Method for reconstructing ground surface model of chinese classical garden, simulating rainwater runoff and identifying risk based on three-dimensional point cloud

Abstract

This invention discloses a method for reconstructing a surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point clouds. The method includes the following steps: Step (1) acquiring millimeter-level 3D point cloud data of all surface elements of classical Chinese gardens using a ground laser scanner and performing data preprocessing; Step (2) using the preprocessed 3D point cloud data, selecting a differentiated point cloud modeling method based on land features, and integrating the 3D surface model of the classical Chinese gardens; Step (3) loading rainfall parameters using hydrological simulation software to generate visualized runoff paths of classical Chinese gardens under different rainfall scenarios, and identifying hydrological risk points using a grid-based cumulative method; Step (4) verifying the reliability of rainwater runoff simulation and the accuracy of hydrological risk point identification by comparing on-site observations and surface model simulation results. This invention can improve the efficiency and accuracy of identifying hydrological risk points in classical Chinese gardens with complex surface features.

Description

Technical Field

[0001] This invention relates to the interdisciplinary technical field of digital preservation of cultural heritage, 3D geographic information modeling, and intelligent hydrological simulation. Specifically, it relates to methods for reconstructing surface models of classical Chinese gardens based on 3D point clouds, simulating rainwater runoff, and identifying risks. Background Technology

[0002] The dynamic evolution of water systems in classical Chinese gardens is highly dependent on regional natural hydrological processes, particularly surface runoff driven by precipitation events. Extreme or prolonged rainfall can easily trigger changes in surface runoff paths, localized waterlogging, soil erosion, and even instability of rock structures, posing a potential threat to historical relics such as garden paving, revetments, platforms, and wooden buildings, and in severe cases, causing irreversible damage to cultural heritage. Although the field of cultural heritage protection has increasingly emphasized the impact of hydrological factors in recent years, existing research on rainwater runoff in classical gardens remains largely at the level of theoretical explanation or qualitative description of traditional garden design principles, with a relative lack of systematic quantitative analysis and dynamic process simulation.

[0003] In particular, regarding technological approaches, most studies have not fully integrated modern spatial information technology, lacking digital modeling and hydrological response simulation methods based on high-precision landform morphology. This results in insufficient understanding of runoff generation, confluence paths, and spatial distribution of risks under complex landform conditions. Currently, traditional technologies suffer from three major pain points, severely restricting the ability to achieve refined understanding and risk assessment of hydrological processes in classical gardens:

[0004] 1. Conventional surface modeling techniques (such as 2D CAD and low-precision point cloud fitting) are difficult to realistically reproduce the geometric details of complex micro-topography such as artificial mountains and stacked rocks. In addition, building roofs are prone to model loss due to the negative angle blind zone of ground scanning, making it difficult to support high-fidelity simulation of rainwater runoff paths.

[0005] 2. For full-element 3D point cloud data with a scale of 1 billion, a single software platform is prone to system crashes or loss of key surface information due to memory overflow or algorithm efficiency bottlenecks during integrated processing.

[0006] 3. Current runoff simulations mostly use general distributed hydrological models suitable for large-scale watersheds. Their parameter settings and algorithm logic do not fully consider the surface characteristics of classical gardens, which are characterized by "small scale, high heterogeneity, and multiple interfaces". This leads to problems such as large path deviations in runoff simulations under complex surface conditions, making it difficult to achieve efficient and accurate identification of the spatial distribution of hydrological risks. Summary of the Invention

[0007] Therefore, the technical problem to be solved by this invention is to provide a complete and customized technical solution, namely, a method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds. Through a complete technical process of "precise point cloud processing - differentiated model construction - multi-scenario runoff simulation - result verification closed loop", the invention solves the above pain points and ultimately improves the efficiency and accuracy of identifying hydrological risk points in classical Chinese gardens.

[0008] This invention provides the following technical solution:

[0009] A method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point clouds includes the following steps:

[0010] Step (1): For classical Chinese gardens with complex surface features, a ground laser scanner is used to acquire millimeter-level three-dimensional point cloud data of all surface elements, and data preprocessing is completed by a combination of machine screening and manual screening.

[0011] Step (2): Using the preprocessed 3D point cloud data, select a differentiated point cloud modeling method based on the features of the land cover, and integrate a complete 3D surface model of the Chinese classical garden through coordinate alignment and topological fusion.

[0012] Step (3): Load rainfall parameters using hydrological simulation software to generate visualized runoff paths of classical Chinese gardens under different rainfall scenarios, and use the grid division and accumulation method to identify hydrological risk points;

[0013] Step (4): By comparing the field observations and the results of the surface model simulation, verify the reliability of the rainwater runoff simulation and the accuracy of the hydrological risk point identification.

[0014] In the above-mentioned method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds, in step (1), the ground laser scanner is one or a combination of two or more of the following: UAV laser scanner, mobile laser scanner, or fixed-point laser scanner.

