A mangrove forest ecological restoration dynamic evaluation method and system based on multi-source data fusion

Abstract

The application provides a mangrove ecological restoration dynamic evaluation method and system based on multi-source data fusion, acquires multi-source time series data of an area to be evaluated; determines time change intensity based on the first derivative absolute value of each time node, and weights the original data sequence to obtain weighted time series data; calculates the trend slope of each spatial unit to generate an initial trend feature map; groups the initial trend feature map according to ecological attributes, extracts a first principal component map, and takes the variance contribution rate as a trend correlation intensity parameter; determines the spatial correction kernel size according to the intensity parameter, and processes to obtain a corrected domain principal component map; performs a second principal component analysis to obtain a global principal component, a score and an eigenvalue; normalizes the eigenvalue to obtain a first weight coefficient set; sets an ecological warning threshold, and when the score exceeds the threshold, the corresponding weight is increased to form a second weight coefficient set; and weights and sums the global principal component score to calculate a comprehensive trend index of mangrove ecological restoration.

Description

Technical Field

[0001] This application belongs to the field of dynamic assessment, and in particular relates to a dynamic assessment method and system for mangrove ecological restoration based on multi-source data fusion. Background Technology

[0002] Mangrove forests have suffered a sharp decline in global area and ecosystem degradation due to human activities and global climate change. Assessments of mangrove ecological restoration rely on comprehensive analysis of multi-source data. Using weighted indexes or linear superposition methods struggles to handle relationships and information redundancy among multi-source data, easily leading to biased assessment results or information loss. Trend analysis methods ignore drastic changes occurring in ecosystems at certain times, failing to fully reflect the decisive impact of high-intensity change events on the overall trend, thus requiring improved assessment sensitivity. While principal component analysis (PCA) is used to handle multidimensional data, treating all data sources as a whole results in principal components that are mixtures of multiple unrelated ecological factors. Furthermore, existing methods rarely consider the strength of correlations within different groups of ecological factors and fail to integrate these correlations with spatial pattern changes, potentially overlooking the spatial heterogeneity of dominant ecological processes in the assessment. When warning signals of certain key ecological indicators exceeding normal fluctuation ranges are detected, existing models cannot adjust their weights in the comprehensive assessment, potentially delaying the identification and response to potential ecological risks. Therefore, there is an urgent need for an assessment method that can deeply integrate spatiotemporal changes, represent the relationships between factors, and combine early warning mechanisms to reveal the true trend of mangrove ecological restoration. Summary of the Invention

[0003] This invention proposes a dynamic assessment method for mangrove ecological restoration based on multi-source data fusion, addressing the problem that existing methods may overlook the spatial heterogeneity of dominant ecological processes, potentially delaying the identification and response to potential ecological risks. The method includes:

[0004] Acquire multi-source time series data of the region to be evaluated; for each data series, determine the intensity of time change based on the absolute value of the first derivative of the series at each time node, and use this as a weight to weight the original data series to obtain weighted time series data; calculate the trend slope of each spatial unit based on the weighted time series data to generate the initial trend feature map of each data source;

[0005] The initial trend feature maps are grouped according to ecological attributes. A first principal component analysis is performed on each group to extract the first principal component map. The variance contribution rate is used as the trend correlation strength parameter for each group. Based on the trend correlation strength parameter of each group, the spatial correction kernel size is determined for the corresponding first principal component map. The spatial correction kernel is then used for convolution processing to obtain the correction domain principal component map.

[0006] All principal component graphs of the modified domain are combined and a second principal component analysis is performed to obtain multiple global principal components, scores, and eigenvalues. The eigenvalues ​​of the second principal component analysis are normalized to obtain a first set of weight coefficients. An ecological warning threshold is set for the scores of each global principal component. When the score of any global principal component exceeds the threshold, the corresponding weight of the global principal component in the first set of weight coefficients is increased to form a second set of weight coefficients.

[0007] Furthermore, this invention also proposes a dynamic assessment system for mangrove ecological restoration based on multi-source data fusion, comprising the following modules:

[0008] The generation module is used to acquire multi-source time series data of the region to be evaluated; for each data series, the intensity of time change is determined based on the absolute value of the first derivative of the series at each time node, and the original data series is weighted using this as a weight to obtain weighted time series data; the trend slope of each spatial unit is calculated based on the weighted time series data to generate the initial trend feature map of each data source;

[0009] The correction module is used to group the initial trend feature map according to ecological attributes, perform a first principal component analysis on each group, extract the first principal component map, and use the variance contribution rate as the trend correlation strength parameter of each group; according to the trend correlation strength parameter of each group, determine the spatial correction kernel size for the corresponding first principal component map, and use the spatial correction kernel to perform convolution processing to obtain the correction domain principal component map.

