A digital twin-based landscape design method

Abstract

The application discloses a garden landscape design method based on digital twinning, and belongs to the technical field of landscape garden design. The method collects target site environment data sets, generates a site thermal distribution cloud map through a geographic space grid processing unit; extracts a site spatial pattern feature matrix based on the same; constructs a landscape effect distribution model, decomposes the feature matrix into terrain undulation, seasonal color and illumination reflection parameters, drives the model to generate a three-dimensional scene rendering sequence, embeds the model into a three-dimensional digital twinning scene of a garden site, and associates the parameters with physical properties of the twinning scene in real time; performs dynamic optimization of a landscape scheme, generates element replacement instructions, and updates a spatial topological relation network; outputs a site design scheme atlas, adjusts facility layout coordinates, and generates a final scheme data packet in combination with path connectivity, so that the accuracy and adaptability of garden landscape design can be improved.

Description

Technical Field

[0001] This invention relates to the field of landscape architecture design technology, specifically a landscape design method based on digital twins. Background Technology

[0002] Traditional landscape architecture design relies heavily on the designer's experience and subjective aesthetic judgment, which has significant limitations in areas such as site analysis and scheme optimization. Traditional landscape design lacks a three-dimensional dynamic mapping to the physical site, making it difficult to verify design schemes in advance in virtual scenarios. This often leads to deviations from the expected results after construction. Furthermore, design data is disconnected from subsequent operation and maintenance data, failing to support the full lifecycle management of the landscape. For example, in areas with significant topographic relief, manual measurement struggles to accurately capture the details of elevation changes, resulting in discrepancies between subsequent topographic design and the actual site. Moreover, inferring the overall distribution of environmental factors such as light and humidity required for plant growth from data from only a few monitoring points can easily lead to mismatches between plant configurations and site conditions, affecting the stability and sustainability of the landscape.

[0003] In the process of scheme generation and optimization, traditional methods lack systematic data support and dynamic simulation capabilities. Designers often develop schemes based on static site data, making it difficult to predict the landscape's appearance under different seasons and climatic conditions, and also unable to timely assess the impact of facility layout and plant configuration on the site's ecological function and user experience. For example, when designing recreational paths, if only visual appeal is considered while ignoring the potential distribution of pedestrian traffic, some areas may become overcrowded or underutilized. In plant selection, if the ecological compatibility between species and the coordination of seasonal changes are not fully considered, the landscape effect may be monotonous or the ecosystem may be unbalanced. This experience-based design model is not only inefficient but also fails to meet the comprehensive requirements of modern landscape architecture for ecology, functionality, and artistry, necessitating the introduction of new technological methods for innovation. Summary of the Invention

[0004] The purpose of this invention is to provide a landscape design method based on digital twins to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a landscape design method based on digital twins, the method comprising:

[0006] Collect a set of environmental data for the target site and input the set of environmental data into a geospatial raster processing unit to generate a site thermal distribution cloud map;

[0007] Based on the site thermal distribution cloud map, extract the site spatial pattern feature matrix;

[0008] A landscape effect distribution model is constructed, and the spatial pattern feature matrix is ​​decomposed into terrain undulation parameters, seasonal color parameters, and light reflection parameters. The landscape effect distribution model is driven to generate a three-dimensional scene rendering sequence. The landscape effect distribution model is embedded in the three-dimensional digital twin scene of the garden site, and the model parameters are correlated in real time with the physical properties of terrain, vegetation, and light in the twin scene.

[0009] The landscape scheme is dynamically optimized by generating element replacement instructions based on the material texture parameters and plant growth simulation data in the three-dimensional scene rendering sequence, and synchronously updating the spatial topology network in the landscape effect distribution model.

[0010] Output site design scheme map, adjust the facility layout coordinates in the 3D scene rendering sequence based on element replacement instructions, and generate the final scheme data package by combining the path connectivity in the spatial topology network.

[0011] Preferably, the target site environmental data set includes:

[0012] The target site environmental data set includes topographic elevation point clouds, vegetation distribution vectors, hydrological monitoring time series and meteorological historical records. The collected data is used to construct a basic geometric and physical attribute library for the three-dimensional digital twin model of the garden site.

[0013] The distribution of land cover types is obtained by using UAV remote sensing equipment, and the land cover types are overlaid with geological structural parameters to generate a geospatial raster.

[0014] Receive soil moisture time series and air particulate matter concentration uploaded by IoT sensors, and calculate the ecological sensitivity index of each cell in the geospatial grid;

[0015] Dynamically match historical tourist trajectory data with real-time population heat map, and mark the coordinates of high-density clustered areas in the geospatial grid;

[0016] By integrating the ecological sensitivity index with the coordinates of high-density clustered areas, a site thermal distribution cloud map containing a geological stability layer, an ecological carrying capacity layer, and an active pressure layer is generated.

[0017] Preferably, the extraction of the site spatial pattern feature matrix includes:

[0018] Visual corridor recognition is performed on the site thermal distribution cloud map, and the intersection of the mountain outline and the water system direction line is extracted as the visual focus coordinates.

[0019] Calculate the spatial correlation between the visual focus coordinates and the sunlight trajectory data to generate a light and shadow intensity gradient map;

[0020] The light and shadow intensity gradient map is input into a spatial clustering algorithm to delineate the boundaries of the ecological core area, the activity-intensive area, and the landscape transition area.

[0021] The database of seasonal vegetation changes is linked to the boundary of the landscape transition zone, and a spatial pattern feature matrix containing spatial perception intensity values, color change frequency values, and material attenuation coefficients is output.

[0022] Preferably, the construction of the landscape effect distribution model includes:

[0023] The terrain undulation parameters in the spatial pattern feature matrix are decomposed into slope vector groups and elevation difference matrix, which drive the three-dimensional spatial topology model to generate a terrain skeleton mesh.

[0024] Input the seasonal color parameters into the material spectral reflectance engine, and output a plant optical attribute table containing leaf reflectance curves and canopy transmittance curves;

[0025] The plant optical attribute table is mapped to the terrain skeleton mesh to generate a 3D point cloud of vegetation distribution with seasonal markers.

[0026] Simultaneously load the light reflection parameters into the landscape perception experience model, and calculate the shadow coverage area data of the three-dimensional point cloud of vegetation distribution when the angle of sunlight incidence changes.

[0027] The 3D point cloud of vegetation distribution is precisely aligned with the terrain skeleton mesh of the digital twin scene, ensuring the visualization accuracy of the twin scene.

[0028] Preferably, the dynamic optimization of the landscape scheme includes:

[0029] Receive the reflectivity data of the paving material in the three-dimensional scene rendering sequence and compare it with the soil coverage area predicted by the plant root growth model;

[0030] When the reflectivity of the paving material is detected to exceed a preset threshold, a material replacement instruction containing the permeable material number and porous structure parameters is generated.

[0031] Input the material replacement command into the spatial topology network and recalculate the path surface runoff permeability coefficient;

[0032] Adjust the surface runoff direction data in the landscape effect distribution model based on the change in permeability coefficient, and update the drainage buffer distance of the facility layout coordinates.

