The construction of new spatiotemporal information infrastructure relies on innovative software middleware and geographic information service models. Driven by cloud computing, artificial intelligence, and big data, traditional spatiotemporal data infrastructure is evolving from data services to intelligent computing services. This paper focuses on a spatiotemporal intelligent computing service platform, aiming to develop a Digital Earth Cube-ready service. It explores key technologies in organizing, computing, and reasoning with geospatial big data, and develops the open geospatial engine (OGE) system, which deeply integrates computing power, data, and algorithms while promoting openness and sharing. The system provides a data-ready, analysis-ready, and decision-ready geospatial data-information-knowledge service framework, forming an open geospatial engine (OGE) and establishing a new type of spatiotemporal information infrastructure based on spatiotemporal cube management and geospatial big data analysis. Based on this foundation, an OGE prototype system has been developed, integrating various types of Earth observation data accumulated by Wuhan University and related institutions. A series of typical spatiotemporal analysis experiments covering raster, vector, and thematic data were conducted, validating OGE's capabilities in managing and analyzing geospatial big data.
Optical remote sensing images (RSIs), which are widely used in various Earth observation tasks due to its rich geoinformation, are often significantly affected by varying degrees of cloud contamination, leading to a significant reduction in data quality and usability. Although extensive research has been conducted on cloud removal from optical RSIs, there is still a lack of systematic review and analysis in this field. To address this gap, this paper first employs bibliometric analysis to investigate the publication trends of relevant literature both domestically and internationally, revealing the long-term development dynamics of cloud removal research in RSIs. Subsequently, the paper then provides a comprehensive and systematic review of research on the removal of thin and thick clouds, thoroughly analyzing the core challenges, underlying assumptions, approaches, and fundamental principles of different cloud removal methods, while evaluating their strengths and weaknesses. Finally, this paper summarizes and discusses the common key challenges and future trends in current optical remote sensing cloud removal research. This paper not only offers crucial insights for readers to fully understand the research progress in optical remote sensing cloud removal over the past three decades but also serves as a valuable reference for grasping the development patterns and trends in this field.
The rapid development of large language models (LLMs) provide a new approach for GIS analysis, leading to the large language model-driven GIS analysis technical architecture (LLM4GIS). Based on the latest research up to October 2024, this paper reviews the evolution of GIS analysis and summarizes the LLM4GIS technical architecture from 3 aspects: application modes, datasets and evaluation methods. It also summarizes the research progress of LLM in GIS analysis tasks such as knowledge question-answering, knowledge extraction, spatiotemporal reasoning, and analyzing and modeling. Finally, the paper prospects the future research directions of GIS4LLM in 5 aspects: collaborative understanding of multimodal spatio-temporal data, balancing generalization with depth, enhancing interpretability and credibility, transitioning to embodied intelligence and edge intelligence, and the development of intelligent and universal GIS analysis. This paper provides inspiration for achieving mutual empowerment between LLM4GIS and GIS4LLM.
China has launched recently its national 3D mapping program to build 3D realistic geospatial landscape model (3DRGLM), and has been considered as an important step towards establishing Digital China and transforming our geospatial industry. It is a huge spatio-temporal information engineering with many complex technical and management factors, and is becoming a challenging task for governmental agencies and academy societies. The paper has introduced the motivation and fundamental concepts of the 3DRGLM, i.e., moving from traditional 2~2.5D cartographic products to 3D realistic geospatial landscape products, from simple data supply to spatio-temporal empowerment, from digital to intelligentized mapping. The overall architecture of national 3DRGLM was then proposed and it has a number of key components, such as the data product system for generating geospatial entities, geospatial scenes and realistic geospatial scenes, the service system designed to support spatio-temporal connection, computing, and intelligence, as well as its digital-intelligentized hybrid technological support system. The 3DRGLM has some special key technological issues, including geospatial entity modeling, stereo reconstruction of muti-dimensional real space, realistic description and geospatial temporal knowledge service. Finally, this paper discussed five typical application scenarios of 3DRGLM, including application of 3DRGLM in digital economy, digital governance, digital living, digital culture, and digital ecological civilization. In order to achieve a successful establishment and an in-depth application of national 3DRGLM, it is necessary to carry out strategic planning, enhance scientific and technological innovation and promote cross-border and multi-disciplinary collaboration.
