Method and device for determining threshold value of lake ecosystem in arid region and electronic equipment

Abstract

The invention discloses a method and device for determining a threshold value of a lake ecosystem in an arid region and electronic equipment, and relates to the technical field of water environment treatment. The method comprises the following steps: based on remote sensing data of a to-be-studied ecosystem, extracting a time sequence matrix comprising a lake area and a multi-dimensional ecological index of the to-be-studied ecosystem; determining a plurality of key ecological indexes based on the time sequence matrix, and constructing a multi-response random forest model; quantizing the driving effect of the lake area on each key ecological index by using a multi-response random forest model, and revealing the nonlinear response relationship between the lake area and each key ecological index; and further based on a nonlinear response relationship, identifying a sensitive response interval of the lake area, and determining a threshold value. According to the method, key response variables and sensitive response intervals can be automatically identified, and the ecological regulation and control precision is improved; a traditional static single-index method is replaced, a dynamic'optimal area point 'can be quantitatively given, and a target value with high operability is provided for actual reservoir regulation and control and ecological water scheduling.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water environment treatment, and in particular to a method and device for determining a threshold of a lake ecosystem in a dry region and an electronic device. BACKGROUND

[0002] Closed endorheic basins account for about one-third of the global land area, and play an important role in climate regulation, biodiversity protection, and desertification prevention. In arid and semi-arid regions, inland lakes, especially tail-lag lakes, as important terminal collectors of water resources and eco-hydrological processes in a watershed, are in a relatively independent water cycle system, and are particularly sensitive to climate change and human activities. As the final receptor of a closed water system, the expansion or shrinkage of tail-lag lakes not only directly reflects the changes in the water inflow process of a watershed, but also directly affects the regional ecological safety and environmental stability, and thus is regarded as an important indicator for measuring the health status of regional water resources utilization and ecosystems. In recent years, with the intensification of climate variability and the increasing strength of human regulation, the area of tail-lag lakes in the global arid region has shown a significant dynamic change trend, and is facing a severe ecological crisis. The expansion of cultivated land and the overexploitation of water resources have led to a series of ecological problems such as river cutoff, lake shrinkage, water quality deterioration, land desertification, grassland degradation, land salinization, and salt-sand storm.

[0003] A "threshold" refers to a point or interval at which a system or a state of matter undergoes a dramatic change. Different ecosystems have ecological threshold phenomena for different ecological factors. The term "threshold" in ecology generally refers to a critical point or interval, when the area of a lake (such as temperature, water level, pollution, and land use intensity) reaches or exceeds this point, the state of the ecosystem will change significantly, rapidly, and even irreversibly. Before this point, the response of the system may be slow, linear, or stable, while once the point is crossed, the system will enter another stable state, showing nonlinear, hysteresis, mutation, and even collapse. When the area of a lake gradually changes, the internal regulation mechanism of the system may be gradually weakened or destroyed, leading to the failure of feedback mechanisms: such as when the water level of a wetland drops below a certain threshold, causing plant death, which in turn exacerbates water loss; state transition is irreversible: such as forest degradation into shrubs or grassland, which is difficult to recover in the short term; system service mutation decreases: such as lake area reduction leading to water quality deterioration and ecological function collapse; ecological response is nonlinear: in some critical intervals, ecological indicators become extremely sensitive to external pressure. Table 1 shows some specific cases.

[0004] Table 1

[0005]

[0006] The study of "threshold" has been widely carried out in forest, grassland, lake, ocean and other ecosystems, especially to evaluate the ecological breakpoints (thresholds) in land use, landscape pattern, biological community and ecosystem services. Therefore, the response relationship between ecological driving factors and lake surface response can be quantified by the position of the breakpoint (threshold).

[0007] Most of the current research on the ecological suitable water level of lakes still relies on static hydrological indicators (such as minimum ecological water level or historical water balance), often ignoring the nonlinear response characteristics of key ecological indicators to lake area changes. For example, in arid regions, insufficient lake area may trigger a chain of ecological disasters, including lake salinization, salt dust storms, and biodiversity collapse, while excessive lake area may exacerbate the water conflict between ecological protection and social and economic development. This ecological paradox highlights the need to build a systematic threshold identification framework to quantify the nonlinear response of key ecological indicators (such as vegetation dynamics, wetland connectivity, and landscape pattern evolution) to lake area changes, and then to build a suitable lake surface threshold identification framework. This is of great significance for achieving ecological water demand management, optimizing water resources allocation, and improving the resilience of lake systems.

[0008] In recent years, the progress of remote sensing technology has greatly improved the ability to monitor lakes. Through water body index (such as NDWI), machine learning classifier, and Otsu automatic threshold segmentation method, lake surface area can be extracted at high frequency. Although these methods have been widely used in reconstructing lake multi-year change trends and analyzing climate and human lake area, there are still some key gaps: (1) Most studies remain at the annual scale, making it difficult to capture seasonal or monthly changes that are critical to ecology; (2) In the face of mixed pixel problems caused by dynamic shorelines, spectral unmixing methods have not been fully utilized; (3) In the assessment of regional ecological effects, the comprehensive response of ecological indicators is rarely considered. In addition, in arid terminal lakes (such as Ebinur Lake in Xinjiang, China), intermittent water pulses are disproportionately important to ecosystem resilience. Although some scholars have determined the water surface area threshold of terminal lakes in arid regions based on the balance between ecological safety and water use efficiency, and analyzed the suitable area of terminal lakes under the constraint of water diversion scheme based on energy theory in arid regions. However, there is still a lack of research on the suitable area of terminal lakes under the mechanism of ecological effects, and a comprehensive research framework for the most suitable lake surface under the nonlinear response of ecological indicators has not been formed. SUMMARY

[0009] To solve the problems in the prior art, the present application provides the following technical solutions.

