A heterogeneous partitioning method and device based on geographic weighted regression and geographic detectors

Abstract

This application discloses a method and apparatus for heterogeneous zoning based on geographic weighted regression and geographic detectors, relating to the field of ecological spatial heterogeneity zoning. The method includes: determining response variables and obtaining annual EVI raster data; constructing a candidate dataset of influencing factors; performing spatial registration on all factors; calculating Sen's slope and performing Mann-Kendall significance discrimination based on the annual EVI raster; fitting a GWR model and outputting local coefficients; correcting the local coefficients; classifying key factors into several categories and obtaining key factor weights; constructing an intensity index for each category and performing normalization and grading to obtain graded indices; and constructing spatial heterogeneity zoning rules based on key factor weights and graded indices to zonify and analyze the spatial heterogeneity of vegetation change, obtaining the analysis results. This method can solve the problem of weak interpretability of zoning results obtained by traditional ecological heterogeneity zoning.

Description

Technical Field

[0001] This application relates to the field of ecological spatial heterogeneity zoning technology, and in particular to a heterogeneous zoning method and apparatus based on geographic weighted regression and geographic detectors. Background Technology

[0002] Remote sensing technology, due to its high spatial and temporal resolution and strong quantitative analysis capabilities, is now widely used in ecological environment monitoring and assessment. For example, the analysis of time series of various ecological indices provides an effective tool for ecological quality monitoring and management. In complex ecological and geological environments such as high mountains and canyons, ecological quality is simultaneously influenced by multiple factors, including ecological climate, human activities, geological geomorphology, and hydrological environment. Related ecological indicators exhibit significant spatial heterogeneity, and their influencing factors often also show obvious spatial differentiation characteristics. To achieve regional ecological assessment and governance, it is usually necessary to consider this spatial heterogeneity and divide the study area into spatially heterogeneous partitions, classifying it into spatial units with relatively consistent ecological processes or driving mechanisms, thereby facilitating analysis and governance.

[0003] However, most current methods for ecological quality analysis focus on trend calculations or regression analysis of single indicators, and many methods fail to effectively explain spatial heterogeneity. To quantify spatial heterogeneity, methods such as Geographically Weighted Regression (GWR) construct spatial weights that decay with geographical distance within spatial cells using kernel functions. Local weighted regression is then performed at each calibration location, allowing regression coefficients to vary with spatial location. This enables local calculation of the impact of spatial nonstationarity and outputs information such as local coefficients and local goodness of fit, making it widely used in the analysis of ecological geographical mechanisms.

[0004] While GWR and its related extensions can effectively describe the spatial heterogeneity of influencing factors and reveal local relationships between them, there is currently no mature solution for constructing spatially heterogeneous partitioning results that are highly interpretable and reproducible. Traditional spatially heterogeneous partitioning methods (such as clustering and empirical threshold segmentation) often suffer from varying degrees of subjectivity and sensitivity issues in parameter setting, category definition, and spatial continuity, making it difficult for partitioning results to maintain good interpretability and reproducibility across different regions and different data. Summary of the Invention

[0005] The purpose of this application is to provide a heterogeneous zoning method and apparatus based on geographic weighted regression and geographic detectors, which can solve the problem of weak interpretability of zoning results obtained by traditional ecological heterogeneous zoning, and thus serve the ecological assessment and governance in complex ecological and geological environments.

[0006] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a heterogeneous partitioning method based on geographic weighted regression and geographic detectors, the method comprising: Acquire remote sensing image data of the study area.

[0007] The response variables are determined based on the remote sensing image data, and the annual EVI raster is obtained.

[0008] Construct a candidate dataset of influencing factors based on several dimensions; and perform spatial registration operation on all factors based on the response variable to obtain key factors; the spatial registration operation includes: uniform projection and uniform resolution.

[0009] Based on the annual EVI raster, Sen's slope calculation and Mann-Kendall significance discrimination are performed to obtain the Sen's slope raster and MK significance classification map.

[0010] After performing spatial correlation analysis on the key factors, a GWR model is fitted with Sen's slope raster as the dependent variable and the key factors as independent variables; local coefficients are output based on the GWR model.

[0011] The local coefficients are then subjected to multiple tests and corrections to obtain the effective local coefficients.

[0012] The key factors are categorized into several classes based on their mechanistic properties and research needs; and the explanatory power index of the optimal parameter geographic detector is used. q The normalized weights within each category are calculated to obtain the weights of the key factors.

[0013] An intensity index for each category is constructed based on the effective local coefficients; and the intensity index for each category is normalized and classified to obtain the classified index.

[0014] Based on the weights of the key factors and the graded indices, spatial heterogeneity partitioning rules are constructed, and the spatial heterogeneity of vegetation change is partitioned to obtain several types of partitions.