[0033] Preferably, the generation of the geospatial raster includes:

[0034] Kriging space interpolation calculations are performed on geological structural parameters to generate a geological basement raster containing rock strata stability coefficients;

[0035] The ecological sensitivity index is input into the inverse distance weighting algorithm, and the output is an ecological carrying capacity grid with an elevation correction factor.

[0036] The activity stress layer data is processed by a convolutional neural network to generate an activity intensity raster containing thermal values ​​of crowd gathering.

[0037] The geological base grid, ecological carrying capacity grid, and activity intensity grid are superimposed in multiple bands to form a site thermal distribution cloud map with a three-dimensional coordinate system.

[0038] Preferably, the delineation of the landscape transition zone boundary includes:

[0039] Extract color change frequency values ​​from the spatial pattern feature matrix and construct seasonal color bands containing hue gradients;

[0040] The chromatic color bands are input into the visual attractiveness measurement model to calculate the predicted visual dwell time for different hue combinations;

[0041] Based on the predicted visual dwell time, spatial perception intensity levels are divided, and coordinate sets of high-intensity perception areas and low-intensity perception areas are generated.

[0042] By fusing the visual focus coordinates with the high-intensity perception zone coordinate set, a landscape transition zone vector boundary with weighted identifiers is output.

[0043] Preferably, the generation of the three-dimensional point cloud of vegetation distribution includes:

[0044] Convert the leaf reflectance curves in the plant optical attribute table into a photon absorptivity matrix;

[0045] Calculate the canopy shading angle data for different tree species based on the photon absorptivity matrix;

[0046] Map the canopy shading angle data to the terrain skeleton grid to generate a set of tree planting points with shading coverage area identifiers;

[0047] The system receives shadow coverage area data from the landscape perception experience model, verifies the sunshine satisfaction rate of the tree planting point set, and outputs a three-dimensional point cloud of vegetation distribution with timestamps.

[0048] Preferably, the updated facility layout coordinates include:

[0049] Obtain the permeable material porosity parameter from the material replacement instruction and calculate the predicted value of rainwater retention per unit area.

[0050] Input the predicted rainwater retention value into the surface runoff simulator to generate a runoff velocity change curve.

[0051] Adjust the path slope parameters in the spatial topology network based on the runoff velocity variation curve;

[0052] Based on the updated path slope parameters, the topological connectivity between the facility layout coordinates and the drainage network is recalculated, and the optimized three-dimensional coordinate set of the facility is output.

[0053] Preferably, the output site design scheme map includes:

[0054] Verify the matching degree between the plant seasonal color parameters in the final solution data package and the seasonal markers of the 3D scene rendering sequence;

[0055] When a color deviation exceeding the tolerance threshold is detected, the material spectral reflection engine is triggered to regenerate the leaf surface reflectivity curve.

[0056] The updated leaf reflectance curves are fed back into the landscape effect distribution model to recalculate the light reflection parameters of the three-dimensional point cloud of vegetation distribution.

[0057] Based on the recalculated light reflection parameters, a scheme maturity assessment report is generated, and a site design scheme map containing facility coordinate correction data and a list of plant configuration changes is output.

[0058] Compared with the prior art, the beneficial effects of the present invention are:

[0059] By collecting and merging target site environmental datasets into geospatial raster processing units to generate site thermal distribution cloud maps, designers can gain a more comprehensive and accurate understanding of the site's environmental characteristics. This data-driven approach overcomes the limitations of traditional manual sampling, presenting site topography, climate, vegetation, and other information in a visual form, providing a detailed foundation for subsequent design stages.

[0060] Extracting the spatial pattern feature matrix of a site based on a site thermal distribution cloud map helps to deeply analyze the site's internal structure and spatial relationships. Compared to the vague understanding of site pattern in traditional design, the extraction of the feature matrix enables a quantitative expression of site information, allowing designers to understand the distribution patterns and interactions of various site elements from a data perspective, thus providing a basis for the scientific construction of landscape schemes.

[0061] A landscape effect distribution model was constructed, decomposing the spatial pattern feature matrix into topographic relief parameters, seasonal color parameters, and light reflection parameters, and generating a 3D scene rendering sequence. This provides a powerful tool for the visualization and effect prediction of design schemes. The 3D scene rendering sequence can dynamically display the landscape's presentation under different conditions, allowing designers to intuitively perceive the impact of topographic changes, seasonal plant changes, and light changes on the landscape before implementation, thus enabling detailed adjustments to the scheme during the design phase.

[0062] Dynamic optimization of the landscape design is implemented by generating element replacement instructions based on material texture parameters in the 3D scene rendering sequence and plant growth simulation data, while simultaneously updating the spatial topology network in the landscape effect distribution model. This enables the design to be dynamically adjusted. This dynamic optimization mechanism can respond promptly to changes in site conditions and design requirements, avoiding the problems of fixed and difficult-to-modify schemes in traditional designs. Through element replacement and updating of the topology network, the coordination and adaptability among various elements in the scheme are ensured, making the landscape design more in line with actual needs in terms of ecological function and user experience.

[0063] When outputting the site design scheme map, the coordinates of facility layouts in the 3D scene rendering sequence are adjusted using element replacement commands, and the path connectivity in the spatial topology network is referenced to generate the final scheme data package, improving the completeness and feasibility of the scheme. Precise adjustment of facility layout coordinates ensures the rational distribution of facilities on the site, while consideration of path connectivity optimizes the pedestrian experience, ensuring the landscape not only has good aesthetic appeal but also fully utilizes its functionality. The final generated scheme data package integrates various data and parameters from the design process, providing comprehensive guidance for the construction and maintenance of the scheme, and promoting the development of landscape architecture design towards a more scientific, efficient, and precise direction. Attached Figure Description

[0064] Figure 1 This is a timeline diagram of the digital twin-based landscape design method described in this invention.

[0065] Figure 2 A flowchart for collecting environmental data from the target site;

[0066] Figure 3 This is a comprehensive analysis diagram of the site's thermal distribution;

[0067] Figure 4 Flowchart for constructing a landscape effect distribution model;

[0068] Figure 5 Visual corridor and light and shadow analysis diagram;

[0069] Figure 6 A flowchart for dynamic optimization of landscape design;

[0070] Figure 7 A flowchart for defining the boundaries of the landscape transition zone. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] Please see Figure 1 This invention provides a landscape design method based on digital twins, the method comprising:

[0073] The system is data-driven at its core. A set of environmental data for the target site is collected, and this data is input into a geospatial raster processing unit to generate a site thermal distribution cloud map. This thermal distribution cloud map is used to extract the site's spatial pattern feature matrix. The landscape effect distribution model defines a 3D numerical model that integrates site topography, vegetation, and illumination data to simulate the landscape presentation effects at different seasons and times. Its core function is to generate a 3D scene rendering sequence and a 3D point cloud of vegetation distribution. This point cloud data is precisely matched with the topographic skeleton grid of the digital twin scene, achieving a 3D visualization layout of vegetation within the twin scene. The model input is a spatial pattern feature matrix (topographic relief, seasonal color, and illumination reflection parameters), and the output is timestamped landscape effect data (such as scene renderings of summer noon and winter dusk). The landscape effect distribution model is constructed, where the spatial pattern feature matrix is ​​decomposed into topographic relief parameters, seasonal color parameters, and illumination reflection parameters. These parameters drive the landscape effect distribution model to generate a 3D scene rendering sequence. The light reflection parameters are calculated using categorized formulas. Data consistency and model adaptation verification were completed before parameter input. These parameters are used to quantify the reflectivity of different surfaces within the site to sunlight. The light reflection parameters need to be calculated separately for three core landscape elements: vegetation surfaces, hard paved surfaces, and water surfaces. This method is suitable for simulating the lighting of 3D point clouds of vegetation distribution. The calculation steps and formulas are as follows:

[0074] Atmospheric attenuation correction factor ( The formula to correct for the weakening effect of air humidity and particulate matter on incident light is:

[0075]

[0076] in, Relative humidity (%) Air particulate matter concentration (unit: );

[0077] Leaf reflectance parameters Combining the physical properties of the leaf surface with geometric angles, the formula is:

[0078]

[0079] in, Leaf surface roughness (dimensionless, values ​​0.1-0.8, from the plant optical properties table). Leaf refractive index (dimensionless, values ​​1.3-1.6, from the plant optical property table). Solar altitude angle (unit: °).

[0080] The calculation steps and formulas for paving material lighting and shadow simulation applicable to 3D scene rendering sequences are as follows: Paving surface absorption coefficient Based on the color brightness and roughness of the paving material, the formula is:

[0081]

[0082] in, The lightness of the paving color (dimensionless, value 0-1, derived from material texture parameters). The surface roughness of the paved surface (dimensionless, value 0.05-0.5, derived from material texture parameters);

[0083] Pavement Reflection Parameters Combining the incident light and the geometric angle, the formula is:

[0084]

[0085] in, The absorption coefficient of the paved surface (dimensionless). Solar altitude angle (unit: °).

[0086] This method is suitable for simulating light and shadow effects on water bodies such as lakes and streams within a site. The calculation steps and formulas are as follows:

[0087] water surface reflection angle According to the law of reflection of light, the formula is:

[0088]

[0089] in, The angle at which sunlight strikes the surface of a body of water (unit: °, equal to) , (Solar altitude angle)

[0090] Water body reflection parameters ( Combining the refractive index of water with the angle of incidence, the formula is:

[0091]

[0092] in, The incident angle at the water surface (unit: °). The angle of reflection from the water surface (unit: °).

[0093] Calculated light reflection parameters ( , , The calculations need to undergo two verification steps to ensure accuracy before being applied to the model. For data consistency verification, the calculated results are compared with the leaf reflectance curve in the plant optical attribute table and the paving reflectance range in the material texture parameters; the deviation must be controlled within ±5%. For model adaptation verification, the parameters are input into the landscape perception experience model to simulate the shadow coverage area at a fixed time period, and compared with the actual shadow area of ​​the site captured by drone remote sensing during the same period; the error must be less than ±10%. After successful verification, the parameters are directly used to drive the landscape effect distribution model to generate a 3D scene rendering sequence, ensuring that the light and shadow presentation of different surfaces is consistent with actual physical laws.

[0094] Dynamic optimization of the landscape scheme is executed, generating element replacement instructions based on material texture parameters in the 3D scene rendering sequence and plant growth simulation data, while simultaneously updating the spatial topology network in the landscape effect distribution model. Finally, a site design scheme atlas is output, with the element replacement instructions used to adjust the facility layout coordinates in the 3D scene rendering sequence, and the final scheme data package generated by combining path connectivity in the spatial topology network. The entire process relies on a big data processing platform, integrating remote sensing equipment, IoT sensors, and spatial analysis algorithms to ensure the accuracy and adaptability of the design scheme.

[0095] Example 1: See Figure 2 The acquisition of the target site environmental data set involves the systematic integration and processing of multi-source heterogeneous data. Topographic elevation point clouds were acquired using a mobile measurement platform equipped with a LiDAR scanner, achieving an accuracy of ±2mm, and were used to construct the basic topographic model for the digital twin scene. The scanner covered the entire target site at a density of no less than 15 points per square meter, generating a discrete point dataset containing 3D coordinates and elevation values. Vegetation distribution vector data utilized high-resolution satellite remote sensing imagery as input, employing an object-oriented image segmentation algorithm to extract the boundaries of different vegetation types. The segmentation process set spectral thresholds and texture rules, outputting a polygonal vector layer containing classifications such as trees, shrubs, and ground cover. Hydrological monitoring time-series data was acquired in real-time by IoT sensor nodes deployed along the edges of rivers, lakes, and wetlands. The sensors recorded water level, elevation, flow velocity, and turbidity parameters at 30-minute intervals, and uploaded them wirelessly to a central database for storage as timestamped serialized data. Historical meteorological records were obtained from regional meteorological observation stations, including formatted datasets of daily extreme temperatures, cumulative rainfall, and wind direction and speed over the past ten years.

[0096] The UAV remote sensing system performs the task of identifying land cover types. A multi-rotor UAV, equipped with a five-channel multispectral camera, flies at an altitude of 100 meters along a preset trajectory, acquiring visible and near-infrared images at a ground resolution of 0.05 meters. The raw images, after radiometric and geometric correction, are input into a pixel-level classification model. This model, based on a random forest algorithm, trains a land cover type recognizer and outputs a classification raster containing eight categories: hard paving, bare soil, water bodies, evergreen vegetation, and deciduous vegetation. Geological structural parameters are extracted from regional geological exploration reports, specifically including rock strata dip measurements, fault line spatial coordinates, and weathering layer thickness data, and stored as geographic entity objects with a spatial reference system.

[0097] Surface cover classification rasters and geological structural parameters were overlaid and analyzed on a GIS data processing platform. The overlay process employed raster algebraic operations: first, the geological structural parameters were converted into a 500×500 mm resolution base raster layer using a vector-to-raster tool; second, the surface cover classification rasters were resampled to match their resolution; finally, pixel-level spatial correlation calculations were performed, whereby each raster cell simultaneously recorded the cover type code and the corresponding geological attribute code, generating a geospatial raster integrating geological and ecological attributes. This raster was stored in GeoTIFF format, and the spatial coordinate system adopted the UTM projection system.

[0098] Soil moisture time-series data were acquired using embedded capacitive sensors. Sensor nodes were arranged in a 20×20 meter grid within the site, measuring the volumetric water content at a depth of 0-30 cm every 60 seconds and recording the timestamp. Air particulate matter concentration monitoring employed a fixed laser scattering particulate counter, mounted on a support 2.5 meters above the ground, continuously recording the minute-by-minute average values ​​of PM2.5 and PM10. These parameters were processed by an ecological risk assessment model: first, a correlation matrix between soil moisture and vegetation water requirement was established; second, a response function between particulate matter concentration and plant stomatal conductance was constructed; and finally, a weighted comprehensive evaluation algorithm was used to calculate the ecological sensitivity index of each grid cell within the range of 1-100, and this index was written as a new attribute field into the geospatial raster file. This raster data was then imported into digital twin modeling software to assign geological and ecological attributes to the twin scene.