[0010] The first aspect of the present application provides a method for determining the threshold of a lake ecosystem in an arid region, comprising:

[0011] Based on remote sensing data of the to-be-studied ecosystem, a time series matrix including a lake area of the to-be-studied ecosystem and multi-dimensional ecological indicators is extracted; wherein the multi-dimensional ecological indicators are pre-selected based on remote sensing data and ecological theory, and the multi-dimensional ecological indicators respond to changes in the lake area;

[0012] Based on the time series matrix, a plurality of key ecological indicators are determined, and a multi-response random forest model is constructed; wherein the key ecological indicators are ecological indicators with a response degree to changes in the lake area higher than a preset value; and the multi-response random forest model is used to jointly model and fit the plurality of key ecological indicators with the lake area as an input variable;

[0013] Based on the fitting result of the multi-response random forest model, a nonlinear response relationship between the lake area and each key ecological indicator is quantified;

[0014] Based on the nonlinear response relationship, a sensitive response interval of the lake area is identified, and a threshold value is determined.

[0015] Preferably, the construction of the multi-response random forest model comprises:

[0016] The data used for modeling is divided into lake area data and key ecological indicator data;

[0017] The key ecological indicator data is dynamically extracted, and a multi-response formula Multivar(y1, y2,...,yn)~x is automatically constructed; wherein x is the lake area; y1, y2,...,yn are n key ecological indicators;

[0018] Based on the multi-response random forest modeling tool and the multi-response formula, a multi-response random forest model is constructed, and the average prediction error rate, block error rate, importance, determination coefficient, and root mean square error of each key ecological indicator are calculated respectively, for fine-grained diagnosis of the multi-response random forest model;

[0019] The variance of the key ecological indicators is taken as a weight to construct a weighted determination coefficient evaluation index; the weighted determination coefficient evaluation index is used to quantify the overall explanatory ability of the lake area to the to-be-studied ecosystem.

[0020] Preferably, based on the time series matrix, a variable screening method is used to determine the plurality of key ecological indicators.

[0021] Preferably, the quantification of the nonlinear response relationship between the lake area and each key ecological indicator comprises: using an improved feature contribution degree explanation method to quantify the response strength and direction of the lake area to each key ecological indicator.

[0022] Preferably, the improved feature contribution degree explanation method is implemented in the following manner:

[0023] construct a single-response random forest model based on the key ecological indicators with nonlinear response relationship and the lake area;

[0024] For any type of single-response random forest model, the prediction data is output through a unified prediction interface;

[0025] Based on the prediction data, the feature contribution degree is calculated and used as a SHAP value to explain the driving effect of the lake area on the key ecological indicators.

[0026] The second aspect of the application provides a device for determining a threshold value of a lake ecosystem in an arid region, comprising:

[0027] A data acquisition module is configured to extract a time series matrix including the lake area and multi-dimensional ecological indicators of the ecosystem to be studied based on remote sensing data of the ecosystem to be studied; wherein the multi-dimensional ecological indicators are preselected based on remote sensing data and ecological theory, and the multi-dimensional ecological indicators respond to changes in the lake area;

[0028] A model construction module is configured to determine a plurality of key ecological indicators based on the time series matrix and construct a multi-response random forest model; wherein the key ecological indicators are ecological indicators with a response degree to the lake area change higher than a preset value; and the multi-response random forest model is used to model and fit the plurality of key ecological indicators with the lake area as an input variable;

[0029] A nonlinear response relationship determination module is configured to quantify the nonlinear response relationship between the lake area and each key ecological indicator based on the fitting result of the multi-response random forest model;

[0030] A threshold value determination module is configured to identify a sensitive response interval of the lake area based on the nonlinear response relationship and determine a threshold value.

[0031] Preferably, in the model construction module, the construction of the multi-response random forest model includes: dividing the data for modeling into lake area data and key ecological indicator data; dynamically extracting the key ecological indicator data to automatically construct a multi-response formula Multivar(y1, y2,..., y n ) ~ x; wherein x is the lake area; y1, y2,..., y nThe n key ecological indicators are determined based on the time series matrix; a multi-response random forest model is constructed based on a multi-response random forest modeling tool and the multi-response formula, and the average prediction error rate, block error rate, importance, coefficient of determination and root mean square error of each key ecological indicator are calculated respectively to perform fine-grained diagnosis on the multi-response random forest model; and a weighted coefficient of determination evaluation index is constructed by taking the variance of the key ecological indicators as the weight, and the weighted coefficient of determination evaluation index is used to quantify the overall explanatory ability of the lake area to the ecological system to be studied.

[0032] Preferably, in the nonlinear response relationship determination module, based on the time series matrix, a plurality of key ecological indicators are determined by using a variable screening method.

[0033] The third aspect of the present application provides a memory storing a plurality of instructions for implementing the method for determining the threshold value of the lake ecological system in the arid region as described in the first aspect.

[0034] The fourth aspect of the present application provides an electronic device comprising a processor and a memory connected to the processor, wherein the memory stores a plurality of instructions which can be loaded and executed by the processor to enable the processor to perform the method for determining the threshold value of the lake ecological system in the arid region as described in the first aspect.

