An ecological clean small watershed achievement evaluation method based on multi-source data

Abstract

The application discloses an ecological clean small watershed effect evaluation method based on multi-source data, relates to the technical field of ecological evaluation, and comprises the following steps: generating a standardized watershed space-time feature dataset; based on a pre-constructed ecological hydrological coupling mechanism model, physical process quantities including water source conservation, soil conservation and water quality purification are calculated, and a multi-dimensional ecological effect state vector is constructed accordingly; a two-way output branch of the deep evaluation decision network is used to respectively calculate a watershed ecological comprehensive score value and a marginal contribution rate of each treatment measure to ecological effect; an ecological value conversion trigger condition is determined according to the marginal contribution rate and treatment input cost data, and the watershed ecological comprehensive score value and the marginal contribution rate are used to dynamically plan and correct the treatment resource allocation of the next stage based on the trigger condition; and the application improves the accuracy of ecological clean small watershed effect evaluation and simultaneously realizes dynamic optimization of treatment resource allocation.

Description

Technical Field

[0001] This invention relates to the field of ecological assessment technology, specifically to a method for evaluating the effectiveness of ecological clean watersheds based on multi-source data. Background Technology

[0002] Ecological clean watershed management is an important measure to ensure regional ecological security, improve water quality, and promote the sustainable use of water and soil resources. Effectiveness evaluation is a crucial step in optimizing management plans and enhancing management efficiency. Currently, the effectiveness evaluation of ecological clean watersheds largely relies on single monitoring data or empirical judgments, which has significant limitations: First, existing evaluation methods often use single data sources, such as ground monitoring or remote sensing imagery, resulting in incomplete data coverage and mismatched spatiotemporal resolution, making it difficult to accurately capture the dynamic changes in the watershed ecosystem. Second, the evaluation process lacks deep coupling with eco-hydrological and physical processes, the indicator system is not scientifically constructed, and the weight allocation is highly subjective, leading to significant deviations between the evaluation results and actual management effectiveness, and failing to effectively guide the optimal allocation of management resources. Furthermore, existing evaluation methods often remain at the level of current status assessment, lacking quantitative analysis of the marginal contribution of management measures, making it difficult to achieve dynamic planning and correction for the next stage of management work, thus hindering the precise and long-term advancement of ecological clean watershed management.

[0003] Therefore, there is an urgent need for a highly efficient effectiveness assessment method that integrates multi-source data, combines eco-hydrological mechanisms, and takes into account both current status assessment and dynamic optimization, in order to address the shortcomings of existing technologies. Summary of the Invention

[0004] To address the aforementioned technical problems, this paper provides a method for evaluating the effectiveness of ecological clean watersheds based on multi-source data. This technical solution solves the problems mentioned above.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data includes: Acquire multi-source heterogeneous monitoring data of ecological clean watersheds within a preset assessment period, and perform spatiotemporal benchmark unification and outlier cleaning on the multi-source heterogeneous monitoring data to generate a standardized watershed spatiotemporal feature dataset. Hydrological and meteorological elements, land use change elements, and non-point source pollution load elements are extracted from the standardized watershed spatiotemporal feature dataset. Based on the pre-constructed eco-hydrological coupling mechanism model, physical process quantities including water conservation, soil retention, and water purification are calculated, and a multidimensional ecological performance state vector is constructed accordingly. The multidimensional ecological effectiveness state vector is input into a pre-trained deep evaluation decision network, and the watershed ecological comprehensive score and the marginal contribution rate of each governance measure to the ecological effectiveness are calculated through the dual output branches of the deep evaluation decision network. The system acquires current governance input cost data and socio-economic driving factors for small watersheds. Based on the marginal contribution rate and governance input cost data, it determines the ecological value transformation trigger conditions. Based on the trigger conditions, it uses the watershed ecological comprehensive score and the marginal contribution rate to dynamically plan and correct the allocation of governance resources for the next stage, generating an ecological clean watershed effectiveness evaluation report that includes spatial layout optimization and measure type recommendations.

[0006] Furthermore, the acquisition of multi-source heterogeneous monitoring data of ecologically clean small watersheds within a preset assessment period, and the spatiotemporal benchmark unification and outlier cleaning of the multi-source heterogeneous monitoring data to generate a standardized watershed spatiotemporal feature dataset, includes: Acquire remote sensing satellite imagery data, drone aerial orthophotos, and real-time monitoring data from ground-based IoT sensors; Using the small watershed boundary as a geographic reference frame, geographic information system technology is used to unify the spatial resolution and perform projection transformation on the remote sensing satellite image data and the UAV aerial orthophoto; Atmospheric correction and radiometric calibration are performed on the transformed image data. Combined with the timestamps in the real-time monitoring data of the ground IoT sensor, interpolation and fitting are performed on the image data and sensor data to remove outliers and generate the standardized watershed feature library with spatiotemporal alignment.

