Ecological corridor identification method and system based on machine learning
The ecological corridor identification method, which combines machine learning and circuit theory, solves the problem of strong subjectivity in the construction of ecological resistance surfaces, realizes the objective quantification of ecological corridors and the accurate characterization of nonlinear relationships, and improves the scientific nature and engineering application value of the identification results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI INST OF TECH
- Filing Date
- 2026-01-30
- Publication Date
- 2026-05-12
AI Technical Summary
Existing methods for identifying ecological corridors rely on expert experience to construct ecological resistance surfaces, which is highly subjective and cannot effectively characterize the nonlinear relationships between multiple sources of factors. This results in poor consistency and repeatability of the results, limiting their applicability.
Machine learning methods are used to train a classification model based on ecosystem service data and multi-source spatial influence factors to predict ecological resistance values. In addition, circuit theory is combined to simulate ecological flow processes and identify ecological corridors.
It improves the objectivity and accuracy of ecological corridor identification, reduces reliance on human-assigned values, can characterize the nonlinear impact of multi-source factors on ecological flow, enhances the accuracy and regional adaptability of the results, and provides a precise basis for protection and restoration decisions.
Smart Images
Figure CN122020306A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ecological protection planning technology, and in particular to a method and system for identifying ecological corridors based on machine learning. Background Technology
[0002] With the accelerating pace of global urbanization, natural ecological spaces are continuously being squeezed, habitat fragmentation is becoming increasingly prominent, and biodiversity conservation and the maintenance of ecosystem services face significant challenges. Against this backdrop, scientifically identifying and constructing ecological corridors to effectively connect scattered and isolated ecological patches is of great significance for ensuring regional ecological security, promoting species migration and gene exchange, and enhancing the overall stability of ecosystems. This has become a key technical aspect of territorial spatial planning and ecological protection and restoration.
[0003] Existing methods for identifying ecological corridors are typically based on a technical framework of "ecological source area – ecological resistance surface – ecological corridor." The construction of the ecological resistance surface, used to characterize the degree to which different spatial units impede ecological flow processes, is a core element affecting the accuracy and reliability of corridor identification results. However, existing resistance surface construction methods generally suffer from the following shortcomings: First, resistance values and weights often rely on subjective assignment based on expert experience, resulting in highly subjective outcomes and difficulty in ensuring consistency and reproducibility among different researchers or regions. Second, they often employ fixed or empirically set weighting systems, which fail to effectively reflect the differences in ecological environment conditions and human disturbance characteristics across different regions, limiting their applicability. Third, they are mostly based on the assumption of linear superposition, making it difficult to characterize the complex nonlinear relationships exhibited by ecological processes.
[0004] For example, invention patent CN119692602A discloses a quantitative method for ecological restoration based on a complex network of watershed ecological security. This method selects five characteristic indicators—watershed topography, land cover, anthropogenic disturbance, vegetation status, and ecological benefits—to construct a watershed ecological resistance surface, and then builds a complex network model of ecological security based on this surface. Although this method integrates multiple influencing factors, it is essentially still a construction method based on a pre-set indicator system. When determining the resistance value of each spatial unit, this method usually relies on classifying the five selected characteristic indicators, assigning empirical values, or performing simple superposition calculations. This approach lacks a training process based on real sample data, making it difficult to objectively quantify the specific contribution of each indicator to ecological resistance. At the same time, the superposition method based on the indicator system makes it difficult to capture the complex and nonlinear joint stress effects of multiple environmental factors (such as the interaction between anthropogenic disturbance and vegetation status) on ecological flow, resulting in room for improvement in the objectivity and accuracy of the generated ecological resistance surface.
[0005] In summary, the current construction of ecological resistance surfaces in ecological corridor identification faces the challenge of relying heavily on subjective assignment based on expert experience and failing to effectively characterize the nonlinear relationships between multiple factors through objective quantification. Summary of the Invention
[0006] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a machine learning-based method and system for identifying ecological corridors.
