Urban green land landscape evaluation method and system based on three-dimensional data point cloud

Abstract

The invention relates to an urban green land landscape evaluation method and system based on a three-dimensional data point cloud. According to the method, a phenological time table containing key phenological events and an uncertain range is generated by constructing a vegetation phenological rhythm prediction model, an index sensitivity weight matrix calculated based on point cloud time sequence data is combined, a non-uniform acquisition strategy is generated by using a joint optimization model, point cloud data acquisition and feature extraction are performed on the basis, and a vegetation phenological rhythm prediction result is obtained. And finally, according to the phenological stage, dynamically selecting a high-sensitivity index for calculation, so that the vegetation morphology data with the most characterization is accurately acquired in the key phenological stage, and the contradiction between data redundancy and insufficient feature capture caused by a traditional fixed interval acquisition mode is effectively solved. And the data acquisition efficiency and the accuracy and pertinence of landscape evaluation are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of urban intelligent management, in particular relates to a kind of urban green space landscape evaluation method and system based on three-dimensional data point cloud. BACKGROUND

[0002] In the current urban green space planning and management, scientific and accurate evaluation of green space landscape is a key link to improve the quality of urban ecological environment. With the progress of three-dimensional information acquisition technology, especially the wide application of ground and airborne laser radar technology, the landscape evaluation method based on three-dimensional point cloud data is gradually becoming the mainstream, because it can provide millimeter to centimeter level precision of vegetation three-dimensional structure information, which surpasses the limitations of traditional manual measurement and two-dimensional remote sensing image. However, the application method of this advanced data source itself still has significant optimization space. The existing technology generally adopts a fixed time interval data collection mode in practice, such as periodic collection according to week, month or quarter. This uniform sampling strategy seems systematic, but fundamentally ignores the inherent, highly dynamic phenology rhythm characteristics of vegetation. Vegetation will go through several key phenological stages such as germination, leaf expansion, flowering and leaf fall in a year, and its morphological structure differs greatly in different stages; for example, the lush leaf period is most suitable for evaluating canopy greenness and canopy density, while the leaf fall period can best show its branch skeleton structure. The fixed collection frequency cannot be synchronized with these biological rhythms, resulting in insufficient data capture in the key time window when the morphological characteristics of vegetation are most representative, and instead a large amount of redundant data is collected in the period when the structure characteristics are stable or have no significant change, which not only wastes storage and computing resources, but also substantially reduces the overall efficiency of information acquisition.

[0003] Further, the existing landscape evaluation index system often presents static characteristics in application, and does not fully consider the sensitivity difference of different evaluation indexes to phenological stages. Whether it is to measure the ecological function of greenness and canopy complexity, or to evaluate the visual aesthetics of transparency and color richness, the effectiveness and discriminability of these indexes are different in different growth stages of vegetation. The existing method usually uses the same set of indexes for evaluation throughout the year, or does not weight the indexes according to the phenological stages, which leads to a significant reduction in the accuracy and reliability of the evaluation results in certain periods (such as leaf fall period still forced to use greenness evaluation). At the same time, due to the selection of collection time points not being linked with the phenological characteristics and index sensitivity, the time series data obtained lacks consistency and comparability in longitudinal comparison, making it difficult to truly and accurately reflect the long-term dynamic change trend and rule of green space landscape.

[0004] Therefore, there is an urgent need in the art for a new data collection and analysis method that can deeply integrate vegetation phenology rules and landscape evaluation theory, to replace passive fixed recording with intelligent scheduling strategies, thereby obtaining the most valuable point cloud data at the lowest cost while ensuring data quality and temporal comparability, and achieving more accurate and efficient evaluation of urban green space landscapes. SUMMARY

[0005] Based on this, it is necessary to provide a city green space landscape evaluation method and system based on three-dimensional data point cloud in view of the above technical problems.

[0006] In a first aspect, the present application provides a city green space landscape evaluation method based on three-dimensional data point cloud, comprising:

[0007] S1, constructing a vegetation phenology rhythm prediction model according to historical phenological observation data, historical meteorological data and historical vegetation species information of a target area; outputting a phenological event time table containing key phenological event types and their predicted occurrence time uncertainty range based on the vegetation phenology rhythm prediction model;

[0008] S2, calculating the information entropy and coefficient of variation of each landscape evaluation index at different phenological stages based on point cloud time series data covering the entire phenological period; calculating the sensitivity weight of each landscape evaluation index at different phenological stages by entropy weight method based on the information entropy and coefficient of variation, to obtain a sensitivity weight matrix;

[0009] S3, inputting the phenological event time table and the sensitivity weight matrix into a joint optimization model, and generating a non-uniformly spaced point cloud data collection time sequence scheduling strategy by solving a multi-objective optimization problem considering both information value maximization and collection cost minimization;

[0010] S4, collecting three-dimensional point cloud data at the time points determined by the point cloud data collection time sequence scheduling strategy to obtain a point cloud data set, and extracting vegetation structure features from the point cloud data set;

[0011] S5, dynamically selecting high-sensitivity landscape evaluation indexes at the current phenological stage according to the sensitivity weight matrix; calculating the high-sensitivity landscape evaluation indexes based on the vegetation structure features to obtain the landscape evaluation results.

[0012] In a second aspect, the present application also provides a city green space landscape evaluation system based on three-dimensional data point cloud, which is used to implement the method described in the first aspect, and the system comprises:

[0013] The phenology dynamic modeling module is configured to construct a vegetation phenology rhythm prediction model according to historical phenology observation data, historical meteorological data and historical vegetation type information of the target region, and output a phenology event schedule containing a key phenology event type and an uncertainty range of a predicted occurrence time of the key phenology event type according to the vegetation phenology rhythm prediction model.

[0014] The index sensitivity quantification module is configured to calculate information entropy and a coefficient of variation of each landscape evaluation index at different phenology stages based on point cloud time series data covering a complete phenology cycle, and calculate a sensitivity weight of each landscape evaluation index at different phenology stages based on the information entropy and the coefficient of variation by using an entropy weight method, to obtain a sensitivity weight matrix.

[0015] The acquisition strategy optimization module is configured to input the phenology event schedule and the sensitivity weight matrix into a joint optimization model, and generate a non-uniformly spaced point cloud data acquisition time sequence scheduling strategy by solving a multi-objective optimization problem considering maximization of information value and minimization of acquisition cost.

[0016] The point cloud feature extraction module is configured to perform three-dimensional point cloud data acquisition at time points determined according to the point cloud data acquisition time sequence scheduling strategy, to obtain a point cloud dataset, and extract vegetation structure features from the point cloud dataset.

[0017] The dynamic evaluation generation module is configured to dynamically select a high-sensitivity landscape evaluation index at a current phenology stage according to the sensitivity weight matrix, and calculate the high-sensitivity landscape evaluation index based on the vegetation structure features, to obtain a landscape evaluation result.

[0018] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the method for evaluating urban green space landscape based on three-dimensional data point cloud according to the first aspect when executing the computer program.

[0019] In a fourth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method for evaluating urban green space landscape based on three-dimensional data point cloud according to the first aspect.

[0020] The urban green space landscape evaluation method and system based on the three-dimensional data point cloud can realize accurate collection of the most representative vegetation morphological data in the key phenological period, effectively solve the contradiction between data redundancy and feature capture deficiency caused by the traditional fixed interval collection method, and significantly improve the data collection efficiency and the accuracy and pertinence of the landscape evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0022] Figure 1 A flowchart of an urban green space landscape evaluation method based on three-dimensional data point cloud provided by the present application is shown in the figure.

[0023] Figure 2 A structural diagram of an urban green space landscape evaluation system based on three-dimensional data point cloud provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0025] REFERENCE Figure 1 A flowchart of an urban green space landscape evaluation method based on three-dimensional data point cloud provided by the present application is shown in the figure. The method comprises the following steps:

[0026] S1, according to the historical phenological observation data, historical meteorological data and historical vegetation species information of the target area, a vegetation phenological rhythm prediction model is constructed; according to the vegetation phenological rhythm prediction model, a phenological event time table containing the uncertainty range of the type and predicted occurrence time of the key phenological event is output.

[0027] Specifically, the first step is to systematically collect three types of core data from the target area to provide a reliable foundation for model construction. The first type is historical phenological observation data, covering key growth event records for all dominant vegetation species in the region. Key growth events include, but are not limited to, the onset of budding, the onset of leaf expansion, the peak of leaf expansion, the onset of flowering, the peak of flowering, the fruiting period, the onset of leaf fall, and the end of leaf fall. The observation period needs to cover multiple complete phenological years to capture the impact of climate fluctuations in different years on phenology. Data sources can prioritize long-term field observation records from urban landscaping management departments and standardized observation data from national or local phenological observation stations. Cross-validation can also be performed using phenological products retrieved from high-resolution remote sensing images. Outliers in the observation data are identified and removed using the Laida criterion. For missing observation data, linear interpolation or machine learning interpolation methods are used to supplement the data based on the phenological change patterns of adjacent years for the same species and the meteorological conditions during the same period, ensuring the integrity of the data sequence.