[0007] Furthermore, the construction of the multidimensional ecological effectiveness state vector includes: Based on the digital elevation model, slope and slope length factors are extracted, and soil loss is calculated by combining rainfall erosion force data to generate soil and water conservation indicators. Extract the spectral reflectance characteristics of the water body, retrieve the concentrations of total nitrogen, total phosphorus, and chemical oxygen demand, and generate the water quality purification index. The normalized vegetation index and landscape fragmentation index are used to assess vegetation cover and ecosystem integrity, and the aforementioned ecological landscape indicators are generated. The soil and water conservation indicators, the water quality purification indicators, and the ecological landscape indicators are vector normalized and then spliced ​​together to generate the multi-dimensional effectiveness evaluation feature set.

[0008] Furthermore, the multidimensional ecological effectiveness state vector is input into a pre-trained deep evaluation decision network. The dual output branches of the deep evaluation decision network are used to calculate the comprehensive ecological score of the watershed and the marginal contribution rate of each governance measure to the ecological effectiveness, including: The multi-dimensional effectiveness evaluation feature set is input into the feature weighting layer of the comprehensive effectiveness evaluation model; The subjective weights of the soil and water conservation indicators, the water quality purification indicators, and the ecological landscape indicators were determined using the analytic hierarchy process. The objective weights of information entropy for each indicator are calculated using the entropy weight method. The subjective weights and objective weights are fused using a game theory-based combined weighting method to obtain the optimal combined weight vector; The multi-dimensional effectiveness evaluation feature set and the optimal combined weight vector are multiplied by a dot product to output the comprehensive watershed ecological index and the individual evaluation scores of each subsystem; wherein, the formula for calculating the comprehensive watershed ecological index is: ; In the formula, As a comprehensive ecological index for the watershed, For the first The optimal combination of weights for the evaluation indicators. For the first The standardized values ​​of the evaluation indicators after vector normalization. To evaluate the total number of indicators.

[0009] Furthermore, the comprehensive effectiveness evaluation model is a machine learning regression model trained based on historical governance data. The specific steps in the training process of the comprehensive effectiveness evaluation model include: Construct a loss function that includes ecological constraints; Obtain data on the input costs and actual output benefits of historical small watershed management projects; The root mean square error between the watershed ecological comprehensive index predicted by the computational model and the actual monitoring value; An ecological response function is introduced to calculate the nonlinear lag effect between the intensity of governance measures and the improvement of environmental quality. The nonlinear lag effect is added as an ecological constraint to the loss function, and the model parameters are optimized using the gradient descent algorithm.

[0010] Furthermore, the construction of the loss function including the ecological constraint term includes: Set the regularization parameters and learning rate decay factor; The cross-entropy loss between the predicted value and the true label is calculated as a basic term; Extract the runoff coefficient and sediment transport ratio in the hydrological cycle process to construct a physical process constraint matrix; The physical process constraint matrix and the basic terms are weighted and summed to construct the final loss function.

[0011] Furthermore, the step of dynamically planning and correcting the allocation of governance resources for the next stage based on the triggering conditions using the comprehensive ecological score of the watershed and the marginal contribution rate includes: If the comprehensive ecological index of the watershed is greater than the first preset threshold, the governance effect is determined to be significant. If the comprehensive ecological index of the watershed is less than the second preset threshold, the governance effect is deemed insufficient. For indicators whose individual evaluation scores are below the passing mark, extract the correlation of the corresponding governance measures. Based on the degree of correlation, targeted engineering or plant-based optimization schemes are generated.

[0012] Furthermore, if the comprehensive ecological index of the watershed is less than the second preset threshold, it includes: The high-resolution image verification mechanism was activated to retrieve dynamic monitoring data from the past three months. Compare and analyze the deviations between the implementation progress of the remediation measures and the design drawings; Generate a supervision list that includes specific location coordinates and rectification deadlines.

[0013] Furthermore, it also includes: Receive future scenario parameters input by the user, the future scenario parameters including different rainfall frequencies and land use change schemes; The future scenario parameters are combined with the current multi-dimensional effectiveness evaluation feature set and input into the comprehensive effectiveness evaluation model; Predict future trends in the comprehensive ecological index of the watershed under different scenarios and generate a report on long-term maintenance strategies.