[0007] The objective of this invention can be achieved through the following technical solutions: According to one aspect of the present invention, a machine learning-based method for identifying ecological corridors is provided, the method comprising the following steps: S1. Obtain land use data for the study area and extract ecological source areas based on the land use data; S2. Obtain ecosystem service data and multi-source spatial impact factors in the study area, train a machine learning classification model using the ecosystem service data and multi-source spatial impact factors, input the multi-source spatial impact factors into the trained machine learning classification model, predict the ecological resistance value of each spatial unit in the study area, and output the ecological resistance surface. S3. Input the ecological source areas and ecological resistance surfaces into the ecological connectivity analysis model based on circuit theory, simulate the ecological flow process, identify the spatial channels connecting different ecological source areas based on the ecological flow probability or cumulative resistance, and determine the ecological corridors within the study area.
[0008] As a preferred technical solution, the specific steps for extracting the ecological source area in S1 include: The land use types in the study area were reclassified into ecological land and non-ecological land, and ecological land patches were extracted as candidate areas. The candidate regions were processed using morphological spatial pattern analysis methods to select patches with core morphological types as candidate ecological source areas. Connectivity evaluation or area threshold screening is performed on candidate ecological source areas, and patches that meet the preset conditions are identified as ecological source areas.
[0009] As a preferred technical solution, the ecosystem service data in S2 includes water production, carbon storage, habitat quality, and soil conservation data.
[0010] As a preferred technical solution, the specific steps for training the machine learning classification model in S2 include: Ecosystem service data are normalized, and the normalized ecosystem service data is used to construct a comprehensive ecosystem service level distribution. The area with the lowest level in the distribution of integrated ecosystem service levels was marked as the negative sample sampling area, and the ecological source area was marked as the positive sample sampling area. Multiple sample points were randomly selected from the negative sample sampling area and the positive sample sampling area respectively. The multi-source spatial influence factor corresponding to the selected sample points was used as the feature variable, and the category attribute of the sample points was used as the label variable to construct a training dataset. The machine learning classification model was trained using this training dataset.
[0011] As a preferred technical solution, the machine learning classification model in S2 is a gradient boosting model, specifically including XGBoost, CatBoost, or LightGBM.
[0012] As a preferred technical solution, an automated hyperparameter optimization method is used to optimize the model parameters during the training of the machine learning classification model.
[0013] As a preferred technical solution, the multi-source spatial influence factors in S2 include natural environmental factors and human activity factors; Natural environmental factors include one or more of the following: elevation, slope, topographic relief, normalized vegetation index, and rainfall. Human activity factors include one or more of the following: distance from roads, distance from water bodies, population density, nighttime light intensity, human footprint index, and gross domestic product.
[0014] As a preferred technical solution, the specific process of predicting the ecological resistance value of each spatial unit within the study area and outputting the ecological resistance surface in S2 includes: The study area is divided into multiple regular grid cells. A prediction point is generated at the center of each grid cell, and a corresponding influence factor value is matched for each prediction point to construct a prediction dataset. The predicted dataset is input into the trained machine learning classification model, which outputs a set of predicted ecological resistance values. Spatial correlation processing is performed on each ecological resistance value in the ecological resistance value set to associate each ecological resistance value with the corresponding regular grid cell, thereby generating spatialized prediction result data; The predicted data is converted into a raster data format to form the ecological resistance surface of the study area.
[0015] As a preferred technical solution, the ecological connectivity analysis model based on circuit theory in S3 is also used to identify ecological pinch points. The specific steps for identifying ecological pinch points are as follows: Based on the ecological flow process simulated by the ecological connectivity analysis model, the ecological flow density of each spatial unit in the study area is calculated. Areas with ecological flow density exceeding a preset threshold are identified as ecological pinch points; The circuit theory-based ecological connectivity analysis model in S3 is also used to identify ecological barrier points. The specific steps for identifying ecological barrier points are as follows: Based on the ecological connectivity analysis model, the ecological flow improvement coefficient of each spatial unit in the study area is calculated; the ecological flow improvement coefficient represents the degree of improvement in ecological flow connectivity after removing the ecological resistance of the spatial unit. Areas with an ecological flow improvement coefficient higher than a preset threshold are identified as ecological barrier points.