[0010] The enhancement module is used to combine all the principal component graphs of the correction domain and perform a second principal component analysis to obtain multiple global principal components, scores, and eigenvalues; normalize the eigenvalues ​​of the second principal component analysis to obtain a first set of weight coefficients; set an ecological warning threshold for the scores of each global principal component, and when the score of any global principal component exceeds the threshold, increase the corresponding weight of the global principal component in the first set of weight coefficients to form a second set of weight coefficients.

[0011] The evaluation module is used to calculate the comprehensive trend index of mangrove ecological restoration by weighted summation of the scores of all global principal components using the first set of weight coefficients and / or the second set of weight coefficients.

[0012] This invention, by weighting raw data using the intensity of time-varying changes, highlights ecosystem changes at key time points, enabling trend analysis to reflect the decisive impact of major events on the ecological restoration process. Employing a two-stage principal component analysis strategy, the data is first analyzed after grouping by ecological attributes, representing the strength of trend correlations within different ecological factor groups. This data is then combined with spatial correction, enhancing not only the ecological interpretability of the principal components but also representing the spatial heterogeneity of dominant ecological processes. By establishing an ecological warning threshold and constructing an alternative weighting system based on it, the calculation of the comprehensive index can focus on ecological processes that have already shown warning signals, giving the assessment results risk warning significance and providing a basis for ecological management decisions. Attached Figure Description

[0013] Figure 1 A flowchart of the first embodiment;

[0014] Figure 2 This is a schematic diagram showing the weighted processing of the original sequence;

[0015] Figure 3 This is a schematic diagram illustrating the trend analysis of a single spatial unit.

[0016] Figure 4 This is a schematic diagram of the first set of weighted coefficients. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] The term "multiple" in this application refers to two or more. Furthermore, it should be understood that the terms "first," "second," etc., used in the description of this application are used only for descriptive purposes and should not be construed as indicating or implying relative importance, nor as indicating or implying order.

[0019] In the first embodiment, the present invention proposes a dynamic assessment method for mangrove ecological restoration based on multi-source data fusion, such as... Figure 1 ,include:

[0020] S1, acquire multi-source time series data of the region to be evaluated; for each data series, determine the intensity of time change based on the absolute value of the first derivative of the series at each time node, and use this as a weight to weight the original data series to obtain weighted time series data; calculate the trend slope of each spatial unit based on the weighted time series data to generate the initial trend feature map of each data source;

[0021] The multi-source time series data includes, but is not limited to, Normalized Difference Vegetation Index (NDVI) data extracted from Landsat series remote sensing images covering, for example, from 2010 to 2020, salinity and turbidity data obtained from field monitoring stations, and rainfall data recorded by meteorological stations. All data are processed into raster data sequences with the same spatial resolution, such as 30 meters, and the same time interval, such as monthly.

[0022] For any spatial unit's NDVI time series, the temporal variation intensity W(t) of the series in month t is calculated using the central difference method. That is, W(t) equals the absolute value of the difference between the NDVI value in month t+1 and the NDVI value in month t-1, divided by 2. Here, the spatial unit is the smallest geographic unit analyzed in this invention, possessing a definite location on a map. If satellite remote sensing imagery is used, a spatial unit is a pixel; if gridded data is used, it is a grid. Each spatial unit is associated with one or more time series data, such as the normalized difference vegetation index (NDVI) values ​​for each pixel over the past ten years. Trend analysis and principal component analysis are performed independently or in association for each spatial unit. The original NDVI value in month t is multiplied by the temporal variation intensity W(t) for that month to obtain the weighted NDVI value for that month. This operation is repeated for all time points to generate the complete weighted NDVI time series.

[0023] For each spatial unit (pixel), weighted NDVI time-series data was used, with time as the independent variable and the weighted NDVI value as the dependent variable, and a univariate linear regression analysis was performed. The slope of the regression equation represents the NDVI trend slope for that pixel. The trend slope values ​​of all pixels were combined to obtain an initial NDVI trend feature map. This process was repeated for other data sources such as salinity and turbidity to generate their respective initial trend feature maps. The value of each pixel in the initial trend feature map represents the overall trend of a specific ecological indicator, such as NDVI or sea surface temperature, within the corresponding spatial unit over the entire study period, specifically expressed as the slope of the linear regression analysis.

[0024] To highlight critical periods of ecosystem change by representing the drastic changes in data over time, in one optional embodiment, for each data sequence, the intensity of temporal change is determined based on the absolute value of the first derivative of the sequence at each time point, and the original data sequence is weighted using this as a weight, including:

[0025] For time node t, using the formula Calculate the first derivative of the node, where V(t+1) and V(t-1) are the data values ​​of the next time node and the previous time node, respectively;

[0026] absolute value of the first derivative at all time points Normalized to the [0,1] interval, it is used as the intensity of time variation W(t);

[0027] The original data sequence V(t) is multiplied point by point with the corresponding time variation intensity W(t) to obtain the weighted data sequence. .

