A data-driven evergreen broad-leaved forest vegetation zoning method and system

Abstract

The present invention discloses a data-driven evergreen broad-leaved forest vegetation zoning method and system, the method comprising: obtaining remote sensing data of a target area; dividing the target area into multiple vegetation regions based on the remote sensing data; performing kernel density analysis on the evergreen broad-leaved forest density in each vegetation region to generate a kernel density surface field, and dividing multiple vegetation groups based on the kernel density surface field and the constituent species composition characteristics of the corresponding region; dividing each vegetation region into multiple vegetation plots based on the vegetation clusters; and integrating the vegetation regions and vegetation plots into a final zoning map. Through the refined application of remote sensing data, the present invention can map the unique ecological conditions and vegetation characteristics of the target region based on the horizontal zonal characteristics of vegetation reflected by heat and moisture conditions; significantly improving the scientific nature and accuracy of vegetation zoning, and is an important supplement and improvement to traditional vegetation zoning methods.

Description

Technical Field

[0001] The present invention relates to the technical field of forest intelligent classification, and in particular to a data-driven evergreen broad-leaved forest vegetation zoning method and system. Background Art

[0002] Vegetation zoning, an important scientific basis for ecological management and protection, focuses on analyzing the coupling mechanisms between vegetation distribution and environmental factors. Subtropical evergreen broad-leaved forests are one of the world's most important forest vegetation types, playing an important role in carbon sequestration, oxygen release, carbon cycle regulation, and maintaining ecosystem stability. Subtropical evergreen broad-leaved forests are not only widely distributed and vast in area, but also, due to complex topography and high environmental heterogeneity, their vegetation communities exhibit significant differences in productivity, biomass, carbon storage, and species composition. These differences directly affect ecosystem function and its response to environmental change. Therefore, conducting high-precision vegetation zoning studies on subtropical evergreen broad-leaved forests is of great significance for the scientific management of forest resources and the formulation of ecological protection policies.

[0003] Traditional vegetation zoning primarily relies on classification methods based on climatic factors, such as the Holdridge life zone theory and the Köppen climate classification system. These methods, which use climate variables to classify vegetation types, have been widely used at the global scale. However, these methods have significant limitations in complex terrain, primarily manifested in the following aspects: First, they neglect local environmental factors: Climate-driven methods primarily focus on large-scale climate changes and fail to fully consider the influence of micro-scale environmental factors, such as topography and soils, on vegetation distribution. For example, factors such as slope, aspect, and soil nutrients play a key role in vegetation distribution in mountainous areas, yet traditional methods often overlook these details. Second, they suffer from insufficient spatial resolution: Vegetation zoning based on traditional climate models typically has low spatial resolution, making it difficult to distinguish ecological differences between different vegetation subtypes or communities. For example, the demarcation between western semi-humid evergreen broad-leaved forests and eastern humid evergreen broad-leaved forests relies primarily on macroclimatic data, which cannot accurately identify the variability of local ecological communities. Third, reliance on manual experience: Traditional vegetation zoning often relies on field surveys and expert knowledge, primarily distinguishing between sub-humid and moist evergreen broad-leaved forests based on topographical features and empirical judgment. This approach is effective when data is limited, but it still has significant limitations in terms of refined and automated zoning. Summary of the Invention

[0004] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the present application is to provide a data-driven evergreen broad-leaved forest vegetation zoning method and system.

[0005] In a first aspect, an embodiment of the present application provides a data-driven evergreen broad-leaved forest vegetation zoning method, comprising:

[0006] Acquire remote sensing data of the target area;

[0007] dividing the target area into a plurality of vegetation areas according to the remote sensing data;

[0008] Performing kernel density analysis on the evergreen broad-leaved forest density in each vegetation area to generate a kernel density surface field, and dividing a plurality of vegetation clusters according to the kernel density surface field and the composition characteristics of the dominant species in the corresponding area;

[0009] Dividing each of the vegetation areas into a plurality of vegetation cells according to the vegetation groups;

[0010] The vegetation areas and the vegetation plots are integrated into a final zoning map.

[0011] In a possible implementation, dividing the target area into a plurality of vegetation areas according to the remote sensing data includes:

[0012] Acquiring spectral features, temporal features, and texture features from the remote sensing data as classification features;

[0013] The vegetation in the target area is divided into a plurality of classification types according to the classification characteristics; the classification types include dry evergreen broad-leaved forest, moist evergreen broad-leaved forest and transitional evergreen broad-leaved forest;

[0014] Calculate the impact weight of each environmental indicator on the classification type based on the classification type and the corresponding environmental indicator;

[0015] The partition boundaries of the vegetation area are generated according to the impact weights.

[0016] In a possible implementation, obtaining spectral features, temporal features, and texture features from the remote sensing data as classification features includes:

[0017] The evergreen broad-leaved forest humidity index is constructed by the following formula, and the evergreen broad-leaved forest humidity index is used as one of the spectral features:

[0018] ;

[0019] Where EBFHI is the evergreen broad-leaved forest wetness index, NIR is the near-infrared reflectance, and SWIR1 is the shortwave infrared reflectance.