[0016] According to another aspect of the present invention, a machine learning-based ecological corridor identification system is provided, the system comprising an ecological source area identification module, an ecological resistance surface construction module, and an ecological corridor identification module; The ecological source area identification module is used to extract and output the ecological source areas based on the land use data of the study area. The ecological resistance surface construction module is used to acquire ecosystem service data and multi-source spatial impact factors. Based on the ecosystem service data and multi-source spatial impact factors, a machine learning classification model is trained, and the trained model is combined with multi-source spatial impact factors to predict the ecological resistance value of each spatial unit and output the ecological resistance surface. The ecological corridor identification module is used to input ecological source areas and ecological resistance surfaces into an ecological connectivity analysis model based on circuit theory, simulate the ecological flow process, and identify ecological corridors connecting ecological source areas.
[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention acquires ecosystem service data and multi-source spatial influencing factors of the study area, and automatically establishes a nonlinear mapping relationship between them using a machine learning model to predict the overall ecological resistance value. Furthermore, it combines the generated objective resistance surface with circuit theory for corridor identification, reducing the subjectivity of ecological resistance assignment and improving the objectivity and accuracy of ecological corridor identification results. This solves the problem of strong subjectivity and lack of scientific basis in existing technologies due to human weighting. This invention achieves objective quantification of ecological resistance through machine learning methods, reducing reliance on empirical assignment and improving the objectivity and repeatability of results. It can effectively characterize the nonlinear influence of multi-source natural environmental factors and human activity factors on the ecological flow process, improving the accuracy of spatial representation of ecological resistance. Simultaneously, by combining circuit theory models, it achieves the collaborative identification of ecological corridors, ecological pinch points, and ecological barrier points, demonstrating good regional adaptability and engineering application value.
[0018] 2. This invention utilizes normalized ecosystem service data to construct a comprehensive grading system, automatically labeling low-grade areas as negative samples and source areas as positive samples, thereby building a training dataset. Input features encompass natural factors such as elevation and human activity factors such as nighttime light pollution, and training is performed using gradient boosting models such as XGBoost combined with hyperparameter optimization. By using objective ecosystem service data as proxy labels, this invention cleverly solves the technical bottleneck of lacking real-label data in ecological resistance inversion, achieving automated and standardized construction of the training set. Simultaneously, by integrating multi-source factors and advanced gradient boosting algorithms, it can accurately capture the combined stress effects of human activities and the natural environment on species migration, significantly improving the model's generalization ability and spatial resolution in predicting resistance values in unknown areas.
[0019] 3. This invention, in extracting ecological source areas, not only bases its approach on land use type but also incorporates morphological spatial pattern analysis methods to screen patches whose morphological type constitutes the core area. A secondary screening is then conducted using connectivity or area thresholds. This effectively overcomes the problem of traditional methods, which rely solely on land use type for screening, resulting in the selection of fragmented or edge-effect-prone, low-quality patches. Starting from the spatial structure of the landscape pattern, this method ensures that the identified ecological source areas possess both high-quality habitat conditions and sufficient internal spatial continuity, providing a reliable starting and ending point for the construction of ecological corridors.