[0015] For each partition, a partition-dominant factor analysis was performed, and the analysis results were obtained.

[0016] Optionally, the response variable is determined based on the remote sensing image data, and an annual EVI raster is obtained, specifically including: The study area was defined, and the interannual rate of change of the Enhanced Vegetation Index (EVI) was selected as the response variable.

[0017] The study timeframe was determined, Landsat surface reflectance products for the corresponding period were obtained, and after uniform projection, resolution matching and radiometric consistency processing and mask generation, the effective pixel EVI was calculated.

[0018] Based on the effective cell EVI, an annual EVI raster is generated using the annual median or medoid synthesis method.

[0019] Optionally, spatial registration is performed on all factors based on the response variable to obtain key factors, specifically including: Explanatory power index of geospatial detectors using optimal parameters q Value screening and multiple influencing factors whose EVI spatial explanatory power is greater than the preset explanatory power threshold.

[0020] The variance inflation factor is used to diagnose multicollinearity of the various influencing factors, and collinear factors that meet the preset threshold are eliminated to obtain the key factors.

[0021] Optionally, the formula for calculating Sen's slope is: ; in, The interannual rate of change is expressed in EVI / year. The length of the time series; for k Enhanced vegetation index at any time; for i Enhanced vegetation index at any time; for k time; for i time.

[0022] Optionally, the expression for MK is: ; in, For MK statistics; sgn (·) is a sign function; The length of the time series; for k Enhanced vegetation index at any time; for i Enhanced vegetation index at any time; for k time; for i time.

[0023] Optionally, the expression for the GWR model is: ; in, For in position The value of the dependent variable at that location; spatial unit i coordinates ( ); These are local regression coefficients; For in position The value of the independent variable at the location; For residuals; To estimate the intercept term; The number of independent variables.

[0024] Optionally, a partition-dominant factor analysis is performed on each partition to obtain the analysis results, which specifically include: Determine the representative value for each factor within each partition; the representative value is the median of the local coefficients for positive and negative effects.

[0025] The absolute values ​​of the representative values ​​are sorted from largest to smallest to obtain the ranking of dominant factors.

[0026] When the representative value is greater than 0, the decision factor has a promoting effect on the response variable in the corresponding partition; when the representative value is less than 0, the decision factor has an inhibiting effect on the response variable in the corresponding partition.

[0027] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the heterogeneous partitioning method based on geographic weighted regression and geographic detectors as described above.

[0028] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the heterogeneous partitioning method based on geographic weighted regression and geographic detectors as described above.

[0029] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the heterogeneous partitioning method based on geographic weighted regression and geographic detectors as described above.