[0014] Furthermore, after generating an ecological clean watershed effectiveness assessment report that includes spatial layout optimization and recommended measure types, it includes: Upload the assessment report to the cloud management platform; Automatically generate report summaries using natural language processing technology; The assessment report and governance optimization suggestions are pushed to management personnel via mobile devices.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By using multi-source heterogeneous monitoring data and unifying the spatiotemporal benchmarks and cleaning outliers, a standardized watershed spatiotemporal feature dataset is generated. This solves the problems of existing assessment data being singular, spatiotemporally mismatched, and of poor quality, and improves the comprehensiveness and accuracy of the assessment data, providing reliable data support for subsequent effectiveness assessments. 2. Based on the eco-hydrological coupling mechanism model, physical process quantities are calculated, and a multi-dimensional ecological performance state vector is constructed. By combining a deep assessment decision network with a game theory-based weighting method, a scientific integration of subjective and objective weights is achieved, avoiding the limitations of single weight allocation. At the same time, the comprehensive ecological index of the watershed is quantified through a clear formula, which improves the objectivity, scientificity, and accuracy of the assessment results and reduces the error of human experience judgment. 3. By using a dual-output branch of the deep evaluation decision network, the comprehensive ecological score of the watershed and the marginal contribution rate of each governance measure are obtained simultaneously. Combined with the governance input cost and socio-economic driving factors, dynamic planning and correction of the allocation of governance resources in the next stage are realized. This not only accurately determines the governance effectiveness, but also provides targeted output of spatial layout optimization and measure type recommendations. It solves the problem of existing assessments emphasizing evaluation but neglecting guidance, and provides precise guidance for the long-term governance of ecologically clean small watersheds. 4. By embedding key technologies such as outlier removal and loss function optimization, the stability and anti-interference ability of the evaluation model are improved. At the same time, the addition of future scenario prediction and cloud push function for evaluation reports broadens the application scenarios of the evaluation method, making it easier for managers to keep abreast of the dynamics of watershed management in real time, improving the efficiency and convenience of management work, and has strong engineering practicality and promotion value. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the present invention. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] Example 1: Refer to Figure 1 As shown, a method for evaluating the effectiveness of ecological clean watersheds based on multi-source data includes: Step S1: Obtain multi-source heterogeneous monitoring data of ecological clean watershed within a preset assessment period, unify the spatiotemporal benchmark and clean outliers of the multi-source heterogeneous monitoring data, and generate a standardized watershed spatiotemporal feature dataset. Step S2: Extract hydrological and meteorological elements, land use change elements, and non-point source pollution load elements from the standardized watershed spatiotemporal feature dataset. Based on the pre-constructed eco-hydrological coupling mechanism model, calculate the physical process quantities including water conservation, soil retention, and water purification, and construct a multi-dimensional ecological performance state vector accordingly. Step S3: Input the multidimensional ecological effectiveness state vector into the pre-trained deep evaluation decision network, and calculate the watershed ecological comprehensive score and the marginal contribution rate of each governance measure to the ecological effectiveness through the dual output branches of the deep evaluation decision network. Step S4: Obtain current governance input cost data and socio-economic driving factors for the small watershed. Determine the ecological value transformation trigger conditions based on the marginal contribution rate and governance input cost data. Based on the trigger conditions, use the watershed ecological comprehensive score and marginal contribution rate to dynamically plan and correct the allocation of governance resources for the next stage, and generate an ecological clean watershed effectiveness evaluation report that includes spatial layout optimization and measure type recommendations.

[0019] The above scheme utilizes multi-source heterogeneous monitoring data to compensate for the deficiencies of insufficient coverage and spatiotemporal mismatch of single data sources. Data quality is improved through spatiotemporal benchmark unification and outlier cleaning. Physical process quantities are calculated based on an eco-hydrological coupling mechanism model to construct a scientific multi-dimensional ecological effectiveness state vector, avoiding the subjectivity of indicator selection. The deep assessment decision network with dual outputs simultaneously realizes effectiveness scoring and quantification of the marginal contribution of governance measures. Combined with cost and socio-economic factors, governance resources are dynamically optimized, overcoming the limitations of existing assessments that "emphasize the status quo and neglect optimization" and "emphasize evaluation and neglect guidance." This effectively improves the accuracy, scientificity, and practicality of ecological clean watershed effectiveness assessment, providing reliable support for long-term watershed governance.

[0020] In some embodiments, step S1 involves acquiring multi-source heterogeneous monitoring data of ecologically clean small watersheds within a preset assessment period, performing spatiotemporal benchmark unification and outlier cleaning on the multi-source heterogeneous monitoring data, and generating a standardized watershed spatiotemporal feature dataset, including the following sub-steps: Step S11: Acquire remote sensing satellite image data, UAV aerial orthophotos, and real-time monitoring data from ground IoT sensors; Specifically, the preset evaluation cycle can be set according to the scale of small watershed management and monitoring needs, usually one quarter, six months or one year. In this embodiment, six months is preferred as the preset evaluation cycle, which can ensure the timeliness of data and avoid evaluation deviation caused by short-term data fluctuations. The remote sensing satellite image data is preferably multispectral image with a resolution of not less than 10m, which is used to obtain macro-spatial information such as watershed land use, vegetation cover and water distribution. The UAV aerial orthophoto is taken by high-definition UAV, with the flight altitude controlled between 50 and 100m, to obtain detailed information such as watershed micro-topography and management project implementation. The shooting range completely covers the boundary of the small watershed. The real-time monitoring data of ground IoT sensors is obtained through multiple types of sensors deployed in the watershed, including rainfall sensors, water level sensors, water quality sensors, soil moisture sensors, etc. The monitoring frequency is set to once every hour to ensure the real-time and continuous data. All sensors are calibrated in advance, and the error is controlled within ±5%.

