Vegetation carbon sink space optimization method and system based on interpretable machine learning

Abstract

The invention discloses a vegetation carbon sink space optimization method and system based on interpretable machine learning. The method comprises the steps that firstly, multi-source environment data such as administrative division, environment factors and target variables of a target area are acquired, and a space table data set is formed; training an XGBoost regression model based on the data set, predicting the vegetation carbon sink amount and evaluating the performance of the model; quantifying the contribution of each environmental factor to the carbon sink by combining an SHAP interpreter, generating a feature importance and interpretability map, fitting through a LOWESS algorithm to obtain an environmental factor and carbon sink response relation curve, and extracting an optimal threshold interval; further predicting a carbon sink gain according to an optimization threshold value, delimiting optimization partitions, and forming a space optimization scheme; and finally outputting a vegetation carbon sink optimization suggestion report. According to the method, scientific quantification of carbon sink space optimization can be realized, and the interpretability and decision support value of a carbon sink improvement strategy are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecological environment, in particular to a vegetation carbon sink space optimization method and system based on interpretable machine learning. BACKGROUND

[0002] As an important way of fixing and storing carbon dioxide in the terrestrial ecosystem, vegetation carbon sink plays a key role in global carbon cycle and climate change mitigation.

[0003] Existing vegetation carbon sink space optimization research mainly relies on traditional statistical methods such as multiple linear regression, correlation analysis, etc. Such methods can reveal the relationship between environmental factors and carbon sinks to some extent, but often assume that there is a linear correlation between variables, making it difficult to capture the complex nonlinear response mechanism of carbon sinks to multi-source environmental factors, resulting in limited prediction accuracy. At the same time, although machine learning methods such as random forest and gradient boosting tree have improved prediction accuracy, their model decision-making process is not transparent, which belongs to the typical "black box" model, making it difficult to extract the interpretable rules of environmental factors, thus failing to provide clear threshold constraints and regulation basis for spatial optimization.

[0004] In addition, most of the existing optimization methods lack a closed-loop design from interpretive modeling to spatial partitioning, often remaining in the result prediction stage, making it difficult to realize the spatialization identification of vegetation carbon sink potential gain and the generation of optimization strategies. This limits its practical application value in regional ecological planning, carbon sink management and policy making.

[0005] Therefore, it is urgent to propose a new method that combines interpretable machine learning and spatial optimization, which can obtain carbon sink distribution through high-precision prediction model, and further determine the threshold interval and optimization partition of key factors by using interpretive algorithm to quantify the action rules of environmental factors, so as to provide a scientific and operable spatial regulation path for vegetation carbon sink improvement. SUMMARY

[0006] In order to solve at least one of the above technical problems, the present application provides a vegetation carbon sink space optimization method and system based on interpretable machine learning.

[0007] The first aspect of the present application provides a vegetation carbon sink space optimization method based on interpretable machine learning, comprising: obtaining multi-source environmental data of a target region, the multi-source environmental data including administrative division data, environmental factor data, and target variable data, performing spatial coordinate unification and resolution standardization processing on the multi-source environmental data to generate a spatial table data set; training an XGBoost regression model according to the spatial table dataset by using a cross-validation and early stopping strategy, predicting vegetation carbon sink capacity of the target region according to the trained XGBoost regression model, and determining model performance data; selecting a preset number of sampling points in the XGBoost regression model training process, calculating contribution values of individual environmental factors to vegetation carbon sink based on a SHAP interpreter, and generating a feature importance ranking graph and a SHAP summary graph; fitting the feature importance ranking graph and the SHAP summary graph based on a LOWESS algorithm, determining a relationship curve between the values of each environmental factor and the SHAP values, and determining an optimized threshold interval for each environmental factor according to the relationship curve; determining optimized feature values of the environmental factors according to the optimized threshold interval, predicting vegetation carbon sink gain of the optimized feature values, obtaining prediction gain data, and determining an optimized partition according to the prediction gain data; generating a vegetation carbon sink optimization suggestion report for the optimized partition according to the model performance data and the optimized threshold interval of each environmental factor.

[0008] In the scheme, the multi-source environmental data of the target region is obtained, the multi-source environmental data includes administrative division data, environmental factor data, and target variable data, the multi-source environmental data is subjected to spatial coordinate unification and resolution standardization processing, and a spatial table dataset is generated, specifically: The administrative division data of the target region is obtained, the administrative division data includes boundary position coordinate information of the target region, and the target region is standardized into a plurality of regular grids according to a preset resolution according to the administrative division data; The environmental factor data of each grid of the target region in a historical preset time period is obtained, the environmental factor data includes precipitation, temperature, height, slope, normalized vegetation index, leaf area index, soil organic carbon content, soil clay content, land use intensity data, and night light data; The target variable data of each grid of the target region is obtained, the target variable data is net primary productivity; The coordinate system of the multi-source environmental data is converted into a WGS84 geographic coordinate system, and the environmental factor data and the target variable data after unifying the coordinate system are extracted into corresponding coordinate grids to generate a spatial table dataset containing environmental factors and target variables.

[0009] In the scheme, the XGBoost regression model is trained according to the spatial table dataset by using a cross-validation and early stopping strategy, the vegetation carbon sink capacity of the target region is predicted according to the trained XGBoost regression model, and model performance data is determined, specifically: performing outlier analysis on the spatial table data based on a quartile range method, calculating a first quartile and a third quartile of each environmental factor variable to obtain a quartile range value, setting an outlier judgment threshold range, and marking data less than the first quartile minus a preset multiple of the quartile range value or greater than the third quartile plus a preset multiple of the quartile range value as a numerical outlier point; performing outlier processing on the numerical outlier point, including replacing an infinite value with a null value NaN, replacing an outlier value with a preset quantile value, and replacing a missing value with a median value; dividing the spatial table data after the outlier processing into an 80% training set and a 20% test set, initializing an XGBoost regression model and setting hyperparameters of the model, including a learning rate, a regularization coefficient, and a maximum depth of a decision tree; training the training set and the test set in the XGBoost regression model based on a k-fold cross-validation strategy, setting a maximum number of iterations and an early stopping number of rounds of a training process, stopping the training when the model does not improve in a preset continuous iteration number of rounds of prediction accuracy of net ecosystem productivity on a validation set, and obtaining a trained XGBoost regression model; predicting vegetation carbon sink capacity of the test set based on the trained XGBoost regression model, calculating a root mean square error, a mean absolute error, and a determination coefficient between a predicted value and an actual value, evaluating model performance according to the root mean square error, the mean absolute error, and the determination coefficient, and obtaining model performance data.