[0035] The beneficial effects of the present application at least include: the method, device and electronic device for determining the threshold value of the lake ecological system in the arid region provided by the present application combine multi-dimensional response with nonlinear modeling, which can truly reflect the complex adaptive mechanism of the ecological system to the fluctuation of the water area; automatically identify key response variables and sensitive response intervals, improve the accuracy of ecological regulation; have high interpretability and migratory ability, and can be applied to terminal lakes in the arid region (such as Bosten Lake and Lop Nur); replace the traditional static single-index method, and can serve the scientific allocation of water resources and the setting of ecological red line at the basin scale; can quantitatively give the "optimal area point", and provide an operable target value for actual reservoir regulation and ecological water regulation. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 It is a flowchart of the method for determining the threshold value of the lake ecological system in the arid region according to the present application;

[0037] Figure 2 It is a schematic diagram of the nonlinear trend of the lake area and key ecological indicators based on SHAP according to the embodiment of the present application;

[0038] Figure 3 It is a schematic diagram of the inflection point detection and optimal contribution interval display based on the SHAP dependence relationship curve according to the embodiment of the present application;

[0039] Figure 4A schematic diagram of the functional structure of the device for determining the threshold of the lake ecosystem in the arid region according to the present application. DETAILED DESCRIPTION

[0040] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in combination with the drawings in the specification and specific embodiments.

[0041] The method provided by the present application can be implemented in a terminal environment, which can include one or more of the following components: a processor, a memory and a display screen. The memory stores at least one instruction, which is loaded and executed by the processor to implement the method described in the following embodiments.

[0042] The processor can include one or more processing cores. The processor connects various parts in the entire terminal through various interfaces and lines, executes various functions of the terminal and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory, and calling data stored in the memory.

[0043] The memory can include random access memory (RAM) and read-only memory (ROM). The memory can be used to store instructions, programs, codes, code sets or instructions.

[0044] The display screen is used to display the user interface of each application program.

[0045] In addition, those skilled in the art can understand that the structure of the terminal described above does not constitute a limitation on the terminal, and the terminal can include more or fewer components, or combine certain components, or different component arrangements. For example, the terminal also includes radio frequency circuit, input unit, sensor, audio circuit, power supply and other components, which will not be described here.

[0046] In the related art, lake ecological threshold identification mainly adopts a static single variable method, such as a method based on ecological water demand, a water level-biological index relationship or historical extreme value analysis. These methods usually have the following shortcomings: ignoring the response relationship: the traditional method often assumes that there is a linear relationship between the ecological variable and the lake area, and cannot identify the mutation or nonlinear sensitive response interval of the ecological system near certain threshold points; single variable, poor adaptability: most methods only consider a certain type of ecological variable (such as vegetation or wetland distribution), and lack of overall understanding of the multi-dimensional structure of the ecological system (such as landscape pattern, water and salt stress, drought index, etc.); lack of dynamic timing support: most existing researches are based on annual scale or fixed time point analysis, ignoring the timing fluctuation characteristics of lake area and its ecological response; weak model interpretability: although some researches introduce machine learning methods can improve the prediction accuracy, but cannot provide interpretable information of the response strength and interval of ecological indicators to lake area change, limiting the practical application in ecological regulation.

[0047] In view of the above problems in the related art, the purpose of the present application is to: by revealing the nonlinear response interval of lake area change to multiple indicators of the ecological system, identifying the "optimal area point", providing quantitative threshold support for ecological water demand control, water resource scheduling, lake area degradation early warning, and providing scientific identification basis for lake area threshold in lake ecology; by constructing a transferable identification framework, providing scientific decision-making basis for ecological priority scheduling and protection red line demarcation of water resources in arid regions in the coupling regulation of lake-wetland-basin system, supporting the ecological water resource management in arid regions; and constructing a future-oriented monitoring and evaluation method.

[0048] In order to achieve the above purpose, in the present application, a transferable lake ecological threshold identification framework in arid regions is constructed; the area-ecological coupling relationship of lakes in arid regions on a monthly scale is revealed, which has guiding significance for ecological water resource management; the RFSRC-SHAP method is used to identify complex ecological trade-off mechanism, which breaks through the traditional single variable ecological threshold identification paradigm. By combining GEE cloud computing and machine learning method, an extensible and automated "remote sensing-ecological threshold identification-management response" technical chain is established, which improves the dynamic evaluation and management ability of the ecological system in arid regions. The variable screening method is used to accurately identify the key ecological indicators. The multi-response random forest (MRF) model is used to consider the joint response of multiple ecological factors to area, which improves the identification accuracy. At the same time, an automatic and transferable efficient identification process is constructed. By improving the feature contribution degree explanation method, the response interval and strength of key ecological indicators to area change are quantified, and the interpretability of the model is improved.

[0049] Embodiment one

[0050] As Figure 1As shown, the embodiment of the present application provides a method for determining the threshold of a lake ecosystem in an arid region, comprising the following steps:

[0051] In step S101, remote sensing data of the ecosystem to be studied is obtained, and a time series matrix including the lake area and multi-dimensional ecological indicators of the ecosystem to be studied is extracted; wherein the multi-dimensional ecological indicators are preselected based on remote sensing data and ecological theory, and the ecological indicators respond to the change of the lake area;

[0052] In step S102, a plurality of key ecological indicators are determined based on the time series matrix, and a multi-response random forest model is constructed; wherein the key ecological indicators are ecological indicators with a response degree to the change of the lake area higher than a preset value; and the multi-response random forest model is used to jointly model and fit the plurality of key ecological indicators with the lake area as an input variable;

[0053] In step S103, based on the fitting result of the multi-response random forest model, the nonlinear response relationship between the lake area and each key ecological indicator is quantified.