[0028] The second category is historical meteorological data, which includes key meteorological factors directly affecting vegetation growth and development. These factors include daily average temperature, daily cumulative precipitation, daily sunshine hours, accumulated temperature above specific temperature thresholds, extreme minimum temperature, and relative humidity. The data has a daily temporal resolution to accurately reflect the driving effect of short-term meteorological changes on phenological processes. The data time span is consistent with historical phenological observation data to ensure temporal compatibility. Data sources include the National Meteorological Data Sharing Platform and regional meteorological observation networks. For target areas with wide spatial coverage, if there is uneven distribution of meteorological stations, spatial interpolation is used to spatially grid the meteorological data, ensuring that the spatial resolution of the meteorological data matches the spatial scale of the target area. Simultaneously, the acquired meteorological data undergoes quality calibration by comparing it with data from neighboring benchmark meteorological stations during the same period to correct systematic errors and ensure the accuracy of the meteorological data.

[0029] The third category is historical vegetation species information, which clarifies the dominant vegetation species composition of each green space patch within the target area, including the species' Chinese name, Latin name, taxonomic attributes, and biological characteristics. Taxonomic attributes cover evergreen or deciduous, broad-leaved or coniferous, tree, shrub, or herbaceous; biological characteristics cover growth cycle, growth rate, and drought or moisture-loving characteristics. In addition, cultivation and management records of species can be collected, such as pruning time, fertilization frequency, irrigation methods, and pest and disease control history. This information can be used to distinguish the impact of natural phenological changes and human management interventions on vegetation morphology. Data acquisition can be combined with urban green space survey databases and landscape plant atlases, and verified one by one through field surveys. Especially for areas with unclear species distribution boundaries, quadrat surveys are used to confirm species and define their distribution range. Finally, a vegetation species database containing species information, distribution coordinates, and management records is constructed, providing a basis for setting species-specific parameters in subsequent model construction.

[0030] Based on the above three types of data, a vegetation phenological rhythm prediction model is constructed. The model can adopt a hybrid architecture of "machine learning algorithm + phenological mechanism", which can not only capture the nonlinear complex correlation between phenology and environmental factors, but also ensure the biological rationality of the prediction results through phenological mechanism constraints. The model construction process first performs input feature processing, performs time series smoothing on meteorological data, and uses Savitzky-Golay filtering to eliminate the interference of short-term random fluctuations on phenological signals; performs time alignment between phenological observation data and meteorological data, matching the occurrence time of phenological events with the corresponding previous meteorological factor data; and uses feature importance analysis to screen key meteorological factors that have a significant impact on phenological events, eliminate redundant features, and reduce model complexity.

[0031] During the model training phase, a stratified cross-validation strategy is employed, stratifying the data according to vegetation species type and year. This ensures that the training and test sets contain an equal proportion of data from different species and years, preventing a decline in model generalization ability due to uneven data distribution. The basic algorithm for the model can be the random forest algorithm, which has advantages in handling high-dimensional data and suppressing overfitting. Model hyperparameters are optimized using grid search or Bayesian optimization to improve prediction accuracy. Simultaneously, phenological mechanism constraints are introduced. For example, constraints on the order and time range of occurrence of phenological events are set based on the biological characteristics of the species. For instance, the order constraint requires leaf unfolding to be later than budding and leaf fall to be later than flowering. The time range constraint requires that the leaf unfolding period of temperate deciduous trees fall during the spring temperature rise phase. Predictions that do not conform to these constraints are corrected to ensure that the predictions align with the natural laws of vegetation growth.

[0032] Model performance is evaluated using a multi-metric approach. Core evaluation metrics include mean absolute error (MAE) and coefficient of determination (COD). The prediction errors for each key phenological event are required to be within a reasonable range, and the COD must reach a high level. If these standards are not met, iterative optimization is performed by increasing feature dimensions, expanding the training dataset, and optimizing the model architecture until the model performance meets the requirements. Increasing feature dimensions could include incorporating soil temperature and moisture data, while optimizing the model architecture could involve replacing random forests with gradient boosting trees.

[0033] After model training, a phenological event timetable is output. The timetable is organized by plant species, with each species corresponding to a record containing a complete phenological event sequence. The core fields of the timetable include species name, phenological event type, predicted occurrence time, uncertainty range, and prediction confidence level. The uncertainty range is quantified using Bootstrap sampling. The training dataset is resampled multiple times, and multiple sub-models are trained. The standard deviation of the prediction results for the same phenological event by all sub-models is calculated. The uncertainty range is defined as the sum or subtraction of the predicted occurrence time and a specific multiple of the standard deviation. This combination corresponds to a certain confidence interval, reflecting the reliability of the prediction result. The prediction confidence level is calculated based on the probability value output by the model, ranging from 0 to 1. A higher confidence level indicates stronger reliability of the phenological event prediction result. The phenological event timetable is stored in a digital format and supports querying and filtering by species name, phenological event type, and predicted time range for subsequent use in conjunction with point cloud data acquisition and scheduling strategies.

[0034] S2. Based on point cloud time series data covering the complete phenological cycle, calculate the information entropy and coefficient of variation of each landscape evaluation index at different phenological stages; based on the information entropy and coefficient of variation, calculate the sensitivity weight of each landscape evaluation index at different phenological stages using the entropy weight method, and obtain the sensitivity weight matrix.

[0035] Specifically, first, point cloud time-series data covering the complete phenological cycle of vegetation is acquired. This data must reflect the dynamic changes in vegetation structure at different phenological stages. Based on this data, the information entropy and coefficient of variation of each landscape evaluation index at different phenological stages are calculated. Information entropy is used to characterize the degree of information dispersion of index data at a specific phenological stage, and its calculation formula is as follows:

[0036]

[0037] In the formula, Represents information entropy. This represents the first time a certain landscape evaluation index is applied within a specific phenological stage. The probability density of a data sample Representing the The natural logarithm of the probability density of a data sample This represents the summation of all data samples within that phenological stage.

[0038] The coefficient of variation is used to characterize the relative dispersion of index data at a specific phenological stage, avoiding the influence of differences in index dimensions on the analysis results. Its calculation formula is as follows:

[0039]

[0040] In the formula, Represents the coefficient of variation. The standard deviation of all data samples representing a certain landscape evaluation index within a specific phenological stage. This represents the mean of all data samples for the same phenological stage.

[0041] Based on the calculated information entropy and coefficient of variation, the entropy weight method is used to calculate the sensitivity weights of each landscape evaluation index at different phenological stages. First, the information entropy and coefficient of variation are standardized to eliminate dimensional differences. Then, both are assigned equal weights, and a comprehensive sensitivity index is calculated. Finally, the comprehensive sensitivity index is normalized to obtain the sensitivity weights of each index at the corresponding phenological stage. The sensitivity weights of all phenological stages and all landscape evaluation indicators together constitute a sensitivity weight matrix, which quantifies the sensitivity differences of different indicators at each phenological stage.

[0042] S3. Input the phenological event timetable and sensitivity weight matrix into the joint optimization model, and generate a non-uniform interval point cloud data acquisition time sequence scheduling strategy by solving a multi-objective optimization problem that simultaneously considers maximizing information value and minimizing acquisition cost.

[0043] Specifically, the phenological event timeline and sensitivity weight matrix are used as input parameters and imported into the joint optimization model. The joint optimization model has two objectives: maximizing information value and minimizing data collection costs. It determines the optimal data collection scheme by solving a multi-objective optimization problem.

[0044] The information value function is used to quantify the contribution of collected data to landscape evaluation, and its calculation formula is as follows:

[0045]

[0046] In the formula, Represents information value, Representing the The first phenological stage Sensitivity weights of each landscape evaluation indicator Representing the Does the time of the second collection fall within the [number]th [period]? A 0-1 variable for each phenological stage; if it falls within that stage, it takes the value 1; otherwise, it takes the value 0. Representing the The second data collection is for the first The information contribution of each landscape evaluation indicator, inner layer The representative sums up all landscape evaluation indicators, outer layer This represents the summation of all phenological stages.

[0047] The acquisition cost function is used to quantify the cost input during the point cloud data acquisition process. Its calculation formula is as follows:

[0048]

[0049] In the formula, Represents the total cost of data collection. Representing the The unit fixed cost of each data collection. Representing the The area of ​​the region collected in this second sampling. Represents the floating cost per unit area. Represents the total number of data collections. This represents the summation of all data collection counts.

[0050] When solving the multi-objective optimization problem, a series of constraints are considered, including constraints on the number of data collections, the time interval between adjacent data collections, and the coverage constraints of phenological stages, to ensure that the optimization results are practically feasible. A Pareto optimal solution set is obtained through a non-dominated sorting genetic algorithm. Based on actual needs, a suitable scheme is selected from the solution set to generate a non-uniformly spaced point cloud data collection timing scheduling strategy. This strategy clarifies the time nodes and related requirements for each data collection session.