[0015] In the above method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point clouds, step (1) involves a customized acquisition scheme for the 3D point cloud data:

[0016] The ground-based laser scanner is a mobile SLAM color 3D laser scanner, model OMNISLAM-R8+, with a relative measurement accuracy of 2mm and a point cloud density of 250,000 points / m². 2 The point cloud thickness is 2mm. This ground laser scanner can meet the millimeter-level data acquisition needs of fine structures such as rock crevices and wooden structures in classical gardens, solving the problem of insufficient accuracy of traditional scanners in small spaces and complex structures.

[0017] The data collection time window was selected during the season when the plant canopy density was less than or equal to 0.4, and the 3D point cloud data was collected after the garden management had cleared the weeds. Clearing the weeds can minimize the obstruction of the ground point information by fallen leaves and weeds, and reduce the workload of subsequent manual screening and data collection.

[0018] The surveying path employs a composite path combining a circular path and a grid path for data acquisition. For three-dimensional structural areas (such as artificial mountains and pavilions), a circular path is used to ensure the integrity of the point cloud on the structural facade. For flat areas (green spaces, paved areas, and areas around water bodies), a grid path is used, maintaining a uniform speed of 0.8-1.2 m / s along the predetermined surveying route to ensure uniform coverage of the ground point cloud. When traversing complex spaces (artificial mountains, caves, etc.), real-time positioning technology is used to record the spatial relationships of the internal structure of the complex space, obtain the three-dimensional coordinates of key control points within the space, and switch to a slow mode (reducing the acquisition speed to 0.5 m / s) to ensure the integrity of the point cloud of the internal structure of the complex space.

[0019] The above-mentioned method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds, in step (1), the data preprocessing method is as follows: the point cloud data is imported into Trimble RealWorks software or GeomagicWrap software for automatic stitching, outliers are removed by point cloud filtering and noise reduction is performed; then, the point cloud classification and recognition function of TrimbleRealWorks12.0 is used to perform machine classification and recognition of the full-element point cloud. TrimbleRealWorks12.0, with its optimized RWP data architecture and block processing technology, has the advantages of high efficiency, controllability, and integrability in engineering scenarios, and is suitable for large-scale point cloud classification tasks in structured environments such as buildings and factories, with better results than other point cloud data platforms; finally, after manual review and screening, a surface point cloud model containing only ground, paving, revetment, building foundation and rocks within the research scope is obtained, and the point cloud data is divided into two major categories: terrain point cloud data and building point cloud data, and artificial mountains are included in the terrain point cloud data.

[0020] The specific method for step (2) of the above-mentioned method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point clouds is as follows:

[0021] (2-1) Import the extracted terrain point cloud data into Trimble RealWorks 12.0 for processing. Use the Create TIN Surface command to generate a terrain surface model. Trimble RealWorks 12.0 can efficiently and stably perform triangulation modeling on point clouds with a scale of billions. In particular, it is significantly better than software such as Geomagic Wrap and Cloud Compare in terms of supporting constraint boundaries, engineering accuracy and memory management. Since non-ground features such as weeds, tree trunks and debris have been removed from the point cloud, the surface model may have local holes in the original ground contact area. Put the terrain surface model with holes into CloudCompare 2.12.2 for processing. Use the Poisson surface reconstruction algorithm to repair the holes and missing areas in the point cloud to complete the construction of the ground model.

[0022] (2-2) The extracted building point cloud data is imported into the 3D modeling software Rhino 8, and geometric correction is achieved through surface reconstruction and spatial position calibration. For the modeling of building point cloud data, due to the characteristics of "flying eaves and upturned corners and multi-layered roofs" in classical Chinese garden architecture, ground laser scanners can only obtain clear and complete outline point clouds, while the roof has missing point clouds due to negative angle blind spots. In this invention, the extracted building point cloud data is imported into the 3D modeling software Rhino 8, and the problem can be optimized more ideally by using manual regular building modeling: First, the main structure of the building (columns, beams, roof frame) follows clear geometric regularity characteristics, and manual modeling (such as using Rhino software) has a natural advantage. Second, manual modeling of the building alone can achieve lightweight optimization of data volume and optimize the global model data volume, ensuring the stability of software operation.

[0023] (2-3) The terrain point cloud data and building point cloud data are aligned and topologically fused in Rhino8 to create a three-dimensional model of the surface of a classical Chinese garden that integrates terrain, rockery and architecture.

[0024] Terrain point clouds are characterized by large amounts of data, numerous noise points, and irregular terrain. Building point clouds, while having a smaller data volume, require high geometric accuracy and need to be regularized into lines, planes, and curved surfaces. This invention employs different software platforms for differentiated processing of terrain and building point cloud data, significantly improving the accuracy of the final 3D surface model of classical Chinese gardens.