[0099] Historical visitor trajectory data is extracted from the scenic area's ticketing system and mobile application backend. After data cleaning to remove GPS drift points, a trajectory network dataset containing visitor path coordinates and dwell time is generated. Real-time pedestrian flow heatmaps are generated using AI video analysis cameras deployed along main passageways, updating the spatial density distribution map every 5 minutes. These two types of data are fused in a spatial analysis engine: the clustering intensity of trajectory points in the spatial grid is calculated using a kernel density estimation algorithm; a spatiotemporal matching algorithm is used to associate the spatial distribution pattern of the video heatmap with historical trajectories; finally, high-density clustered areas with instantaneous pedestrian density exceeding 2 people / square meter are identified, and the set of planar coordinates of their geometric center points is output. These coordinate points are overlaid on the geospatial raster as a point layer.

[0100] The ecological sensitivity index and high-density clustered coordinate point set are fused through spatial weighted overlay. The weighting process assigns different calculation factors: the geological stability factor extracts the shear strength coefficient of rock strata from geological structural parameters; the ecological carrying capacity factor is calculated based on the regression relationship between the ecological sensitivity index and vegetation biomass; and the activity pressure factor constructs a pressure index model based on pedestrian density and frequency. In practice, a raster calculator is used to assign weights of 0.4, 0.3, and 0.3 to the three factor layers respectively, performing multi-band raster overlay operations to output a composite site thermal distribution cloud map containing the geological stability layer, ecological carrying capacity layer, and activity pressure layer. This cloud map is stored in a three-dimensional data structure, with three characteristic layers arranged sequentially along the Z-axis. Each layer's raster cells store standardized values ​​for the corresponding attributes. Spatial reference information is stored in the WGS84 coordinate system, and the raster data is encapsulated in a multi-dimensional HDF5 file format.

[0101] See Figure 3 This study uses high-resolution grid data to present the comprehensive analysis results of the site's thermal distribution. Grayscale cloud maps represent the distribution of comprehensive thermal values; color depth is positively correlated with thermal value, reflecting the weighted combined effect of geological stability, ecological carrying capacity, and human activity pressure. The map includes eight high-density clustered areas (marked with white circles), where human activity intensity is significantly higher than surrounding areas. Black contour lines indicate the gradient of thermal value changes, with numerical labels showing specific thermal value levels. Circular numerical markers display the geological stability coefficient, ranging from 5.2 to 8.7, reflecting differences in geological conditions across different areas. Gray dotted areas represent highly ecologically sensitive areas, mainly distributed along simulated rivers (X=30-70 meters) and forest areas (X=60-80 meters, Y=60-80 meters). The map includes a scale bar (10 meters) and a north arrow for spatial reference. This provides multi-dimensional spatial analysis for design schemes, helping designers identify key areas and develop differentiated strategies.

[0102] Example 2: See Figure 4The site's thermal distribution cloud map serves as the input source to the visual corridor analysis module. This module initiates an automatic terrain feature extraction program, using an edge detection algorithm to identify boundary points of mountain areas within the site. Connecting continuous boundary points forms closed polygons, constituting the mountain outline data, which is stored as vector polyline features. Hydrological direction data, based on previously collected river monitoring time series, undergoes flow direction analysis. The D8 algorithm is used to calculate the main water body flow direction vectors, generating a river system direction line dataset. The spatial geometry calculation engine performs intersection point calculations on the two types of line features: establishing a two-dimensional plane coordinate system to perform spatial intersection analysis of line features, outputting a set of visual focal point coordinates with unique geographic coordinate identifiers. This coordinate set records the plane X / Y values ​​and corresponding elevation information.

[0103] A spatial correlation model was established between the visual focus coordinates and the solar trajectory data generated by the astronomical calculation model. The solar trajectory was calculated using the solar declination angle formula and hour angle to generate a time series of solar altitude angle and azimuth angle, with a time step set to 15 minutes. The spatial correlation calculation adopted a rasterization process: a spatial index grid covering the entire site was established, and the illumination angle variable value under the solar azimuth at different times was calculated for each visual focus coordinate. Based on this, a light and shadow intensity gradient map was generated. Each cell of this raster layer stores the light intensity change value within a specific time period, and the data format is a single-precision floating-point array.

[0104] The intensity gradient map of light and shadow changes is input into the spatial clustering processing unit. An improved K-means clustering algorithm is used for partitioning: three initial values ​​are set for the centroids of the feature categories, and the feature vector includes the light gradient value, elevation standard deviation, and vegetation cover. After five iterations to reach a stable state, the grid cells that meet the attributes of the ecological core area are marked (the main characteristics are that the average light change value is less than 0.5 and the elevation change is gradual), the cells of the densely active area are identified (characterized by the peak frequency of the light gradient exceeding 2 times / hour), and the remaining transitional areas are marked as landscape transition zones. The algorithm outputs a vector dataset containing three closed boundary lines, each boundary consisting of a geometry composed of no less than 200 vertices. The boundary data is imported into a digital twin scene to delineate functional areas in the twin scene.

[0105] The vegetation seasonal change database is linked to the landscape transition zone boundary through spatial join operations. The database consists of phenological observation records of tree species, including leaf hue values ​​(HSV color space), color duration days, and color change sensitivity parameters for typical tree species in spring, summer, autumn, and winter. The join operation establishes the correspondence between seasonal parameters and spatial units: using the landscape transition zone boundary as the spatial query condition, all vegetation distribution vector data within that area are extracted; spatial statistics are performed on the vegetation seasonal parameters according to polygonal features to calculate the seasonal change characteristic values ​​within the unit area. This process outputs a spatial pattern feature matrix: the matrix row index corresponds to the 100×100 meter spatial grid unit code, and the column attributes include spatial perception intensity values ​​(quantized from 0-100 based on visual focus density), color change frequency values ​​(number of color changes per unit area per year), and material attenuation coefficients (surface weathering rate values ​​estimated based on geological structural parameters). The matrix is ​​stored in a sparse matrix compression format, with a file size approximately 30% of the original data.

[0106] Topographic relief parameters are deconstructed and separated from the feature matrix. Parameter decoupling calculations are performed: the slope vector group calculates the grid surface normal vector using DEM data, storing it as a 3D vector set in the form of (ΔX, ΔY, ΔZ); the elevation difference matrix records the absolute value of the vertical elevation difference between adjacent grid cells, forming an N×N symmetric matrix. A 3D spatial topological relationship model is input to generate a terrain skeleton mesh: a constrained Delaunay triangulation algorithm is used to process the terrain point cloud, iteratively optimizing the maximum side length of triangles to no more than 5 meters, generating a mesh dataset containing triangular facet index relationships, with the total number of triangular faces approximately four times the site area.

[0107] Seasonal color parameters are input into the material's spectral reflectance engine. This engine integrates a vegetation optical parameter library and a physical optics model: for maple-like red-leaved trees, it calls a reflectance lookup table for the 300-700 nm wavelength band; for coniferous trees, it loads bidirectional reflectance distribution function parameters. The engine kernel performs radiative transfer calculations and outputs a plant optical attribute table data structure: the primary key of the attribute table is the plant species number; the leaf reflectance curve is stored as a wavelength-reflectance value point sequence (sampling interval 10 nm); and the canopy transmittance curve records the transmittance coefficient variation within the solar altitude angle range of 0-90 degrees. Each record contains approximately 150 data points.