[0020] 4. This invention simulates ecological flow processes based on circuit theory, identifies high-flow ecological pinch points by calculating ecological flow density, and identifies ecological barriers hindering connectivity by calculating ecological flow improvement coefficients. This transforms abstract connectivity analysis into concrete engineering restoration targets. The identified ecological pinch points pinpoint the bottleneck locations of corridors urgently requiring protection, and the identified ecological barriers quantify the improvement in overall connectivity after removing the resistance at those points. This provides precise decision-making basis for environmental spatial planning regarding protected and restored areas, significantly improving the targeting and efficiency of ecological governance. Attached Figure Description
[0021] Figure 1 This is a schematic diagram illustrating the steps of a machine learning-based ecological corridor identification method in this invention. Figure 2 This is a flowchart illustrating the specific implementation process of ecological corridor identification in the example. Figure 3 This is a schematic diagram of the spatial distribution of the ecological resistance surface in the embodiment; Figure 4 This is a schematic diagram of the ecological corridor identification results in the embodiment. Detailed Implementation
[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0023] Existing methods for identifying ecological corridors are typically based on a technical framework of "ecological source area – ecological resistance surface – ecological corridor." The construction of the ecological resistance surface, used to characterize the degree to which different spatial units impede ecological flow processes, is a core element affecting the accuracy and reliability of corridor identification results. However, existing resistance surface construction methods generally suffer from the following shortcomings: First, resistance values and weights often rely on subjective assignment based on expert experience, resulting in highly subjective outcomes and difficulty in ensuring consistency and reproducibility among different researchers or regions. Second, they often employ fixed or empirically set weighting systems, which fail to effectively reflect the differences in ecological environment conditions and human disturbance characteristics across different regions, limiting their applicability. Third, they are mostly based on the assumption of linear superposition, making it difficult to characterize the complex nonlinear relationships exhibited by ecological processes.
[0024] The aforementioned problems, to some extent, restrict the objectivity and precision of ecological resistance surface construction, thus affecting the scientific validity and practical application value of ecological corridor identification results. Therefore, it is necessary to propose a new method for ecological corridor identification that can reduce subjective human interference, effectively characterize multi-factor nonlinear relationships, and possess good regional adaptability.
[0025] Example 1 In this embodiment, a machine learning-based ecological corridor identification method is adopted, and the method steps are as follows: Figure 1 As shown, it specifically includes: S1. Obtain land use data for the study area and extract ecological source areas based on the land use data; S2. Obtain ecosystem service data and multi-source spatial impact factors in the study area, train a machine learning classification model using the ecosystem service data and multi-source spatial impact factors, input the multi-source spatial impact factors into the trained machine learning classification model, predict the ecological resistance value of each spatial unit in the study area, and output the ecological resistance surface. S3. Input the ecological source areas and ecological resistance surfaces into the ecological connectivity analysis model based on circuit theory, simulate the ecological flow process, identify the spatial channels connecting different ecological source areas based on the ecological flow probability or cumulative resistance, and determine the ecological corridors within the study area.
[0026] Based on land use data of the study area, important ecological patches with high ecological functions are identified and extracted as ecological source areas to characterize the starting point of ecological diffusion and material and energy flow.
[0027] A machine learning training sample set is constructed based on multiple ecosystem service data and multi-source spatial influencing factors. The machine learning classification model is used to learn the relationship between each influencing factor and ecological circulation suitability in different spatial units, and a continuously distributed ecological resistance surface is generated.
[0028] By inputting ecological source area data and ecological resistance surface data into an ecological connectivity analysis model based on circuit theory, spatial channels with high ecological flow probability or low cumulative resistance are identified, and ecological corridors within the study area are determined.
[0029] When identifying and extracting ecological source areas based on land use data of the study area, land use / cover data of the study area is acquired and transformed to a unified spatial resolution and coordinate system. Based on the importance of different land use types to ecosystem functions, land use types with high ecosystem service capacity and capable of providing stable habitat conditions for species survival, reproduction, and migration are selected as candidate ecological source area types. Candidate ecological source area types may include, but are not limited to, ecological land use types such as forests, shrubs, water bodies, and grasslands. Candidate areas are screened using spatial analysis methods, combined with connectivity analysis or area threshold screening, to eliminate scattered patches that are too small or have weak ecological functions, obtaining the final ecological source area results, which serve as the basic input data for subsequent analyses.
[0030] The identification of ecological source areas includes: reclassifying the land use types of the study area, identifying forests, shrubs, grasslands, water bodies and wetlands as candidate ecological source areas, and selecting the final ecological source areas through spatial analysis methods.
[0031] Before determining the final ecological source area, a morphological spatial pattern analysis was conducted on the candidate ecological source areas, and spatial patches with the core area as the morphological type were selected as candidate ecological source areas.