[0025] The above-mentioned method for reconstructing the surface model of classical Chinese gardens based on three-dimensional point clouds, simulating rainwater runoff, and identifying risks, in step (3), the surface rainwater runoff simulation method is as follows: based on the three-dimensional surface model, the surface rainwater runoff is simulated using the Groundhog plugin in the Rhino+Grasshopper platform to obtain the visualized runoff path of classical Chinese gardens under different rainfall scenarios.

[0026] The above-mentioned method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point clouds involves randomly distributing 1000, 3000, and 5000 rainwater particles on the mesh surface of the 3D surface model of classical Chinese gardens during simulation. These particles simulate the distribution of surface rainwater runoff under three rainfall scenarios: light rainfall, moderate rainfall, and heavy rainfall, respectively. The number of iterations is greater than or equal to 200, and the fidelity is set to 1 cm.

[0027] The above-mentioned method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds, in step (3), the method for identifying hydrological risk points is as follows: the site occupied by the classical Chinese garden is divided into several rainwater grid units, and the FlowCatchment component in Groundhog (Groundhog is an open-source plugin in the commercial software Rhino, and FlowCatchment is a built-in functional component in Groundhog) is used to identify hydrological risk points in the rainwater grid units.

[0028] The above-mentioned method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point clouds uses rainwater grid cells with a size of less than or equal to 2.5m × 2.5m. During identification, the runoff section loss is set to 0.03, the water volume at the starting position of the runoff path is set to 100, and the total water flow in each rainwater grid cell reaching 700 is used as the basis for judging hydrological risk points.

[0029] In the above-mentioned method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds, in step (4), typical precipitation periods are selected for on-site observation; rainfall data is obtained from the meteorological data center where the classical Chinese gardens are located; during and after rainfall, the distribution and flow characteristics of runoff paths are observed and recorded, and the spatial location of hydrological risk points is marked, while image data is collected.

[0030] The same grid-based cumulative method used for risk point identification was employed to divide the field observation results into grids, and the verification results were mapped onto a plane composed of several pre-divided rainwater grid units. The accuracy of the risk spatial location identification results was verified by comparing the presence of hydrological risks in each rainwater grid unit. Each rainwater grid unit was labeled into two categories: those with risk and those without risk, to determine the accuracy. Recall rate F1 score is a quantitative indicator;

[0031] (1);

[0032] (2);

[0033] (3);

[0034] In equations (1) to (3), This refers to rainwater grid cells where both actual and predicted conditions present risks. To simulate rainwater grid cells where there is no actual risk, Simulate rainwater grid cells that pose a risk when there is none in reality;

[0035] When accuracy Recall rate When both the F1 and F1 values ​​reach 0.65 or above, it indicates that the surface model simulation results are highly reliable and the hydrological risk point identification is accurate.

[0036] The technical solution of the present invention achieves the following beneficial technical effects:

[0037] 1. This invention relates to a method for reconstructing a surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point clouds. It employs terrestrial laser scanning technology to collect 3D point cloud data of the surface of classical Chinese gardens and constructs a 3D model of the garden's surface based on this data. This model is used to simulate visualized runoff paths under different rainfall scenarios, and a grid-based method is employed to identify hydrological risk points. This improves the efficiency and accuracy of identifying hydrological risk points in classical Chinese gardens with complex surface features, thus providing early warning of the threats posed by rainwater erosion and localized rainwater accumulation to garden cultural heritage, and offering a reference for conservation efforts.

[0038] 2. This invention relates to a method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point cloud data. It is a method for reconstructing the complex surface full-element model of classical Chinese gardens and visualizing and simulating rainwater runoff in multiple scenarios based on millimeter-level 3D point cloud data. It is particularly suitable for garden scenes with complex surface conditions (such as artificial mountains and multiple types of land features), such as classical garden heritage sites containing highly complex resort mountain construction and traditional building clusters. It provides technical support for the refined rainwater management, risk prevention and control, and sustainable heritage protection of classical Chinese garden heritage. Attached Figure Description

[0039] Figure 1 A floor plan of Jiqingxuan in an embodiment of the present invention;

[0040] Figure 2 Flowchart of surface model reconstruction and rainwater runoff simulation in the embodiment of this invention;

[0041] Figure 3 A schematic diagram of the surveying trajectory of Jiqingxuan in an embodiment of the present invention;

[0042] Figure 4a The original point cloud data map in this embodiment of the invention;

[0043] Figure 4b Preprocessed point cloud data image in this embodiment of the invention;

[0044] Figure 4c A schematic diagram of model integration of preprocessed point cloud data in Rhino in an embodiment of the present invention;

[0045] Figure 4d A schematic diagram showing the local details of the integrated model in an embodiment of the present invention;

[0046] Figure 5 The runoff path diagram of Jiqingxuan under a light rainfall scenario in this embodiment of the invention;

[0047] Figure 6 Runoff path diagram of Jiqingxuan under moderate rainfall scenario in this embodiment of the invention;