[0108] Spatial mapping is performed between the plant optical attribute table and the topographic skeleton mesh. A spatial correspondence index is created: attribute allocation relationships are established based on the overlay analysis of plant distribution vector data and mesh cells. A bilinear interpolation algorithm is used to distribute optical parameters to mesh vertices: each triangular facet vertex simultaneously records geographic coordinates and binds spectral parameter attributes, forming vegetation distribution point cloud data that integrates three-dimensional spatial attributes and optical features. The system automatically adds seasonal marker attributes: based on a preset phenological calendar, spatiotemporal locations are matched, and each data point in the point cloud is accompanied by an identifier code for the current phenological stage (spring = 0x01, summer = 0x02, autumn = 0x03, winter = 0x04).

[0109] Light reflection parameters are synchronously loaded into the landscape perception experience model. This model initializes the sunlight incident simulation environment: a spherical model of the sun's trajectory is established in a 3D coordinate system, with a time resolution set to 30 minutes. The model kernel initiates ray projection calculations: for vegetation distribution point cloud data, the dynamic changes in the tree canopy projection area boundary under varying solar azimuth angles are calculated. Each calculation cycle outputs shadow coverage area data records, including a timestamp (year, month, day, hour, minute), the coordinate set of the projection polygon vertices, and the coverage area value (square meters). Shadow coverage data is rendered in real-time in the digital twin scene, intuitively presenting the landscape lighting effects at different times. The resulting data is organized and stored in a time series, with each object containing no fewer than 48 observation records within 24 hours. All shadow polygons are associated with the topological identifiers of the vegetation distribution point cloud, forming a complete data chain of lighting effects. The final 3D scene rendering sequence file is encapsulated in a hierarchical data structure, with the upper layer storing the mesh topology and the lower layer associated with the temporal shadow change dataset.

[0110] See Figure 5Professional topographic visualization technology is used to showcase the visual corridors and light and shadow analysis results. The grayscale topographic map represents the distribution of light and shadow intensity variations, with color gradients reflecting the degree of solar radiation variation, ranging from light gray (low variation) to dark gray (high variation). The map includes five visual focal points (marked with stars), which possess optimal viewing conditions and landscape value. Polygonal boundaries of different line types identify the spatial extent of the ecological core area (solid lines), the high-activity area (dashed lines), and the landscape transition area (dotted lines). Text annotations display the specific light intensity values ​​for each visual focal point, ranging from 62.3 to 85.7. Dashed lines represent key visual corridor connections, demonstrating the line-of-sight relationships between visual focal points. The 3D sub-map in the upper right corner displays light and shadow intensity variations from a topographic perspective, intuitively presenting the spatial distribution characteristics of lighting conditions. The ecological core area is located in the northwest of the site (X=20-40 meters, Y=30-50 meters), where light and shadow variations are gentle (average intensity 68.2), making it suitable for establishing an ecological protection zone. The high-activity area is located in the southeast (X=60-90 meters, Y=50-80 meters), where there are significant variations in light intensity (average intensity 78.5), making it suitable for arranging activity spaces. The landscape transition zone connects the two main functional areas, with a clear gradient of light and shadow (intensity range 65.3-82.1), making it an ideal area for creating rich landscape effects. This provides a scientific basis for landscape visual organization and functional zoning.

[0111] Example 3: See Figure 6 The material parameter processing module in the 3D scene rendering sequence receives the reflectivity data stream of the paving material. This data originates from structured records output by the material spectral reflectance engine, with each record containing a material number, wavelength range (380-780 nm), and a corresponding array of reflectivity values. The plant root growth simulator synchronously loads the site soil profile model, which integrates previously acquired spatiotemporal distribution data of soil moisture and geological structural parameters. The simulator predicts the root expansion morphology of woody plants based on a finite diffusion condensation model: setting the time span to a ten-year growth cycle with a three-month step, it calculates the percentage of root coverage area of ​​different tree species in the horizontal plane. The coverage area calculation uses an iterative projection algorithm, updating the set of coordinates of the root projection boundary polygons in each iteration.

[0112] The material reflectivity threshold monitor continuously scans the input data stream. The preset reflectivity threshold is defined as a reflectance value of 0.35 at a wavelength of 550 nanometers, derived from the national standard parameter in the building materials optical properties database. When the reflectivity of a paving material at the reference wavelength exceeds the threshold, the material replacement instruction generator activates the logic unit. The instruction structure contains three core fields: a unique identifier for the permeable material (referencing the international material coding system), a percentage value of the porosity parameter of the porous structure (e.g., 15% ± 2%), and an effective pore size distribution curve (recording the relationship between millimeter-level pore size and quantity). The identifier is associated with the permeable concrete entry "PM-2015" in the material properties database.

[0113] The spatial topology network processor receives material replacement commands. This network is linked to the facility layout model in the digital twin scene, and the material replacement effect is previewed in real time within the twin scene. The network data structure uses an adjacency list storage format: nodes represent the spatial location of landscape facilities, and edges store the topological connections between facilities and path surface attributes. Commands trigger a dynamic update mechanism for network parameters: first, the identifier of the path edge containing the material to be replaced is retrieved; second, the edge attribute fields are rewritten using the permeable material porosity parameter. The runoff permeability coefficient recalculation engine calls the hydraulic conduction model to perform recalculation.

[0114]

[0115] The symbols in the formula are defined as follows: This indicates the dimensionless value of the porosity parameter of the porous structure in the input command. The average pore diameter of the material (in meters). Indicate the water flow density parameter (fixed value is 997 kg / m³); This represents the gravitational acceleration constant (taken as 9.81 m / s²). The viscosity coefficient of the water flow is set to 0.89 × 10^{-3} Pa·s. Output the updated permeability coefficient (in meters per second).

[0116] The formula output is written to the physical attribute field of the spatial topology network edge, overwriting the original parameter values. After the network change monitor detects changes in the permeability coefficient, it activates the surface runoff simulator interface. The simulator loads the initial flow condition dataset, including topographic skeleton grid elevation data, rainfall intensity distribution from historical meteorological records, and the updated permeability coefficient matrix. The numerical simulation uses a two-dimensional shallow water equation solver with a time step of 10 seconds, and the spatial grid resolution is aligned with the size of the site's thermal distribution cloud map cells. Each calculation cycle outputs a surface runoff direction vector field and peak velocity dataset.

[0117] The updated facility coordinate set is written into the final solution data package, and a coordinate change log is generated to record the changed vector values. The entire process is implemented using a distributed computing framework: the material parameter processing module runs on CPU nodes, plant root system simulation is accelerated using GPUs, and runoff calculation is deployed on a high-performance computing cluster. Data transmission uses the Apache Arrow in-memory format to reduce serialization overhead. The iterative optimization termination condition is set to automatically stop outputting the final result when the mean square error of three consecutive coordinate movement distances is less than 0.15 meters.