Geographic entity modeling is a central task and crucial component in 3D realistic geospatial scene reconstruction of China. It involves the abstraction and digitization of real-world geometric spaces, along with its associated attributes and relationships, to generate 3D realistic geospatial scene model. Firstly, geographic entity modeling techniques and methods are thoroughly considered and summarized from the perspective of the construction and application in the 3D realistic geospatial scene of China. The evolution of geographic entity modeling is outlined across three stages: two-dimensional plane, three-dimensional surface, and 3D realistic geospatial scene. Secondly, this paper elaborates on the connotation and extension of the 3D realistic geospatial scene geographic entity modeling. Reviewing research progress in geographic entity modeling, including geometric modeling, attribute modeling, relationship modeling, and temporal modeling in geographic entities. Then, taking the temporal management for geographic entities and the application of spatiotemporal association techniques as examples, the application of 3D realistic geospatial scene geographic entity modeling is presented. Finally, this paper explores future directions for 3D realistic geospatial scene geographic modeling technology, focusing on cross-domain collaborative interaction fusion modeling, pan-spatial integrated and unified modelling, multi-granularity panoramic association modelling, and adaptive intelligent temporal modeling.
3D modeling is one of the core technologies in the digital earth, and it plays a significant role in the development of digital cities, digital earth, and digital economies. Currently, many countries have initiated various 3D modeling projects to support natural resource management, urban planning, emergency response, and sustainable development. At the same time, these projects have also promoted the development of 3D modeling technology. This article analyzes the research and application of 3D mapping across different countries, focusing on four aspects: “needs assessment” “tasks” “technologies” and “applications” based on the fundamental definitions and connotations of 3D mapping. By summarizing and analyzing typical 3D modeling engineering, products, technology, standards, and applications in these regions, the article explores insights into the construction of 3D realistic geospatial scene in China and provides references and insights for the development of realistic model projects domestically.
The existing single-modality pre-trained models are still susceptible to the unorderness and sparisity of point clouds, making it difficult to meet the requirements of diverse downstream tasks for 3D real-scene construction. To further enhance the performance of pre-trained models, this paper proposes a cross-modal contrastive masked autoencoders pre-training method that uses the 2D image modality to assist the 3D point cloud modality, based on multi-modal, contrastive learning, and masked auto-encoding pre-training theories. The network mainly consists of two branches: the intra-modal branch, which is a contrastive masked autoencoder sarchitecture for learning more comprehensive feature information, and the cross-modal branch, which is a 2D/3D cross-modal contrastive learning architecture for improving robustness to unordered and sparse data. To verify the effectiveness of the proposed method, we conduct a series of downstream task experiments, such as masked point cloud reconstruction, classification, few-shot classification, and segmentation, on datasets including ShapeNet, ModelNet40 and ScanObjectNN. The results indicate that the proposed method exhibits superior transferability compared to existing methods.
Guardrails are critical component of highway infrastructure, and their deformation can significantly impair their protective function. Existing methods for detecting guardrail deformation primarily focus on extracting and modeling guardrails from mobile mapping point clouds. However, these methods often lack in-depth semantic feature analysis of the guardrails and fail to accurately reflect the deformation conditions in reverse modeling. To address these limitations, this study proposes a high-precision parametric modeling framework for guardrails driven by mobile laser scanning (MLS) point clouds and guided by building information modeling (BIM). The framework involves: ① Automated extraction and instantiation of guardrail elements from MLS data; ② Solving structural parameters of guardrail using the random sample consensus (RANSAC) algorithm; and ③ Introducing B-spline curve-based parametric modeling of guardrails, creating realistic guardrail models through modular modeling in Dynamo. Moreover, evaluating guardrail deformation mileage using a curvature and vector-constrained trajectory detection mechanism. This approach enhances the precision of component-level guardrail models, providing a safe and efficient solution for maintenance inspection of various guardrail types. Experimental results demonstrate a guardrail recognition accuracy of 98.7% on highway guardrail. All deformed guardrails on the selected test sections were detected, with localization errors less than 2.2 meters, meeting practical inspection requirements and mitigating traffic safety risks associated with guardrail deformation.