[0010] In the scheme, a preset number of sampling points in the training process of the XGBoost regression model are selected, a SHAP interpreter is used to calculate a contribution value of each environmental factor to vegetation carbon sink, and a feature importance ranking graph and a SHAP summary graph are generated. Specifically, randomly extracting a preset number of sample data from training input data of the trained XGBoost regression model as a SHAP analysis sample set, and extracting all samples if the total number of samples is less than the preset number; creating a SHAP interpreter object based on the XGBoost regression model, for each environmental factor of each sample, constructing a benchmark reference value through the SHAP interpreter and evaluating a marginal contribution of the environmental factor to a model output in all feature subsets of the XGBoost regression model based on a Shapley value calculation method, and obtaining a SHAP value of the environmental factor through weighted average of contribution values of all feature subsets; sorting all environmental factors in descending order of average absolute values of the SHAP values to generate a feature importance ranking list of contribution sizes of the environmental factors to vegetation carbon sink prediction; drawing a SHAP summary plot based on the SHAP values, the horizontal axis of the SHAP summary plot being the SHAP values, the vertical axis being the environmental factors sorted by importance, each point in the plot representing the SHAP value distribution of a sample on the environmental factor, and the color of the point reflecting the size of the environmental factor value of the sample; Meanwhile, a feature importance ranking bar chart is generated, the horizontal axis of the bar chart being the average absolute SHAP value, the vertical axis being the environmental factor name, and the length of the bar representing the importance of each environmental factor.

[0011] In this scheme, the relationship curve between the environmental factor value and the SHAP value is determined by fitting the feature importance ranking chart and the SHAP summary plot based on the LOWESS algorithm, and the optimal threshold interval of each environmental factor is determined according to the relationship curve, specifically: The feature values corresponding to each environmental factor in the SHAP summary plot and the SHAP values are data cleaned to remove sample points containing missing values or invalid values; The cleaned feature values and SHAP values are locally weighted regression fitted using the LOWESS algorithm to generate a smooth relationship curve of the feature values and the SHAP values; The Spearman correlation coefficient of the feature values and the SHAP values of each environmental factor is calculated, and the positive and negative influence direction of the environmental factor on the vegetation carbon sink is determined according to the sign of the correlation coefficient; On the smooth relationship curve, a continuous feature value interval with a SHAP value greater than zero is identified, and if there are multiple continuous intervals, the interval with the maximum average SHAP value is selected as the candidate optimal interval; When the length of the candidate optimal interval is less than the preset minimum ecological interval threshold, the first optimal threshold interval is obtained by expanding the interval to the minimum ecological interval length on both sides with the median as the center; If there is no interval with a SHAP value greater than zero on the smooth relationship curve, the second optimal threshold interval is obtained by selecting the feature value point corresponding to the maximum global SHAP value as the center to construct the minimum ecological interval with the feature value point as the center; For each environmental factor, an optimal threshold interval visualization chart is generated, which simultaneously displays the original feature value-SHAP value scatter plot, the LOWESS fitting curve, the SHAP zero value reference line, and the optimal threshold interval boundary marker; The first optimal threshold interval or the second optimal threshold interval of each environmental factor is compared with the original feature value distribution range, and the interval boundary points exceeding the actual feature value range are removed, and the optimal threshold interval of each environmental factor is output.

[0012] In this scheme, the optimal feature value of the environmental factor is determined according to the optimal threshold interval, the vegetation carbon sink gain of the optimal feature value is predicted to obtain prediction gain data, and the optimal partition is determined according to the prediction gain data, specifically: select a preset number of key environmental factors from the feature importance ranking diagram, calculate the deviation distance of each key environmental factor in each grid of the target area from the boundary of the optimization threshold interval, and determine the optimization demand intensity according to the deviation distance; Select the feature value corresponding to the maximum SHAP value of each key environmental factor in each grid of the target area within the optimization threshold interval as the optimization feature value to form an optimization feature data set; The optimization feature data set is imported into the trained XGBoost regression model for vegetation carbon sink prediction to obtain the optimized net ecosystem productivity prediction value, and the difference between the net ecosystem productivity before and after optimization is calculated as the carbon sink gain potential value; If the carbon sink gain potential value is greater than a preset ratio of the net ecosystem productivity before optimization, the environmental factor is selected as the to-be-optimized environmental factor, and the number of to-be-optimized environmental factors is counted; According to the carbon sink gain potential value, the optimization demand intensity, and the number of to-be-optimized environmental factors of each grid of the target area, the optimization partition is determined, and the carbon sink potential gain spatial distribution diagram, the key feature optimization demand heat map, and the optimization partition suggestion diagram are drawn according to the carbon sink potential gain and the to-be-optimized environmental factors of each optimization partition.

[0013] In this scheme, the vegetation carbon sink optimization suggestion report of the optimization partition generated according to the model performance data and the optimization threshold interval of each environmental factor is specifically: According to the model performance data and the optimization threshold interval of each environmental factor, an optimization partition vegetation carbon sink optimization suggestion report is generated; The vegetation carbon sink optimization suggestion report includes the to-be-optimized environmental factors, the optimization direction, the optimization priority, the carbon sink gain potential value, the carbon sink potential gain spatial distribution diagram, the key feature optimization demand heat map, and the optimization partition suggestion diagram of each grid of the target area.

[0014] The second aspect of the present application also provides a vegetation carbon sink space optimization system based on interpretable machine learning, which comprises a memory and a processor, wherein the memory comprises a vegetation carbon sink space optimization method program based on interpretable machine learning, and the vegetation carbon sink space optimization method program based on interpretable machine learning is executed by the processor to realize the following steps: Obtain multi-source environmental data of a target area, wherein the multi-source environmental data comprises administrative division data, environmental factor data, and target variable data, and the multi-source environmental data is subjected to spatial coordinate unification and resolution standardization processing to generate a spatial table data set; An XGBoost regression model is trained according to the spatial table dataset by using a cross-validation and early stopping strategy, the vegetation carbon sink capacity of a target region is predicted according to the trained XGBoost regression model, and model performance data is determined; A preset number of sampling points in the XGBoost regression model training process are selected, a contribution value of each environmental factor to the vegetation carbon sink is calculated based on a SHAP interpreter, a feature importance ranking graph and a SHAP summary graph are generated; The feature importance ranking graph and the SHAP summary graph are fitted based on a LOWESS algorithm, a relationship curve of the value of each environmental factor and the SHAP value is determined, and an optimal threshold interval of each environmental factor is determined according to the relationship curve; An optimal feature value of the environmental factor is determined according to the optimal threshold interval, the vegetation carbon sink gain of the optimal feature value is predicted, prediction gain data is obtained, and an optimized partition is determined according to the prediction gain data; A vegetation carbon sink optimization suggestion report of the optimized partition is generated according to the model performance data and the optimal threshold interval of each environmental factor.