[0051] S4. Perform three-dimensional point cloud data acquisition according to the time points determined by the point cloud data acquisition timing scheduling strategy to obtain a point cloud dataset, and extract vegetation structure features from the point cloud dataset.

[0052] Specifically, 3D point cloud data acquisition is conducted at the time points determined by the point cloud data acquisition timing strategy. During the acquisition process, appropriate acquisition equipment is selected based on regional characteristics and accuracy requirements to ensure that the acquired point cloud data accurately reflects the 3D structural information of vegetation. After acquisition, the data is preprocessed, including coordinate system stabilization, noise reduction, separation of ground and non-ground points, and vegetation point extraction, to obtain a clean vegetation point cloud dataset.

[0053] Based on the preprocessed point cloud dataset, vegetation structure features were extracted. The extraction process was carried out across multiple dimensions, including 3D geometric features, texture features, and topological features. 3D geometric features reflect morphological parameters such as the height, volume, and surface area of ​​the vegetation canopy; texture features capture surface attribute information such as the reflectance intensity and color of the vegetation point cloud; and topological features characterize structural parameters such as the number, length, and angle of vegetation branches. This systematic feature extraction provides fundamental data support for subsequent calculations of landscape evaluation indicators.

[0054] S5. Dynamically select high-sensitivity landscape evaluation indicators for the current phenological stage based on the sensitivity weight matrix; calculate the high-sensitivity landscape evaluation indicators based on vegetation structure characteristics to obtain the landscape evaluation results.

[0055] Specifically, the first step is to determine the current phenological stage of the vegetation. Combining real-time monitoring data with a phenological event timeline, a multi-dimensional feature comparison and comprehensive analysis are used to determine the specific category of the current phenological stage. Based on the sensitivity weight matrix, landscape evaluation indicators with high sensitivity within the current phenological stage are selected. The selection process must comprehensively consider evaluation accuracy and computational efficiency to ensure that the selected indicators can reflect the green space landscape characteristics of the current stage to the greatest extent possible.

[0056] Based on the extracted vegetation structure characteristics, high-sensitivity landscape evaluation indicators were calculated. The calculation process followed unified methods and standards to ensure the accuracy and comparability of the results. After calculation, the results of each indicator were standardized to eliminate the influence of dimensional differences on the comprehensive evaluation. A weighted summation method was used to calculate the comprehensive landscape evaluation score, and the formula is as follows:

[0057]

[0058] In the formula, Represents the overall landscape evaluation score. Representing the The sensitivity weight of each highly sensitive landscape evaluation index in the current phenological stage. Representing the The standardized values ​​of a highly sensitive landscape evaluation index, This represents the summation of all highly sensitive landscape evaluation indicators.

[0059] The comprehensive evaluation scores are graded and mapped to different landscape quality levels, making the evaluation results more readable and practical. To ensure the reliability of the evaluation results, verification work needs to be carried out, including comparison with manual on-site evaluation results, longitudinal stability verification, and error analysis verification. Finally, a complete landscape evaluation report is generated, which should include basic information about the evaluated object, the evaluation process, the evaluation results, result analysis, and optimization suggestions, providing technical support for urban green space management and optimization.

[0060] The aforementioned urban green space landscape evaluation method based on three-dimensional point cloud data generates a phenological timetable containing key phenological events and uncertain ranges by constructing a vegetation phenological rhythm prediction model. Combined with an index sensitivity weight matrix calculated based on point cloud time series data, a non-uniform acquisition strategy is generated using a joint optimization model. On this basis, point cloud data is collected and features are extracted. Finally, highly sensitive indicators are dynamically selected for calculation according to phenological stages. This method achieves accurate collection of the most representative vegetation morphology data during key phenological periods, effectively solving the contradiction between data redundancy and insufficient feature capture caused by traditional fixed-interval collection methods. It significantly improves data collection efficiency and the accuracy and relevance of landscape evaluation.

[0061] In one optional embodiment, based on point cloud time series data covering the complete phenological cycle, the information entropy and coefficient of variation of each landscape evaluation index at different phenological stages are calculated, including the following steps:

[0062] S11. The point clouds from multiple periods are aligned to a unified coordinate system by using the iterative nearest point algorithm. The point cloud time series data is then subjected to time series registration processing to obtain a spatiotemporally consistent point cloud dataset.

[0063] Specifically, point cloud time series data consists of point clouds collected at different times in multiple periods. During the collection process, the spatial locations of the point clouds in each period may be mismatched due to factors such as equipment positioning deviations and differences in coordinate system settings. If used directly for index calculations, it will lead to biased results. Therefore, time series registration processing is required to achieve spatiotemporal consistency of point clouds from multiple periods. First, a unified coordinate system is determined. This coordinate system should be a geodetic coordinate system suitable for urban green space landscape evaluation to ensure that subsequent data accurately correspond to the actual geographic spatial location.

[0064] The Iterative Closest Point (ICP) algorithm is the core method for point cloud registration. Its core idea is to find the optimal spatial transformation matrix through iterative optimization, minimizing the distance error between the point cloud to be registered and the reference point cloud. First, a high-quality point cloud from the point cloud time series is selected as the reference point cloud. Selection criteria typically include uniform point cloud density, low noise, and the ability to fully reflect key vegetation structural features. Generally, point clouds corresponding to stable phenological stages of vegetation growth are chosen as the reference. Then, point clouds from other periods are used as the point clouds to be registered, and registration is performed periodically with the reference point cloud.

[0065] The registration process consists of multiple iterative steps: The first step is corresponding point pair search. For each point in the point cloud to be registered, the nearest point in the reference point cloud is found to form an initial corresponding point pair. The second step is transformation matrix calculation. Based on the initial corresponding point pairs, the spatial transformation matrix is ​​solved using the least squares method. This matrix contains translation and rotation parameters, which describe the spatial position adjustment relationship between the point cloud to be registered and the reference point cloud. The third step is error calculation. After spatially transforming the point cloud to be registered using the solved transformation matrix, the average distance error between the corresponding point pairs of the transformed point cloud and the reference point cloud is calculated. The fourth step is iterative judgment. If the average distance error is greater than the preset convergence threshold, or the number of iterations has not reached the maximum number of iterations, the process returns to the first step to search for corresponding point pairs again and repeats the subsequent steps until the error meets the convergence requirement or the maximum number of iterations is reached, thus completing the registration of a single point cloud with the reference point cloud.

[0066] After registering all point clouds to be registered with the reference point cloud, the registration results need to be verified. By calculating the root mean square error (RMSE) between the registered point cloud and the reference point cloud for each period, it is determined whether the registration accuracy meets the requirements, ensuring that the spatial position deviation of the point clouds in the unified coordinate system is within a reasonable range. After the above temporal registration process, the point clouds in the multiple periods not only form a continuous sequence in the time dimension, but also achieve accurate alignment in the spatial dimension, ultimately obtaining a spatiotemporally consistent point cloud dataset, providing a reliable spatial benchmark for the subsequent extraction and calculation of landscape evaluation indicators.

[0067] S12. Extract landscape evaluation index values ​​from the spatiotemporally consistent point cloud dataset. Calculate the numerical distribution of each landscape evaluation index value within the same phenological stage to obtain the index value sequence. The landscape evaluation index values ​​include green volume, canopy structure complexity, permeability, and species diversity.

[0068] Specifically, spatiotemporally consistent point cloud datasets possess the characteristics of unified spatial location and continuous temporal series. Based on their three-dimensional structural features and spatial distribution information, four types of landscape evaluation indicators—green volume, canopy structure complexity, permeability, and species diversity—can be systematically extracted. The extraction process designs corresponding extraction methods for the characteristics of each type of indicator to ensure that the indicator values ​​accurately reflect the core attributes of green space landscapes.

[0069] The extraction of green volume is based on the three-dimensional volumetric features of point clouds, achieved by quantifying the spatial volume occupied by vegetation point clouds. First, vegetation point clouds are separated from spatiotemporally consistent point cloud datasets to eliminate interference from non-vegetated features (such as buildings and roads). Then, a voxelization method is used to divide the three-dimensional space where the vegetation point clouds are located into a regular voxel grid, and the number of voxels containing vegetation points is counted. Finally, the total volume of vegetation is calculated by combining the volume of individual voxels. This volume is the green volume index value, and its magnitude directly reflects the biomass and ecological function potential of vegetation.

[0070] The extraction of canopy structure complexity is based on the spatial dispersion of the vegetation canopy point cloud, achieved by analyzing the uniformity of the point cloud distribution and the differences in structural layers within the canopy. First, the vertical range of the canopy is determined, i.e., the elevation interval from the bottom to the top of the canopy. Then, this elevation interval is divided into several equidistant vertical layers, and the number and distribution density of vegetation points within each layer are counted. Finally, by calculating parameters such as the degree of variation in density of each layer and the distribution characteristics of porosity within the canopy, a comprehensive quantitative index value of canopy structure complexity is obtained. A higher index value indicates a more complex canopy structure, which is more conducive to improving ecosystem stability and biodiversity.