[0054] In step S104, based on the nonlinear response relationship, the sensitive response interval of the lake area is identified, and the threshold is determined.

[0055] In the embodiment of the present application, the ecosystem to be studied is a lake ecosystem in an arid region. It can include any lake ecosystem in an arid region in the world. For example, a global arid region tail lake, the Ebinur Lake in Xinjiang, China.

[0056] In step S101, the remote sensing data of the ecosystem to be studied can be obtained in the following way:

[0057] Using multi-source remote sensing satellite data, all Landsat 5TM, Landsat 7TM, Landsat 8OLI and Sentinel-2 images covering the entire study area from April to October from 2000 to 2024 are screened. Using the water body spectral unmixing method, a monthly lake area sequence from 2000 to 2024 is generated; wherein the monthly data before 2019 is mainly calculated using Landsat 5, 7 and 8, and the semi-monthly data from 2019 to 2024 is mainly calculated using Sentinel-2 to improve the monitoring accuracy and frequency in recent years.

[0058] The ecological indicators corresponding to the time sequence of the lake area are calculated. Among them, the surface temperature and evapotranspiration indicators are extracted by MOD16A2. The landscape indicators after 2000-2016 are calculated by the land cover data generated by random forest classification. The landscape indicators after 2020 are calculated under the same land cover classification by using the Dynamic World v1 dataset. Dynamic World v1 belongs to the existing known dataset, which is a global high-resolution (10m) dynamic land cover dataset jointly developed by Google and WRI, which uses Sentinel-2 satellite images and deep learning models to generate in real time.

[0059] In order to identify the key ecological indicators affected by the change of lake area, combined with the ecological conditions of the study area and related research, the dry and wet indicators (DSI, NDMI, etc.), salinization indicators (NSI, SI, etc.), vegetation indicators (NDVI, SRI, etc.), landscape indicators (SHDI), evapotranspiration indicators (ET or PET), wind erosion indicators (WEI), sand dust indicators (NDDI) and ecological indicators (CERI, etc.) are selected as multiple ecological response variables, forming a time sequence matrix including lake area and multiple ecological indicators. An example of the time sequence matrix is shown in Table 2.

[0060] Table 2

[0061] Time (months) lake area (km 2 )]]> NDVI DSI NDDI SHDI CERI 2000-01 382.1 0.18 0.62 0.34 1.20 0.75 2000-02 378.4 0.20 0.59 0.36 1.18 0.72 2000-03 370.9 0.23 0.55 0.39 1.21 0.70 ... ... ... ... ... ... ... 2024-12 420.5 0.27 0.48 0.29 1.15 0.81

[0062] In step S102, unsupervised screening can be performed first, including removing variables such as sample names, variables according to missing ratio, and zero variance variables. Then, combined with the data characteristics, i.e. the time sequence matrix including the lake area and multiple ecological indicators of the ecosystem to be studied, the Boruta algorithm is used to screen the key ecological indicators from multiple ecological indicators. Among them, the key ecological indicators can be understood as the ecological indicators with a response degree to the change of lake area higher than a preset value. Among them, the preset value can be set by experience or actual demand. The Boruta algorithm can handle the nonlinear characteristic relationship in the data, and determine the significance of the characteristics by comparing the importance of the original characteristics and the randomly generated shadow characteristics. With the help of random forest training model. Random forest regression prediction method is a kind of ensemble learning method based on decision tree. It includes building and combining multiple decision trees to realize regression prediction. The construction of decision tree contains randomness, including random selection of features and samples for training. This random process helps to reduce the risk of model overfitting and effectively capture the complex relationships and patterns in the data.

[0063] Multivariate Random Forest (MRF) is an extension of traditional Random Forest (RF) to predict multiple response variables simultaneously. Bootstrap resampling is used to build multiple trees, and in each node: a set of candidate features (mtry) is randomly selected, for each candidate split point of each feature, the joint mean squared error (Joint MSE) is calculated to select the split point with the minimum joint MSE, all response variables share the split structure of each tree, but each response variable can have its own prediction value, and finally the average (regression) of each response variable is taken in all trees.

[0064] In an embodiment of the present application, the following deficiencies exist in the original Multivariate Random Forest model: the model construction process is complex and cumbersome: the modeling formula for multiple response variables needs to be manually formulated, especially when there are many variables, it is not convenient to construct the Multivar() function structure; the model performance evaluation is single: the traditional method only supports the determination coefficient or error evaluation of a single response variable, and cannot reflect the overall modeling performance or the coordination of multi-objective fitting; the result is not interpretable: the index weight, feature importance and prediction error result cannot be effectively integrated, and the comparative analysis is insufficient. The following improvements are made:

[0065] 1. Automatic variable separation and pair identification: In the input data frame, the variable grouping is completed by automatically identifying the lake area and key ecological indicators (such as DSI, NDVI, etc.).

[0066] 2. Automatic construction of multivariate formula structure: Multivar(DSI, NDVI, SI, SHDI, NDDI, SRI) ~ lake area.

[0067] 3. Global weighted determination coefficient index design: multi-output evaluation mechanism, fully reflects the fitting coordination between modeling quality and response variables.

[0068]

[0069] In the formula, R2 represents the weighted determination coefficient, i.e. the goodness of fit of the overall model of multiple response variables, n represents the number of key ecological indicators (response variables), R2i represents the determination coefficient of the ith response variable 2 Var(Y i ) represents the variance of the ith response variable.