In recent years, volumetric rendering-based 3D implicit representation methods have achieved notable success in geometry and radiance reconstruction. These methods can simultaneously reconstruct geometric structures and render photorealistic images. However, in outdoor scene reconstruction, the uneven spatial distribution of captured images often leads to suboptimal reconstruction results. To address this issue, this paper proposes a method that represents the spatial distribution of images using sampling points. Through Delaunay triangulation and a centroidal Voronoi tessellation strategy, unevenly distributed sampling points are iteratively transformed into a uniform distribution, thereby achieving uniform allocation of the implicit representation and ensuring the quality of the implicit reconstruction. Furthermore, this concept of spatial transformation of sampling points is embedded into implicit representation frameworks of different paradigms, enabling multi-scale implicit reconstruction of outdoor scenes. Experimental results on implicit 3D reconstruction of real-world outdoor scenes demonstrate that the proposed centroidal Voronoi tessellation transformation ensures the reconstruction accuracy of building structures and significantly improves the multi-scale reconstruction results of existing implicit representation methods.
In this paper, we use the sea surface height data of the China ocean altimetry tandem satellites (COATS) and the SWOT satellite of the United States to compute the disturbed gravity vector (vertical deflection and disturbed gravity). The abnormal data of COATS satellite are filtered by using the regional inverse distance weighted iterative method, and the outliers are eliminated effectively. According to the characteristics of the two-satellite tandem mode and wide-swath mode, a method based on multi-directional sea surface gradient estimation is adopted to solve the grid vertical deflection and the disturbed gravity in the sea area is retrieved. The calculation of the Western Pacific region shows that the COATS satellite makes full use of the near real-time sea surface height in the cross-orbit direction in the calculation of the vertical deflection, and can realize the calculation of the local 1′×1′ vertical deflection in the tandem-following mode, compared with the 1′×1′ vertical deviation calculated by EIGEN-6C4 model, the standard deviation of north-south and east-west component is respectively 1.65″ and 2.3″. SWOT satellite is the first wide-swath interferometric altimetry satellite in the world. The accuracy of its north-south component and east-west component is consistent with that of EIGEN-6C4 model, the standard deviation of north-south and east-west components is about 1.8″, which fully reflects the advantage of uniform observation of wide-swath. Compared with the gravity data of SIO 32.1, the standard deviation of gravity anomaly respectively derived from COATS 5′×5′, SWOT 1′×1′ and two satellite fusion 1′×1′ data is 6.4, 5.2 and 4.9 mGal. Considering that COATS and SWOT are currently the satellites which truly achieve 1′×1′ sea surface height surveying, the combined COATS and SWOT satellite data will get more intensive vertical deflection and disturbed gravity, which will further improve the precision of the gravity field in the sea area.
GNSS-R technology has been applied to monitor soil moisture (SM) and freeze-thaw (F/T) conditions. However, GNSS-R SM retrieval is currently blank in high altitude areas, and GNSS-R F/T results are only available for short time series. Therefore, a five-year GNSS-R SM and F/T record for the Xizang Plateau region was developed in this study. The grid-by-grid average root mean square error (RMSE) and correlation (R) of CYGNSS SM were 0.064 cm3/cm3 and 0.53, respectively. The grid-by-grid average detection accuracy of CYGNSS F/T was 85.5%. The validation results from independent stations show that the RMSE and R for CYGNSS SM are 0.059 cm3/cm3 and 0.56, respectively, and the classification accuracy for CYGNSS F/T is 83.8%. This is comparable to the accuracy of SM and F/T of SMAP in the same period. This study will fill the gap of GNSS-R SM and F/T records at high altitudes. It is worth noting that the number of available days for CYGNSS is significantly higher than that of SMAP. The number of available days for CYGNSS SM is 47.0% higher than that of SMAP SM, and the number of available days for CYGNSS F/T is 14.7% higher than that of SMAP F/T. Moreover, this study also developed an empirical fusion framework for CYGNSS and SMAP. The average RMSE of the fused CYGNSS and SMAP SM (i.e., the CYGNSS-SMAP SM) is 0.056 cm3/cm3, with an R of 0.60. This is an 18.8% improvement in accuracy and a 34.7% improvement in the number of available days compared to the existing SMAP SM. The average accuracy of the fused CYGNSS and SMAP F/T (i.e., CYGNSS-SMAP F/T) is 89.8%. This is a 10.0% improvement in accuracy and a 10.2% improvement in the number of available days over the existing SMAP F/T. Fusing the CYGNSS reflectometer and SMAP radiometer can provide higher accuracy and continuity of SM and F/T in the Xizang Plateau region. The study also demonstrates the fairly good monitoring capability of the spaceborne GNSS-R technique in the high-altitude region.