[0015] The present application discloses a vegetation carbon sink spatial optimization method and system based on interpretable machine learning. The method first acquires multi-source environmental data such as administrative divisions, environmental factors and target variables of a target region, forming a spatial table dataset; an XGBoost regression model is trained based on the dataset to predict the vegetation carbon sink capacity and evaluate the model performance; the SHAP interpreter is combined to quantify the contribution of each environmental factor to the carbon sink, generate a feature importance and interpretability graph, and obtain the environmental factor and carbon sink response relationship curve by fitting through the LOWESS algorithm, and extract the optimal threshold interval; further, the carbon sink gain is predicted according to the optimal threshold, the optimized partition is delimited, and the spatial optimization scheme is formed; finally, the vegetation carbon sink optimization suggestion report is output. The method can realize the scientific quantification of carbon sink spatial optimization and improve the interpretability and decision support value of the carbon sink promotion strategy. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a vegetation carbon sink spatial optimization method based on interpretable machine learning is shown; Figure 2 A flowchart of generating a spatial table dataset is shown; Figure 3 A flowchart of generating a feature importance ranking graph and a SHAP summary graph is shown; Figure 4 A block diagram of a vegetation carbon sink spatial optimization system based on interpretable machine learning is shown. DETAILED DESCRIPTION

[0017] In order to enable a more clear understanding of the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0018] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0019] Figure 1 A flowchart of a vegetation carbon sink space optimization method based on an interpretable machine learning method is shown.

[0020] As Figure 1 shown, the first aspect of the present application provides a vegetation carbon sink space optimization method based on an interpretable machine learning method, comprising: S102, obtaining multi-source environment data of a target region, the multi-source environment data comprising administrative division data, environmental factor data, and target variable data, performing spatial coordinate unification and resolution standardization processing on the multi-source environment data to generate a spatial table data set; S104, training an XGBoost regression model according to the spatial table data set using cross-validation and early stopping strategies, predicting the vegetation carbon sink amount of the target region according to the trained XGBoost regression model, and determining model performance data; S106, selecting a preset number of sampling points in the XGBoost regression model training process, calculating the contribution value of each environmental factor to the vegetation carbon sink based on a SHAP interpreter, generating a feature importance ranking graph and a SHAP summary graph; S108, fitting the feature importance ranking graph and the SHAP summary graph based on a LOWESS algorithm to determine the relationship curve between the environmental factor value and the SHAP value, and determining the optimization threshold interval of each environmental factor according to the relationship curve; S110, determining the optimized feature value of the environmental factor according to the optimization threshold interval, predicting the vegetation carbon sink gain of the optimized feature value to obtain prediction gain data, and determining an optimized partition according to the prediction gain data; S112, generating a vegetation carbon sink optimization suggestion report for the optimized partition according to the model performance data and the optimization threshold interval of each environmental factor.

[0021] It should be noted that by acquiring and uniformly processing multi-source environmental data such as administrative division, environmental factors and target variables of the target area, a spatial table dataset is generated, thereby realizing spatialization fusion and standardization of data of different sources and different accuracies, and ensuring the accuracy and comparability of subsequent modeling; the XGBoost regression model is trained by using cross-validation and early stopping strategy, and carbon sink prediction is performed, which effectively improves the generalization ability and prediction accuracy of the model, avoids overfitting, and makes the estimation of carbon sink more reliable; in the training process, the SHAP interpreter is introduced, the quantitative analysis of the contribution of environmental factors to carbon sink not only reveals the causal effect under the complex nonlinear relationship, but also generates a visual feature importance ranking chart and a summary chart, thereby realizing the transformation of the black box model into an interpretable model; based on the LOWESS algorithm, the SHAP result is fitted to obtain a smooth relationship curve between the values of each environmental factor and the contribution of carbon sink, and the best threshold interval of the factor effect is further identified, thereby providing scientific boundary constraints for environmental regulation and carbon sink improvement; according to the optimized threshold interval, the optimal feature value of the environmental factor is determined and the carbon sink gain is predicted, realizing the quantitative evaluation of the potential carbon sink improvement space, and identifying the optimization priority of different regions through spatial zoning method, and forming a differentiated optimization layout; finally, the carbon sink optimization suggestion report is generated by combining the model performance data and the optimized threshold interval, which can provide intuitive and operable decision basis for regional ecological governance, vegetation restoration and carbon sink management.

[0022] Figure 2 A flowchart for generating a spatial table dataset is shown.

[0023] According to the embodiment of the present application, the multi-source environmental data of the target area is obtained, the multi-source environmental data includes administrative division data, environmental factor data and target variable data, the multi-source environmental data is uniformly processed in spatial coordinates and resolution standardization, and a spatial table dataset is generated, specifically: S202, administrative division data of the target area is obtained, the administrative division data includes boundary position coordinate information of the target area, and the target area is standardized into a plurality of regular grids according to a preset resolution according to the administrative division data; S204, environmental factor data of each grid of the target area in a historical preset time period is obtained, the environmental factor data includes precipitation, temperature, height, slope, normalized vegetation index, leaf area index, soil organic carbon content, soil clay content, land use intensity data and night light data; S206, target variable data of each grid of the target area is obtained, the target variable data is net ecosystem productivity; S208, the coordinate system of the multi-source environmental data is converted into the WGS84 geographic coordinate system, the environmental factor data and the target variable data after unifying the coordinate system are extracted into the corresponding coordinate grid, and spatial table data containing the environmental factor and the target variable are generated.

[0024] It should be noted that the multi-source environmental factor data and the target variable data with different sources, inconsistent resolutions and different coordinate systems are unified into regularized spatial grid units to form a data set with a standardized format and a unified coordinate system, so that the environmental factor and the target variable are accurately matched in the same spatial unit, and calculation errors caused by spatial mismatch or inconsistent scales are avoided.

[0025] According to the embodiment of the present application, the XGBoost regression model is trained according to the spatial table data set by using cross-validation and early stopping strategy, the vegetation carbon sink of the target area is predicted according to the trained XGBoost regression model, and the model performance data is determined, specifically: Based on the quartile range method, the spatial table data is analyzed for outliers, the first quartile and the third quartile of each environmental factor variable are calculated, the quartile range value is obtained, the outlier judgment threshold range is set, and the data less than the first quartile minus the preset multiple of the quartile range value or greater than the third quartile plus the preset multiple of the quartile range value is marked as a numerical outlier point; The numerical outliers are processed, including replacing infinite values with null values NaN, replacing outliers with preset quantile values, and replacing missing values with median values; The spatial table data after outlier processing is divided into 80% training set and 20% test set, the XGBoost regression model is initialized and the hyperparameters of the model are set, including learning rate, regularization coefficient and maximum depth of decision tree; The training set and the test set are imported into the XGBoost regression model based on the k-fold cross-validation strategy for training, the maximum number of iterations and the number of early stopping rounds of the training process are set, the model is stopped when the prediction accuracy of net ecosystem productivity on the validation set does not improve within a preset number of consecutive iterations, and the trained XGBoost regression model is obtained; Based on the trained XGBoost regression model, the vegetation carbon sink of the test set is predicted, the root mean square error, the mean absolute error and the determination coefficient between the predicted value and the true value are calculated, the model performance is evaluated according to the root mean square error, the mean absolute error and the determination coefficient, and the model performance data is obtained.