[0070] In an embodiment of the present application, the sample variance of each response variable Y k on the original data set can be calculated according to the following formula:

[0071]

[0072] in, For variable Y k The sample mean, Let be the value of the i-th sample. This variance reflects the overall volatility of the ecological indicator and serves as an important basis for assessing the explanatory power of the main controlling factor (lake area) for a multi-response variable system.

[0073] In one embodiment of the present invention, a grid search method is used to optimize the hyperparameters of an improved multi-response random forest model: grid search is a method that tries all combinations of hyperparameters within a given search space and finally finds the optimal combination of hyperparameters.

[0074] In one embodiment of the present invention, a multi-response random forest model can be constructed using the following method:

[0075] The data used for modeling is divided into lake area data and key ecological indicator data. Key ecological indicator data is dynamically extracted, and a multi-response formula Multivar(y1, y2, ..., yn) ~ x is automatically constructed. Here, x is the lake area; y1, y2, ..., yn are n key ecological indicators. This multi-response formula uses the independent variable x (lake area) as the explanatory variable and jointly models multiple response variables y1, y2, ..., yn (key ecological indicators). In this formula, the symbol "Multivar()" indicates the joint modeling of multiple response variables; y1, y2, ..., yn are n key ecological indicators, such as the drought index (DSI) and vegetation index (NDVI); the symbol "~" indicates "explained by..." or "a functional relationship about...". A multi-response random forest model was constructed based on the randomForestSRC package and the multi-response formula. The average prediction error rate, block error rate, importance, coefficient of determination, and root mean squared error (RMSE) of each key ecological indicator were calculated to provide fine-grained diagnosis of the multi-response random forest model. The variance of the key ecological indicators was used as weights to construct a weighted coefficient of determination evaluation index. The weighted coefficient of determination evaluation index is used to quantify the overall explanatory power of the lake area for the ecosystem under study.

[0076] In step S103, after constructing a multi-response random forest model using sample data including lake area and key ecological indicators, the nonlinear response relationship between lake area and each key ecological indicator is quantified based on the fitting results of the multi-response random forest model.

[0077] In an embodiment of the present application, the improved feature contribution explanation method (SHAP method) is used to quantify the response strength and direction of each key ecological indicator with a nonlinear response relationship to the lake area.

[0078] The improved SHAP method can be implemented in the following way: based on the key ecological indicators with nonlinear response relationship and the lake area, a single-response random forest model is constructed; for any type of single-response random forest model, the prediction data is output through a unified prediction interface; the feature contribution is calculated based on the prediction data, and the feature contribution is used as a SHAP value (Shapley Additive exPlanations) for explaining the driving effect of the lake area on the key ecological indicators.

[0079] In an embodiment of the present application, the improved SHAP method is used to quantify the response strength and sensitive interval of each ecological indicator to the change of the lake area, and to identify the optimal response interval and the most suitable area point of the ecological system to the lake area. The SHAP method is used to explain the prediction made by the machine learning model. The calculation method of the SHAP value is to use various feature values for prediction and evaluate the influence of each specific feature value on the prediction result. This helps to explain the prediction significance of each feature in the model and the contribution of different feature values to the overall prediction result. In the embodiment of the present application, the existing SHAP method is improved as follows: the existing SHAP framework (such as iml: :Shapley) cannot directly process the randomForest: :randomFores object. In the embodiment of the present application, the shap4reg_onevar() function is constructed to bypass the model structure incompatibility problem and create a SHAP visualization structure for tree models; at the same time, the predfunc() custom interface is used to solve the prediction function adaptation problem, improving the model compatibility.

[0080] In step S104, the threshold value of the lake area is calculated based on the nonlinear response relationship between the lake area and each key ecological indicator. The following implementation can be used:

[0081] (1) Integrate the SHAP value response curve to extract the optimal response interval of the ecological indicator; variable support count and cumulative weighted score, a certain lake area value x, in how many variable optimal interval ranges.

[0082] The cumulative sum of integral values of all ecological indicator variable intervals in which a certain lake area value falls (reflecting the comprehensive support strength of the ecological indicator to the lake area).

[0083]

[0084] where CumScore(x) is the cumulative score; I is the indicator function (1 if the condition is true, 0 otherwise); With are the upper and lower bounds of the optimal integral interval of the i-th ecological indicator variable; n represents the number of key ecological indicators (response variables). Integral (i) is the integral area (contribution strength) of the i-th ecological indicator variable within its optimal interval.

[0085] (2) Cross-Correlation Function (CCF) method

[0086] To analyze the time nonlinear response relationship between lake area change and multiple source ecological indicator variables, the Cross-Correlation Function (CCF) method is used in the embodiments of the present application to quantify the linear correlation degree of two time series under different lag conditions. CCF can reveal the advance or lag effect of one variable relative to another variable in time, and identify the most significant response lag period and direction.

[0087]

[0088] where k ∈ [-K, +K] is the lag period; is the mean of the two sequences; when k > 0, it means that X leads Y; when k < 0, it means that X lags behind Y. N is the total sample length (number of observations) of the time series.

[0089] Example: Suppose X t represents the lake area time series, Y t represents the NDVI time series, and the lag range is set to k = -3 to +3 (unit: month), and the CCF(k) is obtained by calculating the above formula:

[0090] If CCF(k) takes the maximum value at k = +1, it means that in the formula X t+1 has the strongest correlation with Y t , that is, the lake area change leads the NDVI by 1 month;

[0091] If CCF(k) is maximum at k = -2, it means that in the formula X t-2 has the highest correlation with Y t , indicating that the lake area change lags behind the NDVI by 2 months. Specific embodiments:

[0093] The embodiment of the application takes the Ebinur Lake in Xinjiang as an example, and constructs a multi-dimensional ecological threshold identification framework integrating multi-source remote sensing, time series modeling and explanatory machine learning. As a typical tail-lake in arid areas, the Ebinur Lake has reduced by about 60% since the 1950s, and has caused serious ecological crisis.