[0048] Figure 7 The runoff path diagram of Jiqingxuan under heavy rainfall scenario in this embodiment of the invention;

[0049] Figure 8a Risk point diagram of Jiqingxuan under light rainfall scenario in this embodiment of the invention;

[0050] Figure 8b Risk point diagram of Jiqingxuan under moderate rainfall scenario in this embodiment of the invention;

[0051] Figure 8c Risk point diagram of Jiqingxuan under heavy rainfall scenario in this embodiment of the invention;

[0052] Figure 9a Spatial distribution and morphological diagram of the main runoff paths during the simulation process in this embodiment of the invention;

[0053] Figure 9b Spatial distribution and morphological diagram of the main runoff paths in field verification in this embodiment of the invention;

[0054] Figure 9c The embodiments of the present invention include a comparison diagram of runoff path characteristics in simulation and field survey, as well as real-life images of typical differential runoff path 1 and typical differential runoff path 2.

[0055] Figure 10 The risk points identified during the on-site verification of risk points based on natural rainfall events in this embodiment of the invention, along with real-life images of some of the risk points;

[0056] Figure 11 The embodiments of the present invention include real-life images of risk points identified during on-site verification of risk points based on natural rainfall events, as well as other risk points.

[0057] Figure 1 The attached map labels are as follows: ①-Hanging Flower Gate; ②-Jiqing Pavilion; ③-Qingqin Gorge; ④-Octagonal Pavilion; ⑤-Shen Congwen's Former Residence; ⑥-Square Pavilion; ⑦-U-shaped Corridor; ⑧-North Gate; ⑨-Gatehouse; ⑩-Military Affairs Office; ⑪-Bottle Gate; ⑫-Dairy Kitchen. Detailed Implementation

[0058] The following section uses Jiqingxuan as an example to further illustrate the present invention's method for reconstructing the surface model of classical Chinese gardens based on three-dimensional point clouds, simulating rainwater runoff, and identifying risks.

[0059] 1. Overview of the research case

[0060] Jiqingxuan Pavilion is built on the natural terrain at the eastern foot of Longevity Hill in the Summer Palace. The overall terrain slopes from south to north, covering an area of ​​approximately 5360 square meters. 2 Based on a natural, sloping rock face with an elevation difference of approximately 5 meters, a gorge was artificially carved out at the lowest point, forming a stone canyon that runs through the area. The north side of the gorge features a straight section of carved stone, while the south side uses stacked stones to create a cascading stream. Through the ingenious use of carved and stacked stones, Jiqing Pavilion blends the artificial with the natural rock face, creating a complex landscape where artificial stones are interspersed with natural rocks. The garden contains 17 buildings, as well as a U-shaped covered walkway; the overall layout is adapted to the terrain and the specific features of the land. Figure 1 ).

[0061] The Houxi River in the Summer Palace splits into two at its northeast corner. One branch flows out through an underground man-made channel from Qingqin Gorge in the west hall of Jiqing Pavilion. It then flows eastward through the stone gorge and finally exits the Summer Palace from the east wall of Jiqing Pavilion. Therefore, this embodiment uses Jiqing Pavilion as a typical representative of classical Chinese gardens, performing surface model reconstruction, runoff simulation, and risk point identification. It further explains the method of surface model reconstruction, rainwater runoff simulation, and risk identification of classical Chinese gardens based on three-dimensional point clouds, aiming to provide scientific basis and methodological support for the ecological hydrological research and cultural heritage protection of traditional gardens.

[0062] 2. Process and Framework

[0063] This embodiment of the method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds mainly consists of four steps ( Figure 2(1) Use ground laser scanning technology to obtain millimeter-level three-dimensional point cloud data of Jiqingxuan, and perform data processing such as splicing, noise reduction, and classification; (2) Select differentiated modeling methods based on the characteristics of land features, and integrate high-precision surface models of terrain, artificial mountains, and buildings; (3) Analyze and generate visualized runoff paths under different rainfall scenarios based on the model analysis, and identify hydrological risk points based on grid division; (4) Verify the reliability of rainwater runoff simulation and the accuracy of hydrological risk point identification by comparing on-site observation and computer simulation results.

[0064] 3. Data Acquisition and Processing

[0065] (1) Data collection and verification

[0066] In modern cultural heritage digital measurement, close-range photogrammetry and laser scanning are widely used as non-contact measurement technologies for obtaining three-dimensional information of heritage, including unmanned aerial vehicle laser scanning (UAV-LS), mobile laser scanning (MLS), and terrestrial laser scanning (TLS). Based on field research, this embodiment summarizes four main challenges in obtaining three-dimensional data of Jiqingxuan: (1) The artificial mountains in the site are mainly earthen and rocky mountains with complex terrain, which restricts the layout scheme and accuracy of close-range photogrammetry; (2) The area is restricted by airspace flight control policies, making it impossible to use UAV oblique photogrammetry; (3) During the opening period, the flow of tourists is large, and there is continuous interference from the flow of people in some areas, making it difficult to realize the systematic deployment of terrestrial laser scanners; (4) The vegetation canopy is high in spring, summer, and autumn, and the surface of the mountain is covered by shrubs and ground cover plants, making it difficult to obtain data on the texture of the rocks.