[0026] It should be noted that the extreme abnormal points, invalid values and missing values are processed by the quartile range method for detecting and correcting the abnormal values of the spatial table data, the noise interference is significantly reduced, and the stability and reliability of the input data are ensured. Secondly, in the data division stage, the cleaned data set is reasonably divided into a training set and a test set, ensuring that the model can be verified by independent samples while being trained, so as to obtain an objective performance evaluation. In the model construction process, the advantages of the XGBoost regression algorithm are used, the gradient boosting-based ensemble learning framework can automatically capture the complex nonlinear relationship between environmental factors and vegetation carbon sinks, and effectively process the interaction between multi-dimensional variables. By setting the learning rate, regularization coefficient and tree depth and other hyperparameters, the model balances between fitting ability and generalization performance. Combined with the k-fold cross-validation strategy, the limited data can be fully utilized to avoid the bias caused by sample division and enhance the stability of the model. At the same time, the early stopping mechanism is introduced to terminate the training when the performance of the validation set does not improve for a certain number of rounds, thereby effectively preventing overfitting. Finally, based on the trained XGBoost model, the test set is predicted, and the results are evaluated by root mean square error, mean absolute error and coefficient of determination, which not only quantifies the prediction accuracy, but also ensures the high consistency of the prediction results and the true carbon sink. The first quartile is the value at the 25% position after arranging all data from small to large, and the third quartile is the value at the 75% position. The training set includes the validation set, and the validation set is randomly extracted from the training set for model training accuracy verification.

[0027] Figure 3 A flowchart of generating a feature importance ranking graph and a SHAP summary graph is shown.

[0028] According to the embodiment of the present application, the preset number of sampling points in the XGBoost regression model training process is selected, the contribution value of each environmental factor to the vegetation carbon sink is calculated based on the SHAP interpreter, and the feature importance ranking graph and the SHAP summary graph are generated, specifically: S302, randomly extract a preset number of sample data from the training input data of the trained XGBoost regression model as a SHAP analysis sample set, and if the total number of samples is less than the preset number, extract all the samples; S304, create a SHAP interpreter object based on the XGBoost regression model, for each environmental factor of each sample, construct a benchmark reference value through the SHAP interpreter and based on the Shapley value calculation method, evaluate the marginal contribution of the environmental factor to the model output in all feature subsets of the XGBoost regression model, and obtain the SHAP value of the environmental factor by weightedly averaging the contribution values of all feature subsets; S306, rank all environmental factors in descending order according to the average absolute value of their SHAP values, and generate a feature importance ranking list of the contribution of each environmental factor to the prediction of vegetation carbon sink; S308, draw a SHAP summary plot based on the SHAP values, with the horizontal axis of the SHAP summary plot being the SHAP value and the vertical axis being the environmental factor ranked by importance, each point in the plot representing the SHAP value distribution of a sample on the environmental factor, and the color of the point reflecting the size of the environmental factor value of the sample; S310, generate a feature importance ranking bar chart at the same time, with the horizontal axis of the bar chart being the average absolute SHAP value and the vertical axis being the environmental factor name, and the length of the bar representing the importance of each environmental factor.

[0029] It should be noted that first, representative samples are extracted from the model training data to construct a SHAP analysis sample set, ensuring that the typical value space of the environmental factors is covered. Subsequently, the SHAP explainer object created by constructing a benchmark reference value systematically evaluates the marginal contribution of each environmental factor to the model output in all possible feature subsets. Specifically, this method is based on the Shapley value principle of cooperative game theory, which calculates the incremental influence of a certain environmental factor on the prediction result when it is added to different feature combinations, and then weights and averages the contribution values of all subsets to obtain the SHAP value of the factor. This calculation method can accurately reflect the independent action of each environmental factor in model prediction and its interaction with other factors. By statistically analyzing the SHAP values of all samples, not only the ranking list of the contribution of each environmental factor is obtained, but also the nonlinear relationship between the factor value and the SHAP value can be displayed through visual charts. The feature importance ranking bar chart intuitively presents the overall influence of each factor on carbon sink prediction, while the SHAP summary plot further reveals the variation law of each factor's contribution to carbon sink in different value intervals.

[0030] According to an embodiment of the present application, the feature importance ranking chart and the SHAP summary chart are fitted based on the LOWESS algorithm to determine the relationship curve between the value of each environmental factor and the SHAP value, and the optimal threshold interval of each environmental factor is determined according to the relationship curve, specifically: Data cleaning is performed on the feature values and SHAP values corresponding to each environmental factor in the SHAP summary chart to remove sample points containing missing values or invalid values; The cleaned feature values and SHAP values are locally weighted regression fitted using the LOWESS algorithm to generate a smooth relationship curve between the feature values and the SHAP values; The Spearman correlation coefficient between the feature values and the SHAP values of each environmental factor is calculated, and the positive and negative influence directions of the environmental factor on the vegetation carbon sink are determined according to the sign of the correlation coefficient; identify a continuous feature value interval with SHAP value greater than zero on the smoothing relationship curve, and if there are multiple continuous intervals, select the interval with the maximum mean SHAP value as the candidate optimization interval; When the length of the candidate optimization interval is less than the preset minimum ecological interval threshold, expand to the minimum ecological interval length on both sides with the median of the interval as the center to obtain the first optimization threshold interval; If there is no interval with SHAP value greater than zero on the smoothing relationship curve, select the feature value point corresponding to the global maximum SHAP value as the center to construct the minimum ecological interval with the feature value point as the center to obtain the second optimization threshold interval; Generate an optimization threshold interval visualization diagram for each environmental factor, which simultaneously displays the original feature value-SHAP value scatter plot, the LOWESS fitting curve, the SHAP zero value reference line, and the optimization threshold interval boundary marker; Compare the first optimization threshold interval or the second optimization threshold interval of each environmental factor with the original feature value distribution range, eliminate the interval boundary points that exceed the actual feature value range, and output the optimization threshold interval of each environmental factor.