[0030] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a heterogeneous zoning method and apparatus based on geographic weighted regression and geographic detectors. It acquires remote sensing image data of the study area; determines response variables based on the remote sensing image data and generates an annual EVI raster, which can accurately characterize the vegetation growth status and temporal changes in the study area, establishing scientifically reasonable core response indicators for subsequent vegetation trend analysis and driving factor modeling; constructs a multi-dimensional influencing factor candidate dataset and completes spatial registration and screening of key factors through unified projection and resolution, achieving spatial benchmark unification of multi-source factor data, eliminating redundant and irrelevant factors, and improving the data adaptability and accuracy of factor selection in subsequent modeling and analysis; and conducts Sen's... Slope trend calculation and Mann-Kendall significance discrimination can quantitatively characterize the magnitude of vegetation temporal changes and scientifically test the statistical significance of change trends, accurately distinguishing the types and degrees of significance of vegetation change trends. After conducting spatial correlation analysis on key factors, a GWR model is fitted and local coefficients are output, which can break through the homogeneity assumption of traditional global regression, effectively capturing the spatial differentiation characteristics of the influence intensity of vegetation change driving factors, and achieving a refined local characterization of driving relationships. Multiple tests and corrections are performed on local coefficients to effectively avoid false positive errors in spatial statistical analysis, screen out statistically reliable effective local coefficients, and ensure the scientific nature of subsequent coefficient applications and zoning analysis. Key factors are classified according to mechanistic attributes and the optimal parameter geographic detector is used. q The calculation yields normalized weights within each category, enabling objective quantification of each factor's contribution based on its intrinsic mechanism and spatial explanatory power. This avoids subjective weighting bias and enhances the rationality and objectivity of factor weight allocation. Constructing intensity indices for each category based on local coefficients and implementing normalized grading allows for standardized comparability of the influence intensity of factors with different dimensions and types, establishing a unified quantitative grading standard for subsequent heterogeneous zoning. Combining key factor weights and grading indices to construct zoning rules and complete spatial heterogeneous zoning of vegetation change enables the comprehensive and objective, detailed division of vegetation heterogeneity in the study area, driven by multiple factors. The zoning results closely match actual spatial differentiation patterns. Conducting dominant factor analysis on each zone clarifies the dominant mechanisms and differential causes of vegetation change in each region. This application integrates the advantages of local spatial modeling using geographic weighted regression with the factor interpretation and weight quantification advantages of geographic detectors. By combining temporal trend significance discrimination, spatial data registration, statistical error correction, and objective hierarchical zoning processes, it can accurately depict the temporal evolution trend of vegetation and the spatial non-stationarity of driving factors. It can also objectively quantify the contribution weight of factors, achieve fine zoning of vegetation spatial heterogeneity, and accurately identify the dominant driving factors in each zoning. This solves the problems of traditional methods such as global homogeneous analysis, subjective factor weighting, coarse zoning results, and fuzzy analysis of driving mechanisms. It improves the refinement, scientificity, and objectivity of vegetation temporal and spatial evolution heterogeneity analysis and driving mechanism analysis, and can provide reliable technical support and decision-making basis for regional ecological vegetation management and differentiated ecological governance planning. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is an application environment diagram of a heterogeneous partitioning method based on geographic weighted regression and geographic detector in one embodiment of this application; Figure 2 A flowchart illustrating a heterogeneous partitioning method based on geographic weighted regression and geographic detectors, provided as an embodiment of this application; Figure 3 A technical roadmap for a heterogeneous partitioning method based on geographic weighted regression and geographic detectors, provided as an embodiment of this application; Figure 4 This is a schematic diagram of the annual Sen's slope of EVI data in the Lancang River Basin provided in an embodiment of this application; Figure 5 This is a schematic diagram of the MK test results in the Lancang River Basin provided in an embodiment of this application; Figure 6 Key factors of the Lancang River Basin provided in an embodiment of this application β Coefficient distribution diagram; Figure 7 A schematic diagram of vegetation change heterogeneity zoning in the Lancang River Basin provided in an embodiment of this application; Figure 8 A schematic diagram illustrating the contribution analysis of regional influence factors in the Lancang River Basin provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0033] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0034] To address the aforementioned issues, this application uses GWR local regression coefficients. β Explanatory power index of the optimal parameter geographic detector qA heterogeneous zoning method and device based on geographic weighted regression and geographic detectors are proposed, which can provide an effective technical approach for a deeper understanding of the spatial interaction patterns of multidimensional influencing factors and their analysis and interpretation in the fields of regional ecology, environment, planning and governance.

[0035] Thanks to the rapid development of computer and geographic information technologies, various spatial heterogeneity partitioning methods have been developed. The basic idea of ​​ecological spatial heterogeneity partitioning is to divide a given study area into several spatially relatively continuous, but relatively homogeneous within each area, and significantly different between them, based on the consistency of ecological characteristic variables or ecological process mechanisms. According to its technical approach, it can be mainly divided into five types: (1) Knowledge-based rule and threshold-based zoning. This type of method focuses on the construction of an ecological indicator system and the setting of hierarchical thresholds. It combines rule overlay analysis and specified boundary constraints, and uses techniques such as element merging and administrative boundary threshold fitting to construct ecological and environmental control units or functional zones. However, the parameter setting of this type of method relies on expert experience or local regulations, which is highly subjective. In addition, with the rapid updates of remote sensing data and multi-source ecological data, the cost of maintaining the rule system and recalibrating parameters is high, and it is difficult to assess the uncertainty of the zoning results and the impact of rule changes on the zoning results through a unified quantitative method.

[0036] (2) Non-spatial-constrained attribute clustering partitioning. This type of method treats raster or vector spatial units as independent samples and uses clustering algorithms such as k-means, hierarchical clustering, Gaussian mixture model, and fuzzy clustering to perform heterogeneous partitioning based on the similarity of attribute features of ecological indicators or driving factors, ultimately obtaining a set of categories with homogeneous attribute features. However, this type of method only characterizes the distance differences of attribute features during the clustering process and does not explicitly consider the spatial autocorrelation characteristics and the dependence between neighboring units. In addition, its partitioning results are highly sensitive to process parameters such as the cluster number setting, which can easily lead to significant changes in the partitioning boundary as the parameters are adjusted, thereby affecting the reliability of the research conclusions.