[0028] Specifically, for a specific spatial unit, such as the Normalized Difference Vegetation Index (NDVI) time series data for a single pixel, assuming the values ​​for three consecutive months (t-1, t, t+1) are 0.5, 0.7, and 0.6 respectively, then in month t, the first derivative is calculated as D(t) = 0.05. This value represents the rate of change of the vegetation index in that month. This calculation is performed for each time point in the entire time series, resulting in a derivative sequence.

[0029] To convert the rate of change into a weighted intensity index, the absolute values ​​of all derivatives need to be taken and normalized. Assuming the maximum absolute value of all calculated derivatives over the entire study period is 0.2, then the normalized derivative value of 0.05 yields a time-varying intensity W(t) of 0.25. The weight values, ranging from 0 to 1, represent the relative drasticness of change at each time point. Multiplying each value V(t) in the original NDVI sequence by its corresponding time-varying intensity W(t) yields the weighted sequence V'(t), as shown below. Figure 2 For example, the weighted NDVI value for month t becomes 0.175. This weakens the values ​​during periods of gradual change, while preserving the values ​​during periods of dramatic change, allowing trend analysis to detect major shifts in the ecosystem.

[0030] To extract the process of ecological element changes over time in each spatial unit into a numerical value representing the long-term direction and rate of change, i.e., the trend slope, in an optional embodiment, the calculation of the trend slope of each spatial unit based on the weighted time series data includes:

[0031] The weighted time series data of each spatial unit is used as the dependent variable, and the time node series is used as the independent variable. The least squares method is applied to perform univariate linear regression analysis, and the slope value of the obtained regression equation is used as the trend slope of the spatial unit.

[0032] For any spatial unit in the study area, such as a pixel, the weighted NDVI time series data V'(t) of that pixel has been obtained. This series is used as the dependent variable Y, and the time series, such as months 1, 2, 3, ..., N, is used as the independent variable X.

[0033] Based on this, a univariate linear regression model Y = aX + b is constructed, where a is the slope and b is the intercept. The least squares method is used to fit this line, finding the optimal values ​​of a and b that minimize the sum of the squares of the vertical distances from all data points to the line, such as... Figure 3 After calculation, the resulting slope 'a' is the NDVI trend slope for that spatial unit. For example, if the calculated slope for a pixel is 0.005, it indicates that the weighted NDVI value of that pixel shows a slow increasing trend over time. Conversely, if the slope is -0.002, it indicates a decreasing trend. This process is repeated for all spatial units and all ecological attribute data within the study area to generate a series of initial trend feature maps. Each map shows the spatial variation trend distribution of a specific ecological element throughout the entire area.

[0034] S2, group the initial trend feature maps according to ecological attributes, perform the first principal component analysis on each group, extract the first principal component map, and use the variance contribution rate as the trend correlation strength parameter for each group; determine the spatial correction kernel size for the corresponding first principal component map according to the trend correlation strength parameter of each group, and use the spatial correction kernel to perform convolution processing to obtain the correction domain principal component map.

[0035] The initial trend maps of NDVI and vegetation cover were divided into a vegetation ecological group, while the initial trend maps of salinity and turbidity were divided into a hydrological environmental group. Principal component analysis was performed on the two maps of the vegetation ecological group to extract the first principal component. Figure. If the variance contribution rate of this principal component is 0.85, then the trend correlation strength parameter of the vegetation ecogroup is 0.85. The same operation is performed on the hydrological environment group. Each element in the first principal component plot, i.e., the pixel value, is the first principal component score obtained after performing principal component analysis on a certain ecological attribute group, such as the vegetation growth group. The score is a comprehensive new value representing the intensity of the most important change pattern jointly exhibited by all indicators such as NDVI and LAI within the ecological attribute group for that spatial unit.

[0036] The base kernel size is set to 3×3, and the maximum kernel size is set to 11×11. The trend association strength of the vegetation ecogroup is 0.85, so the spatial correction kernel size can be set to 9, i.e., a 9×9 mean filter kernel is used. This 9×9 mean filter kernel is then used to... Convolution operations are performed on the graph to obtain the modified principal component graph of the vegetation ecomes. The lower the correlation strength, the smaller the kernel size.

[0037] To understand ecosystem changes from a functional perspective, in one optional embodiment, grouping the initial trend feature maps according to ecological attributes includes:

[0038] The initial trend feature map was divided into a hydrodynamic group, a vegetation growth group, and an environmental stress group.

[0039] The hydrodynamics group includes trend maps of tides, salinity, and soil moisture content; the vegetation growth group includes trend maps of the normalized difference vegetation index (NDVI) and leaf area index (LAI); and the environmental stress group includes trend maps of sea surface temperature and eutrophication index.