[0071] Transparency is extracted based on the gap ratio of point clouds in the horizontal or vertical directions, reflecting the visual openness of green space landscapes. For horizontal transparency, virtual rays are emitted from a horizontal plane simulating human eye level in multiple evenly distributed horizontal directions, and the proportion of the ray length not obscured by vegetation points to the total ray length is counted. For vertical transparency, virtual rays are emitted vertically upward from the ground, and the proportion of the ray length not obscured by vegetation points to the total ray length is counted. The combined results of horizontal and vertical transparency calculations yield the final transparency index value, which directly affects the visual aesthetic effect and spatial experience of green space landscapes.

[0072] When extracting species diversity, the reflectance characteristics and morphological features of vegetation point clouds are considered. Differences in leaf material and canopy morphology among different vegetation species lead to variations in the reflectance intensity and spatial distribution patterns of their point clouds. First, features such as the mean reflectance intensity, variance of reflectance intensity, and canopy morphology parameters (e.g., canopy height and radius) are extracted from the point clouds of each vegetation patch. Then, a clustering algorithm is used to group vegetation patches with similar characteristics into the same category, with each category corresponding to a potential vegetation species. Finally, the number of species categories is counted, and combined with the coverage area ratio of each category, a species diversity index is calculated. This index reflects the richness and stability of the green space ecosystem.

[0073] After extracting the values ​​of each landscape evaluation index, the index values ​​corresponding to point clouds collected at different times within the same phenological stage are summarized in conjunction with the phenological event timeline. For each landscape evaluation index, all its values ​​within the same phenological stage are organized in chronological order to form an index value sequence for that phenological stage. The index value sequence must ensure that all values ​​come from a spatiotemporally consistent point cloud dataset and strictly correspond to the same phenological stage, providing a continuous and ordered numerical foundation for subsequent calculations of information entropy and coefficient of variation.

[0074] S13. Based on the index value sequence, the information entropy value of each landscape evaluation index value at different phenological stages is calculated using the information entropy formula. The information entropy formula is as follows:

[0075]

[0076] in, Represents information entropy. Indicates the value of the i-th indicator The probability of N is the total number of values ​​that the index can take.

[0077] Specifically, The information entropy is used to quantify the dispersion of landscape evaluation indicators within a specific phenological stage. The larger the information entropy value, the more dispersed the distribution of the indicator values ​​within that stage, reflecting the greater sensitivity of the indicators to phenological changes. This indicates the first time that the landscape evaluation index falls within a specific phenological stage. Each specific value comes from the index value sequence corresponding to the index and is the calculation result of the index at a certain collection time point in this stage. Indicates the value The probability of occurrence is calculated by dividing the frequency of the value in the index value sequence by the total number of all values ​​in the index value sequence. The probability value ranges from 0 to 1, and the sum of the probabilities of all different values ​​is 1. This represents the total number of all different values ​​of the landscape evaluation index within a specific phenological stage, that is, the number of non-repeating values ​​in the index value sequence; This represents the summation of the probability of all distinct values ​​from the 1st to the Nth, multiplied by the base-2 logarithm of that probability. The negative sign in the formula is used to convert the summation result to a non-negative value, ensuring that the information entropy conforms to the non-negativity property defined in mathematics, and avoiding the situation where the result is negative due to logarithmic operations.

[0078] During the calculation process, the first step is to statistically analyze all possible values ​​in the index value sequence and record each distinct value. And its corresponding frequency of occurrence; then calculate the probability of each value based on the frequency. Ensure that the sum of all probabilities is 1; then each and Multiply the values ​​to obtain the product term for each value; finally, sum all the product terms and add a negative sign before the summation to obtain the information entropy value of the landscape evaluation index at a specific phenological stage. This calculation clearly quantifies the information richness of the index at different phenological stages, providing a basis for subsequent sensitivity weight calculations.

[0079] S14. Based on the index value sequence, calculate the coefficient of variation of each landscape evaluation index value at different phenological stages using the coefficient of variation formula; the coefficient of variation formula is:

[0080]

[0081] in, This represents the coefficient of variation at a certain phenological stage. This represents the standard deviation of the landscape evaluation index value during a certain phenological stage. This represents the average value of the landscape evaluation index during a certain phenological stage.

[0082] Specifically, The coefficient of variation is used to measure the relative dispersion of index values ​​within a specific phenological stage after eliminating the influence of the dimensions of the landscape evaluation index. The larger the coefficient of variation, the more significant the relative fluctuation of the index value within that stage, and the higher the sensitivity to phenological changes. The standard deviation of the landscape evaluation index value in a specific phenological stage is a statistical measure that reflects the degree of deviation of the index value from the mean. When calculating it, the mean of all values ​​in the index value sequence must first be calculated, then the difference between each value and the mean is calculated, the difference is squared and the average is calculated, and finally the square root of the average is taken to obtain the standard deviation. The unit of the standard deviation is consistent with the original unit of the index value, reflecting the absolute degree of dispersion. The mean represents the average value of the landscape evaluation index at a specific phenological stage. It is calculated by dividing the sum of all values ​​in the index value sequence by the total number of values. The mean is a statistical measure that reflects the central tendency of the index values, and its unit is consistent with the original unit of the index value.

[0083] The calculation process must follow a fixed logic: first, calculate the mean of the index value sequence. First, sum all the values ​​in the sequence and divide by the total number of values ​​to obtain the mean, which reflects the central tendency of the indicator; then calculate the standard deviation. First, calculate the difference between each value and the mean, then square each difference. Sum all the squared results and divide by the total number of values ​​to get the variance. Next, take the square root of the variance to get the standard deviation, which reflects the absolute dispersion. Finally, calculate the standard deviation... Divide by the mean The coefficient of variation was obtained. Since the coefficient of variation is a dimensionless ratio, it can be used to compare the degree of dispersion between landscape evaluation indicators with different dimensions. For example, the coefficients of variation of green volume (unit of volume) and permeability (unit of proportion) can be directly compared, providing a quantitative basis for cross-indicator comparable dispersion for the subsequent construction of the sensitivity weight matrix.

[0084] In an optional embodiment, based on information entropy and coefficient of variation, the sensitivity weights of each landscape evaluation index at different phenological stages are calculated using the entropy weight method to obtain a sensitivity weight matrix, including the following steps:

[0085] S21. For each phenological stage, normalize the information entropy of each landscape evaluation index to obtain the standardized information entropy. The normalization formula is:

[0086]

[0087] in, The standardized information entropy of index j is represented. The information entropy of index j is represented by... and These represent the minimum and maximum information entropy values ​​of all landscape evaluation indicators, respectively.

[0088] Specifically, for each phenological stage, the information entropy of all landscape evaluation indicators within that stage is first normalized. The aim is to eliminate the impact of differences in the numerical ranges of information entropy values ​​among different indicators, ensuring that the standardized information entropy is horizontally comparable and laying the foundation for subsequent weight calculations. The normalization process must be based on information entropy data within the same phenological stage to avoid data interference across phenological stages, ensuring that the normalization results accurately reflect the relative information dispersion of each indicator within that stage.

[0089] Normalization uses a linear transformation method, and its calculation formula is as follows: In the formula Indicates the first phenological stage within the current phenological phase. The standardized information entropy of each landscape evaluation indicator ranges from 0 to 1. The larger the value, the higher the information dispersion of the indicator at the current stage. Indicates the first phenological stage within the current phenological phase. The original information entropy of each landscape evaluation indicator, which was calculated in the previous stage based on the indicator value sequence; This represents the minimum value among the original information entropy of all landscape evaluation indicators within the current phenological stage, used to determine the lower limit benchmark for normalization. This represents the maximum value among the original information entropy of all landscape evaluation indicators within the current phenological stage, used to determine the upper limit benchmark for normalization. This formula maps the original information entropy of different indicators to the same numerical range, eliminating the impact of differences in dimensions and numerical ranges on subsequent weight calculations.

[0090] S22. Based on standardized information entropy, calculate the weight coefficients of each landscape evaluation index; the formula for calculating the weight coefficients is:

[0091]

[0092] in, This represents the weight of indicator j, and K is the total number of landscape evaluation indicators.

[0093] Specifically, after obtaining the standardized information entropy of each landscape evaluation indicator, it is further converted into weight coefficients to quantify the relative importance of each indicator in the landscape evaluation at the current phenological stage. The core logic for calculating the weight coefficients is: the higher the standardized information entropy of an indicator, the richer the effective evaluation information it contains, the greater its contribution to the landscape evaluation results, and the higher the corresponding weight coefficient should be.