[0118] The generation process of the site thermal distribution cloud map inherits the output results of the geospatial raster processing unit. The geological structural parameter dataset is input into the Kriging interpolation calculation engine: the semi-variogram model is set to a spherical model, the nugget value is set to 0.05, and the range parameter is adaptively adjusted to 120 meters according to the site size. A geological base raster layer is calculated and generated, with each 500×500 mm raster cell storing the rock stratum stability coefficient (0-1 dimensionless value). The ecological sensitivity index matrix is ​​processed using the inverse distance weighting method: based on the monitoring point coordinates, the distance attenuation exponent is set as a quadratic function, and an elevation correction factor η is added.

[0119]

[0120] in: This is the current elevation value. It is the regional average elevation. Represents the maximum and minimum elevations of the site. The output ecological carrying capacity raster cell value ranges from 0 to 100. The activity pressure layer data is input into a convolutional neural network model: the network architecture contains three convolutional layers (kernel size 7×7, stride 2) and one fully connected layer, outputting a population aggregation thermal value prediction matrix. Three sets of rasters are used to generate a site thermal distribution cloud map through a band synthesizer: the geological base raster is assigned a red channel, the ecological carrying capacity raster corresponds to a green channel, and the activity intensity raster is mapped to a blue channel, forming an RGB three-channel 3D raster dataset, with the spatial coordinate system consistent with the input data.

[0121] Example 4: See Figure 7 The data processing workflow is initiated by extracting the color change frequency values ​​from the spatial pattern feature matrix. The feature matrix repository retrieves column vectors containing vegetation seasonal parameters and loads a frequency value dataset from a 15-hectare landscape transition zone. This dataset records the annual color change frequency statistics for each grid cell, with parameter collection intervals of 10×10 meter grid systems per unit area. The color quantification analysis engine performs hue analysis on a grid-by-grid basis: cluster analysis is performed on the seasonal observation data of dominant vegetation within each cell to identify the dominant hue and the proportional relationship of transitional stages.

[0122] The seasonal color band builder receives color analysis results and generates continuous color bands. Taking the North American sweetgum population in the site as an example, the color band generation process includes four processing stages: First, extracting the main color parameters of the tree species in all four seasons: light green in spring (hue angle 85°), dark green in summer (120°), orange-red in autumn (25°), and grayish-brown in winter (40°); Second, calculating the hue gradient step size between adjacent seasonal phases, setting the daily color change rate to 0.25°; Third, creating a hue continuity transition loop in the HSV color space; Fourth, outputting a color band object containing a sequence of 365 gradient color patches. Each color patch is labeled with a date tag and its corresponding hue value, and the data structure is stored as a color lookup table with a time index.

[0123] The visual attractiveness measurement model initializes the eye-tracking simulation environment. This model is a mathematical model that quantifies the visual appeal of landscape colors and spatial layout to the human eye. Its core output is the predicted visual dwell time, used to classify spatial perception intensity levels. The model inputs are seasonal color gamut (hue, saturation) and visual focus coordinates. The output is the predicted dwell time (in seconds) for each spatial unit. The model loads seasonal color gamut and site 3D model data, generating a dynamic color rendering scene at virtual observation points. For the key autumn landscape analysis area, eight observation points were set (each 50 meters apart). Simulated human eye observation parameters included: a visual field angle of 120°, a baseline gaze duration of 2 seconds, and a saccade speed of 30° per second. The model performed hue combination attractiveness calculations: red-orange hues (hue range 15-35°) in the transition zone were designated as high-attention areas, and blue-green hues (170-210°) as low-attention areas. Table 1 shows the predicted visual dwell time values ​​for key color combinations.

[0124] Table 1: Visual Dwell Prediction Values ​​for Key Color Combinations.

[0125]

[0126] The spatial perception intensity grading system processes the model output data. A dwell time threshold is set: dwell times exceeding 2.5 seconds are classified as high-intensity perception zones, and dwell times below 1.8 seconds are classified as low-intensity perception zones. The site's spatial distribution is analyzed using a grid: the 500×500 meter site is divided into 100 analysis units, and the median predicted dwell time within each unit is calculated. The coordinate set of high-intensity perception zones is stored as a point feature layer, containing 28 core coordinate points and their perception intensity weight values ​​(continuous values ​​from 0.7 to 1.0); low-intensity perception zones generate polygon boundary range data.

[0127] The landscape transition zone vector boundary generator initiates spatial fusion calculations. It inputs the visual focus coordinate set (42 points generated previously) and the high-intensity perception zone coordinate set, performing a two-layer spatial overlay: first, a Thiessen polygon spatial weight allocation model is established, with each polygon representing the visual focus influence domain; second, kernel density analysis is performed (search radius 80 meters) to generate a composite weighted heatmap; finally, continuous area boundary lines with weight values ​​greater than 0.85 are extracted, and vector boundary data with weight identifiers are output. The number of boundary line vertices is controlled between 180 and 220, and each node records a local weight value (precision 0.01).

[0128] The photon absorptivity conversion is initiated during the 3D point cloud generation stage of vegetation distribution. The plant optical attribute database retrieves parameters for the *Liquidambar formosana* species and loads a leaf reflectance curve dataset (wavelengths 380-780 nm, 5 nm intervals). The converter performs photon capture efficiency calculations: a reference absorptivity is set in the 550 nm green light band, and the absorption coefficients for each band are calculated based on leaf anatomical parameters. A 22×22 absorptivity matrix is ​​generated, with rows representing wavelength indices and columns representing the depth index of leaf hierarchical structure.

[0129] The canopy shading angle simulator receives absorptivity matrix data. The simulation environment is set with a virtual light source (adjustable solar altitude angle 30-60°), and ray tracing is performed on a typical Chinese sweetgum tree canopy (6-8 meters wide, 12 meters high). The simulation outputs a set of shading angle parameters: the effective shading angle range during midday is 53° (east) to 72° (west), and 24 sets of direction-angle correspondence values ​​are recorded. Each shading angle is associated with a canopy projection area parameter (e.g., a projection area of ​​38.7 square meters at 55°).

[0130] The tree planting point set generation module performs spatial mapping. Elevation data (accuracy 0.1 meters) is loaded onto the terrain skeleton mesh, and shading angle data is converted into a sequence of projected polygons. The spatial allocation algorithm operates in three steps: First, constraints are established at the mesh vertices (slope less than 15° and elevation difference less than 3 meters); second, spatial overlay analysis is performed using the projected polygons as templates; third, the coordinate set of potential planting points that meet the sunlight conditions is output. Each planting point is labeled with shading coverage attributes, including the maximum projected diameter and effective shading period (e.g., 13:00-15:00).

[0131] The 3D point cloud verification process for vegetation distribution initiates real-time sunlight verification. The system loads shadow cover data (30-minute time resolution) generated by the landscape perception experience model and performs ray collision detection on the planting point set. Sunlight satisfaction rate is calculated using a time-cumulative method: daily effective sunlight duration is counted to ensure it exceeds 4 hours, and points that do not meet the standard are iteratively adjusted (movement step size 0.5 meters). The final output is time-stamped point cloud data, which dynamically displays vegetation seasonal changes in a digital twin scenario, simulating the landscape effects of different seasons. Each 3D point contains spatial coordinates (X, Y, Z), associated tree species ID, planting date label, and 96 time-period sunlight status markers (24 hours × 4 seasonal phases). The point cloud density is controlled at 15 points / 100 square meters, with a total data volume of approximately 120,000 records.