[0067] Therefore, this embodiment employs mobile laser scanning technology, selecting winter (December 2024) when plant canopy density is low (approximately 0.3), while ensuring the safety of the garden heritage. Based on this, and according to the elements and environmental characteristics of the Jiqingxuan Garden, a mapping path scheme combining encircling paths and grid paths was systematically planned. Figure 3 ).

[0068] To ensure the lidar system acquires comprehensive and high-precision information on the geometric shape and surface features of garden elements, professionally trained operators collect data while moving at a constant speed along a predetermined surveying route. When traversing complex spaces such as building corridors, artificial hills, and ravines, real-time positioning technology is used to record the internal structural spatial relationships. In narrow rock crevices, a slow-speed mode is switched to improve point cloud density to the millimeter level. The surveying equipment used is the OMNISLAM-R8+ mobile SLAM color 3D laser scanner. This device integrates RTK-SLAM, PPK-SLAM, LIO-PANO, and LiRF technologies, making it suitable for high-precision acquisition of large-space 3D data in multiple scenes. Its main performance parameters are: relative measurement accuracy of 2mm and point cloud density of 250,000 points / m². 2 The point cloud thickness is 2mm.

[0069] Ultimately, over 2,100,000,000 point cloud data points with RGB information were collected. A high-precision total station was used to conduct on-site measurements of key control points of the garden architecture to verify data accuracy. Comparison and analysis of the measured coordinates with corresponding positions in the point cloud data showed that the average error in the horizontal direction was ±2cm, and the error in the vertical direction was ±2cm, which is better than the relevant provisions in the industry technical standard "Technical Specification for Digital Surveying and Mapping of Ancient Building Murals". However, due to the combined influence of complex environmental factors and the characteristics of laser scanning technology, a large number of noise points inevitably exist in the measurement data. Therefore, data preprocessing is necessary to ensure data quality and reliability.

[0070] (2) Data processing and integration

[0071] Data preprocessing mainly includes two steps: point cloud data stitching and noise reduction filtering. For example... Figures 4a to 4b The point cloud data was imported into Trimble RealWorks software for automatic stitching, and outliers were removed and noise was reduced through point cloud filtering. Based on this, non-target point clouds were eliminated, and surface point cloud models containing only ground, paving, revetments, building foundations, and rocks within the research area were manually selected and divided into two categories: terrain (including artificial mountains) and buildings.

[0072] Considering the structural regularity of garden buildings and the potential impact of noise in the point cloud on subsequent analysis, the extracted building point cloud was imported into the mainstream 3D modeling software Rhino 8, where geometric correction was achieved through surface reconstruction and spatial positioning calibration. For the terrain point cloud data, it was imported into CloudCompare 2.12.2 for processing, using the Poisson surface reconstruction algorithm to repair holes and missing areas in the point cloud. Finally, a high-precision surface model integrating terrain, artificial hills, and buildings was integrated into Rhino 8 (see...). Figure 4c and Figure 4d ).

[0073] 4. Data Analysis and Visualization

[0074] (1) Simulation of surface rainwater runoff

[0075] Surface runoff simulation was performed using the Groundhog plugin within the Grasshopper platform. Groundhog, developed by Philip Belesky of RMIT University in Melbourne, Australia, features a Flow Paths component that builds a runoff simulation algorithm based on a particle system. This algorithm treats rainwater as a collection of discrete particles and determines the particle's path based on information such as gravity and grid surface slope. Through multiple iterations, it simulates the trajectory of each rainwater particle to generate a visualized runoff path. The Flow Paths component supports various parameter inputs, providing an efficient and convenient method to observe the dynamic path of rainwater on the ground surface.

[0076] Because buildings on the site significantly influence runoff direction, this embodiment incorporates architectural elements in addition to considering the artificial hills and terrain when simulating surface runoff. By randomly distributing 1000, 3000, and 5000 raindrop particles on the mesh surface of the Jiqingxuan 3D model, runoff distribution under three rainfall scenarios—light, moderate, and heavy rain—was simulated, respectively. Multiple preliminary simulations showed that regardless of the number of raindrop particles, approximately 200 iterations were sufficient for the vast majority, or even all, of the particles to reach stable positions for the Jiqingxuan site. Therefore, this embodiment selected 200 iterations as the input parameter to limit the number of flow iterations. Simultaneously, the fidelity, i.e., the distance the raindrop particles move in each flow iteration, was set to 1 cm to achieve high-precision runoff simulation.