[0021] Step S12: Using the small watershed boundary as a geographic reference frame, use geographic information system technology to unify the spatial resolution and perform projection transformation on remote sensing satellite image data and UAV aerial orthophotos.

[0022] Specifically, firstly, the small watershed boundary vector data is imported using geographic information system software to serve as a geographic reference frame; then, the spatial resolution of remote sensing satellite imagery data and UAV aerial orthophotos is unified to 1m resolution, using a bilinear interpolation algorithm to adjust the resolution and avoid spatial information misalignment caused by resolution differences; finally, UTM projection is used for projection transformation, determining the corresponding projection zone based on the longitude of the small watershed, and converting the coordinate system of all image data to a coordinate system consistent with the small watershed boundary vector data, eliminating coordinate deviations from different data sources and ensuring spatial consistency.

[0023] Step S13: Perform atmospheric correction and radiometric calibration on the transformed image data. Combine the timestamps in the real-time monitoring data of the ground IoT sensors to interpolate and fit the image data and sensor data, remove outliers, and generate a standardized watershed feature library with spatiotemporal alignment.

[0024] Specifically, specialized image processing software was used to perform atmospheric correction on the transformed remote sensing satellite imagery and UAV aerial orthorectified imagery. A suitable atmospheric correction model was selected to eliminate the interference of atmospheric scattering and absorption on the image's spectral information, restoring the image's true reflectance. Radiometric calibration employed an absolute radiometric calibration method, converting the image's DN values ​​into radiance values ​​to ensure the accuracy of the image's spectral information. By combining the timestamps from real-time monitoring data from ground-based IoT sensors, the image data and sensor data were time-aligned. The image data was matched with the sensor data according to the capture time. For data with time mismatches, a linear interpolation algorithm was used to fit the data, filling in the gaps and ensuring that both spatial information from the imagery and quantitative data from sensor monitoring existed at the same time point.

[0025] Outliers were identified using the 3σ criterion to ensure data quality. Finally, all processed data were integrated to generate a standardized watershed feature library that is spatiotemporally aligned, complete, and of acceptable quality, providing data support for subsequent feature extraction and evaluation.

[0026] In some embodiments, the construction of the multidimensional ecological effectiveness state vector in step S2 includes the following sub-steps: Step S21: Extract slope and slope length factors based on the digital elevation model, calculate soil loss by combining rainfall erosion force data, and generate soil and water conservation indicators.

[0027] Specifically, digital elevation model data with a resolution of 1m was extracted from a standardized watershed feature database. Spatial analysis tools in Geographic Information System (GIS) software were used to extract the slope and slope length factors for each grid cell within the small watershed. Slope was calculated in degrees using a slope calculation tool, and slope length was calculated in meters using a slope length calculation tool. Rainfall erosivity data were obtained from the local meteorological department, using multi-year average rainfall erosivity values. Combined with the slope and slope length factors, a general soil loss equation was used to calculate soil loss. The calculated soil loss was used as the core soil and water conservation indicator, supplemented by two auxiliary indicators: vegetation cover and soil erosion control rate, together forming the soil and water conservation indicator system.

[0028] Step S22: Extract the spectral reflectance characteristics of the water body, invert the concentrations of total nitrogen, total phosphorus and chemical oxygen demand, and generate water quality purification indicators.

[0029] Specifically, multispectral data from remote sensing satellite imagery, after atmospheric correction, are extracted from a standardized watershed feature database. Sensitive bands for water bodies are selected, and the spectral reflectance characteristics of all water bodies within the watershed are extracted. An inversion model is constructed using a band combination algorithm. The model is then calibrated using total nitrogen, total phosphorus, and chemical oxygen demand (COD) concentration data monitored by ground-based IoT sensors to ensure the inversion error is controlled within ±10%. Based on the calibrated inversion model, the concentrations of total nitrogen, total phosphorus, and COD in all water bodies within the watershed are calculated. The annual average values ​​and frequency of exceedances of these three parameters are used as core water quality purification indicators. Two auxiliary indicators, water transparency and dissolved oxygen concentration, are also added to form a water quality purification indicator system.

[0030] Step S23: Using the normalized vegetation index and landscape fragmentation index, assess vegetation cover and ecosystem integrity to generate ecological landscape indicators.

[0031] Specifically, the Normalized Difference Vegetation Index (NDI) is calculated using the near-infrared and red bands of remote sensing satellite imagery. The NDI ranges from -1 to 1; a higher value indicates higher vegetation cover. Vegetation cover in a small watershed is calculated using the NDI and categorized into four levels: high cover, medium cover, low cover, and no cover. The Landscape Fragmentation Index (GFCI) is calculated using specialized landscape analysis software to reflect the degree of fragmentation in the watershed's ecological landscape. A higher GFCI value indicates greater landscape fragmentation and poorer ecosystem integrity. Vegetation cover and the GFCI are used as core ecological landscape indicators, supplemented by the biodiversity index, to form an ecological landscape indicator system.