[0132] Example 5: The process of updating facility layout coordinates initiates the material parameter conversion module. The updated coordinates are synchronized to the digital twin scene to verify the adaptability of the facility layout to drainage requirements. The porosity parameter of the permeable material is extracted from the material replacement instruction structure. This parameter exists in the form of a percentage value, typically ranging from 12% to 25%. The system calls the hydraulic property knowledge base to associate the porosity with fluid motion characteristic parameters. The unit for calculating the rainwater retention per unit area performs a physical model conversion: based on the Stokes fluid dynamics principle, it integrates water density parameters and the gravitational acceleration constant, combined with the geometric characteristic parameters of the pore channels, to output the theoretical rainwater retention volume value, in liters / square meter·hour. This predicted value is imported into the surface runoff simulator environment as the core input. The rainwater retention per unit area (Q) is calculated by coupling the hydraulic properties of the permeable material with precipitation conditions, as shown in the following formula:

[0133]

[0134] Where Q is the rainwater retention capacity per unit area (unit: L / m²), K is the saturated hydraulic conductivity of the permeable material (unit: m / s), and t is the duration of a single rainfall event (unit: s). The saturated water content (dimensionless) of the permeable material is derived from the material porosity. =Porosity × 0.95, where 0.95 represents the effective water filling ratio of the material's pores, determined based on common permeable concrete test data); d represents the initial moisture content of the permeable material (dimensionless), which is collected in real time by an IoT soil moisture sensor and is the average value taken one hour before rainfall; d represents the effective thickness of the permeable material (unit: m), which is the thickness of the permeable layer in the paving structure, such as the thickness of permeable concrete layers, which is usually 0.15-0.2m.

[0135] The surface runoff simulator loads 3D terrain mesh data and an updated permeability matrix. The system initializes simulation parameters: the time step interval is set to 120 seconds, and the total simulation duration covers a 24-hour heavy rainfall cycle. The calculation engine executes the water balance equation for each grid cell, generating an independent water flow trajectory for each grid cell. The runoff velocity dynamic monitoring unit records the maximum velocity values ​​at the grid connection boundaries, forming a dataset of velocity change curves with timestamps. Each curve object contains at least 96 sets of time-velocity correspondence values, stored as binary floating-point arrays.

[0136] The spatial topology network editor receives runoff velocity variation curve data. The path slope parameter correction mechanism in the network structure is activated: the initial slope angle value is derived from the surface normal vector calculation data of the terrain skeleton grid. The adjustment algorithm scans the peak points of the velocity curve; when a sustained velocity exceeds a set critical point, the slope parameter is reassigned according to the gradient adjustment formula. Specifically, for each path object: first, the ratio of velocity to slope influence factor is calculated; second, the angle value in steep slope areas is dynamically reduced, with the adjustment range not exceeding 30% of the original value. The corrected slope parameters are updated in real time in the network topology file, and a change log is generated to record the operation timestamp.

[0137] The verification process for the topological connectivity of facility coordinates and drainage network initiates 3D spatial analysis. The initial set of facility layout coordinates uses facility coordinate values ​​from the 3D scene rendering sequence as baseline data. The network analysis module performs connectivity testing: creating virtual connection lines from facility nodes to the nearest drainage network and calculating the spatial relationship between the lines and terrain obstacles. The connectivity index is quantified as a proportionality coefficient of connectable nodes, with the coefficient value determined by both path distance and elevation difference. The optimizer iterates the coordinate positions based on the updated slope parameters: each adjustment step is controlled within the range of 0.2-0.5 meters, moving in the opposite direction to the slope normal. The iteration termination mechanism is set to automatically output results when five consecutive coordinate changes are all less than 0.15 meters. The final 3D coordinate set of facilities is stored according to facility category, including six major categories of coordinate records such as rest facilities, lighting equipment, and service facilities.

[0138] The output system of the site design scheme map loads the final scheme data package. The plant seasonal color matching verification unit starts the automatic detection program: reading seasonal marker attribute data from the 3D scene rendering sequence and associating it with the ideal hue value in the plant configuration database. The color deviation detector workflow includes: collecting hue parameters of vegetation point cloud in the actual scene; converting the hue angle difference in the HSV color model; and calculating the absolute value of the angle difference. The fault tolerance threshold is set to a hue difference not exceeding 8 degrees. When the hue deviation of three consecutive seasonal phases in a specific grid cell exceeds the threshold, the system triggers the spectral engine to recalculate the signal.

[0139] Upon receiving the recalculation command, the material spectral reflectance engine resets its initial parameters. The engine kernel adjusts the chlorophyll reflectance curve generation algorithm: prioritizing parameter sensitivity analysis for vegetation species in deviation areas and adding a light compensation coefficient to the spectral response function. The regenerated leaf reflectance curve uses a higher-resolution sampling method, increasing the wavelength sampling interval from 10 nanometers to 5 nanometers. The new reflectance curve dataset is fed back to the landscape effect distribution model's memory area via a real-time data channel.

[0140] The process of recalculating the illumination parameters of the 3D point cloud of vegetation distribution was initiated simultaneously. The model cleared the original illumination reflection parameter cache and reloaded the location attributes of the vegetation point cloud. The solar incidence angle calculator adopted a higher-precision astronomical algorithm, improving the angular resolution to the 0.1-degree level. The scene rendering engine performed full-time shadow simulation: the time interval was shortened to 15 minutes, covering the four key solar terms of spring equinox, summer solstice, autumn equinox, and winter solstice. Each vegetation point recorded 384 sets of illumination status marker values ​​(24 hours × 16 observation periods), expanding the dataset size to 1.8 times that of the original data.

[0141] The scheme maturity assessment report generator scans all corrected data records. The report data structure includes five core parts: a plant configuration change list detailing specific tree species adjustment suggestions, including increases and decreases for 12 tree species and 8 shrub species; a facility coordinate correction data table recording location changes for three types of facilities; a rendering parameter optimization log tracking parameter changes over five iterations; a summary of material replacement records showing the area of ​​permeable paving updates; and a landscape stability prediction value outputting a credibility index based on parameter convergence. The site design scheme map is packaged and output using the GIS platform's synthesis function: the facility coordinate correction data is converted into point feature layers, the plant configuration change list generates an attribute association table, and the entire package is packaged into a version-marked GDB geodatabase document. Each file block is appended with a digital signature verification code to ensure data integrity during transmission.