[0020] In one possible implementation, calculating the influence weight of each environmental indicator on the classification type according to the classification type and the corresponding environmental indicator includes:

[0021] The linear relationship between the environmental indicators was detected by Pearson correlation analysis, and the degree of multicollinearity between the environmental indicators was confirmed and quantified by VIF;

[0022] Screening selected indicators that meet preset standards from the environmental indicators based on the analysis results;

[0023] Training a machine learning model using the classification type and the corresponding selected indicators to generate a vegetation-environment relationship model; the machine learning model adopts at least one of a logistic regression model, a support vector machine, a gradient boosting tree, and a random forest model;

[0024] The weight of each selected indicator in the vegetation-environment relationship model is used as the first impact weight.

[0025] In a possible implementation, generating the partition boundary of the vegetation area according to the impact weight includes:

[0026] Performing weighted calculation on the environmental index using the first impact weight to obtain a zoning index for each grid in the target area, and forming a spatial distribution of the zoning index for the target area;

[0027] Determining a division threshold of the zoning indicator through a confidence interval, and generating an initial boundary according to the division threshold;

[0028] The resolution of the image of the target area including the initial boundary is upscaled and low-pass filtered to generate the final partition boundary.

[0029] In a possible implementation, the division of the vegetation groups includes:

[0030] Dividing the kernel density surface field into a plurality of density levels using a natural breakpoint method; the density levels include high-density areas and non-high-density areas;

[0031] The evergreen broad-leaved forests in the non-high-density area that meet preset conditions are separately divided into the first vegetation area; the preset conditions include the presence of artificially cultivated non-native evergreen broad-leaved forests and climate change caused by topographical reasons;

[0032] The high-density area is spatially classified according to the terrain characteristics of the high-density area and the corresponding community species composition characteristics to form a plurality of the vegetation clusters.

[0033] In a possible implementation, dividing each vegetation area into a plurality of vegetation cells according to the vegetation cluster includes:

[0034] Calculating the weight of each selected indicator in all vegetation groups as the second impact weight through the vegetation-environment relationship model;

[0035] Performing weighted calculation on the environmental index using the second impact weight to obtain a zoning index for each grid of the second vegetation cell in the target area, and forming a spatial distribution of the zoning index for the target area;

[0036] Determining a division threshold of a zoning indicator of the second vegetation cell through a confidence interval, and generating an initial boundary of the second vegetation cell according to the division threshold;

[0037] Upscaling the resolution of the image of the target area including the initial boundary and performing low-pass filtering to generate a final partition boundary of the second vegetation cell;

[0038] The first vegetation area and the second vegetation area are used as the divided vegetation areas.

[0039] In a possible implementation, integrating the vegetation areas and the vegetation cells into a final zoning map includes:

[0040] The finally divided vegetation areas and vegetation plots are presented in the same image to form the final zoning map.

[0041] In a second aspect, the present application also provides a data-driven evergreen broad-leaved forest vegetation zoning system, including:

[0042] an acquisition unit configured to acquire remote sensing data of a target area;

[0043] a first dividing unit configured to divide the target area into a plurality of vegetation areas according to the remote sensing data;

[0044] The clustering unit is configured to perform kernel density analysis on the density of evergreen broad-leaved forests in each vegetation area to generate a kernel density surface field, and divide the vegetation clusters according to the kernel density surface field and the composition characteristics of the group-forming species in the corresponding area;

[0045] a second dividing unit configured to divide each of the vegetation areas into a plurality of vegetation cells according to the vegetation groups;

[0046] The integration unit is configured to integrate the vegetation areas and the vegetation cells into a final zoning map.

[0047] In a possible implementation, the first dividing unit is further configured to:

[0048] Acquiring spectral features, temporal features, and texture features from the remote sensing data as classification features;

[0049] The vegetation in the target area is divided into a plurality of classification types according to the classification characteristics; the classification types include dry evergreen broad-leaved forest, moist evergreen broad-leaved forest and transitional evergreen broad-leaved forest;

[0050] Calculate the impact weight of each environmental indicator on the classification type based on the classification type and the corresponding environmental indicator;

[0051] The partition boundaries of the vegetation area are generated according to the impact weights.

[0052] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0053] Through the refined application of remote sensing data, the present invention achieves accurate zoning of evergreen broad-leaved forests at two scales. It can not only reflect the horizontal zonal characteristics of vegetation based on heat and moisture conditions, but also map the unique ecological conditions and vegetation characteristics of the target area. It provides an effective technical solution for vegetation zoning in areas with complex terrain and significant regional differences in climate and vegetation distribution. It significantly improves the scientific nature and accuracy of vegetation zoning, and is an important supplement and improvement to traditional vegetation zoning methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0055] Figure 1 This is a schematic diagram of the steps of the method according to the embodiment of the present application;

[0056] Figure 2 This is a schematic diagram of the classification results obtained by the random forest model in an embodiment of the present application;

[0057] Figure 3 The spectral curves of the dry evergreen broad-leaved forest and the moist evergreen broad-leaved forest in the embodiment of this application are as follows;