[0031] It should be noted that firstly, the original data is cleaned to ensure the reliability of the analysis data, and then the LOWESS local weighted regression algorithm is used to fit the smooth relationship curve of the characteristic value and the SHAP value, effectively capturing the complex nonlinear mode of the contribution of the environmental factor to the carbon sink. The positive and negative influence direction of each factor on the carbon sink is determined by calculating the Spearman correlation coefficient. In the curve analysis stage, the continuous characteristic value interval with a SHAP value greater than zero is identified first, indicating that the environmental factor in this interval has a positive promoting effect on the carbon sink; if there are multiple effective intervals, the interval with the maximum average contribution is selected as the candidate to ensure the optimal effect. At the same time, the minimum ecological interval threshold constraint is introduced to avoid the infeasibility of ecological regulation due to the too narrow interval, and the reasonable expansion is carried out when the candidate interval does not meet the length requirement. For the case where there is no obvious positive contribution interval, the optimization interval is constructed based on the global SHAP maximum value to ensure that all factors can be reasonably optimized. Finally, through visualization verification and actual characteristic value range verification, unreasonable interval boundaries are removed to ensure that the output optimization threshold interval not only meets the optimality of the mathematical model, but also meets the feasibility and operability of the actual ecological system. The smooth relationship curve of the characteristic value-SHAP value is a continuous curve that reveals the nonlinear relationship between the specific value of the environmental factor and its corresponding SHAP contribution value through the LOWESS (local weighted regression) algorithm. This curve can reveal the influence mode of the environmental factor on the vegetation carbon sink in different value intervals, such as positive promotion, negative inhibition, or threshold effect. Compared with the discrete distribution of the original scatter plot, the overall trend between the environmental factor and the carbon sink contribution is clearly shown, such as unimodal, monotonic increasing, or complex fluctuation relationship. The rising section of the curve indicates that increasing the factor value can improve the carbon sink, and the decreasing section means that excessive increase may have an inhibitory effect. The minimum ecological threshold interval refers to the minimum effective value range of the environmental factor that can have a significant positive impact on the vegetation carbon sink in the actual ecological system.

[0032] According to the embodiment of the present application, the optimization characteristic value of the environmental factor is determined according to the optimization threshold interval, the vegetation carbon sink gain of the optimization characteristic value is predicted to obtain prediction gain data, and the optimization partition is determined according to the prediction gain data, specifically: The preset number of key environmental factors are selected from the characteristic importance ranking graph, the deviation distance of the current characteristic value of each key environmental factor in each grid of the target area from the boundary of the optimization threshold interval is calculated, and the optimization demand intensity is determined according to the deviation distance; The characteristic value corresponding to the maximum SHAP value of each key environmental factor in each grid of the target area in the optimization threshold interval is selected as the optimization characteristic value to form an optimization characteristic data set; The optimized feature data set is imported into the trained XGBoost regression model for vegetation carbon sink prediction to obtain optimized net ecosystem productivity prediction values, and the difference between the optimized and unoptimized net ecosystem productivity is calculated as the carbon sink gain potential value; If the carbon sink gain potential value is greater than the preset ratio of the unoptimized net ecosystem productivity, the environmental factor is selected as the to-be-optimized environmental factor, and the number of to-be-optimized environmental factors is counted; According to the carbon sink gain potential value, optimization demand intensity and number of to-be-optimized environmental factors of each grid in the target area, the optimization partition is determined, and the carbon sink potential gain spatial distribution map, key feature optimization demand heat map and optimization partition suggestion map are drawn according to the carbon sink gain potential value and to-be-optimized environmental factors of each optimization partition.

[0033] It should be noted that first, the key environmental factors are selected based on the feature importance ranking, the deviation distance of the current feature value from the optimization threshold interval is calculated to quantify the optimization demand intensity of each region, and the urgency of ecological regulation is reflected; at the same time, the feature value corresponding to the maximum SHAP value of each factor is selected to construct the optimization feature data set, ensuring that the optimization value has the maximum carbon sink gain effect. The optimized feature data is input into the trained XGBoost model to predict the carbon sink potential, and the difference between the optimized and unoptimized net ecosystem productivity is compared to objectively evaluate the theoretical improvement space of each region. Further combined with ecological practice demand, the environmental factors with significant gain are selected as to-be-optimized objects and the number is counted to form a complete evaluation of optimization elements. Finally, by constructing a three-dimensional decision model including carbon sink gain potential value, optimization demand intensity and number of to-be-optimized factors, the target area is divided into four levels of no optimization demand area, improvement potential area, key improvement area and key transformation area by using weighted priority algorithm, wherein the priority calculation considers carbon sink income item, ecological demand item and their interaction item to ensure that the partition result reflects both theoretical gain and practical feasibility. The generated spatial distribution map, heat map and partition suggestion map intuitively present the optimization key area and regulation direction. The priority is calculated according to the carbon sink gain potential value and the number of to-be-optimized environmental factors as follows: , wherein, is the carbon sink gain potential value, is the maximum gain potential value (the difference between the current carbon sink value and the maximum carbon sink value of the ecosystem obtained during ecological investigation), is the number of to-be-optimized environmental factors, is the maximum optimization demand quantity (i.e. the total number of the selected preset number of key environmental factors). According to the value, the area is divided into four levels: ; According to the embodiment of the present application, the vegetation carbon sink optimization suggestion report of the optimization partition is generated according to the model performance data and the optimization threshold interval of each environmental factor, in particular: The vegetation carbon sink optimization suggestion report is generated according to the model performance data and the optimization threshold interval of each environmental factor; The vegetation carbon sink optimization suggestion report includes the environmental factor to be optimized, the optimization direction, the optimization priority, the carbon sink gain potential value, the carbon sink potential gain space distribution map, the key feature optimization demand heat map and the optimization partition suggestion map of each grid in the target area.

[0034] It should be noted that the explainability analysis result of the machine learning model is converted into an operable ecological engineering guidance scheme. By integrating the model performance verification data and the environmental factor optimization threshold interval, the generated optimization suggestion report realizes complete conversion from theoretical calculation to practical application, wherein the model performance data ensures the scientific credibility of the suggestion, and the optimization threshold interval provides specific regulation parameters. The report content systematically includes a spatially explicit optimization element list, a quantitative priority order and a visualized potential distribution, so that the decision maker can intuitively identify the optimization focus of different areas: the list of environmental factors to be optimized clearly needs intervention; the optimization direction indicates the regulation trend that the feature value should be increased or decreased; the optimization priority reflects the sequence of implementation in different areas; the carbon sink gain potential value quantifies the expected ecological benefit; and the three types of thematic maps present the optimization resource allocation strategy through spatial visualization.

[0035] According to the embodiment of the present application, the method further comprises: The grid center point coordinates of each data layer are extracted from the multi-source environmental data after spatial coordinate unification, the center point offset between adjacent grids is calculated, and a grid offset distribution map is generated; Based on the grid offset distribution map, an offset compensation matrix is constructed, a bilinear interpolation algorithm is used for resampling the pixel value of the grid whose offset exceeds the threshold, and a compensated grid data layer is generated; The compensated grid data layers and the standard grid template of the target resolution are superimposed and matched, the edge pixel missing area appearing in the matching process is locally interpolated and filled, and a spatially aligned multi-source environmental data set is generated; The environmental factor values of each grid unit are extracted from the spatially aligned multi-source environmental data set, and an environmental factor matrix containing a unified spatial reference system is constructed.