[0037] (3) Spatial Constraint Clustering and Regional Optimization Zoning. The significant feature of this type of method is that it explicitly introduces spatial adjacency connectivity constraints during the clustering and regionalization process. At the same time, additional constraints such as the minimum size threshold of the zoning can be superimposed to ensure that the zoning results are continuous in geographic space, which is more in line with the actual application needs of ecological management and governance units. However, the zoning results of this type of method are quite sensitive to the setting of process parameters, and these methods mostly take the similarity of attribute characteristics as the core optimization objective. When the correlation of ecological driving factors is strong or there are complex interactions, the zoning results mostly reflect statistical similarity and are difficult to reflect the interpretable ecological process differentiation.

[0038] (4) Remote sensing raster segmentation and object-oriented zoning. This type of method is based on remote sensing image segmentation technology. It first divides the raster image into spatial geographic objects with relatively homogeneous internal pixel features, and then conducts classification, clustering or regionalization analysis at the geographic object level to finally obtain zoning results with boundary morphology closer to the physical objects or landscape structure. Among them, image segmentation is the core key link of this type of method, and the quality of subsequent feature extraction, classification and zoning are highly dependent on the quality of image segmentation. However, this type of method has a significant scale effect in terms of segmentation scale and parameter selection. Oversegmentation or undersegmentation will systematically affect the accuracy of subsequent zoning boundaries. More challenging is that in scenarios such as ecological process zoning, the geographic object boundaries formed by remote sensing image segmentation may not match the natural units of ecological processes. There may be problems such as reasonable geometric boundaries but insufficient explanation of ecological mechanisms, making it difficult to accurately reflect the spatial differentiation patterns of ecological processes.

[0039] (5) Spatial Non-Stationary Relationship-Based Partitioning. This type of method first uses models such as geographic weighted regression to characterize the spatial non-stationary variation characteristics of the coefficients or effects of ecological driving factors; then, it conducts clustering and regionalization analysis on the local parameters or local model features output by the model to finally obtain regions with consistent spatial connectivity mechanisms. This type of method is quite sensitive to the quality of local effect estimation, which may lead to unstable partitioning results. Furthermore, when discretizing continuous local parameters into heterogeneous partitions, it is usually necessary to determine the number of clusters, similarity measurement methods, or spatial connectivity constraint strength, etc. Different parameter or method settings will also affect the reproducibility and cross-regional transferability of the results.

[0040] To address the shortcomings of traditional spatial heterogeneous partitioning, such as high parameter dependence and poor interpretability, this application proposes a heterogeneous partitioning method and apparatus based on geographic weighted regression and geographic detectors.

[0041] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] The heterogeneous partitioning method based on geographic weighted regression and geographic detectors provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send remote sensing image data of the study area to server 104. After receiving the remote sensing image data of the study area, server 104 determines the response variable based on the remote sensing image data and obtains the annual EVI raster; constructs a candidate dataset of impact factors based on several dimensions; and performs spatial registration operation on all factors based on the response variable to obtain key factors; the spatial registration operation includes: unified projection and unified resolution; based on the annual EVI raster, performs Sen's slope calculation and Mann-Kendall significance discrimination to obtain the Sen's slope raster and MK significance classification map; after performing spatial correlation analysis on the key factors, fits a GWR model with the Sen's slope raster as the dependent variable and the key factors as independent variables; outputs local coefficients based on the GWR model; performs multiple test correction on the local coefficients to obtain effective local coefficients; classifies the key factors into several categories according to mechanistic attributes and research needs; and uses the optimal parameter geographic detector explanatory power index. q The normalized weights within each category are calculated to obtain the key factor weights; an intensity index for each category is constructed based on the effective local coefficients; and the intensity index of each category is normalized and graded to obtain the graded index; based on the key factor weights and the graded index, spatial heterogeneity partitioning rules are constructed, and the spatial heterogeneity of vegetation change is partitioned to obtain several partitions; partitioning dominant factor analysis is performed on each partition to obtain the analysis results. The server 104 can feed back the obtained analysis results to the terminal 102. In addition, in some embodiments, the heterogeneity partitioning method based on geographic weighted regression and geographic detector can also be implemented by the server 104 or the terminal 102 alone. For example, the terminal 102 can directly perform heterogeneity partitioning based on geographic weighted regression and geographic detector on the remote sensing image data of the study area, or the server 104 can obtain the remote sensing image data of the study area from the data storage system and perform heterogeneity partitioning based on geographic weighted regression and geographic detector on the remote sensing image data of the study area.

[0043] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0044] In one exemplary embodiment, such as Figure 2 and Figure 3 As shown, a heterogeneous partitioning method based on geographic weighted regression and geographic detectors is provided. This method is executed by computer devices, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S1 to S10.

[0045] S1: Acquire remote sensing image data of the study area.

[0046] S2: Determine the response variables based on the remote sensing image data and obtain the annual EVI raster.

