Road lane lines are a key component of high-precision maps. The massive driving recorder videos mounted on online ride-hailing vehicles are real-time observations of road information and are an important data source for more economical lane line data extraction. Based on the massive Didi ride-hailing driving record videos, this paper explores the feasibility of the road lane line data extraction method based on the LaneNet deep network model. This method first uses the LaneNet network model to perform semantic segmentation on each frame of the video image, and then predicts The perspective transformation matrix realizes the fitting and extraction of the lane line pixel position. In the experimental analysis, simulation data and Didi driving recorder data in complex scenes were used to evaluate the experimental results. The experimental results show that the model used in this article has better lane line extraction performance in vehicle video images.

A method for extracting underground target based on local entropy of discrete point clouds after voxelization of 3D ground penetrating radar (3D GPR) data is proposed to address the problem of insufficient exploration of 3D spatial information by 3D ground penetrating radar and the data processing mainly based on analysis and interpretation of 2D slice images. Firstly, the obtained 3D ground penetrating radar data was voxelized into discrete 3D point clouds. Then, the local entropy of the voxelized point cloud for the entire region were calculated. The soil background and underground targets were distinguished by classifying them from multiple dimensions through support vector machine (SVM). Finally, taking the underground environment of urban roads as the research object, it is used to conduct experimental analysis using measured data. The experimental results show that the accuracy of this method in extracting underground targets is as high as 84.2%, and the missed detection rate of underground targets is as low as 9.8%. This method is accurate and effective, providing a new approach for 3D ground penetrating radar to extract underground target.

SLAM in complex environments is one of the challenging tasks in the field of robot autonomous navigation research. The drastic changes in the surrounding spatial environment can lead to drift and overlap in SLAM mapping, thereby reducing mapping accuracy. To address this issue, this paper proposes an effective adaptive optimization method for point cloud thresholding, improving the applicability of SLAM algorithms in complex environments. The algorithm calculates the depth information of the point cloud in real-time and adaptively optimizes the effective point cloud threshold based on the fluctuation of depth information and the coefficient of variation of point cloud distribution, thereby achieving closed-loop control. Experiments show that the proposed threshold adaptive optimization method significantly improves the mapping performance of fast and direct LiDAR with inertial odometry algorithms in complex environments. It corrects the odometer coordinate errors of this algorithm in narrow environments and reduces loop closure positioning errors by 7.5%.

In the SAR image ship target detection task, due to factors such as complex coastal background and multi-scale ship targets, ship targets may have low detection accuracy and missed detections during the detection process. In response to the above issues, this article proposes an improved SAR image ship target detection model based on YOLOv8s, and conducts experimental verification by SSDD and HRSID datasets, with better performance than other classical algorithms.

Surface deformation, a common geological hazard, frequently triggers landslides and debris flow disasters in complex mountainous terrain, posing a long-term threat to personal and property safety. However, the effective prevention of surface deformation remains challenging due to the lack of comprehensive monitoring and quantitative analysis of its driving factors. To accurately identify the dominant factors of surface deformation in mountainous areas, this study uses InSAR technology with ascending and descending orbit SAR datasets to determine the spatiotemporal distribution of surface deformation in Liuku county, Yunnan province, from January 2016 to February 2022. After verifying the reliability of the monitoring results, a geographic detector quantifies the contributions of driving factors and reveals their interactive mechanisms. The research results show that:①Surface deformation in Liuku county is primarily subsidence, mainly concentrated in the main urban area, with deformation rates ranging from -44 to 30 mm/a. ②Rainfall, elevation, and soil type have a strong spatial correlation with surface deformation in the study area, and interactions among driving factors enhance this spatial correlation. ③Deformation in the main urban area is primarily characterized by vertical subsidence, while horizontal deformation occurs in Lijiatian and Damikou village.

This article proposes an optimization algorithm based on the wavelet gated recurrent neural network autoregressive moving average model(GRU-ARMA)on the basis of the long short term memory neural network autoregressive moving average model(LSTM-ARMA).Firstly,it decomposes the original InSAR time series into trend and noise components using wavelet denoising,employs the GRU recurrent neural network for rolling prediction of the trend component,and utilizes the ARMA model for forecasting the noise component.Subsequently,the sum of the predicted values of the trend and noise components is used as the total time series prediction value,thereby enhancing the prediction accuracy at each monitoring point.Finally, this paper selects multiple points(CP0001,CP0007,and CP0009)in the most severe subsidence area of the Argan mining area from 2020 to 2023 as examples for study.It demonstrates that the prediction accuracy of the wavelet-optimized combination model surpasses that of the traditional single models GRU/LSTM.Furthermore,compared to the LSTM-ARMA model,the predictive performance of the wavelet-optimized GRU-ARMA model is more stable,indicating it as an effective approach and method for surface subsidence prediction.