[0058] Figure 4 This is a violin plot of the distribution of zoning indicators for the dry evergreen broad-leaved forest and the moist evergreen broad-leaved forest in the embodiment of the present application;

[0059] Figure 5 This is a comparison of the data before and after upscaling and filtering in the embodiment of this application;

[0060] Figure 6 This is a map of the evergreen broad-leaved forest vegetation area in the embodiment of this application;

[0061] Figure 7 This is a kernel density analysis result diagram of the embodiment of this application. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of illustration and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate the operations implemented according to some embodiments of the embodiments of the present application. It should be understood that the operations of the flowcharts can be implemented out of sequence, and steps without logical context can be reversed or implemented simultaneously. In addition, those skilled in the art, under the guidance of the contents of this application, can add one or more other operations to the flowchart, or remove one or more operations from the flowchart.

[0063] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0064] Please refer to Figure 1 , which is a flow chart of a data-driven evergreen broad-leaved forest vegetation zoning method provided in an embodiment of the present invention. Furthermore, the data-driven evergreen broad-leaved forest vegetation zoning method may specifically include the contents described in the following steps S1 to S5.

[0065] S1: Acquire remote sensing data of the target area;

[0066] S2: Dividing the target area into a plurality of vegetation areas according to the remote sensing data;

[0067] S3: performing kernel density analysis on the evergreen broad-leaved forest density in each vegetation area to generate a kernel density surface field, and dividing a plurality of vegetation clusters according to the kernel density surface field and the composition characteristics of the dominant species in the corresponding area;

[0068] S4: Dividing each vegetation area into a plurality of vegetation cells according to the vegetation groups;

[0069] S5: Integrate the vegetation areas and the vegetation plots into a final zoning map.

[0070] When implementing the embodiments of this application, a two-level classification approach was first adopted, with the first level being vegetation regions and the second being vegetation plots. Classification using remote sensing data allows the target area to be divided into multiple vegetation regions, primarily based on the horizontal zonation of evergreen broad-leaved forests within the region, which is determined by differences in latitude, longitude, and altitude. Vegetation plot classification, performed after the vegetation clusters have been divided, serves as a more detailed unit within the vegetation region classification and focuses more on mesoscale terrain characteristics and specific combinations of dominant species. Mesoscale terrain refers to the various terrain types found within a specific area, such as mountains, hills, and plains. These terrains, due to their unique topographic conditions (e.g., altitude, aspect, and slope), significantly influence local microclimate conditions, thereby determining the distribution and composition of vegetation types. Secondly, the composition of different dominant species forms a key basis for the classification of vegetation plots. A dominant species combination refers to a collection of plants with a specific structure and composition that forms due to similar ecological conditions within a given geographic area. Different combinations of dominant species reflect the unique ecological conditions and vegetation characteristics of the area. Through this classification method, it is possible to accurately zonate evergreen broad-leaved forests at two scales. It can not only reflect the horizontal zonal characteristics of vegetation based on heat and water conditions, but also map the unique ecological conditions and vegetation characteristics of the target area.

[0071] In a possible implementation, dividing the target area into a plurality of vegetation areas according to the remote sensing data includes:

[0072] Acquiring spectral features, temporal features, and texture features from the remote sensing data as classification features;

[0073] The vegetation in the target area is divided into a plurality of classification types according to the classification characteristics; the classification types include dry evergreen broad-leaved forest, moist evergreen broad-leaved forest and transitional evergreen broad-leaved forest;

[0074] Calculate the impact weight of each environmental indicator on the classification type based on the classification type and the corresponding environmental indicator;

[0075] The partition boundaries of the vegetation area are generated according to the impact weights.

[0076] When the embodiment of the present application is implemented, a specific vegetation area division scheme is provided, in which spectral features, temporal features and texture features can be used as classification features for classification. The specific classification can be carried out using SVM, random forest and gradient boosting tree in the existing technology, and the embodiment of the present application does not limit it. The final classification result is to divide the vegetation into dry evergreen broad-leaved forest, wet evergreen broad-leaved forest and transitional evergreen broad-leaved forest. The classification process is a large-scale classification based on classification features, which is based on the horizontal zonal characteristics of vegetation reflected by heat and moisture conditions. For evergreen broad-leaved forests of different vegetation types, the characteristics they show under multiple indicators are also different, so performing large-scale classification first can effectively improve the accuracy of subsequent subdivision classification, that is, improve the accuracy of subsequent vegetation cell division. For examples, please refer to Figure 2 , showing the classification results obtained by the random forest model. After obtaining the classification results, the environmental indicators corresponding to each classification result can be obtained. Generally, there are many types of environmental indicators used, which can be processed and filtered out to obtain the most suitable indicators. For example, the filtered indicators include the minimum sunshine hours in the dry season, the average precipitation value in the dry season, the evapotranspiration in the dry season, the temperature in the dry season, the temperature in the wet season, the slope, the aspect, the soil pH, the soil moisture content, the soil type, the soil texture grade, the soil organic carbon content and the elevation. Through calculation and analysis, the influence weight of each indicator on the classification type can be obtained, and then the zoning boundary of the vegetation area can be determined based on the weight.