[0032] Ecosystem service data in S2 include at least one or more of the following indicators: water production, carbon storage, habitat quality, and soil conservation.
[0033] In S2, the study area is divided into multiple ecological function levels based on the normalized ecosystem service data. The area with the lowest ecological function level is identified as the negative sample sampling area, and the ecological source area is identified as the positive sample sampling area.
[0034] Sample points were randomly selected from both the negative and positive sample areas, with negative samples marked as 0 and positive samples marked as 1, to construct a binary classification machine learning training sample set.
[0035] Spatial impact factors include one or more of the following indicators: Normalized Difference Vegetation Index (NDVI), land use type, distance from road, distance from water body, population density, nighttime light intensity, elevation, slope, topographic relief, human footprint index, GDP, and rainfall.
[0036] The machine learning classification model is a gradient boosting-based model, including one or more of XGBoost, CatBoost, or LightGBM.
[0037] When training machine learning classification models, an automated hyperparameter optimization method is introduced to optimize the model parameters in order to improve the model's predictive performance and generalization ability.
[0038] In S3, the ecological connectivity analysis model based on circuit theory is further used to identify spatial units with highly concentrated ecological flows as ecological pinch points, and spatial units that significantly hinder ecological flows as ecological barrier points.
[0039] The implementation process of this method is as follows: Figure 2 As shown, the details are as follows: S1.1 Unify all types of spatial data required in the research process into the same spatial reference system and spatial resolution. The data includes influencing factors such as land use, elevation, human footprint, population density, GDP, normalized vegetation index, distance from roads, distance from water bodies, rainfall, slope, topographic relief and nighttime light, so as to ensure spatial consistency and comparability between different data. S1.2 Based on land use data within the study area, land use types are reclassified, and forests, shrubs, grasslands, water bodies, and wetlands are identified as preliminary candidate areas for ecological source areas; S1.3 Morphological spatial pattern analysis (MSPA) was conducted on the preliminary candidate ecological source areas, and spatial patches with the morphological type "core" were selected from the analysis results as candidate ecological source areas to ensure that the identified source areas have high ecological integrity and stability. S1.4 Further, the area threshold method is used to screen the candidate ecological source areas, and scattered patches with too small an area or weak ecological function are removed to obtain the final ecological source areas, which are then saved as vector data for subsequent analysis.
[0040] S2.1 Obtain data on ecosystem service indicators such as water production, carbon storage, habitat quality, and soil conservation in the study area, and normalize each indicator; based on this, use principal component analysis to determine the weight of each ecosystem service indicator and construct the spatial distribution of the normalized integrated ecosystem service capacity. S2.2 Based on the numerical distribution of integrated ecosystem service capacity, the study area was divided into five levels: "very low, low, medium, high, and very high" using an equal-interval grading method, and the area at the "very low" level was identified as the negative sample sampling area. S2.3 In both the ecological source area and the negative sample sampling area, a predetermined number of sample points are selected using a random sampling method. In this embodiment, the number of positive and negative sample points is 10,000 each. Sample points in the ecological source area are assigned a value of 1, while sample points in the negative sample sampling area are assigned a value of 0. Subsequently, the values of each influencing factor from S1.1 are matched to each sampling point to form a sample feature dataset, which is then exported as a structured data file. S2.4 Divide the study area into regular grids, generate prediction points at the center of each grid cell, and match the corresponding influence factor values for each prediction point in the same way as in S2.3 to construct a dataset for regional-scale prediction. S2.5 The sample feature dataset of S2.3 is divided into a training sample set and a validation sample set in a 7:3 ratio. The XGBoost binary classification model based on gradient boosting is used to train the training sample set, and the model performance is validated using the validation sample set. After the model training is completed, the model is applied to the regional prediction dataset of S2.4 to predict the ecological resistance value of each spatial unit in the study area. S2.6 The predicted ecological resistance values are spatially correlated to correspond one-to-one with the corresponding regular grid cells, and spatialized prediction result data is generated. S2.7 Convert the regular grid data into a raster data format to form the ecological resistance surface of the study area.