[0032] Step S24: Perform vector normalization on the soil and water conservation indicators, water quality purification indicators, and ecological landscape indicators, and splice them to generate a multi-dimensional effectiveness evaluation feature set.

[0033] Specifically, the min-max normalization method was used to normalize all indicators, eliminating the dimensional differences between different indicators. After normalization, the indicator values ​​were all in the range [0, 1]. The normalized soil and water conservation indicators, water quality purification indicators, and ecological landscape indicators were then concatenated in sequence to form a multi-dimensional effectiveness evaluation feature set. The dimension of the feature set was equal to the total number of indicators in the three indicator systems, and each feature vector corresponded to the comprehensive characteristics of an evaluation unit in a small watershed.

[0034] In some embodiments, step S3 involves inputting the multidimensional ecological effectiveness state vector into a pre-trained deep evaluation decision network, and calculating the comprehensive ecological score of the watershed and the marginal contribution rate of each governance measure to the ecological effectiveness through the dual output branches of the deep evaluation decision network, including the following sub-steps: Step S31: Input the multi-dimensional effectiveness evaluation feature set into the feature weighting layer of the comprehensive effectiveness evaluation model.

[0035] Specifically, the comprehensive effectiveness evaluation model adopts a deep evaluation decision network, whose structure includes an input layer, a feature weighting layer, a hidden layer, and a dual-path output layer. The dimension of the input layer is consistent with the dimension of the multi-dimensional effectiveness evaluation feature set. The feature weighting layer is used to assign weights to the input feature vectors, enhance the influence of key indicators, and lay the foundation for subsequent weight fusion.

[0036] Step S32: Determine the subjective weights of soil and water conservation indicators, water quality purification indicators, and ecological landscape indicators using the analytic hierarchy process.

[0037] Specifically, 5-7 experts in the fields of ecology, environment, and soil and water conservation were invited to construct a judgment matrix based on the impact of each indicator on the effectiveness of ecological clean watersheds. The analytic hierarchy process (AHP) was used to calculate the maximum eigenvalue and its corresponding eigenvector of the judgment matrix. The eigenvectors were then normalized to obtain the subjective weights of each indicator. The subjective weights of each specific indicator within each indicator system were further refined to ensure that the weight allocation aligns with actual governance needs.

[0038] Step S33: Calculate the objective weight of information entropy for each indicator using the entropy weight method.

[0039] Specifically, based on the standardized multi-dimensional performance evaluation feature set, the information entropy of each indicator is calculated. The smaller the information entropy, the more information the indicator contains and the greater its weight; conversely, the larger the information entropy, the smaller its weight. Objective weights are calculated based on information entropy to ensure that the weight allocation conforms to the actual data patterns.

[0040] Step S34: Use game theory combined weighting method to fuse subjective weights and objective weights to obtain the optimal combined weight vector.

[0041] Specifically, subjective weight vector, objective weight vector, and combined weight vector are set. Based on the principles of game theory, an objective function is constructed. Under constraints, the optimal combined weight vector is solved to ensure that the combined weights take into account both expert experience and actual data patterns, thus avoiding the limitations of a single weight.

[0042] Step S35: Perform a dot product operation on the multi-dimensional effectiveness evaluation feature set and the optimal combined weight vector to output the comprehensive watershed ecological index and the individual evaluation scores of each subsystem; wherein, the formula for calculating the comprehensive watershed ecological index is: ; In the formula, As a comprehensive ecological index for the watershed, For the first The optimal combination of weights for the evaluation indicators. For the first The standardized values ​​of the evaluation indicators after vector normalization. To assess the total number of indicators, The value range is [0, 100]. The higher the value, the better the ecological cleaning effect of the small watershed. The individual evaluation score of each subsystem is calculated by the dot product of the index weight and the standardized value in the corresponding index system. The scores of the soil and water conservation subsystem, water quality purification subsystem, and ecological landscape subsystem are obtained respectively, which are used for subsequent identification of the weak indicators.

[0043] In some embodiments, the comprehensive effectiveness evaluation model is a machine learning regression model trained based on historical governance data. The specific steps in the training process of the comprehensive effectiveness evaluation model include: Step S301: Construct a loss function that includes ecological constraints.

[0044] Specifically, regularization parameters and learning rate decay factors are set; the cross-entropy loss between the watershed ecological comprehensive index predicted by the model and the actual monitoring value is calculated as the basic term of the loss function; the runoff coefficient and sediment transport ratio in the hydrological cycle process are extracted to construct the physical process constraint matrix; the physical process constraint matrix and the basic term are weighted and summed to construct the final loss function, ensuring that the model prediction results conform to the ecological hydrological physical laws and avoiding the disconnect between the predicted values ​​and the actual ecological processes.

[0045] Step S302: Obtain data on the input costs and actual output benefits of historical small watershed management projects.