[0036] According to the embodiment of the present application, the offset compensation matrix is constructed based on the grid offset distribution map, the bilinear interpolation algorithm is used for resampling the pixel value of the grid whose offset exceeds the threshold, and the compensated grid data layer is generated, in particular: According to the grid offset distribution map, the lateral and longitudinal offsets of each data layer relative to the standard grid template are calculated, and the offsets are decomposed into integer pixel offsets and decimal pixel offsets; The integer pixel offsets are compensated by using a grid overall translation strategy, and the decimal pixel offsets are compensated by constructing a bilinear interpolation weight matrix; According to the bilinear interpolation weight matrix, the original grid pixel values are calculated by weighting, to generate compensated pixel values, and the compensated grid data layer is subjected to edge pixel checking to detect invalid values or abnormal values introduced in the compensation process; According to the detection result, the invalid values are filled with the mean value of the neighboring pixels to generate the final compensated grid data layer; The compensated grid data layer is compared with the standard grid template pixel by pixel, and an alignment accuracy report is output, and the compensation parameters are iteratively optimized according to the alignment accuracy report.

[0037] It should be noted that in the research on the spatial optimization of vegetation carbon sinks, the spatial standardization processing of multi-source environmental data is a key basic link, but in actual application, due to the different collection equipment and coordinate systems used by remote sensing images and ground observation data from different data sources, after coordinate unification conversion, there are still slight misalignment phenomena in the grid boundaries of each data layer. This grid misalignment problem will directly affect the accuracy of subsequent environmental factor extraction, especially when different resolution data are fused, the offset of the edge pixels will cause the spatial matching error of key parameters such as vegetation index and terrain factor, and thus reduce the reliability of the carbon sink prediction model. The dynamic grid alignment compensation method proposed in the present application accurately calculates the offsets of each data layer relative to the standard template, and effectively solves this problem by using a hierarchical compensation strategy. The method first generates an offset distribution map through grid center point coordinate analysis, then compensates the integer and decimal pixel offsets by using overall translation and bilinear interpolation respectively, not only ensures the consistency of the data spatial position, but also ensures the integrity of the compensated data through edge pixel checking and neighborhood filling mechanism, which can significantly improve the spatial alignment accuracy of multi-source data.

[0038] Figure 4 A block diagram of a vegetation carbon sink spatial optimization system based on an interpretable machine learning is shown.

[0039] The second aspect of the present application also provides a vegetation carbon sink spatial optimization system based on an interpretable machine learning, which comprises a memory 401, a processor 402 and a communication interface 403, the memory comprises a vegetation carbon sink spatial optimization method program based on an interpretable machine learning, the communication interface is used for data connection communication between the memory and the processor, and when the vegetation carbon sink spatial optimization method program based on an interpretable machine learning is executed by the processor, the following steps are realized: Obtaining multi-source environmental data of a target region, the multi-source environmental data including administrative division data, environmental factor data, and target variable data, performing spatial coordinate unification and resolution standardization processing on the multi-source environmental data to generate a spatial table dataset; Training an XGBoost regression model according to the spatial table dataset using a cross-validation and early stopping strategy, predicting the vegetation carbon sink capacity of the target region according to the trained XGBoost regression model, and determining model performance data; Selecting a preset number of sampling points in the XGBoost regression model training process, calculating the contribution value of each environmental factor to the vegetation carbon sink based on a SHAP interpreter, and generating a feature importance ranking graph and a SHAP summary graph; Fitting the feature importance ranking graph and the SHAP summary graph based on a LOWESS algorithm to determine the relationship curve between the environmental factor value and the SHAP value, and determining the optimization threshold interval of each environmental factor according to the relationship curve; Determining the optimized feature value of the environmental factor according to the optimization threshold interval, predicting the vegetation carbon sink gain of the optimized feature value to obtain prediction gain data, and determining an optimized partition according to the prediction gain data; Generating a vegetation carbon sink optimization suggestion report for the optimized partition according to the model performance data and the optimization threshold interval of each environmental factor.

[0040] The present application discloses a vegetation carbon sink spatial optimization method and system based on interpretable machine learning. The method first obtains multi-source environmental data of a target region, including administrative division, environmental factors, and target variables, forming a spatial table dataset. Based on this dataset, an XGBoost regression model is trained to predict the vegetation carbon sink capacity and evaluate the model performance. The SHAP interpreter is used to quantify the contribution of each environmental factor to the carbon sink, generating a feature importance and interpretability graph. The LOWESS algorithm is used to fit the environmental factor and carbon sink response relationship curve, and the optimal threshold interval is extracted. Further, the carbon sink gain is predicted based on the optimization threshold to delineate the optimization partition and form a spatial optimization scheme. Finally, a vegetation carbon sink optimization suggestion report is output. This method can realize the scientific quantification of carbon sink spatial optimization and improve the interpretability and decision support value of the carbon sink improvement strategy.

[0041] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other manners. The described device embodiments are merely illustrative. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed coupling, or direct coupling or communication connection between the components can be indirect coupling or communication connection through some interfaces, devices, or units, and can be electrical, mechanical, or in other forms.

[0042] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units; they can be located in one place, or distributed on multiple network units; and some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0043] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0044] Those of ordinary skill in the art can understand that all or part of the steps of the above-described method embodiments can be completed by a program instructing related hardware, and the foregoing program can be stored in a computer readable storage medium, and when the program is executed, the steps of the method embodiments are executed; and the foregoing storage medium includes: mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks or optical disks, and various media that can store program codes.

[0045] Alternatively, the integrated units of the present application, if implemented in the form of software functional modules and sold or used as independent products, can also be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the embodiments of the present application. The foregoing storage medium includes: mobile storage devices, ROMs, RAMs, magnetic disks or optical disks, and various media that can store program codes.