[0077] In a possible implementation, obtaining spectral features, temporal features, and texture features from the remote sensing data as classification features includes:

[0078] The evergreen broad-leaved forest humidity index is constructed by the following formula, and the evergreen broad-leaved forest humidity index is used as one of the spectral features:

[0079]

[0080] Where EBFHI is the evergreen broad-leaved forest wetness index, NIR is the near-infrared reflectance, and SWIR1 is the shortwave infrared reflectance.

[0081] When implementing the embodiment of this application, an innovative key feature, the evergreen broad-leaved forest wetness index, is introduced as a classification parameter. The main reason for this is that compared with the dry evergreen broad-leaved forest, the wet evergreen broad-leaved forest has a higher reflectance value in the near infrared (NIR) and shortwave infrared (SWIR1) bands. Figure 3, showing the spectral curves of dry evergreen broad-leaved forest and moist evergreen broad-leaved forest; the vertical axis in the figure is reflectance, the horizontal axis is spectral band, the colored areas identify the spectral distribution of different evergreen broad-leaved forests, the blue line is the boundary curve of dry evergreen broad-leaved forest, and the red line is the boundary curve of moist evergreen broad-leaved forest; this difference indicates that evergreen broad-leaved forest may have higher vegetation water content or denser vegetation structure in the rainy season, resulting in stronger reflectivity in these spectral bands; the evergreen broad-leaved forest humidity index is used to comprehensively characterize the difference in these two bands, which can enhance the spectral contrast between the two forest subtypes and effectively capture the different characteristics within these spectral ranges. Based on this feature, the dry evergreen broad-leaved forest and the moist evergreen broad-leaved forest can be effectively classified and distinguished.

[0082] For example, when the vegetation in the target area is divided into multiple classification types by classification features, a pre-trained classification model is used to implement the classification; the input data of the classification model is the classification feature, and the output data is the classification type after classification; the classification model can be obtained by training SVM, random forest and gradient boosting tree. In specific applications, the spectral features in the classification features can include the blue band, red band, green band, near infrared band, shortwave infrared band 1 and shortwave infrared band 2 of Landsat8OLI, as well as the evergreen broad-leaved forest humidity index, normalized vegetation index, ratio vegetation index, normalized infrared index, water stress index, brightness, greenness and yellowness of vegetation expressed by tasseled cap transformation, and other data; and the texture features can be angular second moment (AngularSecondMomentASM), contrast (Contrast), correlation (Correlation), variance (V The classification features include variability, homogeneity, sumentropy, entropy, dissimilarity, mean, cluster prominence (CP), and cluster skewness (ClusterShadeCS). Temporal features include the difference between the dry and wet seasons of the ratio vegetation index, normalized infrared index, and water stress index. Based on this, a classification model can be trained, and the corresponding classification types output during classification model training can be obtained through field surveys. For example, by obtaining classification features in a sample area and constructing different classification types corresponding to the sample area using field survey data and an existing vegetation classification system, a sample library can be formed through the corresponding relationship. Using the classification features in the sample library as input data and the classification types in the sample library as output data, a classifier can be trained to ultimately form a classification model. The classifier can be at least one of an SVM, a random forest model, and a gradient descent model.

[0083] In one possible implementation, calculating the influence weight of each environmental indicator on the classification type according to the classification type and the corresponding environmental indicator includes:

[0084] The linear relationship between the environmental indicators was detected by Pearson correlation analysis, and the degree of multicollinearity between the environmental indicators was confirmed and quantified by VIF;

[0085] Screening selected indicators that meet preset standards from the environmental indicators based on the analysis results;

[0086] Training a machine learning model using the classification type and the corresponding selected indicators to generate a vegetation-environment relationship model; the machine learning model adopts at least one of a logistic regression model, a support vector machine, a gradient boosting tree, and a random forest model;

[0087] The weight of each selected indicator in the vegetation-environment relationship model is used as the first impact weight.

[0088] During the implementation of the present application, due to the large number of actual environmental indicators, many of which may be highly correlated, multicollinearity may occur during model analysis, leading to deviations in model estimation and affecting the accuracy of statistical classification. Therefore, in the present application, Pearson correlation analysis and VIF (variance inflation factor) can be used to effectively identify highly correlated indicators and optimize them. The higher the VIF value, the greater the impact of the collinearity of other indicators on the indicator. Generally speaking, a VIF value exceeding 10 indicates a serious multicollinearity problem. For example, precipitation and sunshine hours among the climate factors are subdivided into maximum, mean, and minimum values ​​for the dry season, and maximum, mean, and minimum values ​​for the wet season. Given the high correlation between the maximum, minimum, and mean values ​​within each season, and to avoid multicollinearity, the variables with the highest correlation coefficients within each season, precipitation and sunshine duration, were selected for further analysis. Wet-season precipitation and sunshine duration showed only moderate correlations with the spatial distribution of both dry and wet evergreen broad-leaved forests and were therefore excluded. Surface temperature and evaporation showed strong correlations with the spatial distribution of both dry and wet evergreen broad-leaved forests and were therefore included in the model. The soil factor also contained data from six different depths, and it is clear that data from different depths of the same soil type are highly correlated. Therefore, only the soil depth with the highest correlation with vegetation distribution was selected for analysis. Given the potential influence of topography on vegetation distribution, all topographical indicators were included in the correlation analysis. The selected indicators were finally selected as minimum sunshine hours in the dry season, average precipitation in the dry season, evapotranspiration in the dry season, temperature in the dry season, temperature in the wet season, slope, aspect, soil pH, soil moisture content, soil type, soil texture grade, soil organic carbon content and elevation. As can be seen from the table below, all the selected indicators can pass the multiple commonality test and can be used for subsequent analysis.