[0041] S3.1 Input the ecological source vector data obtained in S1.4 and the ecological resistance surface grid data constructed in S2.7 into the ecological connectivity analysis model based on circuit theory to construct the ecological network structure and identify the spatial channels that connect different ecological source areas with high probability of ecological flow or low cumulative resistance. The spatial channels are the ecological corridors of the study area. S3.2 Based on the calculation results of the circuit theory model, key spatial node areas with high concentration of ecological flow are further identified and ecological pinch points are determined; at the same time, spatial units that significantly hinder ecological flow are identified and ecological obstacle points are determined, where ecological pinch points are areas that urgently need key protection and ecological obstacle points are areas that urgently need priority restoration.
[0042] Spatial distribution of ecological resistance surfaces as follows Figure 3 As shown, Figure 3 In the diagram, red to blue represents a gradual decrease in ecological resistance. The red area represents the region with the highest ecological resistance, with a maximum value of 0.653; the blue area represents the region with the lowest ecological resistance, with a minimum value of 0.0135. Figure 3 In the study area, ecological resistance values showed a pattern of high correlation with the intensity of human activities and the quality of natural habitats. Low-resistance areas were mainly concentrated in contiguous forests, water bodies, and areas with good natural vegetation cover, which had a relatively small obstructive effect on species migration and ecological flow. Medium- and high-resistance areas were mostly distributed around farmland and construction land, reflecting the strong impact of human disturbance on ecological connectivity. In contrast, densely populated areas of human activity, such as towns, roads, and industrial and mining areas, had the highest resistance values, forming obvious ecological barrier zones.
[0043] Ecological corridor identification results as follows Figure 4 As shown, Figure 4 The green areas represent ecological sources, and the orange lines represent ecological corridors. Figure 4 Based on the ecological source area identification results in S1.4 and the ecogroups obtained through machine learning in S2.7, the ecological corridors of the study area were derived using circuit theory. These corridors represent regions with high current density and low resistance in the simulation, effectively demonstrating the connectivity between two ecological patches. This clearly reveals the potential paths and spatial patterns of ecological flow within the study area, providing clear spatial guidance for the construction of regional ecological security patterns and the ecological restoration of national land space.
[0044] In summary, this scheme achieves objective quantification of ecological resistance through machine learning methods, reducing reliance on empirical assignments and improving the objectivity and repeatability of the results. It can effectively characterize the nonlinear impact of multiple natural environmental factors and human activity factors on the ecological flow process, improving the accuracy of the spatial representation of ecological resistance. At the same time, combined with circuit theory models, it realizes the collaborative identification of ecological corridors, ecological pinch points, and ecological barrier points, and has good regional adaptability and engineering application value.
[0045] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A machine learning-based method for identifying ecological corridors, characterized in that, The method steps include: S1. Obtain land use data for the study area and extract ecological source areas based on the land use data; S2. Obtain ecosystem service data and multi-source spatial impact factors in the study area, train a machine learning classification model using the ecosystem service data and multi-source spatial impact factors, input the multi-source spatial impact factors into the trained machine learning classification model, predict the ecological resistance value of each spatial unit in the study area, and output the ecological resistance surface. S3. Input the ecological source area and the ecological resistance surface into the ecological connectivity analysis model based on circuit theory, simulate the ecological flow process, identify the spatial channels connecting different ecological source areas according to the ecological flow probability or cumulative resistance, and determine the ecological corridors in the study area.
2. The method for identifying ecological corridors based on machine learning according to claim 1, characterized in that, The specific steps for extracting the ecological source area in S1 include: The land use types in the study area were reclassified into ecological land and non-ecological land, and ecological land patches were extracted as candidate areas. The candidate regions were processed using morphological spatial pattern analysis methods to select patches with core morphological types as candidate ecological source areas. Connectivity evaluation or area threshold screening is performed on candidate ecological source areas, and patches that meet the preset conditions are identified as ecological source areas.
3. The method for identifying ecological corridors based on machine learning according to claim 1, characterized in that, The ecosystem service data in S2 includes water production, carbon storage, habitat quality, and soil conservation data.