[0046] Specifically, we collected 30-50 cases of ecologically clean small watersheds in China that have been completed and have complete monitoring data. We extracted historical governance input cost data and actual output benefit data for each case. At the same time, we collected multi-source monitoring data and indicator data for each case to construct a training sample set, of which 70% was used as the training set and 30% as the test set.

[0047] Step S303: Calculate the root mean square error between the watershed ecological comprehensive index predicted by the model and the actual monitoring value.

[0048] Specifically, the root mean square error is used to measure the deviation between the model's predicted value and the actual monitored value. The smaller the value, the higher the accuracy of the model's prediction. In this embodiment, it is required that the root mean square error of the test set be ≤5% after the model is trained.

[0049] Step S304: Introduce the ecological response function and calculate the nonlinear lag effect between the intensity of governance measures and the improvement of environmental quality.

[0050] Specifically, the ecological response function adopts the Logistic regression function. The function parameters are determined by fitting historical data, and the nonlinear lag effect corresponding to different governance measures intensity is calculated. That is, the time and magnitude of environmental quality improvement required to reach a stable state after the implementation of governance measures. This lag effect is added to the loss function as an ecological constraint to ensure that the model takes into account the lag of the ecosystem and improves the rationality of the prediction.

[0051] Step S305: Add the nonlinear hysteresis effect as an ecological constraint to the loss function, and optimize the model parameters using the gradient descent algorithm.

[0052] Specifically, the calculated nonlinear lag effect is transformed into a constraint term and incorporated into the loss function. The objective of the loss function is adjusted so that the model training process not only fits the deviation between the predicted and actual values ​​but also takes into account the ecological lag effect. The stochastic gradient descent algorithm is used to optimize the model parameters. The initial learning rate is set to 0.001, and the learning rate decays once every 100 iterations. The number of iterations is set to 1000. Training stops when the loss function converges or reaches the maximum number of iterations, resulting in a comprehensive evaluation model of the trained model.

[0053] In some embodiments, step S4, based on triggering conditions, uses the comprehensive ecological score of the watershed and the marginal contribution rate to dynamically plan and correct the allocation of governance resources for the next stage, including the following sub-steps: Step S41: If the comprehensive ecological index of the watershed is greater than the first preset threshold, the governance effect is determined to be significant; if the comprehensive ecological index of the watershed is less than the second preset threshold, the governance effect is determined to be insufficient.

[0054] Specifically, the first preset threshold and the second preset threshold are set according to the small watershed management goals and regional ecological environment standards. In this embodiment, the first preset threshold is set to 80 points, and the second preset threshold is set to 60 points; if A score >80 indicates significant governance effectiveness; the next phase will maintain the existing governance resource allocation, with a focus on promoting long-term operation and maintenance. If the score is 60 or below... A score of ≤80 indicates generally poor governance effectiveness, requiring optimization of resource allocation to address weaknesses in key indicators; if A score of less than 60 indicates insufficient effectiveness in governance, requiring a comprehensive adjustment of the governance plan and increased resource investment.

[0055] Step S42: For indicators whose individual evaluation scores are below the passing grade, extract the correlation degree of the corresponding governance measures.

[0056] Specifically, the passing score for each evaluation item is set at 60 points. Shortcoming indicators with scores below 60 points are selected from the three subsystems of soil and water conservation, water purification, and ecological landscape. By constructing a correlation matrix between governance measures and indicators, the correlation degree of governance measures corresponding to each shortcoming indicator is extracted. The correlation degree ranges from [0, 1]. The higher the correlation degree, the more significant the effect of the governance measure on improving the shortcoming indicator. For example, the governance measures corresponding to excessive soil loss include terrace construction, vegetation planting, and grain dam construction. Among them, vegetation planting has the highest correlation degree and is given priority for implementation.

[0057] Step S43: Generate targeted engineering or plant management optimization schemes based on the degree of correlation.

[0058] Specifically, the extracted governance measures are ranked by their correlation, with priority given to those with high correlation. Targeted optimization plans are then generated by considering the actual topography, soil conditions, and governance costs of the small watershed. For water and soil conservation-related shortcomings, priority is given to measures such as vegetation planting and terrace construction. For water quality purification-related shortcomings, priority is given to measures such as artificial wetland construction and ecological revetment laying. For ecological landscape-related shortcomings, priority is given to measures such as vegetation replanting and landscape patch restoration, ensuring the feasibility and relevance of the plans.

[0059] In some embodiments, if the comprehensive ecological index of the watershed is less than a second preset threshold, the following steps are included: A high-resolution image review mechanism was initiated to retrieve dynamic monitoring data from the past three months; the implementation progress of the remediation measures was compared and analyzed to identify deviations from the design drawings; and a supervision list containing specific location coordinates and rectification deadlines was generated.