[0089] Table 1 Calculation results of variance inflation factors between environmental indicators

[0090]

[0091] When implementing the embodiment of the present application, after screening out suitable selected indicators, it is necessary to calculate the influence weights of these selected indicators on the classification results through a machine learning model; in the embodiment of the present application, at least one of the logistic regression model, support vector machine, gradient boosting tree and random forest model is preferably used for weight analysis; the core problem of evergreen broad-leaved forest vegetation zoning is how to transform discrete and discontinuous vegetation classification data into continuous and clear vegetation zoning in combination with its environmental factors. The embodiment of the present application adopts a data-driven approach, the core of which is to construct a vegetation environment association model, which extracts environmental factors that have a significant impact on vegetation distribution from multidimensional data through statistics and machine learning techniques, and constructs a linear and comprehensive vegetation environment association model based on the importance value obtained by the model and the indicator. This model not only considers a single environmental factor alone, but also integrates multiple factors, such as climate, topography, soil, etc., as well as the complex interactions that may exist between them.

[0092] For example, when performing weight analysis through a logistic regression model, it specifically includes:

[0093] Let the probability of dry evergreen broad-leaved forest be P, and the probability of wet evergreen broad-leaved forest be 1-P, and the function expression of logistic regression is:

[0094]

[0095]

[0096] Where, P represents the probability of dry evergreen broad-leaved forest; Z represents the sum of weights obtained after superimposing independent variables; Is a constant, which is obtained by logistic regression through training samples; is the regression coefficient obtained by logistic regression operation of the training sample; Represents the various factors affecting dry and wet evergreen broad-leaved forests, which are the independent variables of the logistic regression model;

[0097] Taking the natural logarithm of the above formula gives:

[0098]

[0099] The calculated results P i Areas with a value greater than 0.5 are considered dry evergreen broad-leaved forests, while areas with a value less than 0.5 are considered moist evergreen broad-leaved forests. The distinction is made based on the following formula:

[0100]

[0101] In the process of applying the logistic regression model to determine the index weights of evergreen broad-leaved forest vegetation zoning, the results of the calculation areB The value is the regression coefficient, which can represent the importance of each indicator on the spatial distribution of dry evergreen broad-leaved forest and moist evergreen broad-leaved forest. B The larger the value, the greater the contribution of this evaluation factor to its environmental factors.

[0102] For example, when conducting weight analysis, it specifically includes:

[0103] The spatial distribution of dry evergreen broad-leaved forest and moist evergreen broad-leaved forest was randomly divided into training set and validation set in a ratio of 7:3, and the classification label of dry evergreen broad-leaved forest was set to 1, and the classification label of moist evergreen broad-leaved forest was set to 0.

[0104] Based on the above content, when further calculations are performed using the logistic regression model, the C value range of the logistic regression model is (0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000), and the grid search method combined with 5-fold cross validation is used for parameter optimization. The final optimal parameter C is 0.1.

[0105] Based on the above content, when further calculation is performed through the support vector machine, the radial basis function is used as the kernel function, and the penalty coefficient C ranges from (2 -5 , 2 5 ), the value range of γ is (2 -5 , 2 5 ), and the grid search method combined with 5-fold cross validation was used for parameter optimization. The final optimal parameters C were 1 and γ were 2;

[0106] Based on the above, further calculations were performed using the random forest model and gradient boosted tree model. The number of trees was set to a range of (100, 200), the learning rate was set to (0.005, 0.001, 0.1, 1), the maximum tree depth was set to (3, 5), and the minimum number of samples required for an internal node to be repartitioned was 10. Parameter optimization using grid search combined with 5-fold cross-validation revealed that the optimal parameters for the gradient boosted tree model were 100 trees, a learning rate of 0.1, and a maximum tree depth of 5. The optimal parameters for the random forest model were 200 trees, a learning rate of 0.1, and a maximum tree depth of 5.