[0040] Specifically, the grouping method reflects several interconnected core subsystems within the mangrove ecosystem. For example, the hydrodynamic group brings together elements that directly affect the physical basis of the mangrove's living environment. Tidal trends determine the frequency and duration of flooding, salinity trends affect the osmotic pressure regulation of plants, and soil moisture trends are related to root respiration and nutrient absorption.

[0041] The vegetation growth group focuses on the biological responses of mangrove communities themselves. The Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI) are internationally recognized remote sensing indicators of vegetation health and cover; their trends collectively represent the overall growth, decline, or stability of the vegetation community. The environmental stress group includes external environmental factors that put pressure on the ecosystem. A sustained rise in sea surface temperature may lead to heat stress, while an increase in the eutrophication index of water bodies may cause water quality deterioration and ecological imbalance. By grouping these factors, ecological change issues can be analyzed from three relatively independent yet interactive dimensions: hydrology, vegetation, and stress.

[0042] To spatially smooth the dominant change pattern maps of each ecological functional group, eliminate local noise, and enhance regional spatial patterns, in an optional embodiment, the step of determining the spatial correction kernel size for the corresponding first principal component map based on the trend correlation strength parameters of each group, and performing convolution processing using the spatial correction kernel, includes:

[0043] The size S of the spatial correction kernel is determined by the formula S=3+2round(10k), where k is the trend correlation strength parameter of each group and round is the rounding function;

[0044] A Gaussian smoothing kernel of size S×S is used as the spatial correction kernel to perform convolution smoothing on the first principal component map.

[0045] A basic kernel size of 3×3 is set. The more synchronized the changing trends of elements within a group, i.e., the higher the trend correlation strength k, the larger the spatial scope of the change pattern may be. Therefore, a larger range of smoothing is needed to reflect the macroscopic pattern. The size of the convolution kernel is determined by a formula. For example, for the vegetation growth group, the first principal component explains 85% of the total variance within the group, so the trend correlation strength parameter k is 0.85. Substituting into the formula, the kernel size S is calculated to be 21. A 21×21 convolution kernel will be used.

[0046] After determining the dimensions, a Gaussian smoothing kernel of corresponding size is generated. The Gaussian smoothing kernel is a weight matrix with the largest value at the center, decreasing outwards according to a Gaussian function distribution. The 21×21 Gaussian kernel is used to perform a convolution operation on the first principal component map of the vegetation growth group. The Gaussian kernel is moved pixel by pixel on the principal component map, and the new value of each pixel is replaced by the weighted average of the pixel's own value and the values ​​of all pixels in its 21×21 neighborhood. The weights are the values ​​at the corresponding positions of the Gaussian kernel. If the k-value of another group, such as the environmental stress group, is only 0.5, the kernel size S will be calculated as 13, and a smaller 13×13 Gaussian kernel will be used for smoothing. This convolution process smooths and integrates variation patterns with strong internal consistency over a broad spatial scale, while preserving local details of patterns with large internal differences.

[0047] S3, combine all the principal component graphs of the correction domain and perform a second principal component analysis to obtain multiple global principal components and their scores and eigenvalues; normalize the eigenvalues ​​of the second principal component analysis to obtain a first set of weight coefficients; set an ecological warning threshold for the scores of each global principal component, and when the score of any global principal component exceeds the threshold, increase the corresponding weight of the global principal component in the first set of weight coefficients to form a second set of weight coefficients.

[0048] The modified principal component maps of the vegetation ecogroup and the hydrological environment group were used as input data for a second principal component analysis. The analysis generated two global principal components, denoted as GPC1 and GPC2. Simultaneously, the scores of each pixel within the study area were calculated on GPC1 and GPC2, and the eigenvalues ​​of GPC1 and GPC2 were obtained, for example, eigenvalues ​​of 1.5 and 0.5 respectively. The eigenvalues ​​of GPC1 and GPC2, 1.5 and 0.5, summed to 2.0. After normalization, the first set of weight coefficients was obtained. =0.75, =0.25, such as Figure 4Calculate the mean and standard deviation of all pixel scores in the GPC1 score map, and set a warning threshold of the mean plus twice the standard deviation. Iterate through the GPC1 score map; if any pixel score exceeds this threshold, it indicates an abnormal vegetation degradation trend. At this point, adjust the weights... Increase by 0.1 to 0.85, and The values ​​are renormalized to 0.15, forming a second set of weight coefficients. In one embodiment, the score refers to the new coordinate value obtained by projecting the values ​​of each spatial unit on all modified domain principal component maps onto the new global principal component coordinate axis. The calculation method is as follows: for a specific global principal component, there is a corresponding feature vector containing a set of weight coefficients; the values ​​of a spatial unit on all input modified domain principal component maps are weighted and summed with this set of weight coefficients, and the result is the score of the spatial unit on that global principal component. The score measures the comprehensive change pattern of the location at the entire ecosystem level. For example, the eigenvector of GPC1 = [0.8, 0.6] represents the composition of 0.8 times the vegetation group trend and 0.6 times the hydrological group trend, and the eigenvector of GPC2 = [0.6, -0.8] represents the composition of the difference between 0.6 times the vegetation group trend and -0.8 times the hydrological group trend. If a pixel P in the study area has a modified principal component map value of 1.2 for the vegetation ecogroup and a modified principal component map value of 0.9 for the hydrological environment group, then the GPC1 score of P = (1.2 × 0.8) + (0.9 × 0.6) = 1.50. Similarly, the GPC2 score of P = (1.2 × 0.6) + (0.9 × -0.8) = 0.0.