[0060] Specifically, a high-resolution image review mechanism was initiated, retrieving orthophotos from UAV aerial photography and remote sensing satellite imagery from the past three months, and combining them with monitoring data from ground-based IoT sensors to conduct a comprehensive review of the small watershed management area. The implementation progress of actual management measures was compared with the design drawings one by one to identify areas where implementation was lagging behind or measures were not being implemented effectively. The reasons for the deviations were analyzed, including insufficient construction personnel, delayed material supply, and unreasonable construction techniques. For the areas with deviations, the specific location coordinates, rectification content, responsible persons, and rectification deadlines were clearly defined, a supervision list was generated, and the rectification progress was tracked regularly to ensure that the rectification work was implemented effectively and to promote the improvement of management results.

[0061] In some embodiments, the method further includes the following steps: The system receives future scenario parameters input by users, including different rainfall frequencies and land use change schemes; it combines these future scenario parameters with the current multi-dimensional effectiveness assessment feature set and inputs them into the comprehensive effectiveness assessment model; it predicts the future trend of watershed ecological comprehensive index changes under different scenarios and generates a long-term maintenance strategy report.

[0062] Specifically, the system receives future scenario parameters input by users, including three scenarios with different rainfall frequencies: high-water years, normal-water years, and low-water years. Land use change schemes include various options such as farmland protection, forest expansion, and construction land control. These future scenario parameters are then fused with the current multi-dimensional effectiveness evaluation feature set to supplement scenario-related parameter data and ensure the completeness of the input data. The fused data is then input into a trained comprehensive effectiveness evaluation model. The model outputs the trend of the watershed's comprehensive ecological index changes over the next 1-5 years under different scenarios, analyzing the impact of different scenarios on the watershed's ecological effectiveness. Based on the changing trends, a long-term maintenance strategy report is generated, clarifying the governance priorities, adjustment directions of measures, and resource input plans under different scenarios, providing forward-looking guidance for the long-term ecological governance of small watersheds.

[0063] In some embodiments, after generating an ecological clean watershed effectiveness assessment report that includes spatial layout optimization and recommended measure types, the following steps are included: The assessment report is uploaded to the cloud management platform; a report summary is automatically generated using natural language processing technology; and the assessment report and governance optimization suggestions are pushed to managers via mobile devices.

[0064] Specifically, the generated ecological clean watershed effectiveness assessment reports are standardized in format and uploaded to a cloud management platform for centralized storage, retrieval, and management, facilitating sharing and viewing by personnel from different departments. Natural language processing technology is used to extract core content from the reports, including the watershed's comprehensive ecological index, shortcomings indicators, and governance optimization suggestions, automatically generating report summaries to shorten reading time for managers. The assessment reports and governance optimization suggestions are then pushed to watershed governance managers via mobile apps and SMS, ensuring they promptly grasp the watershed governance effectiveness and optimization direction, quickly deploy subsequent governance work, and improve the efficiency and convenience of governance efforts.

[0065] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data, characterized in that, include: Acquire multi-source heterogeneous monitoring data of ecological clean watersheds within a preset assessment period, and perform spatiotemporal benchmark unification and outlier cleaning on the multi-source heterogeneous monitoring data to generate a standardized watershed spatiotemporal feature dataset. Hydrological and meteorological elements, land use change elements, and non-point source pollution load elements are extracted from the standardized watershed spatiotemporal feature dataset. Based on the pre-constructed eco-hydrological coupling mechanism model, physical process quantities including water conservation, soil retention, and water purification are calculated, and a multidimensional ecological performance state vector is constructed accordingly. The multidimensional ecological effectiveness state vector is input into a pre-trained deep evaluation decision network, and the watershed ecological comprehensive score and the marginal contribution rate of each governance measure to the ecological effectiveness are calculated through the dual output branches of the deep evaluation decision network. The system acquires current governance input cost data and socio-economic driving factors for small watersheds. Based on the marginal contribution rate and governance input cost data, it determines the ecological value transformation trigger conditions. Based on the trigger conditions, it uses the watershed ecological comprehensive score and the marginal contribution rate to dynamically plan and correct the allocation of governance resources for the next stage, generating an ecological clean watershed effectiveness evaluation report that includes spatial layout optimization and measure type recommendations.

2. The method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data according to claim 1, characterized in that, The process involves acquiring multi-source heterogeneous monitoring data of ecologically clean small watersheds within a preset assessment period, performing spatiotemporal benchmark unification and outlier cleaning on the multi-source heterogeneous monitoring data, and generating a standardized watershed spatiotemporal feature dataset, including: Acquire remote sensing satellite imagery data, drone aerial orthophotos, and real-time monitoring data from ground-based IoT sensors; Using the small watershed boundary as a geographic reference frame, geographic information system technology is used to unify the spatial resolution and perform projection transformation on the remote sensing satellite image data and the UAV aerial orthophoto; Atmospheric correction and radiometric calibration are performed on the transformed image data. Combined with the timestamps in the real-time monitoring data of the ground IoT sensor, interpolation and fitting are performed on the image data and sensor data to remove outliers and generate the standardized watershed feature library with spatiotemporal alignment.