[0077] (2) Identification of hydrological risk points

[0078] The erosion and localized accumulation of rainwater can pose a threat to garden cultural heritage. Therefore, based on runoff simulation analysis, this embodiment divides the entire study site into 862 rainwater grid cells on a 2.5m × 2.5m scale, and uses the FlowCatchment component in Groundhog to identify hydrological risk points within Jiqingxuan. The FlowCatchment component (Groundhog is an open-source plugin for the commercial software Rhino, and FlowCatchment is a built-in functional component of Groundhog) can estimate the total amount of rainwater under specific topographic conditions based on the surface runoff path generated by the analysis, taking into account surface infiltration characteristics. The flow and infiltration of water on the ground are affected by both environmental conditions and material properties. Considering the geographical characteristics of Jiqingxuan, this embodiment sets the runoff segment loss to 0.03, that is, assuming that when rainwater flows along the runoff path across the surface, 3% of the rainwater will infiltrate and accumulate in each calculation segment. The water volume at the starting point of the runoff path is set to 100, and the total water flow in each grid cell reaches 700 (preliminary experimental statistical interval) as the basis for judging the hydrological risk point.

[0079] 5. Analytical Verification Methods

[0080] (1) Field observation records

[0081] First, the condition of the garden's surface rocks was recorded through on-site observation to explore the potential impact and risks of rainwater runoff on the garden surface over a long period. Based on this, on-site observations were conducted during a typical rainfall period (July 27, 2025, 15:00-16:00, with a total hourly rainfall of 3.6 mm) to observe the rainwater runoff characteristics and water collection within Jiqingxuan Garden. Rainfall data were obtained from the Beijing Meteorological Data Center and determined to be moderate rain according to relevant standards. During and after the rainfall, professionals conducted a systematic inspection of the park, focusing on recording the distribution and flow direction of runoff paths and marking the spatial locations of hydrological risk points. Simultaneously, image data was collected to assist in subsequent comparative analysis and verification.

[0082] (2) Accuracy verification

[0083] Referring to the risk point identification simulation grid division method based on the Rhino+Grasshopper platform, the above verification results were also mapped onto a plane composed of 862 grid cells. The number of grid cells was obtained by adding edge residual grid correction to the actual area of ​​5360㎡. The accuracy of the risk spatial location identification results was verified by comparing whether hydrological risks existed in each grid cell. Each grid cell was labeled into two categories: those with risk and those without risk, using precision, recall, and F1 score as quantitative indicators. Their specific definitions and calculation formulas are as follows:

[0084] (1);

[0085] (2);

[0086] (3);

[0087] In the formula, TP represents a grid with risk in both the actual and predicted scenarios, FP represents a grid with risk in the actual scenario but no risk in the simulation, and FN represents a grid with no risk in the actual scenario but risk in the simulation.

[0088] 6. Results Analysis

[0089] (1) Characteristics of the runoff path of Jiqingxuan under three rainfall scenarios

[0090] The visualization of runoff visually presents the direction and distribution of rainwater flow by fading the color as it moves away from the "source" along the path. Simulation results of site rainwater runoff under three rainfall scenarios show that runoff from each rainfall level is concentrated in similar areas, mainly including the south side of the stone gorge and the area around the promenade, with most eventually flowing into the low-lying stone gorge. This reflects the core channel function of the stone gorge in the site's hydrological system. Building roofs divert rainwater according to their slope (four or two), forming relatively concentrated linear runoff cascading down the eaves. Due to their vertical structure above the ground, building platforms intercept runoff, creating runoff paths along the platforms in multiple areas. Some artificial hills also participate in the drainage process. Notably, the paved areas on the east and west sides of the site also exhibit a clear convergence of runoff lines, which becomes more pronounced with increasing rainfall, demonstrating the guiding role of paving on runoff direction. Figures 5 to 7 ).

[0091] (2) Identification of hydrological risk points at Jiqingxuan under three rainfall scenarios

[0092] Computer simulation results indicate that under three rainfall scenarios, the hydrological risk points at Jiqingxuan are mainly distributed in the stone gorge, around the buildings, and the paved area of ​​the road to the east, with a high degree of spatial overlap. Under light rain, the stone gorge effectively functions as a drainage system; under moderate rain, the building's foundation provides some protection against the spread of rainwater. However, under heavy rain, the number of hydrological risk points within the park increases significantly, reflecting its limited ability to regulate runoff under these conditions. Figures 8a to 8c ).

[0093] The number of risk points varied significantly under different rainfall scenarios. Statistically, based on grid units, the number of hydrological risk points at Jiqingxuan was 6 under light rain, accounting for 0.70% of the total grids. This increased to 81 under moderate rain, accounting for 9.35% of the total grids, approximately 13.36 times that of the light rain scenario; and reached 202 under heavy rain, accounting for 23.33% of the total grids, 33.33 times that of the light rain scenario, significantly reflecting the impact of rainfall intensity on the park's hydrological safety.