[0049] In an optional embodiment, the step of normalizing the eigenvalues ​​of the second principal component analysis to obtain the first set of weight coefficients includes:

[0050] The eigenvalues ​​of each global principal component Divide by the sum of all eigenvalues ​​to obtain the normalized weights. ,all Together they form the first set of weighted coefficients.

[0051] Specifically, after performing a second principal component analysis on the spatially modified principal component maps of the three dimensions of hydrodynamics, vegetation growth, and environmental stress, several new, mutually orthogonal global principal components are obtained, along with eigenvalues ​​corresponding to each principal component. Each eigenvalue... The value of eigenvalue represents the amount of total variance of the input data that the i-th global principal component can explain. The larger the eigenvalue, the more important the comprehensive change pattern represented by the principal component is in the entire ecosystem in time and space.

[0052] Suppose the second principal component analysis produces three global principal components, with corresponding eigenvalues ​​as follows: , , Calculate the sum of all eigenvalues, which is 8.0. Divide the eigenvalues ​​of each principal component by this sum to obtain the normalized weights. For example, the weights of the first principal component... The weight of the second principal component The weight of the third principal component The three weight values ​​of 0.6, 0.2625, and 0.1375 together constitute the first set of weight coefficients, reflecting the basic contribution of each global change pattern to the overall trend of the ecosystem.

[0053] To give high priority to ecological patterns exhibiting extreme or anomalous changes, in an optional embodiment, an ecological warning threshold is set for the scores of each global principal component. When the score of any global principal component exceeds the threshold, the corresponding weight of the global principal component in a first weight coefficient set is increased to form a second weight coefficient set, including:

[0054] For each global principal component score map, calculate the mean score of all spatial units. and standard deviation ;

[0055] Ecological warning threshold ;

[0056] If the first In the principal component score map, there exists a spatial unit whose score exceeds a threshold. Then define temporary weights. Where g is a gain coefficient greater than 1, otherwise ;

[0057] Normalize all temporary weights to obtain the weights in the second weight coefficient set. .

[0058] For each global principal component score map, the value of each pixel on the map represents the local intensity of the integrated variation pattern. The average score of all pixels in the map is calculated. and standard deviation Based on statistical principles, an ecological warning threshold is set. This value typically identifies regions of abnormally high intensity in the data distribution. Check the score map of the i-th principal component to see if any spatial unit's score exceeds the corresponding threshold. If such a phenomenon exists, even if only one pixel exceeds the limit, it indicates that the overall trend represented by that principal component is exhibiting drastic changes in a localized area, potentially foreshadowing ecological risks. In this case, the weight of that principal component will be amplified. For example, assuming a gain coefficient g of 1.5, the original weight of the i-th principal component... If the value is 0.2625, then the temporary weight is... Adjust to 0.39375. If other principal components do not trigger the threshold, the temporary weight remains unchanged.

[0059] Since the increase in some weights violates the constraint that the total weight sum is 1, all temporary weights need to be renormalized. All temporary weights are summed to obtain a new sum, and then each temporary weight is divided by this new sum to obtain the second set of weight coefficients. This new set of weights, compared to the first set of weights which is entirely determined by variance contribution, places greater emphasis on reflecting the changing patterns of extreme ecological signals, making the composite index more sensitive to potential ecological problems.

[0060] S4. Using the first set of weighted coefficients and / or the second set of weighted coefficients, the scores of all global principal components are weighted and summed to calculate the comprehensive trend index of mangrove ecological restoration.

[0061] If no score exceeds the threshold as described above, the first set of weighting coefficients is used. For each pixel, in one embodiment, the pixel comprehensive trend index = the pixel's GPC1 score × 0.75 + the pixel's GPC2 score × 0.25. If the GPC1 score exceeds the threshold as described above, the second set of weighting coefficients is used, and the comprehensive trend index = the pixel's GPC1 score × 0.85 + the GPC2 score × 0.15. A spatial distribution map of the comprehensive trend index of mangrove ecological restoration covering the entire area to be evaluated is generated.