[0094] With the aid of the cloud computing capability of Google Earth Engine, the embodiment of the application uses Landsat-5 / 7 / 8 and Sentinel-2 images to construct monthly and semi-monthly lake area sequences from 2000 to 2024, among which semi-monthly data from 2019 to 2024 are generated using Sentinel-2 data; a multi-dimensional ecological time series matrix composed of dryness index, salinization index, vegetation index, landscape index, evapotranspiration index, ecological index, wind erosion index and dust index is constructed, monthly land cover data is generated through random forest classification, and landscape pattern and ecological index time series are supplemented; based on the variable screening method, variables with high response degree to lake area are found, a multi-response random forest model is constructed, the improved SHAP explanation random forest method is used to quantify the optimal response interval of each ecological index, and the response curve integral maximization method is used to determine the suitable range, and finally the suitable threshold of the lake is determined.

[0095] In the embodiment of the application, the Boruta algorithm and the multi-response random forest (MRF) model are further introduced to identify key ecological variables, and the SHAP method is used to quantify the ecological response intensity and nonlinear sensitive interval. Research shows that the change of lake area significantly drives the processes of regional drought stress, landscape fragmentation and salinization stress, among which DSI, NDDI, SHDI, SRI and NDVI are key response variables; the optimal response interval of ecological index to lake area is concentrated in 310-480km 2 , 418km 2 is the point of the strongest ecological regulation effect. The application breaks through the traditional single-variable static identification paradigm and proposes a transferable multi-dimensional ecological threshold identification path, which has important reference value for lake ecological management and water resource regulation in arid areas.

[0096] The results obtained by the specific embodiment are as follows:

[0097] 1. Model construction and performance evaluation

[0098] Multivariate Random Forest (MRF) method was used to model multiple ecological variables including drought index, landscape pattern index, vegetation index, etc. to evaluate their response strength and importance to lake area. Table 3 shows six variables that show high responsiveness in modeling, and R-squared (determination coefficient), RMSE (root mean square error) and variable importance rank are used as comprehensive evaluation indicators of model performance and variable explanation ability.

[0099] The results show that lake area change has a significant driving effect on multiple ecological variables. Among them, DSI (drought sensitivity index) and NDDI (sand dust index) R 2 Both reached 0.5731, ranking the top two in variable importance, indicating that lake dynamics has a core regulatory effect on drought and dust. SRI and SHDI represent soil salinity and landscape structure change, and their explanation and stability in the model are also good. In addition, SI and NDVI reflect soil salinity and vegetation coverage, although R 2 is relatively low, but the RMSE is the smallest, and the impact of lakes on vegetation is more limited to the edge area. Overall, lake change shows a strong coupling relationship with ecological factors such as drought, water salinity, and landscape diversity. The constructed MRF model can effectively capture the multi-dimensional ecological response caused by lake area change and show good generalization ability and prediction stability in different ecological dimensions.

[0100] Table 3 Modeling accuracy of key ecological indicators

[0101] Variables R-squared RMSE Importance rank DSI 0.5731 0.0881 1 NDDI 0.5731 0.0881 2 SHDI 0.4489 0.1343 4 NDVI 0.3391 0.0216 7 SRI 0.4992 0.0938 3 SI 0.4328 0.0452 5 Overall Weighted 0.5114 0.0785

[0102] 2. SHAP explainability analysis

[0103] To enhance the explainability of the Multivariate Random Forest model, the invention improves the SHAP analysis method to quantitatively model and visualize the contribution mechanism of lake area (Lakes) to key ecological indicator variables. Figure 2The SHAP value change curve of Lakes on the six ecological variables of DSI, NDDI, SHDI, NDVI, SRI and SI is shown in FIG. 1, and the red smooth line is the fitting trend of local weighted regression, which is used to reveal the nonlinear response characteristics of the variable. The SHAP value of DSI increases significantly with the increase of lake area, showing a significant positive driving relationship. The expansion of the lake may regulate regional water cycle by increasing the humid environment and the evaporation amount of the area, thereby inhibiting drought; the increase of lake area is negatively correlated with the degree of sand dust (NDDI), indicating its ecological regulation effect of mitigating the risk of desertification; the medium lake area is most conducive to vegetation growth, and further expansion may reduce NDVI due to wetland flooding or water salt stress; the SHAP value of SRI continues to decrease with the increase of lake area, and the expansion of the lake is helpful to flush the salt on the surface of the soil, thereby reducing the saline-alkali stress; the response trend of SI is similar to that of SRI, showing a nonlinear monotone decrease.

[0104] 3. Inflection point detection and optimal contribution interval identification based on SHAP Dependence curve

[0105] Figure 3 The inflection point detection and optimal contribution interval identification (integral maximum interval) based on the SHAP Dependence curve are shown in FIG. 2. The identified inflection points (P1, P2...) are marked, and the interval with the maximum integral of the SHAP contribution value of each variable is highlighted in green background, so as to identify the nonlinear sensitive zone of the influence of the change of lake area (Lakes) on the key ecological index variable. Among them, the increase of lake area in the medium interval has the strongest effect on drought relief, and the marginal effect decreases when the lake area is too large or too small; the sensitive interval of NDDI is small, but the contribution continues to decrease, indicating that the change of the lake area in a small range can significantly affect the wind erosion and desertification process; there are multiple inflection points in SHDI, and the change is frequent. The expansion of the lake will break the original landscape pattern, and SHDI will decrease sharply, which prompts that the lake area needs to be accurately regulated in this interval to maintain the stability of the ecological pattern; the expansion of the lake area is conducive to the growth of vegetation, and too large or too small is not conducive to the maintenance of vegetation; the expansion of the lake area can effectively regulate the regional salt, but the regulation ability will weaken after exceeding a certain critical point. The specific inflection point position and interval are shown in Table 4. Figure 3 and Table 4.

