A high-frequency storm event driven urban river ecological health dynamic evolution tracking method

Abstract

The application discloses a high-frequency rainstorm event driven urban river ecological health dynamic evolution tracking method, relates to the technical field of urban river water ecological monitoring and evaluation, and performs mutation detection and trend decomposition on the sequence to obtain a first intermediate result; the intermediate result is substituted back to the weight distribution and mutation detection link to perform secondary operation to obtain a second intermediate result; the second intermediate result is compared with a qualified threshold, and when the second intermediate result does not pass, multi-stage iteration is triggered until the result is qualified or the iteration upper limit is reached; the contribution degree and influence mode of each rainstorm characteristic parameter are analyzed through a random forest and a SHAP method, and the tracking results of an evolution track, a risk threshold and driving attribution are output; the sampling density of a dominant driving factor is reversely checked, and the sampling frequency is doubled when the sampling density is insufficient, a closed loop of monitoring, analysis and feedback to monitoring is formed, and dynamic tracking and law identification of urban river ecological health evolution under rainstorm disturbance are realized.

Description

Technical Field

[0001] This invention relates to the field of urban river water ecology monitoring and assessment technology, specifically a method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events. Background Technology

[0002] Urban rivers, as key carriers of urban water security and ecological balance, often bear the dual impact of drastic hydrological fluctuations and pulsed pollutant inputs during rainstorm events. With the intensification of global climate change and the increasing frequency of high-frequency rainstorm events, urban river ecosystems are subjected to further impacts before they have fully recovered from one disturbance. This cumulative effect may lead to irreversible degradation such as a sharp decline in biodiversity, a decline in self-purification capacity, and even the breakage of ecological corridors.

[0003] However, existing methods for monitoring and assessing the ecological health of urban rivers are mostly based on low-frequency sampling during regular periods or fixed cycles, making it difficult to capture dramatic changes in water quality, abrupt hydrological shifts, and biological responses throughout the entire process before, during, and after rainstorm events. At the evaluation level, traditional methods often rely on static indicator systems or health grading at single time points, failing to characterize the dynamic response trajectory, recovery process, and cumulative degradation effects of ecosystems under continuous rainstorm disturbances. At the driving force analysis level, existing analytical methods are mostly limited to linear statistical models, making it difficult to reveal the complex nonlinear relationships and interactive driving mechanisms between rainstorm characteristics and ecological health. These shortcomings collectively lead to a lack of systematic understanding of the evolutionary patterns of urban river ecological health under high-frequency rainstorm disturbances, hindering the provision of effective dynamic tracking and diagnostic tools for risk warning and resilience management of urban rivers under extreme climate conditions.

[0004] The present invention aims to provide a method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events, in order to solve the technical problem in the prior art that it is impossible to achieve dynamic tracking and evolution law identification of the entire process of urban river ecological health under high-frequency rainstorm disturbance. Summary of the Invention

[0005] The purpose of this invention is to provide a method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events, in order to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events, comprising the following steps: S1. Collect multi-dimensional monitoring data of the entire process of the rainstorm event. The multi-dimensional monitoring data includes hydrological dynamic data, water environment quality data and water ecological response data. The entire process includes three periods: before the rainstorm, during the rainstorm and after the rainstorm. S2. Integrate the multidimensional monitoring data into a time series of the ecological health index; S3. Perform the first operation on the time series to obtain a first intermediate result containing candidate mutation point location information and preliminary fitting parameters of trend components; S4. The first intermediate result is used as a correction term and substituted back into the weight allocation stage of the ecological health index and the mutation point detection stage of the first calculation to perform a second calculation and obtain the second intermediate result. S5. Compare the second intermediate result with the preset qualified threshold. When the second intermediate result meets the qualified threshold, output the second intermediate result. When the second intermediate result does not meet the qualified threshold, trigger multi-stage iteration and repeat the back substitution and the second operation until the qualified threshold is met or the preset iteration cycle limit is reached. S6. Based on the second intermediate result that meets the qualified threshold, establish a nonlinear quantitative relationship between rainstorm characteristic parameters and ecological health response, and identify the risk threshold of system state change. S7. Based on the random forest model and SHAP method, analyze the contribution and impact pattern of each rainstorm characteristic parameter on the evolution of ecological health, and output the tracking results including evolution trajectory, risk threshold and driving attribution. The SHAP method, or SHapley Additive exPlanations method, is a concept of cooperative game solutions based on game theory, used to quantify the contribution of each input feature to the model's prediction results.