[0142] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

Claims

1. A digital-twin-based garden landscape design method, characterized in that, The method comprises the following steps: Collecting a target site environment data set, inputting the environment data set into a geospatial grid processing unit to generate a site thermal distribution cloud map; Based on the site thermal distribution cloud map, extracting a site spatial pattern feature matrix; Constructing a landscape effect distribution model, decomposing the spatial pattern feature matrix into a terrain undulation parameter, a seasonal color parameter, and a light reflection parameter, driving the landscape effect distribution model to generate a three-dimensional scene rendering sequence, the landscape effect distribution model being embedded in a three-dimensional digital twin scene of a garden site, model parameters being in real-time association with physical properties of terrain, vegetation, and light in the digital twin scene; Performing dynamic optimization of a landscape scheme, generating element replacement instructions according to material texture parameters in the three-dimensional scene rendering sequence and plant growth simulation data, and synchronously updating a spatial topological relationship network in the landscape effect distribution model; Outputting a site design scheme atlas, adjusting facility layout coordinates in the three-dimensional scene rendering sequence based on the element replacement instructions, and generating a final scheme data package in combination with path connectivity in the spatial topological relationship network; The collecting of the target site environment data set comprises: The target site environment data set contains terrain elevation point cloud, vegetation distribution vector, hydrological monitoring time series, and meteorological history record, and the collected data is used to construct a basic geometry and physical property library of a three-dimensional digital twin model of a garden site; Obtaining ground cover type distribution through a UAV remote sensing device, superimposing and analyzing the ground cover type and geological structure parameters to generate a geospatial grid; Receiving soil moisture time series and air particulate matter concentration uploaded by a networked sensor, and calculating an ecological sensitivity index of each unit in the geospatial grid; Dynamically matching historical tourist trajectory data and real-time people flow heat map, and marking high-density aggregation area coordinates in the geospatial grid; Fusing the ecological sensitivity index and the high-density aggregation area coordinates to generate a site thermal distribution cloud map containing a geological stability layer, an ecological bearing layer, and an activity pressure layer. 2.The digital-twin-based garden landscape design method according to claim 1, wherein, The extracting of the site spatial pattern feature matrix comprises: Performing visual corridor identification on the site thermal distribution cloud map, extracting intersection points of mountain contour lines and water system trend lines as visual focus coordinates; Calculating spatial correlation of the visual focus coordinates and sunshine trajectory data to generate a light and shadow change intensity gradient map; Inputting the light and shadow change intensity gradient map into a spatial clustering algorithm to divide boundaries of an ecological core area, an activity intensive area, and a landscape transition area; Associating a vegetation seasonal change database with the landscape transition area boundaries to output a spatial pattern feature matrix containing spatial perception intensity values, color change frequency values, and material attenuation coefficients. 3.The digital-twin-based garden landscape design method according to claim 1, wherein, The constructing of the landscape effect distribution model comprises: Decomposing a terrain undulation parameter in the spatial pattern feature matrix into a slope vector group and an elevation difference matrix, and driving a three-dimensional spatial topological relationship model to generate a terrain skeleton grid; Inputting a seasonal color parameter into a material spectral reflection engine to output a plant optical property table containing a leaf albedo curve and a tree crown transmittance curve; Mapping the plant optical property table to the terrain skeleton grid to generate a vegetation distribution three-dimensional point cloud with seasonal markers. Synchronously loading the lighting reflection parameters to the landscape perception experience model, calculating the shadow coverage area data of the vegetation distribution three-dimensional point cloud when the sunlight incident angle changes; Precise alignment of the vegetation distribution three-dimensional point cloud and the terrain skeleton grid of the digital twin scene, ensuring the visualization accuracy of the twin scene. 4.The digital-twin-based garden landscape design method according to claim 1, wherein, The execution of the landscape scheme dynamic optimization includes: Receiving the paving material reflectivity data in the three-dimensional scene rendering sequence, comparing the soil coverage area predicted by the plant root system growth model; When detecting that the paving material reflectivity exceeds the preset threshold, generating a material replacement instruction containing the number of water permeable materials and the porous structure parameters; Inputting the material replacement instruction into the spatial topological relationship network, recalculating the path surface runoff permeability coefficient; According to the permeability coefficient change amount, adjusting the surface runoff flow direction data in the landscape effect distribution model, and updating the drainage buffer distance of the facility layout coordinates. 5.The digital-twin-based garden landscape design method according to claim 1, wherein, The generation of the geographic space grid includes: Kriging spatial interpolation calculation of the geological structure parameters, generating a geological basement grid containing the rock layer stability coefficient; Inputting the ecological sensitivity index into the inverse distance weighting algorithm, outputting an ecological carrying capacity grid with an elevation correction factor; Processing the active stress layer data through a convolutional neural network to generate an activity intensity grid containing crowd gathering heat values; Multi-band superposition of the geological basement grid, ecological carrying capacity grid, and activity intensity grid to form a site heat distribution cloud chart with a three-dimensional coordinate system. 6.The garden landscape design method based on digital twinning according to claim 2, wherein, The division of the landscape transition zone boundary includes: Extracting the color change frequency value in the spatial pattern feature matrix to construct a seasonal color spectrum band containing a color phase gradient; Inputting the seasonal color spectrum band into a visual attraction measurement model to calculate the visual stay duration prediction value of different color phase combinations; According to the visual stay duration prediction value, dividing the spatial perception intensity level to generate a high-intensity perception area coordinate set and a low-intensity perception area coordinate set; Fusing the visual focus coordinates and the high-intensity perception area coordinate set to output a landscape transition zone vector boundary with a weight identifier. 7.The digital-twin-based garden landscape design method according to claim 3, wherein, The generation of the vegetation distribution three-dimensional point cloud includes: Converting the leaf reflectivity curve in the plant optical property table into a photon absorption rate matrix; According to the photon absorption rate matrix, calculating the crown layer shading angle data of different tree species; Mapping the crown layer shading angle data to the terrain skeleton grid to generate a set of arbor planting points with shading coverage range identifiers; Receiving the shadow coverage area data output by the landscape perception experience model, performing sunlight satisfaction rate verification on the set of arbor planting points, and outputting a vegetation distribution three-dimensional point cloud with a timestamp marker. 8.The digital-twin-based garden landscape design method according to claim 4, wherein, The update of the facility layout coordinates includes: Obtaining the water permeable material porosity parameters in the material replacement instruction, calculating the unit area rainwater retention prediction value; Inputting the rainwater retention prediction value into the surface runoff simulator to generate a runoff flow rate change curve; According to the runoff flow rate change curve, adjusting the path slope parameters in the spatial topological relationship network; Based on the updated path slope parameters, recalculating the topological connectivity between the facility layout coordinates and the drainage pipe network, and outputting an optimized facility three-dimensional coordinate set. 9.The digital-twin-based garden landscape design method according to claim 1, wherein, The output of the site design scheme atlas includes: Verifying the matching degree of the plant seasonal color parameters in the final scheme data packet and the season marker of the three-dimensional scene rendering sequence; When detecting that the color deviation exceeds the fault tolerance threshold, triggering the material spectrum reflection engine to regenerate the leaf reflectance curve; Feed the updated leaf reflectance curve to the landscape effect distribution model to recalculate the light reflection parameters of the vegetation distribution three-dimensional point cloud; Generate a maturity assessment report based on the recalculated light reflection parameters, and output a site design scheme atlas containing facility coordinate correction data and plant configuration change list.

Images