[0107] The first impact weights of the selected indicators based on the above are shown in the following table:

[0108] Table 2 The first impact weight obtained from the vegetation-environment relationship model

[0109]

[0110] The performance of each model based on the above content is shown in the following table:

[0111] Table 3 Classification accuracy results of different models

[0112]

[0113] In a possible implementation, generating the partition boundary of the vegetation area according to the impact weight includes:

[0114] Performing weighted calculation on the environmental index using the first impact weight to obtain a zoning index for each grid in the target area, and forming a spatial distribution of the zoning index for the target area;

[0115] Determining a division threshold of the zoning indicator through a confidence interval, and generating an initial boundary according to the division threshold;

[0116] The resolution of the image of the target area including the initial boundary is upscaled and low-pass filtered to generate the final partition boundary.

[0117] When implementing the embodiment of the present application, the zoning index of each grid in the target area can be calculated by using the first impact weight that has been obtained to achieve the assignment of values ​​to each grid; after the replication is completed, all grids in the target area will express the spatial distribution of the zoning index; in the work of vegetation zoning, scientifically determining the zoning index threshold is a key step to ensure the accuracy and reliability of the zoning results. The embodiment of the present application adopts a confidence interval-based method to determine the zoning index threshold, aiming to ensure that the statistical significance of the zoning results is consistent with ecological reality. Please refer to Figure 4 , showing violin plots of the distribution of regional indicators for dry and wet evergreen broad-leaved forests, where **** indicates a highly statistically significant difference between the two data sets. Combined with the analysis of the elevational distribution of dry and wet evergreen broad-leaved forests, the boundary between dry and wet evergreen broad-leaved forest areas was found to be most consistent with the 95% confidence level. Based on this, specific numerical ranges were selected to define different regions: the range of [0-0.4447] was defined as the wet evergreen broad-leaved forest area, the range of [0.5034-1] as the dry evergreen broad-leaved forest area, and the range of [0.4447-0.5034] as the transition zone between dry and wet evergreen broad-leaved forests.

[0118] In the implementation of the present application, after initial boundaries are generated by dividing thresholds, although determining the thresholds for zoning indicators using confidence intervals can statistically ensure the accuracy of vegetation zoning results, for small-scale resolutions, this often results in a large number of scattered enclaves, or isolated groups of pixels, in the vegetation zoning results. Therefore, to achieve a more spatially coherent vegetation zoning map, it is necessary to further optimize the method for determining zoning boundaries. Upscaling and low-pass filtering are introduced here. By analyzing images at different scales, the upscaling method can grasp the overall trends of vegetation zoning at a macro level and smooth local texture details at a micro level, effectively reducing the appearance of enclaves and obtaining more continuous and consistent zoning boundaries at a large scale. The core of this method is to identify the sensitive scale of zoning by adjusting the observation scale and reveal the main patterns of ecological distribution at this scale. The vegetation zoning results are then smoothed using low-pass filtering to remove high-frequency information introduced by factors such as object classification, remote sensing image resolution, and noise, while retaining low-frequency components reflecting the main ecological gradients and trends. This step is to eliminate the interference of high-frequency noise and make the demarcation lines clearer and more reasonable. For an example, please refer to Figure 5 Due to the interlaced distribution of humid evergreen broad-leaved forests and dry evergreen broad-leaved forests in the transitional area, it was difficult to determine the boundary for demarcation at a spatial resolution of 30 m. Therefore, the resolution of the original image was upscaled to 300 m, and the boundary between the humid evergreen broad-leaved forest and the dry evergreen broad-leaved forest was initially clear. Considering that there were still a few interfering independent grid patches in the image, low-pass filtering was used to effectively filter out these independent grid patches, making the boundary clearer. Finally, a map of the evergreen broad-leaved forest vegetation area was obtained. See the details for details. Figure 6 .

[0119] In a possible implementation, the division of the vegetation groups includes:

[0120] Dividing the kernel density surface field into a plurality of density levels using a natural breakpoint method; the density levels include high-density areas and non-high-density areas;

[0121] The evergreen broad-leaved forests in the non-high-density area that meet preset conditions are separately divided into the first vegetation area; the preset conditions include the presence of artificially cultivated non-native evergreen broad-leaved forests and climate change caused by topographical reasons;

[0122] The high-density area is spatially classified according to the terrain characteristics of the high-density area and the corresponding community species composition characteristics to form a plurality of the vegetation clusters.

[0123] When the embodiment of the present application is implemented, in order to accurately further subdivide the vegetation area, the kernel density surface field is divided into multiple density levels by kernel density analysis. It should be understood that different vegetation areas need to be divided separately. For example, for dry evergreen broad-leaved forests and moist evergreen broad-leaved forests, they can be divided into five levels: low-density area, medium-low-density area, medium-density area, medium-high-density area and high-density area; among them, low-density area, medium-low-density area, medium-density area and medium-high-density area are all non-high-density areas; and for transitional evergreen broad-leaved forests, they can be divided into low-density area, medium-density area and high-density area; among them, low-density area and medium-density area are non-high-density areas. The reason is that the transition area is long and narrow, and the environmental gradient along the length of the area varies significantly while the width direction varies less. This division method more effectively captures the environmental variation along the main gradient direction and avoids the increase in intra-class differences that may be caused by excessive classification. The kernel density analysis results after the division are shown in FIG. Figure 7 .