[0047] S3: Construct a candidate dataset of influencing factors based on several dimensions; and perform spatial registration operation on all factors based on the response variable to obtain key factors; the spatial registration operation includes: uniform projection and uniform resolution.

[0048] S4: Based on the annual EVI raster, perform Sen's slope calculation and Mann-Kendall significance discrimination to obtain the Sen's slope raster and MK significance classification map.

[0049] S5: After performing spatial correlation analysis on the key factors, fit a GWR model with Sen's slope raster as the dependent variable and the key factors as independent variables; output local coefficients based on the GWR model.

[0050] S6: Perform multiple tests and corrections on the local coefficients to obtain the effective local coefficients.

[0051] S7: The key factors are categorized into several classes according to their mechanistic properties and research needs; and based on the explanatory power index of the optimal parameter geographic detector... q The normalized weights within each category are calculated to obtain the weights of the key factors.

[0052] S8: Construct an intensity index for each category based on the effective local coefficients; and normalize and classify the intensity index of each category to obtain the classified index.

[0053] S9: Based on the weights of the key factors and the graded index, construct spatial heterogeneity partitioning rules and partition the spatial heterogeneity of vegetation change to obtain several types of partitions.

[0054] S10: Perform a partition-dominant factor analysis on each partition and obtain the analysis results.

[0055] Implementing steps S1 to S10 above has the following advantages: (1) The partitioning results are interpretable. Traditional clustering and partitioning of local coefficients often relies on the number of clusters, initial conditions and distance metrics, and the meaning of boundaries and categories is difficult to reproduce stably. This application aggregates and normalizes the multi-factor local effects according to the mechanism dimension into three-dimensional influence coordinates, thereby transforming the relative contributions of different influence dimensions into structures with real physical meaning for expression.

[0056] (2) The partitioning results are reproducible. This application determines the partitioning type through the inequality threshold rule. The partitioning category is uniquely determined by a clear criterion. Therefore, under the same input and parameters, completely consistent partitioning results can be obtained, and each category corresponds to a clear physical meaning and interpretation.

[0057] As an optional implementation, in step S2, the response variable is determined based on the remote sensing image data, and the annual EVI raster is obtained, specifically including: S21: Determine the scope of the study area and select the interannual rate of change of the Enhanced Vegetation Index (EVI) as the response variable.

[0058] S22: Determine the time frame of the study, obtain Landsat surface reflectance products for the corresponding period, and calculate the EVI of each effective pixel after uniform projection, resolution matching and radiometric consistency processing and mask generation.

[0059] S23: Based on each effective cell EVI, generate an annual EVI raster using the annual median or medoid synthesis method.

[0060] This step falls under the category of response variable identification and acquisition. Specifically, it involves defining the study area and selecting the interannual rate of change of the Enhanced Vegetation Index (EVI). To determine the response variable, the time frame of the study was defined, and Landsat surface reflectance products for the corresponding period were obtained. All imagery underwent uniform projection, resolution matching, and radiometric consistency processing, and cloud, shadow, snow, and water body masks were generated based on quality band or cloud detection algorithms. EVI was calculated for each effective pixel. With an annual time step, effective EVI observations after masking were aggregated within each year's growing season window, and annual EVI raster (annual EVI dataset) was generated using either the annual median or medoid synthesis method.

[0061] As an optional implementation, in step S3, spatial registration is performed on all factors based on the response variable to obtain key factors, specifically including: S31: Explanatory power index of geospatial detectors using optimal parameters q Value screening and multiple influencing factors whose EVI spatial explanatory power exceeds the preset explanatory power threshold; S32: Use the variance inflation factor to perform multicollinearity diagnosis on the multiple types of influencing factors, remove collinear factors that meet the preset threshold, and obtain the key factors.

[0062] This step involves identifying key factors. Specifically, it involves constructing a candidate dataset of influencing factors covering dimensions such as human activities, ecology, climate, and geology, and performing spatial registration operations such as unified projection and resolution on all factors. The explanatory power index of the optimal parameter geographic detector is then used. q The system screens multiple influencing factors with high explanatory power in the EVI space, and uses the variance inflation factor (VIF) to diagnose multicollinearity. Based on actual needs, it further eliminates highly collinear factors with a preset threshold, thereby controlling multicollinearity among factors and obtaining key factors.

[0063] As an optional implementation, step S4 pertains to EVI trend extraction and MK significance determination. Specifically, an interannual EVI time series is established for each spatial pixel. Sen's slope is used as the rate of interannual variation. Sen's slope is calculated using the median of pairwise slopes: (1); in, The interannual rate of change is expressed in EVI / year. The length of the time series; for k Enhanced vegetation index at any time; for i Enhanced vegetation index at any time; for k time; for i time.

