate adaptation strategies, and cross-border environmental management initiatives.

[Objective] Traditional indoor measurement methods exhibit significant limitations in complex and confined spaces. In intricate indoor layouts, their precision often suffers, and data collection can be incomplete. For instance, in areas with complex furniture arrangements or narrow corridors, obtaining accurate and comprehensive data becomes challenging. Similarly, in tight spaces like ducts or between closely spaced shelves, the operation of traditional measuring tools is restricted, leading to potential data omissions. This research aims to develop an indoor UAV-based autonomous exploration and mapping method integrated with a software-hardware system. The goal is to achieve highly accurate indoor mapping without relying on GNSS signals, which is essential for applications such as indoor navigation, facility inspection, and emergency response in buildings. By doing so, it seeks to provide effective solutions for various indoor-related tasks and enhance the overall efficiency and accuracy of indoor spatial data acquisition.
[Method] This method is founded on the information gain maximization strategy and the Euclidean signed distance field (ESDF) map. Initially, exploration points are evenly distributed across the map to ensure comprehensive coverage. Data from the LiDAR and inertial measurement unit (IMU) are then fused. During this process, distortions in the LiDAR data are corrected to ensure accuracy. The IMU data is pre-integrated to predict the UAV's pose changes between LiDAR scans. Factor graph optimization is employed to incorporate LiDAR odometry constraints, IMU pre-integration results, and environmental constraints, effectively reducing drift in parameters like inertial navigation biases. The UAV's hardware includes a semi-solid-state LiDAR for high-resolution environmental perception and a flight controller for stable flight management. The ROS-based software features algorithms for precise positioning, real-time dynamic obstacle avoidance, and optimal path planning, enabling the UAV to navigate and explore the indoor environment autonomously.
[Result] Simulation results demonstrate notable improvements. Compared to the FUEL algorithm, the proposed method reduces exploration time by an average of 13.28%, indicating a more efficient exploration process. Additionally, the trajectory length is decreased by 14.89%, suggesting less redundant movement. In a basement corridor field experiment, point cloud measurements show a maximum deviation of 4.87 cm, a minimum of 0.37 cm, and a standard deviation of 2.66 cm. These results meet the high-precision requirements for indoor mapping. Furthermore, the UAV efficiently explores abandoned train tunnels, demonstrating its adaptability to complex and challenging environments.
[Conclusion] The developed UAV system has proven its capability to explore autonomously in indoor environments without GNSS signals. However, in highly complex and dynamic indoor scenarios, such as crowded areas with constantly moving objects, preset exploration points may face challenges. For example, new obstacles can render preset points ineffective, reducing exploration efficiency. Future research will focus on optimizing the algorithm, potentially integrating more advanced sensor fusion techniques and intelligent decision-making algorithms. This enhancement will improve the UAV's adaptability in complex environments and further advance the development of indoor mapping technology.