[0094] The formula for calculating the weighting coefficient is as follows: In the formula Indicates the first phenological stage within the current phenological phase. The initial weight coefficients of each landscape evaluation indicator range from 0 to 1, and the sum of the initial weight coefficients of all indicators is 1. Indicates the first phenological stage within the current phenological phase. The effective information contribution of each landscape evaluation indicator, due to The larger the value, the higher the dispersion of the indicator information. The relative content of effective information in the inverse characterization index; This represents the sum of the effective information contributions of all landscape evaluation indicators within the current phenological stage, where... This represents the total number of landscape evaluation indicators participating in the evaluation during the current phenological stage. This summation operation is used to normalize the effective information contribution of each indicator, ensuring that the final weight coefficients meet the constraint that the sum is 1, which conforms to the mathematical definition of weight.

[0095] During the calculation process, each indicator is calculated first. Then all indicators Summation, and finally using the sum of each indicator. Divide by this sum to obtain the corresponding initial weight coefficients, thus completing the conversion from standardized information entropy to weight coefficients.

[0096] S23. By multiplying by the coefficient of variation adjustment factor, the weights of landscape evaluation indicators with coefficients of variation higher than a preset threshold are adjusted to obtain the adjusted weight coefficients; the coefficient of variation adjustment factor is... ,in This represents the coefficient of variation of index j.

[0097] Specifically, the initial weighting coefficients were calculated solely based on information entropy, without fully considering the fluctuations of indicators within phenological stages. The coefficient of variation, as a core parameter characterizing the fluctuation of indicators, indicates that a higher value signifies more significant numerical changes in the indicator within the current phenological stage, greater sensitivity to phenological stage characteristics, and thus warrants higher weighting in landscape evaluation. Therefore, a coefficient of variation adjustment factor is introduced to adjust the weights of indicators with coefficients of variation exceeding a preset threshold, thereby optimizing the rationality of weight allocation.

[0098] First, determine the preset threshold for the coefficient of variation. The threshold is set based on historical data statistics or industry-standard evaluation criteria. By analyzing the distribution characteristics of the coefficient of variation of similar green spaces and similar indicators in multiple phenological cycles, the mean or median can be taken as the preset threshold to ensure that the threshold can effectively distinguish between indicators with significant and insignificant fluctuations.

[0099] The formula for calculating the coefficient of variation adjustment factor is as follows: In the formula Indicates the first phenological stage within the current phenological phase. The coefficient of variation adjustment factor for each landscape evaluation indicator is greater than 1 (due to the coefficient of variation). (non-negative values) The larger, The larger the value, the greater the adjustment range for the weight; Indicates the first phenological stage within the current phenological phase. The coefficient of variation of each landscape evaluation indicator is calculated based on the indicator value sequence in the previous stage and reflects the degree of numerical fluctuation of the indicator in the current stage.

[0100] The specific procedure for weight adjustment is as follows: First, adjust the weight of each indicator... Compared with a preset threshold, if If the value is higher than the preset threshold, it indicates that the indicator fluctuates significantly, and the initial weighting coefficients obtained in S22 are used to determine its value. Multiply by the corresponding adjustment factor The adjusted weighting coefficients are obtained; if If the value is below or equal to the preset threshold, it indicates that the indicator fluctuates moderately and requires no adjustment. The adjusted weighting coefficient will retain its initial weighting coefficient. The weighting remains unchanged. This adjustment process gives higher weight to indicators that fluctuate significantly and are sensitive to phenological stages, further enhancing the scientific and targeted nature of the weighting allocation.

[0101] S24. Organize the adjusted weight coefficients of all landscape evaluation indicators for each phenological stage into a matrix form and output the sensitivity weight matrix.

[0102] Specifically, after calculating the adjusted weight coefficients of each landscape evaluation index for all phenological stages, these weight coefficients are organized into a matrix according to a unified rule to form a sensitivity weight matrix, which can then be directly called in subsequent point cloud data collection scheduling and landscape evaluation index selection.

[0103] The core rule of matrix organization is to construct a two-dimensional matrix with phenological stages as the row dimension and landscape evaluation indicators as the column dimension. Each row of the matrix corresponds to a phenological stage, and each element in that row represents the adjusted weight coefficient of the landscape evaluation indicator in the corresponding column within that phenological stage. Each column of the matrix corresponds to a landscape evaluation indicator, and each element in that column represents the adjusted weight coefficient of that indicator within the phenological stage of its corresponding row. For example, if the research involves... Each phenological stage For each landscape evaluation index, the sensitivity weight matrix is: A dimensional matrix, the matrix in which the dimensional matrix is... Line number Column elements That is to say, the first Within the first phenological stage Adjusted weighting coefficients for each landscape evaluation indicator.

[0104] In one optional embodiment, a vegetation phenological rhythm prediction model is constructed based on historical phenological observation data, historical meteorological data, and historical vegetation species information of the target area; the vegetation phenological rhythm prediction model outputs a phenological event schedule containing the uncertainty range of key phenological event types and their predicted occurrence times, including the following steps:

[0105] S31. By removing outliers and filling in missing data, the historical phenological observation data is subjected to quality control processing to obtain a standardized phenological observation dataset.

[0106] Specifically, historical phenological observation data may contain outliers due to observation errors, equipment malfunctions, or human error during long-term recording. Data may also be missing for certain time periods. These issues can affect the accuracy of subsequent model construction; therefore, quality control processing is implemented first. Outlier removal is based on statistical methods, combined with the natural laws of phenological events to determine the judgment criteria: First, the statistical characteristics (mean, standard deviation) of observation data for a single phenological event (such as the leaf unfolding date of a species) are calculated. The Laida criterion is used to identify observations exceeding the mean ± 3 times the standard deviation, which are initially judged as outliers. Then, a second verification is performed based on the temporal logic of the phenological event. For example, if the flowering period of a deciduous tree is earlier than the budding period, or the leaf fall period is later than the leaf unfolding period of the following year, even if the value does not exceed the statistical threshold, it is still judged as an outlier and removed, ensuring that the removal results conform to the natural rhythm of vegetation growth.

[0107] For imputation of missing data, the appropriate method is selected based on the duration of the missing data and the data distribution characteristics: If the missing data is an isolated missing data point at a single time point, and the adjacent observation data is continuous and complete, the linear interpolation method is used. A linear function is constructed using two adjacent valid observation values ​​before and after the missing point to calculate the estimated value of the missing point; If the missing data is missing at multiple consecutive time points, or the adjacent data fluctuates greatly, a regression imputation method based on meteorological data is used. The key meteorological factors affecting the phenological event at the same time (such as daily average temperature and sunshine duration) are used as independent variables, and the time of occurrence of the phenological event is used as the dependent variable to construct a linear regression model. The model is then used to predict the missing phenological data; For phenological data missing with strong species specificity, the imputation value can also be obtained by combining the phenological data of closely related species at the same time and adjusting the similarity coefficient of physiological characteristics between species.

[0108] After outlier removal and missing data imputation, the data underwent standardization: The definition of phenological events was standardized; for example, "leaf unfolding start" was defined as the date when 5% of the leaves on the plant were fully unfolded, and "peak flowering" was defined as the date when 50% of the flowers on the plant were fully open, ensuring consistency in data recorded by different years and observers. The time recording format was also standardized, converting all phenological event times to the standard "year-month-day" format to avoid data misuse due to format differences. The resulting standardized phenological observation dataset includes fields such as species identifier, phenological event type, observation time, observation location, and data quality level (e.g., "raw valid," "interpolated imputation," "regression prediction"), providing a high-quality data foundation for subsequent model training.

[0109] S32. Based on the daily average temperature data in the meteorological data, the cumulative heat value is calculated using the growth degree day model; by fitting the relationship between the occurrence time of historical phenological events and the cumulative heat value, the growth degree day threshold of each phenological event is determined.

[0110] Specifically, the growth-day model is a core tool for quantifying the driving effect of heat accumulation on phenological events. Its core logic is that the occurrence of vegetation phenological events requires a certain amount of heat accumulation; when the accumulated heat reaches a specific threshold, the phenological event is initiated. First, daily average temperature data is extracted from historical meteorological data (…). ), and determine the temperature baseline values ​​for each vegetation species ( This baseline value represents the minimum temperature at which vegetation begins to grow and develop, and needs to be determined based on the physiological characteristics of the species, such as temperate deciduous trees. Typically set at a fixed temperature, tropical plants The answer is relatively high.

[0111] The formula for calculating growing days (GDD, i.e., cumulative heat value) is:

[0112]

[0113] In the formula, GDD represents the start date. By target date The cumulative heat value; The starting date for annual heat accumulation is usually set after the end of the winter cold season (e.g., when the average daily temperature remains consistently above a certain level for five consecutive days). (the first day) For any target date; Let t be the average daily temperature on day t. This serves as the temperature baseline for vegetation growth; the max function ensures that only values ​​above this threshold are accumulated. Temperature contribution, when If the heat contribution for that day is 0, it will not be included in the cumulative calculation.