In response to the common issue of extensive missing data in optical remote sensing data and the insufficient consideration of the spatial correlation of geographic data in existing algorithms for data reconstruction, this paper fully utilizes the spatio-temporal correlation between geographic spatial data and proposes a reconstruction method that combines random forest(RF) and geographically weighted regression(GWR), termed as RF+GWR. Using normalized difference vegetation index (NDVI) from GF-4, MODIS land surface temperature (LST), and GF-4 reflectance data as experimental materials, the universality and missing data reconstruction performance of the RF+GWR method are evaluated. Experimental results show that, under different cloud masking levels as set in the paper, compared to K-nearest neighbor (KNN) and RF methods, the RF+GWR method exhibits varying degrees of improvement in reconstructing missing data of GF-4 NDVI, MODIS LST, and GF-4 band reflectance data. The maximum improvements in root mean square error, mean absolute error, and coefficient of determination are 33.07%, 30.19%, and 7.06%.

Aiming at the problem that the current map data model is difficult to effectively represent the complex correlation and dynamic change characteristics of cyberspace elements, this paper analyzes the concepts and characteristics of cyberspace map and cyberspace spatio-temporal entity, and takes cyberspace spatio-temporal entity as the basic unit of cyberspace map representation. Secondly, the cyberspace spatio-temporal entity representation model is proposed to be established at three levels, namely entity layer, feature layer and change layer. Finally, it focuses on the visualization method of dynamic changes and relationship characteristics of cyberspace spatio-temporal entities, and takes typical national cybersecurity events as an example for visualization technology implementation, which verifies that the representation model proposed in this paper can be effectively applied to the representation of multi-dimensional, correlated, and dynamic characteristics of cyberspace, and provides a reference for the subsequent research and application of visualization of cyberspace maps.

Accurate and reliable prediction of PM_2.5 concentrations is important for the public to effectively avoid air pollution and for governmental strategy development. However, due to the dynamic nature of atmospheric flows, the prediction of PM_2.5 concentration is characterized by great uncertainty and instability, making it difficult for a single model to efficiently extract spatio-temporal correlations. In this paper, a robust prediction system is proposed to realize accurate single-step, multi-step and trend prediction of PM_2.5 concentration. First, the article adopts a correlation analysis method to screen the meteorological and pollutant spatial information that can help predict the pollutant concentration in the target city. Then, the feature extraction capability of DenseNet is utilized to extract spatially relevant features from pollution and meteorological datasets from multiple cities; the ConvLSTM layer combines the temporal and spatial features of the pollutant data, and extracts the spatial and temporal features in order to achieve accurate pollutant prediction. Finally, the performance of the proposed prediction system is comprehensively evaluated by four accuracy indicators and three prediction experiments. In addition, the pilot study shows that the prediction system has good application prospects in early warning, regional prevention and control of air pollution, and its accuracy and stability are better than those of various baseline models.

Loess sinkholes are a unique type of geological hazard that is widespread across the Loess Plateau. Their prevention and control are essential considerations in construction projects within this region. Based on a modified RUSLE, this study extracts 12 different types of feature factors from multiple data sources, including DEM, precipitation, surface cover, and vegetation index. Two prediction models, CNN and DNN are constructed to predict areas prone to loess sinkhole development. The results of the two models are compared and analyzed to provide reference for the prevention and control of sinkhole hazards, construction projects, and soil and water conservation in loess areas. The findings show that both the CNN and DNN models achieve an accuracy rate of over 80% and an F1 score of over 83%, indicating their effectiveness in predicting areas prone to loess sinkholes. The CNN model achieves an accuracy of 83.25% and an F1 score of 85.18%, which are 2.63% and 1.56% higher than those of the DNN model, respectively. This demonstrates the superior generalization ability and detailed performance of the CNN model. Analysis of the prediction results indicates that loess sinkholes develop more strongly in valley areas, less so on flat terrain, and are influenced to some extent by human activities.