[0064] right The significance of the trend is determined using the Mann-Kendall (MK) test. The MK statistic can be expressed as: (2); in, For MK statistics; sgn (·) is a sign function; The length of the time series; for k Enhanced vegetation index at any time; for i Enhanced vegetation index at any time; for k time; for i time.

[0065] When considering the existence of commutative values ​​in the sequence, the variance Var(S) can be calculated with commutative correction, and the standardized statistic Z can be constructed accordingly: (3); Based on significance level Calculate the significance probability p The values ​​are calculated, and the Sen's slope raster and MK saliency classification map are output.

[0066] As an optional implementation, step S5 involves GWR fitting. Specifically, let the study area have a total of Each spatial observation unit, whose spatial location is denoted as . The interannual Sen's slope of EVI was used as the response variable, and key factors were used as independent variables. Construct the GWR model: (4); in, For in position The value of the dependent variable at that location; spatial unit i coordinates ( ); These are local regression coefficients; For in position The value of the independent variable at the location; For residuals; To estimate the intercept term; The number of independent variables.

[0067] Parameter estimation uses weighted least squares: (5); in, For parameter estimation; The spatial weight matrix is ​​determined by the kernel function and bandwidth, with the bandwidth selected through AICc or cross-validation. This is the sampling matrix for the independent variables, with the first column containing a value of 1, used to estimate the intercept term. ; This is a column vector of sampled values ​​for the dependent variable. The output includes local coefficients, t-statistics, and local R-squared values. 2(s) and residual raster.

[0068] As an optional implementation, steps S6-S8 pertain to the construction and normalization grading of the three-dimensional intensity index. Specifically, let the set of all key factors be... A The key factors are classified into several categories according to their mechanistic properties and research needs.

[0069] For ease of explanation, the following related information will be divided into three categories, which will be referred to as Category I. Category 2 and the third category The three types of data sets are mutually exclusive and satisfy the following conditions: After obtaining the local coefficients of the univariate variables, to ensure their reliability, multiple tests are first performed to correct the local significance of each factor, thus obtaining the effective local coefficients. Then, let's assume... For the first Key factors are explained by the optimal parameter geographic detector explanatory power index. The calculated weights. To eliminate the impact of differences in the number of factors and total weights within different categories on the three-dimensional intensity index, weight normalization is performed within each category, as shown in the following formula: (6); (7); (8); Based on this, three categories of intensity indices are constructed. The first category intensity index is as follows: (9); Second category intensity index: (10); Third category intensity index: (11); In the formula, Index for spatial units; spatial units The coordinates are .

[0070] A higher intensity index indicates a stronger overall influence of that category on the EVI variation of the unit. Based on this, the three intensity indices were min-max normalized within the study area: (12); (13); (14); Subsequently, the normalized , , Calculate the 25th, 50th, and 75th percentiles respectively. Etc. Taking the first category as an example, Divided into four levels: 1-4 (15); The other two categories follow the same pattern, resulting in... and Their values ​​range from 1 to 4. After completing the exponential classification, each spatial unit... It can be represented by a level triple: (16); As an optional implementation, step S9 involves constructing an interpretability threshold system to complete the spatial heterogeneity partitioning. Specifically, based on the three categories of indicators after the above classification, spatial heterogeneity partitioning is performed according to the following rules: ①When At that time, The designation as a comprehensive high-impact area indicates that all three categories of factors have a strong effect at the local scale, and the spatial differences in vegetation index (EVI) variation are shaped by multiple strong influencing factors.

[0071] ②When and At that time, Setting it as the first category dominant region means that the spatial difference of EVI is more affected by the first category than by the other two categories of factors.

[0072] ③When and At that time, The fact that the second category is designated as the dominant region indicates that the second category factor has a stronger effect on EVI changes in this region.

[0073] ④ When and At that time, It is designated as the third category dominant area, where the third category factors have a strong influence on vegetation change.

[0074] ⑤ At that time, The area is designated as a low-impact zone to indicate that the overall impact of the three categories of factors on the EVI is generally weak or the regional process is relatively stable, and vegetation change may be more affected by factors not explicitly included or random disturbances.

[0075] ⑥ Spatial units other than the above 5 categories are designated as multi-factor composite transition zones, reflecting that at least two types of influence dimensions have a high or medium-high level of influence at the local scale, but have not formed a pattern of high influence across all categories or dominance by a single category.

[0076] As an optional implementation, in step S10, a partition dominance factor analysis is performed on each partition to obtain the analysis results, specifically including: S101: Determine the representative value of each factor within each partition; the representative value is the median of the local coefficients for positive and negative effects.

[0077] S102: Sort the absolute values ​​of the representative values ​​from largest to smallest to obtain the ranking of dominant factors.