[0046] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A vegetation carbon sink space optimization method based on interpretable machine learning, characterized in that, The method comprises the following steps: Obtaining multi-source environmental data of a target area, the multi-source environmental data comprising administrative division data, environmental factor data and target variable data, performing spatial coordinate unification and resolution standardization processing on the multi-source environmental data to generate a spatial table dataset; Training an XGBoost regression model according to the spatial table dataset using a cross-validation and early stopping strategy, predicting the vegetation carbon sink of the target area according to the trained XGBoost regression model, and determining model performance data; Selecting a preset number of sampling points in the XGBoost regression model training process, calculating the contribution value of each environmental factor to the vegetation carbon sink based on a SHAP interpreter, and generating a feature importance ranking graph and a SHAP summary graph; Fitting the feature importance ranking graph and the SHAP summary graph based on a LOWESS algorithm to determine the relationship curve between the environmental factor value and the SHAP value, and determining the optimization threshold interval of each environmental factor according to the relationship curve, specifically: Performing local weighted regression fitting on the cleaned feature value and the SHAP value using the LOWESS algorithm to generate a smooth relationship curve of the feature value-SHAP value; Calculating the Spearman correlation coefficient of the environmental factor feature value and the SHAP value, and determining the positive and negative influence direction of the environmental factor on the vegetation carbon sink according to the correlation coefficient sign; Identifying the continuous feature value interval with a SHAP value greater than zero on the smooth relationship curve, and selecting the interval with the maximum SHAP value mean as the candidate optimization interval if there are multiple continuous intervals; When the length of the candidate optimization interval is less than a preset minimum ecological interval threshold, expanding to the minimum ecological interval length on both sides with the interval median as the center to obtain a first optimization threshold interval; If there is no interval with a SHAP value greater than zero on the smooth relationship curve, selecting the feature value point corresponding to the global maximum SHAP value as the center to construct a minimum ecological interval with the feature value point as the center to obtain a second optimization threshold interval; Determining the optimized feature value of the environmental factor according to the optimization threshold interval, predicting the vegetation carbon sink gain of the optimized feature value to obtain prediction gain data, and determining the optimized partition according to the prediction gain data; Generating a vegetation carbon sink optimization suggestion report for the optimized partition according to the model performance data and the optimization threshold interval of each environmental factor.

2. The vegetation carbon sink spatial optimization method based on interpretable machine learning according to claim 1, characterized in that, The method comprises the following steps: Obtaining multi-source environmental data of a target area, the multi-source environmental data comprising administrative division data, environmental factor data, target variable data, performing spatial coordinate unification and resolution standardization processing on the multi-source environmental data to generate a spatial table dataset, specifically: Obtaining administrative division data of the target area, the administrative division data comprising boundary position coordinate information of the target area, and standardizing the target area into a plurality of regular grids according to a preset resolution based on the administrative division data; Acquire environmental factor data of each grid in the target area within a preset historical time period, including precipitation, temperature, height, slope, normalized vegetation index, leaf area index, soil organic carbon content, soil clay content, land use intensity data, and nighttime light data; Acquire target variable data of each grid in the target area, which is net ecosystem productivity; Convert the coordinate system of the multi-source environmental data to WGS84 geographic coordinate system, and extract the environmental factor data and target variable data in the unified coordinate system to the corresponding coordinate grid to generate spatial table data containing environmental factors and target variables.

3. The vegetation carbon sink spatial optimization method based on interpretable machine learning according to claim 1, characterized in that, Train an XGBoost regression model based on the spatial table data set using cross-validation and early stopping strategy, predict the vegetation carbon sink of the target area based on the trained XGBoost regression model, and determine the model performance data, specifically: Perform outlier analysis on the spatial table data based on the interquartile range method, calculate the first quartile and third quartile of each environmental factor variable, obtain the interquartile range value, set the outlier judgment threshold range, and mark the data less than the first quartile minus the preset multiple of the interquartile range value or greater than the third quartile plus the preset multiple of the interquartile range value as numerical outliers; Perform outlier processing on the numerical outliers, including replacing infinite values with NaN, replacing outliers with a preset quantile value, and replacing missing values with the median; Divide the spatial table data after outlier processing into 80% training set and 20% test set, initialize the XGBoost regression model and set the hyperparameters of the model, including learning rate, regularization coefficient, and maximum depth of decision tree; Import the training set and test set into the XGBoost regression model based on the k-fold cross-validation strategy for training, set the maximum number of iterations and the number of early stopping rounds in the training process, stop training when the model's prediction accuracy on the validation set does not improve within a preset number of consecutive iterations, and obtain the trained XGBoost regression model; Based on the trained XGBoost regression model, predict the vegetation carbon sink of the test set, calculate the root mean square error, mean absolute error, and determination coefficient between the predicted value and the true value, evaluate the model performance based on the root mean square error, mean absolute error, and determination coefficient, and obtain the model performance data.

4. The vegetation carbon sink spatial optimization method based on interpretable machine learning according to claim 1, characterized in that, Select a preset number of sampling points in the XGBoost regression model training process, calculate the contribution value of each environmental factor to the vegetation carbon sink based on the SHAP interpreter, and generate a feature importance ranking chart and a SHAP summary chart, specifically: Randomly extract a preset number of sample data from the training input data of the trained XGBoost regression model as a SHAP analysis sample set, and if the total number of samples is less than the preset number, extract all samples; creating a SHAP explainer object based on the XGBoost regression model, for each sample and each environmental factor, constructing a benchmark reference value through the SHAP explainer, and based on the Shapley value calculation method, evaluating the marginal contribution of the environmental factor to the model output in all feature subsets of the XGBoost regression model, and obtaining the SHAP value of the environmental factor by weighted averaging the contribution values of all feature subsets; sorting all environmental factors in descending order according to the average absolute value of the SHAP value to generate a feature importance ranking list of the contribution of each environmental factor to the prediction of vegetation carbon sink; drawing a SHAP summary plot based on the SHAP value, with the SHAP value as the horizontal axis and the environmental factor ranked by importance as the vertical axis, each point in the plot representing the SHAP value distribution of the sample on the environmental factor, and the color of the point reflecting the size of the environmental factor value of the sample; At the same time, a feature importance ranking bar chart is generated, with the average absolute SHAP value as the horizontal axis and the environmental factor name as the vertical axis, and the length of the bar representing the importance of each environmental factor.

5. The vegetation carbon sink spatial optimization method based on interpretable machine learning according to claim 1, characterized in that, Fitting the feature importance ranking chart and the SHAP summary plot based on the LOWESS algorithm to determine the relationship curve between the environmental factor value and the SHAP value, and determining the optimization threshold interval of each environmental factor according to the relationship curve, further comprising: cleaning the feature value and SHAP value corresponding to each environmental factor in the SHAP summary plot to remove sample points containing missing values or invalid values; generating an optimization threshold interval visualization plot for each environmental factor, which simultaneously displays the original feature value-SHAP value scatter plot, the LOWESS fitting curve, the SHAP zero value reference line, and the optimization threshold interval boundary markers; comparing the first or second optimization threshold interval of each environmental factor with the original feature value distribution range, removing the interval boundary points that exceed the actual feature value range, and outputting the optimization threshold interval of each environmental factor.