As an important ecological security barrier area in China, regional vegetation cover changes directly reflect the health of the ecosystem, so monitoring regional vegetation cover changes and analyzing their driving factors are conducive to realizing the sustainable development of the Three Rivers Headwaters region ecosystem. This study is based on the MODIS NDVI dataset from 2000 to 2022, combined with the mean annual temperature and mean annual precipitation data, and explores the spatial and temporal evolutionary characteristics of vegetation cover and its relationship with the climate driver during the last 23 years in the Three Rivers Headwaters region by using the image element dichotomy, Sen+Mann-Kendall trend analysis, and bias correlation coefficient analysis. Results show that: ①The vegetation cover of the Three Rivers Headwaters region show a high spatial distribution in the southeast and a low distribution pattern in the northwest. The vegetation cover show a significant linear growth trend from 2000 to 2022, with a growth rate of 0.001/a. ②Over the past 23 years, the vegetation cover of the Three Rivers Headwaters region has increased by 55.06% and decreased by 31.36%, with the increase in vegetation cover significantly higher than that of the decrease, and the overall vegetation cover is in the stage of gradual recovery.③The partial correlation analysis showe that the regional vegetation cover is positively correlated with the average annual temperature and average annual precipitation, with the correlation with the average annual temperature of 0.02 and the correlation with the average annual precipitation of 0.29, and precipitation is the main factor for the increase of vegetation cover. This study reveals the dynamic change and distribution pattern of vegetation cover in the Three Rivers Headwaters region over the past 23 years, which can provide a scientific basis for regional ecological protection and restoration.

In view of the current practical problems such as poor consistency,uneven quality,and weak customization and derivation ability of multi-source fundamental geographic information data,on the basis of analyzing the current situation of “one map” related to geographic information,the concept of “one map” data model of fundamental geographic entity base is proposed,and the mathematical definition of the model and the data structure with geographic entities as the core are given,and the model application architecture covering key links such as analysis and collation of data,merit-based fusion of data and flexible customization of products is designed. The experimental verification shows that the model is feasible and effective. The research results of the model can realize the efficient integration and utilization of multi-source geographic information data,and at the same time provide an efficient and intensive technical path for the construction of new fundamental surveying and mapping,which is expected to promote the innovation of geographic information data production,updating,management and application mode,and provide accurate,reliable,unified and efficient geographic information services for the development of national economy and national defense construction.

This paper proposes a vehicle target tracking and localization algorithm using roadside LiDAR based on Kalman filter. Firstly, the position calibration of roadside LiDAR is performed based on reflective strips; the background point cloud map is constructed, and the background differential filtering is performed on the point cloud data collected by roadside LiDAR, and only the foreground point cloud data such as motor vehicles, non-motor vehicles and pedestrians are retained; the density-based DBSCAN clustering is performed on the foreground point cloud, and the point cloud is divided into point cloud clusters; then the SVM classifier is trained, and the extracted point cloud cluster features are used for SVM-based target classification recognition; finally, the inter-frame association of targets is realized by the nearest neighbor algorithm, and the trajectories of target objects are predicted and updated by linear Kalman filtering. In the part of accuracy evaluation, the calibration results and the algorithm positioning results are analyzed for accuracy using relative position error and absolute trajectory error. From the analysis results, it can be seen that the optimized smooth trajectory obtained by the trajectory estimation method with Kalman filter is more consistent with the actual operation of the vehicle and can significantly improve the target tracking accuracy.

In recent years, there have been an increasing number of high-rise heterogeneous buildings in coal mining subsidence areas. Traditional point measurement methods can no longer meet the requirements for the deformation monitoring accuracy, mainly due to the following two reasons: ①It is not easy to select feature points; ②It is difficult to accurately measure the feature points. Regarding these two problems, based on point cloud data and by using the target central axis fitting method, this paper proposes a tilt measurement method for high-rise heterogeneous buildings based on surface mapping. Verified by practical examples, the tilt obtained by this algorithm for the structures is 22.97 mm/m, while the tilt obtained by the traditional measurement method is 21.53 mm/m, with a relative error of 6.9%, which meets the accuracy requirements for tilt measurement. The research results can provide technical support for the rapid and efficient tilt monitoring of similar high-rise structures in mining areas.