[0094] (3) Comparison of runoff characteristics

[0095] According to the aforementioned computer simulation results, under moderate rainfall conditions, the relatively long and stable runoff paths formed on the surface of Jiqingxuan are mainly concentrated around the buildings and on hard surfaces such as road paving, and eventually converge in the low-lying Shixia area within the park. Figure 9a Field observations based on natural precipitation events have also verified this overall trend. Figure 9b However, it also revealed discrepancies between some runoff paths and simulation results. Figure 9c For example, typical runoff path 1, located at the southwest corner of Jiqingxuan, was found by on-site investigation to have an actual runoff path that is U-shaped and not the shortest path to Shixia. Figure 9c This phenomenon may be attributed to the long-term erosion of rainwater diverted from the building roof, resulting in small grooves forming at the junction of the building foundation and the natural terrain. To prevent the continued impact of rainwater erosion on the building structure, the park management used cement to repair the surface of this area, thus forming a diversion channel, allowing rainwater to flow down along the channel. In addition, typical path 2, located northeast of Jiqing Pavilion, shows that rainwater forms a significant runoff in front of the stone gorge, which may be due to the combined effects of vegetation blocking and micro-topographical undulations. The stone gorge is equipped with an inlet to collect rainwater. The originally spreading runoff is blocked by vegetation at the entrance, and then converges in the low-lying area to form a relatively stable runoff channel. Figure 9c ).

[0096] (5) Verification of hydrological risk points

[0097] During field observations, the research team identified 71 hydrological risk points, and their spatial distribution showed a high degree of consistency with the computer simulation results. Figure 10 (See Figure a in the text). The identification criteria for these risk points mainly include two aspects: first, the areas of cumulative damage to artificial mountains, real rocks, and building foundations caused by long-term surface runoff (hereinafter referred to as "surface damage"); second, the areas of concentrated runoff scouring and water accumulation that can be directly identified during rainfall.

[0098] Among them, surface damage can be divided into four categories: (1) the grouting of some artificially stacked rocks has fallen off and stones have fallen, which is presumed to be related to rainwater erosion along the cracks in the rocks. Figure 10(Figure b in the figure); (2) Settlement occurred on the foundation of buildings in some areas (Figure b in the figure); Figure 10 (Figure c in the figure) It is speculated that it is related to long-term water accumulation. This phenomenon has been verified in field observations under natural precipitation conditions; (3) Surface runoff carries sediment to promote vegetation growth. The root-splitting action of vegetation roots further leads to the cracking and falling off of natural rock surfaces. Figure 10 (d in the figure); (4) Some of the building foundation joints have broken locally, which is speculated to be related to the impact force generated by the vertical drop of the eaves runoff ( Figure 10 (Figure e in the diagram).

[0099] The main cause of the risk points of concentrated runoff scouring areas and waterlogged areas that can be directly identified during rainfall is the water-blocking effect of building foundations. Figure 11 (Figure b in the text) The erosion effect of rainwater on both sides of the gorge ( Figure 11 (Figure c in the diagram) and the traffic guidance function of road paving, etc. Figure 11 (See Figure d in the diagram). The distribution of some risk points identified by computer simulations differs from the results of on-site surveys, which is considered to be due to the water-blocking effect of plants. In summer, dense vegetation, while absorbing water, may also hinder rainwater flow to the drainage ditch, thus affecting the spatial location of risk points. Figure 11 (Figure e in the diagram).

[0100] Using the results of risk point identification from on-site observations as a validation benchmark, the results of computer simulations based on the Rhino+Grasshopper platform were compared and analyzed to quantitatively evaluate the accuracy of the computer simulation method in hydrological risk point identification. The results show that the computer simulations exhibit high accuracy in hydrological risk point identification, with all three core accuracy indicators (precision, recall, and F1 score) exceeding 0.65, indicating that simulations possess basic risk point identification capabilities (Table 1). Specifically, the recall and F1 score, among other key indicators, both exceeded 0.7. Overall, computer simulations can effectively identify hydrological risk points and demonstrate high feasibility and applicability.

[0101] Table 1

[0102]