[0078] S103: When the representative value is greater than 0, the decision factor has a promoting effect on the response variable in the corresponding partition; when the representative value is less than 0, the decision factor has an inhibiting effect on the response variable in the corresponding partition.

[0079] This step belongs to the heterogeneous partitioning factor contribution analysis. Specifically, for each factor within each partition, the median of the local coefficients with positive and negative effects is taken as the representative value of the positive and negative effects of that factor. The median is calculated as follows: (17); In the formula, The median; For partitioning The set of spatial units it contains.

[0080] For the same partition All candidate factors according to Sort the partitions by absolute values ​​from largest to smallest to obtain the ranking of the dominant factors. Further, use the average action coefficient... The direction of influence is determined by the sign: when At that time, it was considered that the factor This partition has a promoting effect on the response variable; when When the coefficients were considered to have an inhibitory effect, in order to make the variables comparable, the local effective coefficients were first standardized by Min–Max to scale them to the interval [-1,1], so as to perform factor analysis plotting.

[0081] The present application will be further described below through specific embodiments.

[0082] Step 1: Determine the spatial unit and input response variables. The Lancang River basin is selected as the study area, with a 1km raster pixel as the basic spatial unit, and the unit center coordinates are: The interannual variation rate of EVI Sen's slope from 2000 to 2020 was used as the response variable. The EVI data were obtained from Landsat data. After preprocessing, the annual EVI raster data were generated using the medoid synthesis method, and the annual EVI raster was resampled to a resolution of 1 km.

[0083] Step two: Establish a candidate impact factor library and perform preprocessing. Construct a candidate impact factor set. Spatial registration was performed between the response variable and the candidate impact factors, and the candidate impact factors were standardized to ensure that the effects of different factors were comparable.

[0084] Step 3: EVI trend extraction and significance determination. For the EVI data from 2000 to 2020, the Theil-SenMedian algorithm was used to calculate the annual slope of change, and the Mann-Kendall test was used for significance testing to obtain the annual Sen's slope of EVI. Figure 4 ) and MK test results ( Figure 5 ).

[0085] Step four: Key factor screening and spatial correlation analysis. The optimal parameter geographic detector OPGD is used to calculate the pairs of each candidate factor. y Explanatory power q And sort, and combine VIF (by VIF Key factors with high explanatory power and clear mechanisms were retained (with a threshold of <10): distance from fault, distance from hydropower station, distance from landslide, GDP, nighttime light, population, river network density, road network density, slope, soil erosion intensity, average annual temperature, and comprehensive curvature. Spatial correlation analysis was performed on the key factors to confirm their spatial distribution differentiation characteristics, and then they were divided into three categories according to mechanism: H (human activities), C (ecological climate), and G (geological geomorphology) (equivalent to the first, second, and third categories mentioned above, respectively).

[0086] Step 5, calculate GWR. A GWR model was fitted with the dependent variable and key factors as independent variables. The bandwidth was determined using AICc, the kernel function was chosen to be of the Bisquare form, and the bandwidth search used the Golden Section to obtain the local coefficients. ( Figure 6 ).

[0087] Step 6: Construct spatial heterogeneity partitioning rules. Calculate the effective local coefficients within each of the three categorical factors after multiple test corrections. The average value, combined with the OPGD-based explanatory power index. The normalized weights within each category are calculated using the values. , Based on this, a weighted synthesis is performed within the three category factors to construct three comprehensive driving force intensity indices. and Furthermore, it is normalized and classified into three-dimensional influence coordinates. Based on mutually exclusive partitioning rules, spatial heterogeneity is divided into 6 types of partitions ( Figure 7 The zones are categorized as follows: high-impact zone, human-dominated zone, ecological-climate-dominated zone, geological-geomorphological-dominated zone, low-impact zone, and multi-factor composite transition zone.

[0088] Step 7, Partition Factor Contribution Analysis. For each partition, calculate the positive and negative effect sample points for each variable, and scale them to the interval [-1, 1] using Min–Max standardization to eliminate scale differences between variables. Then, define the promoting strength as the median of the positive effect value and the inhibiting strength as the median of the negative effect value, thereby creating a bidirectional stacked bar chart that simultaneously displays the direction and magnitude of the driving factor contributions within the spatial region. Figure 8 (This is used to support the explanation and management of mechanisms for changes in the ecological quality of watersheds.)

[0089] In summary, this application constructs a spatial heterogeneity partitioning rule based on the local regression coefficients of geographic weighted regression, which considers the weight of the explanatory power index of the optimal parameter geographic detector, and can achieve interpretable spatial heterogeneity partitioning.

[0090] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 9 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores remote sensing image data of the study area. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a heterogeneous partitioning method based on geographic weighted regression and a geographic detector.