[0114] After calculating the GDD sequence over many years, the correlation between the occurrence time of historical phenological events and GDD was established. For each phenological event (such as the leaf unfolding date of a species), the GDD value corresponding to the occurrence date of the event in previous years was extracted (i.e., from...). The cumulative GDD value up to the date of the event is used as the sample data. A nonlinear fitting method (such as Logistic regression or multinomial regression) is employed to construct a fitting curve, with the goal of minimizing the sum of squared residuals between the sample GDD values ​​and the predicted values ​​of the fitted curve. The fitting effect is evaluated using the coefficient of determination (R²), for example, requiring R² ≥ 0.8, to ensure that the fitted curve accurately reflects the correlation between the two.

[0115] The growth day threshold for each phenological event is determined by fitting curves. For example, the GDD value corresponding to a 50% probability of occurrence of a phenological event in the fitted curve is defined as the growth day threshold for that phenological event. For instance, if the fitted curve for the leaf-opening stage of a certain tree shows that the probability of the leaf-opening stage occurring is 50% when the accumulated GDD reaches a certain value, that value is the growth day threshold for the leaf-opening stage of that tree. The growth day thresholds for different phenological events (budding stage, leaf-opening stage, flowering stage, and leaf-falling stage) need to be calculated separately, and the thresholds for the same phenological event across different species also need to be determined individually based on the fitting results to ensure the species-specific and phenological event-specific nature of the thresholds.

[0116] S33. Combining the physiological characteristic parameters of vegetation species information, a plant biological clock model is used to simulate the influence of photoperiod on phenological events; based on the influence pattern, the phenological correction amount of photoperiod regulation is output by solving a set of differential equations describing the internal physiological rhythms of plants.

[0117] Specifically, photoperiod is another key environmental factor influencing vegetation phenological events. Its core lies in the regulation of internal physiological rhythms by the periodic changes in day length, thus affecting the timing of phenological events. First, physiological characteristic parameters related to photoperiod response are extracted from vegetation species information. These include the species' critical day length (i.e., the minimum or maximum day length that induces phenological events; for example, short-day plants require day length ≤ critical day length to flower, while long-day plants require day length ≥ critical day length), light sensitivity coefficient (characterizing the intensity of a plant's response to changes in day length; the larger the coefficient, the more significant the impact of day length changes on phenological events), and circadian rhythm (the inherent cycle of a plant's internal physiological rhythms, typically close to 24 hours). These parameters are obtained through species physiological experimental data or authoritative plant databases to ensure their accuracy.

[0118] The plant circadian rhythm model employs an oscillator-based simulation framework. Its core is to describe, through a system of differential equations, the concentration changes of internal plant circadian rhythm components (such as circadian rhythm gene expression products) and their interaction with external photoperiodic signals. The system of differential equations describing the internal physiological rhythms of plants can be expressed as:

[0119]

[0120]

[0121] In the formula, A and B represent the concentrations of two key biological clock components, respectively; t represents time. and These represent the rates of change of concentrations A and B over time, respectively. The activation coefficient of the light signal to the concentration of A is denoted as . Let be the solar radiation intensity at time t (takes a value of 0 or 1, where 0 represents darkness and 1 represents light). This indicates the positive modulation of the concentration of A by the light signal; Let be the natural decay coefficient of concentration A. This indicates the natural degradation of concentration A; Let be the interaction coefficient between A and B. This indicates that the combination of A and B negatively regulates the concentration of A. Let A be the activation coefficient of B concentration. This indicates that an increase in concentration A positively promotes an increase in concentration B. Let be the natural decay coefficient of B concentration. This indicates the natural degradation of B concentration.

[0122] The above differential equations were solved using numerical methods (such as the Runge-Kutta method) to obtain the concentration change curves of biological clock components A and B under different photoperiod conditions (i.e., different day lengths and different day start times). By analyzing the correlation between the curve characteristics (such as the time of peak concentration and the amplitude of concentration change) and photoperiod conditions, the influence of photoperiod on phenological events can be simulated: for example, when the day length exceeds the critical day length for a certain species, the peak concentration of biological clock component A appears earlier, and the corresponding phenological event (such as the flowering period) also occurs earlier; conversely, when the day length is less than the critical value, the peak concentration is delayed, and the phenological event is also delayed.

[0123] Based on the aforementioned influence patterns, the phenological correction amount for photoperiod regulation is calculated. Using the predicted time of phenological events under a constant environment with no change in light intensity (e.g., daylight duration always equal to the critical day length) as the baseline time, the difference between the predicted time of phenological events under actual photoperiod conditions and the actual time is the photoperiod correction amount. If the actual photoperiod causes the phenological event to occur earlier, the correction amount is negative; if it delays the phenological event, the correction amount is positive. The magnitude of the correction amount is positively correlated with the degree to which the photoperiod deviates from the critical day length and the light sensitivity coefficient of the species. The greater the deviation and the higher the light sensitivity coefficient, the larger the absolute value of the correction amount, ensuring that the correction amount accurately quantifies the regulatory effect of photoperiod on phenological events.

[0124] S34. Input the growth degree-day threshold and phenological correction into the integrated prediction model, combine it with the phenological observation dataset, and generate a phenological event timetable by weighted combination of the results of multiple prediction methods, including the predicted occurrence time and confidence interval of key phenological events such as germination period, leaf expansion period, flowering period, and leaf fall period.

[0125] Specifically, the core of the ensemble prediction model is to combine the advantages of multiple prediction methods, reduce the prediction error of a single model through weighted combination, and improve the accuracy of phenological event prediction. The ensemble model includes the following prediction methods: First, the prediction result of the growth-day model (based on the growth-day threshold determined in S32; when the real-time accumulated GDD reaches the threshold, it is predicted as the time of the phenological event) serves as the first prediction method; second, the prediction result of the plant circadian rhythm model (based on the phenological correction amount calculated in S33, the predicted time of the growth-day model is corrected to obtain the corrected time of the phenological event) serves as the third prediction method; and third, a machine learning prediction method (such as the random forest model) is introduced. This method uses the phenological observation dataset as training samples, key meteorological factors (daily average temperature, sunshine duration, precipitation) and species physiological parameters as input features, and the time of the phenological event as the output label.

[0126] The key to weighted combination is determining the weight coefficient for each prediction method, which is based on the historical prediction accuracy of each method. Standardized phenological observation data from the past 3-5 years can be selected as the validation set. The prediction errors (e.g., mean absolute error, MAE) of the three prediction methods on the validation set are calculated separately. The reciprocal of the prediction error is used as the basis for weight allocation; the smaller the error, the larger the weight coefficient. The calculation formula is as follows:

[0127]

[0128] In the formula, The weight coefficients for the m-th prediction method are represented (m=1,2,3, corresponding to the growth day model, biological clock model, and machine learning model, respectively). This represents the mean absolute error of the m-th prediction method on the validation set; the denominator is the sum of 1 / MAE of the three prediction methods, ensuring that the sum of all weight coefficients is 1, satisfying the weight normalization requirement.

[0129] The prediction times for the three prediction methods (denoted as ) , , ) and the corresponding weighting coefficients ( , , By weighting and combining the results, the final predicted time of occurrence of phenological events can be obtained. The calculation formula is:

[0130]

[0131] To quantify the uncertainty of the prediction results, a confidence interval for the predicted occurrence time is calculated. A bootstrap sampling method is used, for example, resampling the phenological observation dataset 1000 times. After each resampling, the ensemble prediction model is retrained to obtain a predicted time. The 1000 predicted times are sorted in ascending order, and the 2.5th and 97.5th percentile values ​​are taken as the lower and upper limits of the confidence interval, respectively. This interval corresponds to a 95% confidence level, meaning the probability that the actual occurrence time of the phenological event falls within this interval is 95%. This interval represents the range of uncertainty regarding the predicted occurrence time.

[0132] The final generated phenological event timetable is organized by species, with each species corresponding to a record containing a complete sequence of key phenological events. Record fields include species name, Latin name of the species, phenological event type (germination period, leaf unfolding period, flowering period, leaf fall period, each type further subdivided into beginning, peak, and end), predicted occurrence time (formatted as "year-month-day"), uncertainty range (formatted as "lower limit date-upper limit date"), confidence level (labeled as 95%), weighting coefficients of each prediction method, and prediction error. The timetable is stored in digital formats (such as Excel and CSV), supporting queries and filtering by species name, phenological event type, and prediction time range. It also includes explanations of model parameters (such as growth degree-day thresholds and the basis for calculating phenological corrections) to ensure that subsequent data collection and scheduling processes clearly understand the source and reliability of the prediction results.

[0133] In one alternative embodiment, extracting vegetation structure features from a point cloud dataset includes the following steps:

[0134] S41. Based on the point cloud dataset, the leaf area density is calculated using the voxelization analysis method. The contact frequency of the point cloud within each voxel is statistically analyzed. The leaf area density and contact frequency are used as characteristic parameters of the canopy structure.