4. The method for identifying ecological corridors based on machine learning according to claim 3, characterized in that, The specific steps for training the machine learning classification model in S2 include: Ecosystem service data are normalized, and the normalized ecosystem service data is used to construct a comprehensive ecosystem service level distribution. The area with the lowest level in the distribution of integrated ecosystem service levels was marked as the negative sample sampling area, and the ecological source area was marked as the positive sample sampling area. Multiple sample points were randomly selected from the negative sample sampling area and the positive sample sampling area respectively. The multi-source spatial influence factor corresponding to the selected sample points was used as the feature variable, and the category attribute of the sample points was used as the label variable to construct a training dataset. The machine learning classification model was trained using this training dataset.
5. The method for identifying ecological corridors based on machine learning according to claim 4, characterized in that, The machine learning classification model in S2 is a gradient boosting model, specifically including XGBoost, CatBoost, or LightGBM.
6. The method for identifying ecological corridors based on machine learning according to claim 4, characterized in that, In the process of training the machine learning classification model, an automated hyperparameter optimization method is used to optimize the model parameters.
7. The method for identifying ecological corridors based on machine learning according to claim 1, characterized in that, The multi-source spatial influence factors in S2 include natural environmental factors and human activity factors. The natural environmental factors include one or more of the following: elevation, slope, topographic relief, normalized vegetation index, and rainfall. The human activity factors include one or more of the following: distance from roads, distance from water bodies, population density, nighttime light intensity, human footprint index, and gross domestic product.
8. The method for identifying ecological corridors based on machine learning according to claim 1, characterized in that, The specific process of predicting the ecological resistance value of each spatial unit within the study area and outputting the ecological resistance surface in S2 includes: The study area is divided into multiple regular grid cells. A prediction point is generated at the center of each grid cell, and a corresponding influence factor value is matched for each prediction point to construct a prediction dataset. The predicted dataset is input into the trained machine learning classification model, which outputs a set of predicted ecological resistance values. Spatial correlation processing is performed on each ecological resistance value in the ecological resistance value set to associate each ecological resistance value with the corresponding regular grid cell, thereby generating spatialized prediction result data; The predicted data is converted into a raster data format to form the ecological resistance surface of the study area.
9. The method for identifying ecological corridors based on machine learning according to claim 1, characterized in that, The circuit theory-based ecological connectivity analysis model in S3 is also used to identify ecological pinch points. The specific steps for identifying ecological pinch points are as follows: Based on the ecological flow process simulated by the ecological connectivity analysis model, the ecological flow density of each spatial unit within the study area is calculated. Areas with ecological flow density exceeding a preset threshold are identified as ecological pinch points; The circuit theory-based ecological connectivity analysis model in S3 is also used to identify ecological barrier points. The specific steps for identifying ecological barrier points are as follows: Based on the ecological connectivity analysis model, the ecological flow improvement coefficient of each spatial unit in the study area is calculated; the ecological flow improvement coefficient represents the degree of improvement in ecological flow connectivity after removing the ecological resistance of the spatial unit. Areas with an ecological flow improvement coefficient higher than a preset threshold are identified as ecological barrier points.
10. An ecological corridor identification system based on machine learning, characterized in that, The system operates using a machine learning-based ecological corridor identification method as described in any one of claims 1-9. The system includes an ecological source area identification module, an ecological resistance surface construction module, and an ecological corridor identification module. The ecological source area identification module is used to extract and output the ecological source area based on the land use data of the study area. The ecological resistance surface construction module is used to acquire ecosystem service data and multi-source spatial impact factors, train a machine learning classification model based on the ecosystem service data and multi-source spatial impact factors, and use the trained model combined with multi-source spatial impact factors to predict the ecological resistance value of each spatial unit and output the ecological resistance surface. The ecological corridor identification module is used to input the ecological source area and the ecological resistance surface into an ecological connectivity analysis model based on circuit theory, simulate the ecological flow process, and identify the ecological corridors connecting the ecological source areas.