6. The vegetation carbon sink spatial optimization method based on interpretable machine learning according to claim 1, characterized in that, determining the optimized feature value of the environmental factor according to the optimization threshold interval, predicting the vegetation carbon sink gain of the optimized feature value, obtaining the prediction gain data, and determining the optimization partition according to the prediction gain data, specifically: selecting a preset number of key environmental factors from the feature importance ranking chart, calculating the deviation distance of the current feature value of each key environmental factor in each grid of the target area from the boundary of the optimization threshold interval, and determining the optimization demand intensity according to the deviation distance; selecting the feature value corresponding to the maximum SHAP value of each key environmental factor in each grid of the target area within the optimization threshold interval as the optimized feature value to form an optimized feature data set; importing the optimized feature data set into the trained XGBoost regression model to predict the vegetation carbon sink, obtaining the optimized net ecosystem productivity prediction value, and calculating the difference between the optimized net ecosystem productivity and the original net ecosystem productivity as the carbon sink gain potential value; if the carbon sink gain potential value is greater than a preset ratio of the original net ecosystem productivity, the environmental factor is selected as the environmental factor to be optimized, and the number of environmental factors to be optimized is counted. According to the carbon sink gain potential value of each grid of the target area, the optimization demand intensity, and the number of environmental factors to be optimized, an optimization partition is determined, and a carbon sink potential gain spatial distribution map, a key feature optimization demand heat map, and an optimization partition suggestion map are drawn according to the carbon sink gain potential value and the environmental factors to be optimized of each optimization partition.

7. The vegetation carbon sink spatial optimization method based on interpretable machine learning according to claim 1, characterized in that, The vegetation carbon sink optimization suggestion report of the optimization partition generated according to the model performance data and the optimization threshold interval of each environmental factor, specifically comprises: The vegetation carbon sink optimization suggestion report of the optimization partition generated according to the model performance data and the optimization threshold interval of each environmental factor, specifically comprises: The vegetation carbon sink optimization suggestion report includes the environmental factors to be optimized, the optimization direction, the optimization priority, the carbon sink gain potential value, the carbon sink potential gain spatial distribution map, the key feature optimization demand heat map, and the optimization partition suggestion map of each grid of the target area.

8. An explainable machine learning based vegetation carbon sink spatial optimization system, characterized in that, The vegetation carbon sink space optimization system based on explainable machine learning comprises a storage and a processor, the storage comprises a vegetation carbon sink space optimization method program based on explainable machine learning, and the vegetation carbon sink space optimization method program based on explainable machine learning is executed by the processor to realize the following steps: Obtain multi-source environmental data of a target area, the multi-source environmental data comprising administrative division data, environmental factor data, and target variable data, perform spatial coordinate unification and resolution standardization processing on the multi-source environmental data to generate a spatial table data set; According to the spatial table data set, train an XGBoost regression model using cross-validation and early stopping strategies, predict the vegetation carbon sink of the target area according to the trained XGBoost regression model, and determine model performance data; Select a preset number of sampling points in the XGBoost regression model training process, calculate the contribution value of each environmental factor to the vegetation carbon sink based on a SHAP interpreter, generate a feature importance ranking chart and a SHAP summary chart; Based on the LOWESS algorithm, fit the feature importance ranking chart and the SHAP summary chart to determine the relationship curve of the environmental factor value and the SHAP value, and determine the optimization threshold interval of each environmental factor according to the relationship curve, specifically comprising: Using the LOWESS algorithm, locally weighted regression fitting is performed on the cleaned feature values and SHAP values to generate a smooth relationship curve of the feature values and SHAP values; Calculate the Spearman correlation coefficient of the feature values and SHAP values of each environmental factor, and determine the positive and negative influence direction of the environmental factor on the vegetation carbon sink according to the sign of the correlation coefficient; Identify the continuous feature value interval with a SHAP value greater than zero on the smooth relationship curve, and if there are multiple continuous intervals, select the interval with the maximum SHAP value mean as the candidate optimization interval; When the length of the candidate optimization interval is less than the preset minimum ecological interval threshold, expand to the minimum ecological interval length from both sides with the median as the center to obtain the first optimization threshold interval; If there is no interval on the smooth relationship curve with a SHAP value greater than zero, the feature value point corresponding to the maximum global SHAP value is selected as the center, a minimum ecological interval is constructed with the feature value point as the center, and a second optimized threshold interval is obtained; According to the optimized threshold interval, the optimized feature value of the environmental factor is determined, the vegetation carbon sink gain of the optimized feature value is predicted to obtain prediction gain data, and the optimization partition is determined according to the prediction gain data; According to the model performance data, the optimized threshold interval of each environmental factor, a vegetation carbon sink optimization suggestion report of the optimization partition is generated.

9. The vegetation carbon sink spatial optimization system based on interpretable machine learning of claim 8, wherein, According to the optimized threshold interval, the optimized feature value of the environmental factor is determined, the vegetation carbon sink gain of the optimized feature value is predicted to obtain prediction gain data, and the optimization partition is determined according to the prediction gain data, specifically: From the feature importance ranking chart, a preset number of key environmental factors are selected, the deviation distance of the current feature value of each key environmental factor in each grid of the target area from the boundary of the optimized threshold interval is calculated, and the optimization demand intensity is determined according to the deviation distance; Select the feature value corresponding to the maximum SHAP value of each key environmental factor in each grid of the target area within the optimized threshold interval as the optimized feature value to form an optimized feature data set; The optimized feature data set is imported into the trained XGBoost regression model for vegetation carbon sink amount prediction to obtain an optimized net ecosystem productivity prediction value, and the difference between the net ecosystem productivity before and after optimization is calculated as a carbon sink gain potential value; If the carbon sink gain potential value is greater than a preset ratio of the net ecosystem productivity before optimization, the environmental factor is selected as a to-be-optimized environmental factor, and the number of to-be-optimized environmental factors is counted; According to the carbon sink gain potential value, the optimization demand intensity, and the number of to-be-optimized environmental factors of each grid of the target area, the optimization partition is determined, and the carbon sink potential gain spatial distribution chart, the key feature optimization demand heat map, and the optimization partition suggestion chart are drawn according to the carbon sink gain potential value and the to-be-optimized environmental factor of each optimization partition.

10. The vegetation carbon sink spatial optimization system based on interpretable machine learning of claim 8, wherein, According to the model performance data, the optimized threshold interval of each environmental factor, a vegetation carbon sink optimization suggestion report of the optimization partition is generated. According to the model performance data, the optimized threshold interval of each environmental factor, a vegetation carbon sink optimization suggestion report of the optimization partition is generated. The vegetation carbon sink optimization suggestion report includes the to-be-optimized environmental factor, the optimization direction, the optimization priority, the carbon sink gain potential value, the carbon sink potential gain spatial distribution chart, the key feature optimization demand heat map, and the optimization partition suggestion chart of each grid of the target area.

Images