[0062] To integrate the multi-dimensional, abstract global principal component information obtained from the preceding analysis into a comprehensive index, in another optional embodiment, the scores of all global principal components are weighted and summed using the first set of weight coefficients and / or the second set of weight coefficients to calculate the comprehensive trend index of mangrove ecological restoration, including:

[0063] Through formula Calculate the comprehensive trend index for each spatial unit p, where This represents the overall trend index, where i is the index of the global principal component. Let i be the weight of the i-th principal component in the selected set of weight coefficients. Let be the score of the i-th global principal component in spatial cell p.

[0064] Specifically, decision-makers can choose which set of weighting coefficients to use based on the assessment objectives. If a routine, objective trend assessment is desired, the first set of weighting coefficients is used; if the goal is to highlight potential risks and areas of extreme change, the second set of weighting coefficients is used. The calculation process is performed on each spatial unit p. Taking a specific pixel p as an example, its score on all N global principal component score maps is obtained, denoted as... At the same time, the selected set of weight coefficients The overall trend index of this pixel. The composite index is obtained by multiplying the score of each principal component by its corresponding weight and then summing all the products. For example, assuming there are three principal components, and a pixel has scores of 2.0, -1.5, and 0.8, with selected weights of 0.6, 0.26, and 0.14, then the composite index for that pixel is... This calculation is repeated for every pixel within the study area to generate a comprehensive trend index map of mangrove ecological restoration. On this map, high-value areas represent an overall positive trend in ecological condition, while low-value or negative-value areas represent a poor overall trend or ongoing degradation, thus providing clear spatial guidance for assessing the effectiveness of ecological restoration and making management decisions.

[0065] In the second embodiment, the present invention also proposes a dynamic assessment system for mangrove ecological restoration based on multi-source data fusion, comprising the following modules:

[0066] The generation module is used to acquire multi-source time series data of the region to be evaluated; for each data series, the intensity of time change is determined based on the absolute value of the first derivative of the series at each time node, and the original data series is weighted using this as a weight to obtain weighted time series data; the trend slope of each spatial unit is calculated based on the weighted time series data to generate the initial trend feature map of each data source;

[0067] The correction module is used to group the initial trend feature map according to ecological attributes, perform a first principal component analysis on each group, extract the first principal component map, and use the variance contribution rate as the trend correlation strength parameter of each group; according to the trend correlation strength parameter of each group, determine the spatial correction kernel size for the corresponding first principal component map, and use the spatial correction kernel to perform convolution processing to obtain the correction domain principal component map.

[0068] The enhancement module is used to combine all the principal component graphs of the correction domain and perform a second principal component analysis to obtain multiple global principal components, scores, and eigenvalues; normalize the eigenvalues ​​of the second principal component analysis to obtain a first set of weight coefficients; set an ecological warning threshold for the scores of each global principal component, and when the score of any global principal component exceeds the threshold, increase the corresponding weight of the global principal component in the first set of weight coefficients to form a second set of weight coefficients.

[0069] The evaluation module is used to calculate the comprehensive trend index of mangrove ecological restoration by weighted summation of the scores of all global principal components using the first set of weight coefficients and / or the second set of weight coefficients.

[0070] In an optional embodiment, the step of determining the intensity of time change for each data sequence based on the absolute value of the first derivative of the sequence at each time node, and using this as a weight to weight the original data sequence, includes:

[0071] For time nodes Using the formula Calculate the first derivative of the nodes, where V(t+1) and V(t-1) are the data values ​​of the next and previous time nodes, respectively; calculate the absolute values ​​of the first derivatives of all time nodes. Normalize to the [0,1] interval and use it as the time variation intensity W(t); multiply the original data sequence V(t) point by point with the corresponding time variation intensity W(t) to obtain the weighted data sequence. .

[0072] In an optional embodiment, calculating the trend slope of each spatial unit based on the weighted time series data includes:

[0073] The weighted time series data of each spatial unit is used as the dependent variable, and the time node series is used as the independent variable. The least squares method is applied to perform univariate linear regression analysis, and the slope value of the obtained regression equation is used as the trend slope of the spatial unit.

[0074] In an optional embodiment, grouping the initial trend feature maps according to ecological attributes includes:

[0075] The initial trend feature map was divided into a hydrodynamic group, a vegetation growth group, and an environmental stress group.

[0076] The hydrodynamics group includes trend maps of tides, salinity, and soil moisture content; the vegetation growth group includes trend maps of the normalized difference vegetation index (NDVI) and leaf area index (LAI); and the environmental stress group includes trend maps of sea surface temperature and eutrophication index.