[0091] Those skilled in the art will understand that Figure 9The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0092] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0093] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0094] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of the relevant data are carried out in compliance with the relevant data protection laws and policies of the country where the location is located, and with the authorization granted by the owner of the corresponding device.

[0095] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0096] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0097] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A heterogeneous partitioning method based on geographic weighted regression and geographic detectors, characterized in that, The method includes: Acquire remote sensing image data of the study area; The response variables are determined based on the remote sensing image data, and the annual EVI raster is obtained; Construct a candidate dataset of influencing factors based on several dimensions; and perform spatial registration operation on all factors based on the response variable to obtain key factors; the spatial registration operation includes: uniform projection and uniform resolution; Based on the annual EVI raster, Sen's slope is calculated and Mann-Kendall significance is determined to obtain the Sen's slope raster and MK significance classification map; After performing spatial correlation analysis on the key factors, a GWR model was fitted with Sen's slope raster as the dependent variable and the key factors as independent variables; local coefficients were output based on the GWR model. The local coefficients are then subjected to multiple tests and corrections to obtain the effective local coefficients. The key factors are categorized into several classes based on their mechanistic properties and research needs; and the explanatory power index of the optimal parameter geographic detector is used. q The values ​​are used to calculate the normalized weights within each category, thus obtaining the key factor weights; The intensity index of each category is constructed based on the effective local coefficients; and the intensity index of each category is normalized and classified to obtain the classified index. Based on the weights of the key factors and the graded index, spatial heterogeneity partitioning rules are constructed, and the spatial heterogeneity of vegetation change is partitioned to obtain several types of partitions. For each partition, a partition-dominant factor analysis was performed, and the analysis results were obtained.

2. The heterogeneous partitioning method based on geographic weighted regression and geographic detectors according to claim 1, characterized in that, Based on the remote sensing image data, response variables are determined and annual EVI gratings are obtained, specifically including: The study area was defined, and the interannual rate of change of the enhanced vegetation index (EVI) was selected as the response variable. The time frame of the study was determined, Landsat surface reflectance products for the corresponding period were obtained, and after uniform projection, resolution matching and radiometric consistency processing and mask generation, the effective pixel EVI was calculated. Based on the effective cell EVI, an annual EVI raster is generated using the annual median or medoid synthesis method.

3. The heterogeneous partitioning method based on geographic weighted regression and geographic detectors according to claim 1, characterized in that, Based on the aforementioned response variable, spatial registration is performed on all factors to obtain key factors, specifically including: Explanatory power index of geospatial detectors using optimal parameters q Value screening and multiple influencing factors whose EVI spatial explanatory power exceeds the preset explanatory power threshold; The variance inflation factor is used to diagnose multicollinearity of the various influencing factors, and collinear factors that meet the preset threshold are eliminated to obtain the key factors.

4. The heterogeneous partitioning method based on geographic weighted regression and geographic detectors according to claim 1, characterized in that, The formula for calculating Sen's slope is: ； in, The interannual rate of change is expressed in EVI / year. The length of the time series; for k Enhanced vegetation index at any time; for i Enhanced vegetation index at any time; for k time; for i time.

5. The heterogeneous partitioning method based on geographic weighted regression and geographic detectors according to claim 1, characterized in that, The expression for MK is: ； in, For MK statistics; sgn (·) is a sign function; The length of the time series; for k Enhanced vegetation index at any time; for i Enhanced vegetation index at any time; for k time; for i time.

6. The heterogeneous partitioning method based on geographic weighted regression and geographic detectors according to claim 1, characterized in that, The expression for the GWR model is: ； in, For in position The value of the dependent variable at that location; spatial unit i coordinates ( ); These are local regression coefficients; For in position The value of the independent variable at the location; For residuals; To estimate the intercept term; The number of independent variables.

7. The heterogeneous partitioning method based on geographic weighted regression and geographic detectors according to claim 1, characterized in that, For each partition, a partition-dominant factor analysis was performed, and the analysis results were obtained, including: Determine the representative value for each factor within each partition; the representative value is the median of the local coefficients for positive and negative effects. The absolute values ​​of the representative values ​​are sorted from largest to smallest to obtain the ranking of dominant factors; When the representative value is greater than 0, the decision factor has a promoting effect on the response variable in the corresponding partition; when the representative value is less than 0, the decision factor has an inhibiting effect on the response variable in the corresponding partition.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the heterogeneous partitioning method based on geographic weighted regression and geographic detectors as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the heterogeneous partitioning method based on geographic weighted regression and geographic detectors as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the heterogeneous partitioning method based on geographic weighted regression and geographic detectors as described in any one of claims 1-7.

Images