[0106] Table 4

[0107] Variables Optimal interval / km 2 ]] Maximum integral value DSI [306.37，896.47] 340.6053 NDDI [263.31，306.37] -17.0379 SHDI [100.63，274.47] 171.3142 NDVI [174.00，338.27] 6.1549 SRI [309.56，896.47] 194.7942 SI [100.63，418.01] 106.4458

[0108] 4. Comprehensive identification of ecological threshold interval

[0109] Based on the SHAP analysis results of the six key response variables, the present application comprehensively considers the sensitive response interval and integral response intensity of each variable, and constructs a lake area-ecological index variable coupling integral model. The results show that when the lake area is 310-480 km2 In the interval, most ecological variables show a synergistic response. Among them, 310km 2 is identified as the minimum area threshold of variable response activation, while 418km 2 is the optimal area point with the highest response integral superposition, representing the "optimal area" of the lake regulating ecosystem function.

[0110] The embodiment of the present application faces the typical tail-lake in the arid region, and constructs a lake ecological threshold identification framework integrating remote sensing multi-source data, nonlinear modeling and explanatory analysis. The results show that the ecological system produces complex responses in multiple dimensions such as dry and wet conditions, vegetation coverage, salt stress, landscape structure and wind erosion intensity. On this basis, combined with Boruta variable screening and multi-response random forest (MRF) modeling, the response intensity and importance of key ecological variables are quantified, and the results show that the aridity index, dust index, landscape diversity and salt stress index are most sensitive to lake area changes. With the help of SHAP method, the model interpretability is enhanced, and the nonlinear contribution interval and ecological regulation efficiency of each ecological variable under the change of lake area are clarified. Based on the multi-variable integral response results, the "optimal lake area threshold interval" of Ebinur Lake ecosystem is identified as 310-418km 2 , among which 418km 2 is the optimal area point with the strongest ecological response.

[0111] Overall, the multi-dimensional threshold identification framework proposed in the present application has good generalization ability and interpretability in the ecological management of lakes in arid regions, providing a scientific basis for ecological water replenishment, lake regulation and comprehensive management of the basin, and breaking through the problem of single variable and one-sided response in traditional ecological threshold research. It has important theoretical and practical value for the steady-state regulation of arid region ecosystems.

[0112] Embodiment two

[0113] As Figure 4 shown, another aspect of the present application also includes a functional module architecture completely consistent with the foregoing method flow, that is, the present application embodiment also provides a determination device for the threshold of the lake ecosystem in the arid region, comprising:

[0114] The data acquisition module 401 is used for extracting a time series matrix including the lake area of the to-be-studied ecosystem and multi-dimensional ecological indicators based on the remote sensing data of the to-be-studied ecosystem; wherein the multi-dimensional ecological indicators are preselected based on the remote sensing data and ecological theory, and the ecological indicators respond to the change of the lake area.

[0115] The model construction module 402 is configured to determine a plurality of key ecological indexes based on the time series matrix, and construct a multi-response random forest model; wherein the key ecological index is an ecological index with a response degree to lake area change higher than a preset value; and the multi-response random forest model is configured to take the lake area as an input variable, and jointly model and fit the plurality of key ecological indexes.

[0116] The nonlinear response relationship determination module 403 is configured to quantize the nonlinear response relationship between the lake area and each key ecological index based on the fitting result of the multi-response random forest model.

[0117] The threshold determination module 404 is configured to identify a sensitive response interval of the lake area based on the nonlinear response relationship, and determine a threshold.

[0118] Further, in the model construction module, the construction of the multi-response random forest model comprises: dividing the data for modeling into lake area data and key ecological index data; dynamically extracting the key ecological index data, and automatically constructing a multi-response formula Multivar(y1, y2,..., y n )~x; wherein x is the lake area; y1, y2,..., y n are n key ecological indexes; constructing the multi-response random forest model based on a multi-response random forest modeling tool and the multi-response formula, and respectively calculating the average prediction error rate, the block error rate, the importance, the determination coefficient and the root mean square error of each key ecological index, for fine-grained diagnosis of the multi-response random forest model; taking the variance of the key ecological index as a weight, and constructing a weighted determination coefficient evaluation index; and the weighted determination coefficient evaluation index is configured to quantize the overall explanatory ability of the lake area to the ecological system to be studied.

[0119] Further, in the nonlinear response relationship determination module, a plurality of key ecological indexes are determined based on the time series matrix, and a variable screening method is adopted.

[0120] The device can be implemented by the method for determining the threshold of the lake ecological system in the arid region provided in the above embodiment one, and the specific implementation method can be referred to the description in the embodiment one, which will not be described here.

[0121] The application further provides a memory storing a plurality of instructions for implementing the method for determining the threshold of the lake ecological system in the arid region.

[0122] The application further provides an electronic device comprising a processor and a memory connected to the processor, wherein the memory stores a plurality of instructions, and the instructions can be loaded and executed by the processor to enable the processor to perform the method for determining the threshold of the lake ecological system in the arid region.