The cableway towers is a critical component of cableway systems. In the realm of cableway inspection systems, monitoring focuses primarily on wire ropes and elevator cars. However, gaps persist in the processing and analysis of monitoring data specific to cableway towers. This paper presents findings from the real-time monitoring platform of the Wanghailou passenger cableway of Yimengshan, which continuously observes cableway towers. Script tools are employed for static calculations of extensive GNSS observation files. We apply the K-nearest neighbor anomaly detection algorithm to evaluate the static calculation results. By integrating data from sensors such as inclinometers and meteorological instruments, and considering anomaly scores, we validate the abnormal detection dates. The validated static results are then utilized as the displacement data for the bracket, which when combine with the inclinometer's attitude data, are allowed for an analysis of the bracket's long-term changes. The findings reveal significant diurnal variations in the data from the cableway towers, with varying degrees of change at different positions. These results offer valuable insights into the deformation patterns and safety monitoring of cableway towers.

Scene semantic segmentation based on deep learning and computer vision technology is currently a hot research topic. This paper proposes a 3D real scene semantic segmentation framework that includes “scene input-preprocessing-model inference-semantic segmentation”. By transforming the 3D real scene as input into multi-view 2D images through orthogonal projection, segmentation inference is carried out, and segmentation masks are generated and further processed,achieving the object selection, singulation, and semantic processing of 3D real scene.The experiments show that the method has good semantic segmentation performance and efficiency.

It has carried out pilot work on new fundamental surveying and mapping,and has completed more than 10000 km of holographic roads in the city in recent year,covering the main roads in Shanghai.With the rapid development of intelligent driving,accurate detection and identification of road traffic signs is essential in constructing intelligent driving road framework data.In actual scenarios,many factors will bring challenges to the detection and recognition of traffic signs,such as motion blur,sunlight conditions,and shooting angles.So,the paper proposes an improved traffic signs detection algorithm based on YOLOv8.The GAM attention mechanism is introduced in the Neck part of the model to enhance the characteristic information of traffic signs.The Wise_IoU loss function has improved the training performance of the dataset compared to the original loss function.Compared with the model without any optimization,the accuracy and mean average precision increased by 6.5% and 4.1% respectively,which has practical application value.

Addressing the issues of the lack of formal expression in existing achievements of the bulildings 3D model, the uniformity of facade information scales acquired by different disciplines, and the absence of specific norms for collection content, this paper constructs a normative description of building facade feature elements and their relationships from three scales: block, individual building, and wall. It proposes a multi-scale graph structure representation method for building facade features based on the hybrid graph model, which realizes multi-scale feature description of building facades. An experimental area is selected, and relying on Neo4j graph database technology, the construction of a multi-scale feature model for buildings is achieved through spatial multi-scale construction, facade feature entity construction, and relationship construction. The results indicate that by storing relevant information about the exterior facades of urban buildings in a graph database, data integration can be achieved, making originally dispersed and difficult-to-manage data centralized and orderly. The establishment of a graph structure database promotes data sharing and improves the efficiency and utilization of information flow.

Shouning area in Fujian province is prone to geological hazards, severely affecting the lives and properties of the local residents. In this study, persistent scatterer interferometric synthetic aperture radar (PS-InSAR) technology is applied to process data from 100 ascending orbit Sentinel images covering the study area from January 2018 to May 2020. Deformation information and deformation rate profiles of the study area are obtained, identifying deformation anomalies and confirming 6 deformation targets through field investigations. The results indicate that during the study period, the deformation rate ranged from -16 to 10 mm/a, with cumulative deformation ranging from -54 to 34 mm. The average deformation rate along profiles varied from -10 to 5 mm/a. Overall, the study area's surface remains relatively stable, with deformation possibly occurring in some localized areas. The identified 6 target areas mainly exhibit deformation characteristics such as landslides, unstable slopes, and collapses, with consistent deformation trends in the surrounding areas, generally sinking away from the line of sight. Precipitation emerges as the primary factor influencing deformation. The research findings provide valuable data support and decision-making basis for geological hazard prevention and control in the Shouning area of Fujian province.

Food security is the foundation for the long-term stability and prosperity of a country.Leveraging the advantages of high spatial resolution remote sensing imagery, which covers a wide range and possesses both spectral and textural information, this paper combines textural feature extraction based on gray-level co-occurrence matrices with geometric feature extraction using the Hough transform line detection algorithm.Focusing on the characteristic manifestations of agricultural greenhouses in the imagery, a local area in Ningbo is selected as the study area to conduct information extraction experiments and validations.For sub-meter high spatial resolution remote sensing imagery, an average extraction accuracy of approximately 90% can be achieved, effectively reducing the impact of spectral reflectance differences on recognition accuracy.Compared to object-oriented extraction methods based on image segmentation and neural network extraction methods based on deep learning, the method proposed in this paper exhibits higher feature intuitiveness and comprehensibility, reduces computational complexity and requirements for sample volume, and is conducive to rapid and accurate interpretation and extraction of agricultural greenhouses within cultivated land areas.