[0124] In the embodiments of the present application, special reasons that may exist in different regions may cause changes in the distribution of evergreen broad-leaved forests, so these areas need to be divided separately; the reasons may be due to human activities or changes in terrain; for example, a certain area as a whole belongs to the subtropical evergreen broad-leaved forest area, and the human interference caused by human activities at the bottom of the basin has led to a reduction in the distribution of evergreen broad-leaved forests, forming medium-low or low-density clusters. In view of the widespread existence of artificially cultivated non-native evergreen broad-leaved forests in the area, the medium-low clusters and low clusters shown in the kernel density analysis are defined as the first vegetation community at the bottom of the basin. Similarly, due to the cooling of the climate caused by the uplift of the plateau terrain in a certain area, there has also been a reduction in the distribution of evergreen broad-leaved forests, forming medium-low or low-density clusters. Subalpine coniferous forests, shrubs and meadows are widely distributed in the area, and the medium-low clusters and low clusters in the kernel density analysis are classified as another first vegetation community. Combining the topographic features of high-density areas from the kernel density analysis results with the composition characteristics of dominant species recorded in field surveys, high-density areas were spatially classified into vegetation clusters for subsequent vegetation plot division. For example, the vegetation cluster division is shown in the table below:

[0125] Table 4 Vegetation group division table

[0126]

[0127]

[0128]

[0129] In a possible implementation, dividing each vegetation area into a plurality of vegetation cells according to the vegetation cluster includes:

[0130] Calculating the weight of each selected indicator in all vegetation groups as the second impact weight through the vegetation-environment relationship model;

[0131] Performing weighted calculation on the environmental index using the second impact weight to obtain a zoning index for each grid of the second vegetation cell in the target area, and forming a spatial distribution of the zoning index for the target area;

[0132] Determining a division threshold of a zoning indicator of the second vegetation cell through a confidence interval, and generating an initial boundary of the second vegetation cell according to the division threshold;

[0133] Upscaling the resolution of the image of the target area including the initial boundary and performing low-pass filtering to generate a final partition boundary of the second vegetation cell;

[0134] The first vegetation area and the second vegetation area are used as the divided vegetation areas.

[0135] When implementing the embodiment of the present application, the vegetation-environment relationship model used in the above embodiment can be used to calculate the second impact weight and the corresponding zoning index; at the same time, the subsequent initial boundary and partition boundary generation process is also the same as the above embodiment, except that the corresponding impact weights are different.

[0136] In a possible implementation, integrating the vegetation areas and the vegetation cells into a final zoning map includes:

[0137] The finally divided vegetation areas and vegetation plots are presented in the same image to form the final zoning map.

[0138] In a second aspect, the present application also provides a data-driven evergreen broad-leaved forest vegetation zoning system, including:

[0139] an acquisition unit configured to acquire remote sensing data of a target area;

[0140] a first dividing unit configured to divide the target area into a plurality of vegetation areas according to the remote sensing data;

[0141] The clustering unit is configured to perform kernel density analysis on the density of evergreen broad-leaved forests in each vegetation area to generate a kernel density surface field, and divide the vegetation clusters according to the kernel density surface field and the composition characteristics of the group-forming species in the corresponding area;

[0142] a second dividing unit configured to divide each of the vegetation areas into a plurality of vegetation cells according to the vegetation groups;

[0143] The integration unit is configured to integrate the vegetation areas and the vegetation cells into a final zoning map.

[0144] In a possible implementation, the first dividing unit is further configured to:

[0145] Acquiring spectral features, temporal features, and texture features from the remote sensing data as classification features;

[0146] The vegetation in the target area is divided into a plurality of classification types according to the classification characteristics; the classification types include dry evergreen broad-leaved forest, moist evergreen broad-leaved forest and transitional evergreen broad-leaved forest;

[0147] Calculate the impact weight of each environmental indicator on the classification type based on the classification type and the corresponding environmental indicator;

[0148] The partition boundaries of the vegetation area are generated according to the impact weights.

[0149] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0150] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or can be electrical, mechanical or other forms of connection.

[0151] The units described as separate components may or may not be physically separated. As units, it is obvious that a person of ordinary skill in the art can realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0152] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0153] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or grid device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0154] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A data-driven evergreen broad-leaved forest vegetation zoning method, characterized in that: include: Acquire remote sensing data of the target area; dividing the target area into a plurality of vegetation areas according to the remote sensing data; Performing kernel density analysis on the evergreen broad-leaved forest density in each vegetation area to generate a kernel density surface field, and dividing a plurality of vegetation clusters according to the kernel density surface field and the composition characteristics of the dominant species in the corresponding area; Dividing each of the vegetation areas into a plurality of vegetation cells according to the vegetation groups; The vegetation areas and the vegetation plots are integrated into a final zoning map.

2. The data-driven evergreen broad-leaved forest vegetation zoning method according to claim 1, characterized in that: Dividing the target area into a plurality of vegetation areas according to the remote sensing data includes: Acquiring spectral features, temporal features, and texture features from the remote sensing data as classification features; The vegetation in the target area is divided into a plurality of classification types according to the classification characteristics; the classification types include dry evergreen broad-leaved forest, moist evergreen broad-leaved forest and transitional evergreen broad-leaved forest; Calculate the impact weight of each environmental indicator on the classification type based on the classification type and the corresponding environmental indicator; The partition boundaries of the vegetation area are generated according to the impact weights.