Claims

1. A method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on 3D point clouds, characterized in that... Includes the following steps: Step (1): Use a ground laser scanner to acquire millimeter-level three-dimensional point cloud data of all elements of the surface of classical Chinese gardens, and complete data preprocessing by combining machine screening and manual screening. Step (2): Using the preprocessed 3D point cloud data, select a differentiated point cloud modeling method based on the features of the land cover, and integrate a complete 3D surface model of the Chinese classical garden through coordinate alignment and topological fusion. Step (3): Load rainfall parameters using hydrological simulation software to generate visualized runoff paths of classical Chinese gardens under different rainfall scenarios, and use the grid division accumulation method to identify hydrological risk points; Step (4): By comparing the field observation results with the surface model simulation results, verify the reliability of the rainwater runoff simulation and the accuracy of the hydrological risk point identification. In step (1), the data acquisition time window is selected when the plant canopy density is less than or equal to 0.4 and the three-dimensional point cloud data is acquired after the garden management has completed the weed removal. The surveying path uses a composite path combining a circular path and a grid path for data acquisition; a circular path is used for three-dimensional structural areas. For flat areas, a grid path is used, and the travel speed is maintained at 0.8-1.2 m / s along the predetermined survey route; when traversing complex spaces, the three-dimensional coordinates of key control points in the space are recorded through real-time positioning technology, and the travel speed is reduced to 0.5 m / s. In step (1), the data preprocessing method is as follows: the point cloud data is imported into Trimble RealWorks software or GeomagicWrap software for automatic stitching, outliers are removed by point cloud filtering and noise reduction is performed; then the point cloud classification and recognition function of TrimbleRealWorks12.0 is used to perform machine classification and recognition of the full-element point cloud; finally, after manual review and screening, a surface point cloud model containing only ground, paving, revetment, building foundation and rocks within the research scope is obtained, and the point cloud data is divided into two categories: terrain point cloud data and building point cloud data, and the artificial mountain is included in the terrain point cloud data; The specific method for step (2) is as follows: (2-1) Import the extracted terrain point cloud data into Trimble RealWorks 12.0 for processing, and use the CreateTIN Surface command to generate a terrain surface model; put the terrain surface model with holes into CloudCompare 2.12.2 for processing, and use the Poisson surface reconstruction algorithm to repair the holes and missing areas in the point cloud to complete the construction of the ground model. (2-2) Import the extracted building point cloud data into the 3D modeling software Rhino 8, and achieve geometric correction through surface reconstruction and spatial position calibration; (2-3) The terrain point cloud data and building point cloud data are aligned and topologically fused in Rhino8 to create a three-dimensional surface model of Chinese classical gardens that integrates terrain, rockery and architecture. In step (3), the surface rainwater runoff simulation method is as follows: based on the three-dimensional surface model of the classical Chinese garden, the surface rainwater runoff is simulated using the Groundhog plugin in the Rhino+Grasshopper platform to obtain the visualized runoff path of the classical Chinese garden under different rainfall scenarios. In step (3), the method for identifying hydrological risk points is as follows: the site occupied by the classical Chinese garden is divided into several rainwater grid units, and the FlowCatchment component in Groundhog is used to identify hydrological risk points in the rainwater grid units; The size of the rainwater grid unit is less than or equal to 2.5m × 2.5m; during identification, the runoff section loss is set to 0.03, the water volume at the starting position of the runoff path is set to 100, and the total water flow in each rainwater grid unit reaching 700 is used as the basis for judging the hydrological risk point.

2. The method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds as described in claim 1, is characterized in that... In step (1), the ground laser scanner is one or a combination of two or more of the following: UAV laser scanner, mobile laser scanner, or fixed-point laser scanner.

3. The method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds as described in claim 2, is characterized in that... In step (1), a customized acquisition scheme is adopted for the 3D point cloud data acquisition: The ground-based laser scanner is a mobile SLAM color 3D laser scanner, model OMNISLAM-R8+, with a relative measurement accuracy of 2mm and a point cloud density of 250,000 points / m². 2 The point cloud thickness is 2mm.

4. The method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds as described in claim 1, is characterized in that... During the simulation, 1000, 3000, and 5000 raindrop particles were randomly distributed on the mesh surface of the 3D model of the Chinese classical garden to simulate the distribution of surface rainwater runoff under three rainfall scenarios: light rainfall, moderate rainfall, and heavy rainfall, respectively. The number of iterations was greater than or equal to 200, and the fidelity was set to 1 cm.

5. The method for reconstructing the surface model of classical Chinese gardens, simulating rainwater runoff, and identifying risks based on three-dimensional point clouds as described in claim 1, is characterized in that... In step (4), on-site observations are conducted during the precipitation period; rainfall data is obtained from the meteorological data center of the location of the classical Chinese garden; during and after the rainfall, the distribution and flow characteristics of the runoff path are observed and recorded, the spatial location of hydrological risk points is marked, and image data is collected at the same time. The same grid-based cumulative method used for risk point identification was employed to divide the field observation results into grids, and the verification results were mapped onto a plane composed of several pre-divided rainwater grid units. The accuracy of the risk spatial location identification results was verified by comparing the presence of hydrological risks in each rainwater grid unit. Each rainwater grid unit was labeled into two categories: those with risk and those without risk, to determine the accuracy. Recall rate F1 score is a quantitative indicator; （1）； （2）； （3）； In equations (1) to (3), This refers to rainwater grid cells where both actual and predicted conditions present risks. To simulate rainwater grid cells where there is no actual risk, Simulate rainwater grid cells that pose a risk when there is none in reality; When accuracy Recall rate When both the F1 and F1 values ​​are greater than or equal to 0.65, it indicates that the surface model simulation results are highly reliable and the hydrological risk point identification is accurate.

Images