Point cloud segmentation is an important part of airborne LiDAR point clouds processing. The regional growth method is a traditional classical method of point cloud segmentation, but it usually takes the point as the unit to grow, which leads to the problems of slow segmentation speed and unstable segmentation performance. To solve these problems, this paper proposes a point cloud optimization fast segmentation algorithm combining K-means clustering method and regional growth method. First, K-means clustering is carried out for point cloud to obtain object primitives and calculate centroid points, judge whether the centroid points of each object element meet the angle and height difference threshold, and realize point cloud filtering based on centroid points. Then, the ground object primitives are traversed, and the normal vector angle and distance are calculated for the adjacent points within the object primitives to determine whether they meet the growth conditions of the regional growth threshold. The iteration is repeated until the end of the segmentation. Three groups of point cloud data from different regions are used for experimental analysis. The experimental results shows that the segmentation accuracy of this method could reach 86.19%, which is greatly improved compared with the traditional K-means clustering method and regional growth method airborne LiDAR point cloud segmentation accuracy. In addition, this method can significantly improve the computational efficiency compared with the traditional regional growth method.

In response to the need for refined management of island development and utilization in Guangdong province, the application of high-resolution satellite remote sensing images and unmanned aerial vehicle images, combined with GIS methods and artificial intelligence interpretation technology, has been incorporated to carry out research and application of remote sensing intelligent interpretation platform for island development and utilization. The platform constructs an intelligent interpretation sample library and a change detection sample library based on island-based methods. It trains remote sensing intelligent interpretation models for island development and utilization, as well as change detection models for island development and utilization, using deep learning. By monitoring and analyzing remote sensing data from different periods, the platform outputs thematic products and analysis reports. The research results improve the automated processing efficiency and monitoring accuracy of massive remote sensing data, and provide strong technical support for the development, protection of sea areas and islands in Guangdong province.

In order to quickly identify the suitable site selection range for engineering construction in the field environment and better serve emergency support needs, this article starts with the impact of key geological elements on engineering construction. By analyzing the mechanical properties of rock and soil, physical and chemical reactions of water bodies, structural development laws, types and susceptibility of geological disasters, resource distribution and availability, the influence mechanism of key geological elements in engineering construction site selection is explored. Based on the suitability evaluation of engineering construction, the range of terrain and geological element indicators is quantified into 5 levels, and the analytic hierarchy process is adopted to quantitatively evaluate the impact weight of each element on surface engineering construction. The construction of geological element grid data structure has been achieved, and the division and optimization of engineering construction site selection areas have been completed through suitability evaluation. The experimental results show that the site selection method for surface engineering construction in this article fully considers the influence of geological factors on the site selection of engineering construction. The model is scientifically reasonable and can efficiently complete the site selection range planning under the premise of known engineering demand parameters.

Forest land is an important contributor to green carbon in terrestrial ecosystems and plays a crucial role in achieving the “dual carbon” goals. This article focuses on forest land types and collects multi-source remote sensing data from around 26.67 hm² of forest land in Jinze town. Combined with ground survey sampling and detection techniques, it evaluates the dynamic carbon sequestration capacity of typical forest land, clarifies the carbon storage account, estimates carbon sequestration potential, and combines digital twin technology to build a visual demonstration system, providing a quantifiable, traceable, and visual theoretical and practical method for decision-making.

As typical fragile ecosystems, dryland systems exhibit nonlinear evolution of their surface states, profoundly affecting ecological balance and human well-being. However, existing studies have yet to fully uncover their complex dynamic evolution processes. Based on monthly time-series products of vegetation and soil endmember fractions from 2001 to 2022, this study employs a trend and breakpoint detection algorithm based on discrete wavelet transform to identify the evolution paths of vegetation-soil states in dryland systems and analyze their dynamic changes. The results indicated that: ①From 2001 to 2022, the fractions of photosynthetic vegetation and non-photosynthetic vegetation endmembers in China's dryland systems showed a significant increasing trend, while the fractions of soil endmembers exhibited a significant decreasing trend.②The sudden increase, A-shaped increase, and V-shaped increase state evolution paths for photosynthetic vegetation accounted for 9.9%, 16.8%, and 21.9% of the total dryland area, respectively, while the corresponding state evolution paths for non-photosynthetic vegetation accounted for 9.1%, 12.6%, and 28.8%, respectively.③Overall, vegetation restoration in dryland had promoted the reduction of soil exposure, but in local areas, water scarcity and human activities had led to vegetation degradation. This study provides a new perspective for understanding the surface state evolution and its response mechanisms in China's dryland systems and offers scientific evidence for land degradation monitoring and ecological restoration.