3. The data-driven evergreen broad-leaved forest vegetation zoning method according to claim 2, characterized in that: Acquiring spectral features, temporal features, and texture features from the remote sensing data as classification features includes: The evergreen broad-leaved forest humidity index is constructed by the following formula, and the evergreen broad-leaved forest humidity index is used as one of the spectral features: ； Where EBFHI is the evergreen broad-leaved forest wetness index, NIR is the near-infrared reflectance, and SWIR1 is the shortwave infrared reflectance.

4. The data-driven evergreen broad-leaved forest vegetation zoning method according to claim 2, characterized in that: The influence weight of each environmental indicator on the classification type is calculated based on the classification type and the corresponding environmental indicator, including: The linear relationship between the environmental indicators was detected by Pearson correlation analysis, and the degree of multicollinearity between the environmental indicators was confirmed and quantified by VIF; Screening selected indicators that meet preset standards from the environmental indicators based on the analysis results; Training a machine learning model using the classification type and the corresponding selected indicators to generate a vegetation-environment relationship model; the machine learning model adopts at least one of a logistic regression model, a support vector machine, a gradient boosting tree, and a random forest model; The weight of each selected indicator in the vegetation-environment relationship model is used as the first impact weight.

5. The data-driven evergreen broad-leaved forest vegetation zoning method according to claim 4, characterized in that: Generating the partition boundary of the vegetation area according to the impact weight includes: Performing weighted calculation on the environmental index using the first impact weight to obtain a zoning index for each grid in the target area, and forming a spatial distribution of the zoning index for the target area; Determining a division threshold of the zoning indicator through a confidence interval, and generating an initial boundary according to the division threshold; The resolution of the image of the target area including the initial boundary is upscaled and low-pass filtered to generate the final partition boundary.

6. The data-driven evergreen broad-leaved forest vegetation zoning method according to claim 4, characterized in that: The division of vegetation groups includes: Dividing the kernel density surface field into a plurality of density levels using a natural breakpoint method; the density levels include high-density areas and non-high-density areas; The evergreen broad-leaved forests in the non-high-density area that meet preset conditions are separately divided into the first vegetation area; the preset conditions include the presence of artificially cultivated non-native evergreen broad-leaved forests and climate change caused by topographical reasons; The high-density area is spatially classified according to the terrain characteristics of the high-density area and the corresponding community species composition characteristics to form a plurality of the vegetation clusters.

7. The data-driven evergreen broad-leaved forest vegetation zoning method according to claim 6, characterized in that: Dividing each vegetation area into a plurality of vegetation cells according to the vegetation groups includes: Calculating the weight of each selected indicator in all vegetation groups as the second impact weight through the vegetation-environment relationship model; Performing weighted calculation on the environmental index using the second impact weight to obtain a zoning index for each grid of the second vegetation cell in the target area, and forming a spatial distribution of the zoning index for the target area; Determining a division threshold of a zoning indicator of the second vegetation cell through a confidence interval, and generating an initial boundary of the second vegetation cell according to the division threshold; Upscaling the resolution of the image of the target area including the initial boundary and performing low-pass filtering to generate a final partition boundary of the second vegetation cell; The first vegetation area and the second vegetation area are used as the divided vegetation areas.

8. The data-driven evergreen broad-leaved forest vegetation zoning method according to claim 7, characterized in that: Integrating the vegetation areas and the vegetation plots into a final zoning map includes: The finally divided vegetation areas and vegetation plots are presented in the same image to form the final zoning map.

9. A data-driven evergreen broad-leaved forest vegetation zoning system using the method according to any one of claims 1 to 8, characterized in that: include: an acquisition unit configured to acquire remote sensing data of a target area; a first dividing unit configured to divide the target area into a plurality of vegetation areas according to the remote sensing data; The clustering unit is configured to perform kernel density analysis on the density of evergreen broad-leaved forests in each vegetation area to generate a kernel density surface field, and divide the vegetation clusters according to the kernel density surface field and the composition characteristics of the group-forming species in the corresponding area; a second dividing unit configured to divide each of the vegetation areas into a plurality of vegetation cells according to the vegetation groups; The integration unit is configured to integrate the vegetation areas and the vegetation cells into a final zoning map.

10. The data-driven evergreen broad-leaved forest vegetation zoning system according to claim 9, characterized in that: The first dividing unit is further configured to: Acquiring spectral features, temporal features, and texture features from the remote sensing data as classification features; The vegetation in the target area is divided into a plurality of classification types according to the classification characteristics; the classification types include dry evergreen broad-leaved forest, moist evergreen broad-leaved forest and transitional evergreen broad-leaved forest; Calculate the impact weight of each environmental indicator on the classification type based on the classification type and the corresponding environmental indicator; The partition boundaries of the vegetation area are generated according to the impact weights.

Images