[0135] Specifically, the preprocessed vegetation point cloud dataset is first voxelized. The core of voxelization is to divide the three-dimensional space containing the vegetation into regular cubic voxel units. The division principle must be adapted to the structural scale of the vegetation canopy—ensuring that the voxel size can capture the details inside the canopy (such as the difference between leaf cluster areas and pore areas) while avoiding data redundancy due to voxels that are too small or information loss due to voxels that are too large. Spatial continuity is maintained during voxel division, with no overlap between adjacent voxels and complete coverage of the entire three-dimensional range of the vegetation canopy. After division, the point cloud data is assigned to the corresponding voxel units according to spatial location, forming a "voxel-point cloud" association mapping relationship. Each voxel unit contains only the vegetation point cloud within its spatial range.

[0136] Leaf area density (LAD) is calculated based on voxelization results. LAD is a key parameter characterizing the leaf area per unit volume of vegetation, reflecting the canopy density and photosynthetic capacity. The calculation process is based on the spatial correspondence between point cloud data and leaves—assuming that the vegetation point cloud within a voxel uniformly represents the leaf distribution within that region, the leaf area is indirectly estimated through the number of point clouds. The formula for calculating leaf area density is:

[0137]

[0138] In the formula, LAD represents the leaf area density of a voxel unit; k is a conversion coefficient used to correlate the number of vegetation point clouds within a voxel with the actual leaf area. Its value needs to be determined in conjunction with the leaf morphology of the vegetation species (e.g., broadleaf, coniferous) and the scanning characteristics of the point cloud acquisition equipment (e.g., point cloud density, laser wavelength). Its core function is to eliminate the differences in point cloud-leaf area mapping caused by different species and different equipment; N is the number of vegetation point clouds contained in the voxel unit; V is the volume of the voxel unit, calculated by multiplying the length, width, and height of the voxel. After the calculation is completed, the LAD values ​​of all voxels are statistically analyzed to form a leaf area density distribution matrix at different heights and horizontal positions of the canopy, fully reflecting the vertical and horizontal structural differences of the canopy.

[0139] Simultaneously count the contact frequency of point clouds within each voxel ( Contact frequency is a parameter characterizing the density of point cloud distribution within a voxel, reflecting the spatial filling characteristics of the canopy—a higher contact frequency indicates a denser leaf distribution and fewer pores in that area; conversely, a lower contact frequency indicates more pores. The contact frequency is calculated based on the spatial distribution characteristics of point clouds within a voxel, defined as the number of vegetation point clouds per unit volume, using the following formula:

[0140]

[0141] In the formula, This represents the point cloud contact frequency of a specific voxel unit; N is the number of vegetation point clouds within that voxel unit; and V is the volume of that voxel unit. Unlike leaf area density, contact frequency directly reflects the spatial density of point clouds, requiring no conversion coefficient and can be obtained directly from the ratio of the number of point clouds to the volume within a voxel. After calculation, a spatial distribution matrix of contact frequency is formed, which, together with leaf area density, constitutes the canopy structure characteristic parameters—leaf area density focuses on the quantification of "leaf area," while contact frequency focuses on the quantification of "point cloud density." Both reflect the structural characteristics of the canopy from different dimensions, providing basic data for subsequent landscape assessments (such as canopy closure and ecological function evaluation).

[0142] S42. Based on the point cloud dataset, calculate the canopy height model by the difference between the digital elevation model and the digital surface model, and extract tree height feature parameters and canopy width feature parameters based on the canopy height model.

[0143] Specifically, a Digital Surface Model (DSM) and a Digital Elevation Model (DEM) are first generated from the point cloud dataset. The DSM is a three-dimensional model reflecting the top elevations of all surface features (including vegetation, buildings, and ground). During the generation process, spatial interpolation is performed on the entire point cloud dataset (including ground points and non-ground points). Kriging interpolation or inverse distance weighted interpolation is used to interpolate the discrete point cloud elevation values ​​into a regular two-dimensional grid. The grid resolution is adapted to the scale of the study area to ensure that the canopy outlines of different vegetation can be clearly distinguished. The DEM is a three-dimensional model that only reflects the ground elevation. Before generation, ground points are separated from the point cloud dataset (progressive morphological filtering can be used to remove non-ground points such as vegetation and buildings through morphological operations, while retaining ground points). Then, the same interpolation method as the DSM is used to interpolate the elevation values ​​of the ground points into a two-dimensional grid consistent with the DSM, ensuring that the spatial grids of the DSM and DEM are completely matched, laying the foundation for subsequent difference calculations.

[0144] The canopy height model (CHM) is calculated by subtracting the ground elevation (DEM) from the elevation of the top of the vegetation cover (DSM) at the location of the ground elevation (DSM). The core logic is that the canopy height at a given location equals the difference between the top elevation of the vegetation cover (DSM value) and the ground elevation (DEM value). The difference is the actual height of the vegetation canopy at that location. The formula for calculating the canopy height model is as follows:

[0145]

[0146] In the formula, Indicates the coordinates in a two-dimensional grid. The canopy height corresponding to the pixel; This represents the elevation value of the digital surface model for that pixel. This represents the digital elevation model (DEM) elevation value of that pixel. If the calculation result... This indicates that there is no vegetation cover at this location (such as bare ground or the base of a building), and its value should be assigned to 0 to ensure that CHM only reflects the height information of the vegetation canopy. After calculation, CHM is presented in the form of a grayscale image or a 3D grid. Different grayscale values ​​or height values ​​correspond to different canopy heights, intuitively showing the vegetation height distribution of the entire study area.

[0147] Tree height feature parameters are extracted based on the Canopy Height Model (CHM). Tree height is a key parameter characterizing vegetation growth and landscape visual hierarchy. The extraction process first segments the CHM into canopy objects using a watershed algorithm or region growing algorithm. Based on the height differences and connectivity of the CHM, continuous high-value areas (vegetation canopy) are segmented into independent "canopy objects," each corresponding to a single tree or a cluster of vegetation. For each segmented canopy object, the extracted tree height feature parameters include two core indicators: first, the maximum CHM value within the canopy object, defined as "maximum tree height," reflecting the highest growth height of the vegetation; and second, the average CHM value within the canopy object, defined as "average tree height," reflecting the overall height level of the vegetation canopy. After extraction, the maximum and average tree heights of all canopy objects are statistically analyzed to form the tree height distribution characteristics of the study area (such as the proportion of tree height grades, average tree height, and maximum tree height), providing data support for "layer richness" and "visual aesthetic assessment" in landscape evaluation.

[0148] Simultaneously, canopy width characteristic parameters are extracted. Canopy width is a parameter characterizing the horizontal expansion range of vegetation canopy, reflecting the vegetation's ability to occupy horizontal space and the landscape's coverage. The extraction process is based on segmented canopy objects. First, the projected outline of each canopy object on the CHM horizontal plane is determined. By setting a height threshold (e.g., the CHM value is greater than half the maximum tree height), the effective horizontal range of the canopy is filtered out, excluding low branches or noise interference at the bottom of the canopy. Then, geometric analysis is performed on the projected outline, calculating the circumscribed rectangle of the outline. The average length and width of the rectangle are defined as the "average canopy width." The minimum circumscribed circle of the outline is calculated, and the diameter of the circle is defined as the "maximum canopy width." Simultaneously, the area of ​​the outline is calculated and defined as the "canopy projected area." These canopy width characteristic parameters reflect the horizontal expansion characteristics of the vegetation canopy from different dimensions. Together with tree height characteristic parameters, they constitute the core morphological characteristics of vegetation, providing key inputs for subsequent landscape evaluation index calculations (such as canopy complexity and permeability).

[0149] The aforementioned urban green space landscape evaluation method based on three-dimensional point cloud data generates a phenological timetable containing key phenological events and uncertain ranges by constructing a vegetation phenological rhythm prediction model. Combined with an index sensitivity weight matrix calculated based on point cloud time series data, a non-uniform acquisition strategy is generated using a joint optimization model. On this basis, point cloud data is collected and features are extracted. Finally, highly sensitive indicators are dynamically selected for calculation according to phenological stages. This method achieves accurate collection of the most representative vegetation morphology data during key phenological periods, effectively solving the contradiction between data redundancy and insufficient feature capture caused by traditional fixed-interval collection methods. It significantly improves data collection efficiency and the accuracy and relevance of landscape evaluation.

[0150] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0151] Based on the same inventive concept, this application also provides a system for implementing the urban green space landscape evaluation method based on three-dimensional point cloud data as described above. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the urban green space landscape evaluation system based on three-dimensional point cloud data provided below can be found in the limitations of the urban green space landscape evaluation method based on three-dimensional point cloud data described above, and will not be repeated here.