3. The method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data according to claim 1, characterized in that, The construction of the multidimensional ecological effectiveness state vector includes: Based on the digital elevation model, slope and slope length factors are extracted, and soil loss is calculated by combining rainfall erosion force data to generate soil and water conservation indicators. Extract the spectral reflectance characteristics of the water body, retrieve the concentrations of total nitrogen, total phosphorus, and chemical oxygen demand, and generate the water quality purification index. The normalized vegetation index and landscape fragmentation index are used to assess vegetation cover and ecosystem integrity, and the aforementioned ecological landscape indicators are generated. The soil and water conservation indicators, the water quality purification indicators, and the ecological landscape indicators are vector normalized and then spliced ​​together to generate the multi-dimensional effectiveness evaluation feature set.

4. The method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data according to claim 1, characterized in that, The multidimensional ecological effectiveness state vector is input into a pre-trained deep evaluation decision network. The dual output branches of the deep evaluation decision network are used to calculate the comprehensive ecological score of the watershed and the marginal contribution rate of each governance measure to the ecological effectiveness, including: The multi-dimensional effectiveness evaluation feature set is input into the feature weighting layer of the comprehensive effectiveness evaluation model; The subjective weights of the soil and water conservation indicators, the water quality purification indicators, and the ecological landscape indicators were determined using the analytic hierarchy process. The objective weights of information entropy for each indicator are calculated using the entropy weight method. The subjective weights and objective weights are fused using a game theory-based combined weighting method to obtain the optimal combined weight vector; The multi-dimensional effectiveness evaluation feature set and the optimal combined weight vector are multiplied by a dot product to output the comprehensive watershed ecological index and the individual evaluation scores of each subsystem; wherein, the formula for calculating the comprehensive watershed ecological index is: ； In the formula, As a comprehensive ecological index for the watershed, For the first The optimal combination of weights for the evaluation indicators. For the first The standardized values ​​of the evaluation indicators after vector normalization. To evaluate the total number of indicators.

5. The method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data according to claim 4, characterized in that, The comprehensive effectiveness evaluation model is a machine learning regression model trained based on historical governance data. The specific steps in the training process of the comprehensive effectiveness evaluation model include: Construct a loss function that includes ecological constraints; Obtain data on the input costs and actual output benefits of historical small watershed management projects; The root mean square error between the watershed ecological comprehensive index predicted by the computational model and the actual monitoring value; An ecological response function is introduced to calculate the nonlinear lag effect between the intensity of governance measures and the improvement of environmental quality. The nonlinear lag effect is added as an ecological constraint to the loss function, and the model parameters are optimized using the gradient descent algorithm.

6. The method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data according to claim 5, characterized in that, The construction of the loss function including ecological constraints includes: Set the regularization parameters and learning rate decay factor; The cross-entropy loss between the predicted value and the true label is calculated as a basic term; Extract the runoff coefficient and sediment transport ratio in the hydrological cycle process to construct a physical process constraint matrix; The physical process constraint matrix and the basic terms are weighted and summed to construct the final loss function.

7. The method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data according to claim 1, characterized in that, The dynamic planning and correction of the next stage of governance resource allocation based on the triggering conditions, using the comprehensive ecological score of the watershed and the marginal contribution rate, includes: If the comprehensive ecological index of the watershed is greater than the first preset threshold, the governance effect is determined to be significant. If the comprehensive ecological index of the watershed is less than the second preset threshold, the governance effect is deemed insufficient. For indicators whose individual evaluation scores are below the passing mark, extract the correlation of the corresponding governance measures. Based on the degree of correlation, targeted engineering or plant-based optimization schemes are generated.

8. The method for evaluating the effectiveness of ecological clean watersheds based on multi-source data according to claim 7, characterized in that, If the comprehensive ecological index of the watershed is less than the second preset threshold, it includes: The high-resolution image verification mechanism was activated to retrieve dynamic monitoring data from the past three months. Compare and analyze the deviations between the implementation progress of the remediation measures and the design drawings; Generate a supervision list that includes specific location coordinates and rectification deadlines.

9. The method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data according to claim 1, characterized in that, Also includes: Receive future scenario parameters input by the user, the future scenario parameters including different rainfall frequencies and land use change schemes; The future scenario parameters are combined with the current multi-dimensional effectiveness evaluation feature set and input into the comprehensive effectiveness evaluation model; Predict future trends in the comprehensive ecological index of the watershed under different scenarios and generate a report on long-term maintenance strategies.

10. The method for evaluating the effectiveness of ecological clean-up small watersheds based on multi-source data according to claim 1, characterized in that, After generating an ecological clean watershed effectiveness assessment report that includes spatial layout optimization and recommended measure types, it includes: Upload the assessment report to the cloud management platform; Automatically generate report summaries using natural language processing technology; The assessment report and governance optimization suggestions are pushed to management personnel via mobile devices.

Images