[0077] In an optional embodiment, determining the spatial correction kernel size for the corresponding first principal component map based on the trend correlation strength parameters of each group, and performing convolution processing using the spatial correction kernel, includes:

[0078] The size S of the spatial correction kernel is determined by the formula S=3+2round(10k), where k is the trend correlation strength parameter of each group and round is the rounding function;

[0079] A Gaussian smoothing kernel of size S×S is used as the spatial correction kernel to perform convolution smoothing on the first principal component map.

[0080] In an optional embodiment, the step of normalizing the eigenvalues ​​of the second principal component analysis to obtain the first set of weight coefficients includes:

[0081] The eigenvalues ​​of each global principal component Divide by the sum of all eigenvalues ​​to obtain the normalized weights. ,all Together they form the first set of weighted coefficients.

[0082] In an optional embodiment, the step of setting an ecological warning threshold for each global principal component score, and increasing the corresponding weight of the global principal component in the first weight coefficient set to form a second weight coefficient set when the score of any global principal component exceeds the threshold, includes:

[0083] For each global principal component score map, calculate the mean score of all spatial units. and standard deviation ;

[0084] Ecological warning threshold ;

[0085] If the first In the principal component score map, there exists a spatial unit whose score exceeds a threshold. Then define temporary weights. Where g is a gain coefficient greater than 1, otherwise ;

[0086] Normalize all temporary weights to obtain the weights in the second weight coefficient set. .

[0087] In an optional embodiment, the step of using the first set of weighted coefficients and / or the second set of weighted coefficients to perform a weighted summation of the scores of all global principal components to calculate the comprehensive trend index of mangrove ecological restoration includes:

[0088] Through formula Calculate the comprehensive trend index for each spatial unit p, where This represents the overall trend index, where i is the index of the global principal component. Let i be the weight of the i-th principal component in the selected set of weight coefficients. Let be the score of the i-th global principal component in spatial cell p.

[0089] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0090] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0091] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A mangrove ecological restoration dynamic evaluation method of multi-source data fusion, characterized in that, The method comprises the following steps: obtaining multi-source time series data of an area to be evaluated; for each data sequence, determining the time variation intensity based on the absolute value of the first derivative of the sequence at each time node, and weighting the original data sequence with the time variation intensity to obtain weighted time series data; calculating the trend slope of each spatial unit based on the weighted time series data to generate an initial trend feature map of each data source; grouping the initial trend feature map according to ecological attributes, performing first principal component analysis on each group to extract a first principal component map, and taking the variance contribution rate as a trend correlation intensity parameter of the group; determining the spatial correction kernel size for the corresponding first principal component map according to the trend correlation intensity parameter of each group, and performing convolution processing using the spatial correction kernel to obtain a corrected domain principal component map; combining all corrected domain principal component maps and performing second principal component analysis to obtain a plurality of global principal components, scores and characteristic values of the global principal components; normalizing the characteristic values of the second principal component analysis to obtain a first weight coefficient set; setting an ecological warning threshold for the score of each global principal component, and when the score of any global principal component exceeds the threshold, increasing the corresponding weight of the global principal component in the first weight coefficient set to form a second weight coefficient set; using the first weight coefficient set and / or the second weight coefficient set to perform weighted summation on the scores of all global principal components to calculate a comprehensive trend index of mangrove ecological restoration; the calculation of the trend slope of each spatial unit based on the weighted time series data comprises: using the weighted time series data of each spatial unit as the dependent variable and the time node sequence as the independent variable, performing one-dimensional linear regression analysis by applying the least square method, and taking the slope value of the obtained regression equation as the trend slope of the spatial unit; the grouping of the initial trend feature map according to ecological attributes comprises: dividing the initial trend feature map into a hydrodynamic group, a vegetation growth group and an environmental stress group; wherein the hydrodynamic group contains trend feature maps of tide, salinity and soil moisture content; the vegetation growth group contains trend feature maps of normalized vegetation index NDVI and leaf area index LAI; and the environmental stress group contains trend feature maps of sea surface temperature and water body eutrophication index; the determination of the spatial correction kernel size for the corresponding first principal component map according to the trend correlation intensity parameter of each group and the convolution processing using the spatial correction kernel comprises: the size S of the spatial correction kernel is determined by the formula S=3+2round(10k), wherein k is the trend correlation intensity parameter of the group, and round is a rounding function; a Gaussian smoothing kernel with a size of S×S is used as the spatial correction kernel to perform convolution smoothing processing on the first principal component map.

2. The method of claim 1, wherein, the determination of the time variation intensity for each data sequence based on the absolute value of the first derivative of the sequence at each time node and the weighting of the original data sequence with the time variation intensity comprises: For a time node t, the first derivative of the node is calculated using the formula where V(t+1) and V(t-1) are the data values at the next time node and the previous time node, respectively. normalizing the absolute value of the first derivative at all time nodes to the interval [0,1] as the time variation intensity W(t). The original data sequence V(t) is multiplied point by point with the corresponding time-varying intensity W(t) to obtain a weighted data sequence .