[0152] In one exemplary embodiment, such as Figure 2 As shown, an urban green space landscape evaluation system 20 based on three-dimensional point cloud data is provided to implement the methods in the above embodiments. The system includes:

[0153] The phenological dynamic modeling module 21 is used to construct a vegetation phenological rhythm prediction model based on historical phenological observation data, historical meteorological data and historical vegetation type information of the target area; and output a phenological event timetable containing the uncertainty range of key phenological event types and their predicted occurrence times based on the vegetation phenological rhythm prediction model.

[0154] The indicator sensitivity quantification module 22 is used to calculate the information entropy and coefficient of variation of each landscape evaluation indicator at different phenological stages based on point cloud time series data covering the complete phenological cycle; based on the information entropy and coefficient of variation, the sensitivity weight of each landscape evaluation indicator at different phenological stages is calculated by the entropy weight method to obtain the sensitivity weight matrix.

[0155] The acquisition strategy optimization module 23 is used to input the phenological event timetable and sensitivity weight matrix into the joint optimization model. By solving a multi-objective optimization problem that simultaneously considers maximizing information value and minimizing acquisition cost, a non-uniform interval point cloud data acquisition time sequence scheduling strategy is generated.

[0156] The point cloud feature extraction module 24 is used to collect three-dimensional point cloud data according to the time point determined by the point cloud data acquisition timing scheduling strategy, obtain a point cloud dataset, and extract vegetation structure features from the point cloud dataset.

[0157] The dynamic evaluation generation module 25 is used to dynamically select high-sensitivity landscape evaluation indicators in the current phenological stage based on the sensitivity weight matrix; and to calculate the high-sensitivity landscape evaluation indicators based on vegetation structure characteristics to obtain the landscape evaluation results.

[0158] Embodiments of this application also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the aforementioned method embodiments.

[0159] Embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0160] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0161] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for evaluating urban green space landscape based on a three-dimensional data point cloud, characterized in that, The method comprises: S1, constructing a vegetation phenology rhythm prediction model according to historical phenological observation data, historical meteorological data and historical vegetation species information of a target area; outputting a phenological event schedule containing a key phenological event type and an uncertainty range of a predicted occurrence time of the key phenological event type according to the vegetation phenology rhythm prediction model; S2, calculating information entropy and a coefficient of variation of each landscape evaluation index at different phenological stages based on point cloud time series data covering a complete phenological period; calculating a sensitivity weight of each landscape evaluation index at different phenological stages by an entropy weight method based on the information entropy and the coefficient of variation, and obtaining a sensitivity weight matrix; S3, inputting the phenological event schedule and the sensitivity weight matrix into a joint optimization model, and generating a non-uniformly spaced point cloud data acquisition time sequence scheduling strategy by solving a multi-objective optimization problem considering maximum information value and minimum acquisition cost; S4, acquiring three-dimensional point cloud data at time points determined according to the point cloud data acquisition time sequence scheduling strategy, obtaining a point cloud data set, and extracting vegetation structure features from the point cloud data set; S5, dynamically selecting a high-sensitivity landscape evaluation index at a current phenological stage according to the sensitivity weight matrix; and calculating the high-sensitivity landscape evaluation index based on the vegetation structure features to obtain a landscape evaluation result.

2. The method of claim 1, wherein, The calculation of information entropy and a coefficient of variation of each landscape evaluation index at different phenological stages based on point cloud time series data covering a complete phenological period comprises: S11, aligning multi-period point clouds to a unified coordinate system by an iterative closest point algorithm, performing time sequence registration processing on the point cloud time series data, and obtaining a spatiotemporally consistent point cloud data set; S12, extracting landscape evaluation index values from the spatiotemporally consistent point cloud data set, and obtaining an index value sequence by calculating the numerical distribution of each landscape evaluation index value within the same phenological stage; the landscape evaluation index values include green amount, crown structure complexity, permeability and species diversity; S13, calculating information entropy values of each landscape evaluation index value at different phenological stages based on the index value sequence by using an information entropy formula; the information entropy formula is: wherein, denotes the information entropy, denotes the probability of the i-th index taking the value N is the total number of index values. S14, calculating a coefficient of variation value of each landscape evaluation index value at different phenological stages based on the index value sequence by using a coefficient of variation formula; the coefficient of variation formula is: wherein, represents the coefficient of variation of the landscape evaluation index value at a certain phenological stage, represents the standard deviation of the landscape evaluation index value at a certain phenological stage, represents the average value of the landscape evaluation index value at a certain phenological stage.

3. The method of claim 1, wherein, The calculation of a sensitivity weight of each landscape evaluation index at different phenological stages based on the information entropy and the coefficient of variation by an entropy weight method to obtain a sensitivity weight matrix comprises: S21, for each phenological stage, normalizing the information entropy of each landscape evaluation index to obtain a standardized information entropy; the normalization formula is: wherein, denotes the normalized information entropy of index j, denotes the information entropy of index j, and denote the minimum and maximum values of the information entropy of all the landscape evaluation indices, respectively. S22, calculating a weight coefficient of each landscape evaluation index based on the standardized information entropy; the calculation formula of the weight coefficient is: wherein, K represents the weight of index j, K is the total number of the landscape evaluation indexes; S23, adjusting the weight of the landscape evaluation index with the coefficient of variation higher than the preset threshold by multiplying a coefficient of variation adjustment factor, to obtain an adjusted weight coefficient; the coefficient of variation adjustment factor is wherein Cj represents the coefficient of variation of the index j. S24, organizing the adjusted weight coefficients of all landscape evaluation indexes at each phenological stage into a matrix form, and outputting the sensitivity weight matrix.

4. The method of claim 1, wherein, The vegetation phenology rhythm prediction model is constructed according to historical phenological observation data, historical meteorological data and historical vegetation species information of a target area; a phenological event timetable containing a key phenological event type and an uncertainty range of a predicted occurrence time of the key phenological event type is output according to the vegetation phenology rhythm prediction model, and the phenological event timetable comprises the following steps: S31, quality control processing is performed on the historical phenological observation data by removing outliers and filling missing data to obtain a standardized phenological observation data set; S32, based on daily average temperature data in the meteorological data, a heat accumulation value is calculated by using a growing degree day model; by fitting the relationship between the occurrence time of a historical phenological event and the heat accumulation value, a growing degree day threshold value of each phenological event is determined; S33, in combination with physiological characteristic parameters of the vegetation species information, a plant biological clock model is used to simulate the influence law of photoperiod on a phenological event; in combination with the influence law, a phenological correction amount regulated by photoperiod is output by solving a differential equation set describing the internal physiological rhythm of a plant; S34, the growing degree day threshold value and the phenological correction amount are input into an integrated prediction model, the phenological observation data set is combined, the results of multiple prediction methods are combined by weighting, and the phenological event timetable including the predicted occurrence time and the confidence interval of the key phenological events of the germination period, the leaf unfolding period, the flowering period and the leaf fall period is generated.

5. The method according to any one of claims 1 to 4, characterized in that, The vegetation structure features are extracted from the point cloud data set, and the extraction comprises the following steps: S41, based on the point cloud data set, a leaf area density is calculated by using a voxelization analysis method, and a contact frequency of point clouds in each voxel is counted; the leaf area density and the contact frequency are taken as crown structure characteristic parameters; S42, based on the point cloud data set, a crown height model is calculated by using a difference value of a digital elevation model and a digital surface model, and tree height characteristic parameters and crown width characteristic parameters are extracted according to the crown height model.

6. A system for evaluating urban green space landscape based on three-dimensional data point cloud, for implementing the method of any one of claims 1 to 5, characterized in that, The system comprises: a phenological dynamic modeling module, configured to construct a vegetation phenology rhythm prediction model according to historical phenological observation data, historical meteorological data and historical vegetation species information of a target area; and output a phenological event timetable containing a key phenological event type and an uncertainty range of a predicted occurrence time of the key phenological event type according to the vegetation phenology rhythm prediction model; an index sensitivity quantification module, configured to calculate information entropy and a variation coefficient of each landscape evaluation index at different phenological stages based on point cloud time series data covering a complete phenological period; and calculate a sensitivity weight of each landscape evaluation index at different phenological stages by using an entropy weight method based on the information entropy and the variation coefficient, to obtain a sensitivity weight matrix; a collection strategy optimization module, configured to input the phenological event timetable and the sensitivity weight matrix into a joint optimization model, and generate a non-uniformly spaced point cloud data collection time sequence scheduling strategy by solving a multi-objective optimization problem considering maximization of information value and minimization of collection cost; a point cloud feature extraction module, configured to perform three-dimensional point cloud data collection at time points determined according to the point cloud data collection time sequence scheduling strategy, to obtain a point cloud data set, and extract vegetation structure features from the point cloud data set. The dynamic evaluation generation module is configured to dynamically select a high-sensitivity landscape evaluation index in the current phenology stage according to the sensitivity weight matrix, and calculate the high-sensitivity landscape evaluation index based on the vegetation structure characteristics to obtain a landscape evaluation result. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The computer program, when executed by the processor, implements the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the method of any one of claims 1 to 5.

Images