[0123] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the foregoing description. It is therefore intended that the appended claims shall cover all such modifications and changes as fall within the true spirit and scope of the application. It is further understood that the application can be used in a variety of applications and that the application is not limited to the applications described above. Accordingly, many modifications and variations of the application are possible in light of the above teachings without departing from the spirit and scope of the application. It is therefore intended that the application be covered all such modifications and variations provided they come within the scope of the claims and their equivalents.

Claims

1. A method for determining the threshold of a lake ecosystem in an arid region, characterized in that, include: Based on remote sensing data of the ecosystem under study, a time series matrix including the lake area and multidimensional ecological indicators of the ecosystem under study is extracted; wherein, the multidimensional ecological indicators are pre-selected based on remote sensing data and ecological theory, and the multidimensional ecological indicators respond to the changes in the lake area; Based on the time series matrix, several key ecological indicators are determined, and a multi-response random forest model is constructed. The key ecological indicators are those with a higher responsiveness to changes in lake area than a preset value. The multi-response random forest model is used to jointly model and fit multiple key ecological indicators with lake area as the input variable. Based on the fitting results of the multi-response random forest model, the nonlinear response relationship between lake area and various key ecological indicators is quantified. Based on the aforementioned nonlinear response relationship, the sensitive response interval of the lake area is identified, and the threshold is determined.

2. The method for determining the threshold of arid zone lake ecosystems as described in claim 1, characterized in that, The construction of the multi-response random forest model includes: The data used for modeling is divided into lake area data and key ecological indicator data; Dynamically extract key ecological indicator data and automatically construct multi-response formulas (Multivar(y1, y2, ..., y...)). n )~x; where x is the lake area; y1, y2, ..., y n There are n key ecological indicators; A multi-response random forest model is constructed based on the multi-response random forest modeling tool and the multi-response formula. The average prediction error rate, block error rate, importance, coefficient of determination and root mean square error of each key ecological indicator are calculated respectively to perform fine-grained diagnosis of the multi-response random forest model. The variance of key ecological indicators is used as weights to construct a weighted coefficient of determination evaluation index; the weighted coefficient of determination evaluation index is used to quantify the overall explanatory power of lake area for the ecosystem under study.

3. The method for determining the threshold of arid zone lake ecosystems as described in claim 1, characterized in that, Based on the aforementioned time series matrix, a variable screening method was used to determine several key ecological indicators.

4. The method for determining the threshold of arid zone lake ecosystems as described in claim 1, characterized in that, The nonlinear response relationship between the quantified lake area and each key ecological indicator includes: using an improved characteristic contribution interpretation method to quantify the intensity and direction of the lake area's response to each key ecological indicator.

5. The method for determining the threshold of arid zone lake ecosystems as described in claim 4, characterized in that, The improved feature contribution interpretation method is implemented in the following manner: Based on the key ecological indicators with nonlinear response relationships and lake area, a single-response stochastic forest model is constructed. For any type of single-response random forest model, prediction data is output through a unified prediction interface; The feature contribution is calculated based on the predicted data, and the feature contribution is used as a SHAP-like value to explain the driving effect of lake area on the key ecological indicators.

6. A device for determining the threshold of a lake ecosystem in an arid region, characterized in that, include: The data acquisition module is used to extract a time series matrix including the lake area and multidimensional ecological indicators of the ecosystem under study based on remote sensing data of the ecosystem under study; wherein, the multidimensional ecological indicators are pre-selected based on remote sensing data and ecological theory, and the multidimensional ecological indicators respond to the changes in the lake area; The model building module is used to determine multiple key ecological indicators based on the time series matrix and construct a multi-response random forest model; wherein, the key ecological indicators are ecological indicators whose responsiveness to changes in lake area is higher than a preset value; the multi-response random forest model is used to jointly model and fit multiple key ecological indicators with lake area as input variable. The nonlinear response relationship determination module is used to quantify the nonlinear response relationship between lake area and various key ecological indicators based on the fitting results of the multi-response random forest model. The threshold determination module is used to identify the sensitive response interval of the lake area based on the nonlinear response relationship and determine the threshold.

7. The apparatus for determining the threshold of arid zone lake ecosystems as described in claim 6, characterized in that, In the model building module, building the multi-response random forest model includes: The data used for modeling is divided into lake area data and key ecological indicator data; Dynamically extract key ecological indicator data and automatically construct multi-response formulas (Multivar(y1, y2, ..., y...)). n )~x; where x is the lake area; y1, y2, ..., y n There are n key ecological indicators; A multi-response random forest model is constructed based on the multi-response random forest modeling tool and the multi-response formula. The average prediction error rate, block error rate, importance, coefficient of determination and root mean square error of each key ecological indicator are calculated respectively to perform fine-grained diagnosis of the multi-response random forest model. The variance of key ecological indicators is used as weights to construct a weighted coefficient of determination evaluation index; the weighted coefficient of determination evaluation index is used to quantify the overall explanatory power of lake area for the ecosystem under study.

8. The apparatus for determining the threshold of arid zone lake ecosystems as described in claim 6, characterized in that, In the nonlinear response relationship determination module, based on the time series matrix, key ecological indicators with a high response relationship to the lake area are determined by a variable screening method.

9. A memory, characterized in that, The system stores multiple instructions for implementing the method for determining thresholds for arid zone lake ecosystems as described in any one of claims 1-5.

10. An electronic device, characterized in that, The device includes a processor and a memory connected to the processor, the memory storing multiple instructions that can be loaded and executed by the processor to enable the processor to perform the method for determining the threshold of arid lake ecosystems as described in any one of claims 1-5.

Images