3. The method of claim 1, wherein, The characteristic value of the second principal component analysis is normalized to obtain a first weight coefficient set, comprising: The eigenvalue of each global principal component is divided by the sum of all eigenvalues to obtain a normalized weight , all commonly constitute a first weight coefficient set.​ 4. The method of claim 3, wherein, The ecological warning threshold of each global principal component score is set, and when the score of any global principal component exceeds the threshold, the corresponding weight of the global principal component in the first weight coefficient set is increased to form a second weight coefficient set, comprising: for each global principal component score map, calculating the mean of the scores of all spatial units of the global principal component score map and the standard deviation ; Ecological warning threshold ; If the score of any spatial unit in the i-th principal component score map exceeds a threshold value then define a temporary weight where g is a gain factor greater than 1, otherwise i is the serial number of the global principal component; All temporary weights are normalized to obtain the weights in the second weight coefficient set.

5. The method of claim 1, wherein, The first weight coefficient set and / or the second weight coefficient set are used to perform weighted summation on the scores of all global principal components to calculate a mangrove ecological restoration comprehensive trend index, comprising: The global trend index for each spatial unit p is calculated by the formula wherein is the global trend index, i is the index of the global principal component, is the weight of the i-th principal component of the set of weight coefficients used, is the score of the i-th global principal component in spatial unit p.

6. A mangrove forest ecological restoration dynamic evaluation system based on multi-source data fusion, characterized in that, The following modules are included: The generation module is used to obtain multi-source time series data of a region to be evaluated; for each data sequence, the absolute value of the first derivative of the sequence at each time node is determined to determine the time change intensity, and the original data sequence is weighted with the weight to obtain weighted time series data; The trend slope of each spatial unit is calculated based on the weighted time series data to generate an initial trend feature map of each data source; The correction module is used to group the initial trend feature map according to ecological properties, perform first principal component analysis on each group to extract a first principal component map, and use the variance contribution rate as a trend correlation strength parameter of the group; The spatial correction kernel size of the corresponding first principal component map is determined according to the trend correlation strength parameter of each group, and convolution processing is performed using the spatial correction kernel to obtain a corrected domain principal component map; The improvement module is used to combine all corrected domain principal component maps and perform second principal component analysis to obtain a plurality of global principal components, as well as the scores and characteristic values of the global principal components; The characteristic values of the second principal component analysis are normalized to obtain a first weight coefficient set; the ecological warning threshold of each global principal component score is set, and when the score of any global principal component exceeds the threshold, the corresponding weight of the global principal component in the first weight coefficient set is increased to form a second weight coefficient set; The evaluation module is used to use the first weight coefficient set and / or the second weight coefficient set to perform weighted summation on the scores of all global principal components to calculate a mangrove ecological restoration comprehensive trend index; The trend slope of each spatial unit is calculated based on the weighted time series data, comprising: The weighted time series data of each spatial unit is used as the dependent variable, and the time node sequence is used as the independent variable. A one-dimensional linear regression analysis is performed using the least squares method, and the slope value of the obtained regression equation is used as the trend slope of the spatial unit; The initial trend feature map is grouped according to the ecological properties, comprising: The initial trend feature map is divided into a hydrodynamic group, a vegetation growth group, and an environmental stress group; The hydrodynamic group includes trend feature maps of tide, salinity, and soil moisture content; the vegetation growth group includes trend feature maps of normalized vegetation index NDVI and leaf area index LAI; and the environmental stress group includes trend feature maps of sea surface temperature and water body eutrophication index; The spatial correction kernel size of the corresponding first principal component map is determined according to the trend correlation strength parameter of each group, and convolution processing is performed using the spatial correction kernel, comprising: The size S of the spatial correction kernel is determined by the formula S=3+2round(10k), wherein k is the trend correlation strength parameter of the group, and round is a rounding function; A Gaussian smoothing kernel with a size of S×S is used as the spatial correction kernel to perform convolution smoothing on the first principal component map.

7. The system of claim 6, wherein, For each data sequence, the absolute value of the first derivative of the sequence at each time node is determined as the time variation strength, and the original data sequence is weighted by using the time variation strength as the weight, including: For a time node t, the first derivative of the node is calculated using the formula where V(t+1) and V(t-1) are the data values at the next time node and the previous time node, respectively. The absolute values of the first derivatives of all time nodes are normalized to the interval [0, 1] as the time variation strength W(t); and The original data sequence V(t) is multiplied point by point with the corresponding time-varying intensity W(t) to obtain a weighted data sequence .

Images