The optical texture material of non-lambertian artifacts such as porcelain is a critical foundation for the realistic visualization and rendering of 3D models. Due to the color differences such as highlights caused by non-lambertian surfaces from different viewing angles, existing photogrammetric texture reconstruction methods crudely alleviate these color differences through color equalization and feathering; Neural radiance field methods based on implicit spatial representation rely solely on multi-layer perceptron to blur spatial fitting for different reflection directions, leading to insufficient realism. Therefore, this study proposes a differentiable rendering method for non-lambertian material reconstruction using close-range photogrammetry geometry images. By employing a handheld stereo scanner and controlled lighting conditions, the method accurately acquires oriented parameters, model geometry, and initial color texture of images. Material properties such as albedo, roughness, and normal perturbation are treated as planar tensors to be optimized, mapped to screen space via rasterization to obtain rich geometric images with coordinates, normal, texture coordinates, and material attributes; Drawing on deferred shading techniques, the geometric image simulates pixel shaders in screen space, achieving differentiable rendering and inverse feedback optimization of unknown material information using the Cook-Torrance microfacet model for highlights. Experimental results on typical non-lambertian targets demonstrate that the proposed method achieves a structural similarity index (SSIM) better than 0.85 compared to real images, improving rendering accuracy by 8.7% and 7.2% over 3D Gaussian points and multi-resolution hash encoding based on implicit space representation, respectively, significantly enhancing the realism of the model.
Temperature field modelling is crucial for bridge construction and refined management. However, the temperature difference in mountainous areas varies greatly, and the existing temperature field simulation methods based on finite element analysis face challenges due to sparse results and inhomogeneous spatio-temporal distributions, making it difficult to describe bridges' temperature changes accurately. This article proposes a 3D modelling method for temperature field of mountain bridges coupled with numerical simulation and spatio-temporal interpolation fusion. First, a numerical simulation model of the solar radiation temperature field for mountainous bridges is established; Second, the temperature simulation results are registered with the bridge voxel model, and spatio-temporal interpolation fusion is adopted to model the temperature field of mountainous bridges. Next, efficient visualization of the temperature field of mountainous bridges is achieved based on ray casting. Finally, a large-scale steel truss cable-stayed bridge in Ganzi, Sichuan province, China, is selected as a case study for experimental analysis. The experimental results show that the proposed method can effectively complement the temperature field simulation results to accurately depict the spatio-temporal distribution and variation patterns of the temperature field of mountainous bridges. The interpolation fusion accuracy was improved by 22.11% and 7.38% compared to spatial and temporal interpolation, respectively. The visualization frame rate has increased by 36.4% through volume rendering, which can provide key data support for subsequent intelligent construction and refined management of mountainous bridges driven by digital twins. Also, it offers crucial reference value for the realistic representation of environmental parameters of the component-level 3D real scene model.
The onboard mobile mapping system is affected by pre-calibration errors, installation errors, and camera projection errors, leading to mismatches between the acquired point clouds and panoramic image sequence data. In highway scenarios, existing calibration methods face challenges due to the linear distribution of highways and the significant depth-of-field differences in panoramic images. These factors cause uneven distribution of control points with high correlation, making it difficult to adequately calibrate the camera's exterior orientation parameters, especially the translation components related to the vehicle's roll angle and travel direction. To address this issue, this paper proposes a self-calibration method for panoramic cameras in highway scenarios, integrating point-line features with depth information. The method improves the sensitivity of the adjustment model to translation components through inverse distance weighting, extracts point-line features from highway infrastructure, and constructs a joint adjustment model to reduce the impact of depth differences on uneven feature distribution, ensuring high sensitivity to errors in all directions. Experimental results show that the proposed method can accurately calibrate the exterior orientation parameters of the panoramic camera, with calibration accuracy better than 3 pixels, and the calibration model demonstrates high responsiveness to errors in all directions.