Core function of nature reserves is to provide ecosystem services,thus proposing suggestions on hierarchical management for nature reserves from the perspective of ecosystem service functions is of great significance for improving the hierarchical agency system of nature reserves. Current status of hierarchical management of nature reserves at or above the provincial-level and its response to ecosystem service functions in Zhejiang province from 2010 to 2020 are researched in this paper,and the results indicate that:①ecosystem service functions are basically consistent with the hierarchical management of nature reserves,various ecosystem service functions of national parks,nature reserves,and nature parks show a clear downward trend; ②growth rate of ecosystem service functions of national nature reserves show a downward trend,growth rate of ecosystem service functions of provincial nature reserves show an upward trend; ③ecosystem service functions per unit area in 3 provincial nature parks,including Pingyang Mantian provincial forest park,etc.,is higher than that of national parks,thus it is suggested to consider upgrading its management level.

The paper elaborates on the technical route for rapid construction of terrain level realistic 3D models, developes a multi-source feature fusion based urban level realistic 3D reconstruction technology,improves model method of point cloud feature extraction and topological construction of building tops and stereotypes,automatic matching of tilted images and top surface models,realizes 3D reconstruction of complex roofs,improves the refinement and geometric accuracy of model textures,describes the process of component level realistic 3D automated modeling,indicating the real-life 3D model application under the public security emergency management,national spatial planning,natural resource management and other service scenarios of Ningbo, and the next steps of thinking.

Under the background of imperfect current real estate registration inventory data and inadequate information sharing in China, this paper proposes a technical method for cleaning and integrating real estate inventory data. Based on the actual situation of the comprehensive cleaning project of Changsha real estate inventory data, this paper clarifies the current situation of real estate registration data, proposes the idea and technical scheme of real estate inventory data cleaning and integration, adopts the data cleaning method combining “offline+online”, and uses database SQL, ArcGIS and other technical methods to realize the whole process of data extraction and task distribution, data cleaning, quality inspection and inventory storage. The whole process of cleaning work is standardized and managed, greatly improving the efficiency of real estate inventory data integration, and providing reference for other cities to carry out real estate inventory data cleaning and integration.

Due to the limitations of poor data completeness and low knowledge utilization, it remains a challenge to accurately integrate multiple sources of heterogeneous disaster information to comprehensively evaluate the susceptibility of landslides. Therefore,a mixed framework for mapping the susceptibility of landslides with regional statistical constraints is proposed. Firstly,environmental factors such as topography,geomorphology,geological structure,meteorology and hydrology,and human activities were selected based on expert knowledge,and the relevant data is pre-processed such as cleaning and normalization. Then,from the perspective of “mechanism-geography-physics-mathematics”,the feature mapping of disaster data is constructed,and a balanced sampling of positive and negative samples is carried out for historical disaster sites and non-disaster sites. On this basis,the administrative units are adaptively aggregated using a multi-scale spatial unit division method,and the study area is divided into a series of homogeneous sub-regions according to slope and average annual rainfall. Finally,under the constraint of regional statistics,a random forest model is constructed for sample training and susceptibility mapping. The experiments show that the proposed hybrid framework improves the accuracy of landslide susceptibility assessment by at least 9%.

Realistic 3D China, as a novel means and method for comprehensive analysis, multi-dimensional display, and trend analysis of economic and social factors under modern technological conditions, is playing a greater role in accelerating the construction of an economic layout and national spatial system that reflects the requirements of high-quality development, especially in the development of new productive forces. The 3D spatial information of urban underground pipelines is a component of the construction of realistic 3D China. This article combines multi-source data fusion technology based on spatial analysis, integrated spatio-temporal data collaborative collection technology of above ground and underground, and component level pipeline model construction technology based on BIM to comprehensively consider the technical means of 3D construction of underground pipelines. It has certain reference and application value for supporting and empowering the construction and development of 3D scenes for the digital transformation of super large cities.