[0007] Preferably, the collection frequency for the three periods before, during, and after the rainstorm is dynamically calibrated with the occurrence time of the rainstorm event as the zero point. The collection frequency before the rainstorm is lower than the collection frequency during the rainstorm, and the collection after the rainstorm continues until the parameters of the multidimensional monitoring data recover to the baseline level.

[0008] Preferably, the hydrological dynamic data includes water level, flow velocity and flow rate; the water environmental quality data includes turbidity, pH value and electrical conductivity; and the water ecological response data includes species composition, diversity index and EPT species ratio of macrobenthic animals. The EPT species ratio refers to the ratio of the sum of the number of individuals in the three taxa of Ephemeroptera, Pterygoptera and Trichoptera to the total number of individuals collected.

[0009] Preferably, the method for fusing the multidimensional monitoring data into the time series of the ecological health index is as follows: Principal component analysis is used to reduce the dimensionality of candidate evaluation indicators in the multidimensional monitoring data, retaining indicators with loading values ​​greater than a set threshold to form a core evaluation indicator set. Entropy weighting is used to assign weights to the core evaluation indicator set, and the weighted core evaluation indicators are synthesized according to three dimensions: hydrological integrity, physicochemical integrity, and biological integrity, to obtain the ecological health index corresponding to each monitoring time point. The set threshold is determined based on the cumulative variance contribution rate of the principal components, with the minimum loading value corresponding to the cumulative variance contribution rate reaching a preset requirement serving as the set threshold.

[0010] Preferably, the initial calculation includes a mutation point detection subprocess and a trend decomposition subprocess. The mutation point detection subprocess uses the Pettitt nonparametric test method to scan the time series point by point and identify the time corresponding to the maximum value of the statistic as a candidate mutation point. The trend decomposition subprocess uses a seasonal term trend decomposition method based on local weighted regression to decompose the time series into trend components, seasonal components and residual components, and extracts preliminary fitting parameters from the trend components.

[0011] Preferably, the back-substitution method is as follows: extract the candidate mutation point location information from the first intermediate result, backtrack to the multidimensional monitoring data to extract the response amplitude of each single indicator at the corresponding time, use the response amplitude of each single indicator at the candidate mutation point as the measured sensitivity value, reverse correct the weight assigned by the entropy weight method, resubstitute the corrected weights into the synthesis process of the ecological health index to obtain an updated ecological health index sequence, re-input the updated ecological health index sequence into the mutation point detection stage, perform secondary mutation point detection, and obtain the second intermediate result.

[0012] Preferably, the qualified thresholds include a location deviation threshold, a trend consistency threshold, and a recovery identifiable threshold. The location deviation threshold requires that the time deviation between the mutation point location in the first intermediate result and the second intermediate result does not exceed two monitoring time points. The trend consistency threshold requires that the updated ecological health index sequence and the original ecological health index sequence have the same trend direction during non-mutation periods. The recovery identifiable threshold requires that a monotonic upward trend can be identified in the time series after the rainstorm event for more than three consecutive time points.

[0013] Preferably, in each iteration of the multi-stage iteration, the back substitution and the second operation are re-executed with the latest obtained candidate mutation point position as the correction term. After each iteration is completed, the comparison with the qualified threshold is re-executed until all threshold conditions are met or the preset iteration cycle limit is reached.

[0014] Preferably, the method for establishing the nonlinear quantitative relationship between rainstorm characteristic parameters and ecological health response is as follows: using rainfall intensity, rainfall duration, rainfall interval, and peak runoff as explanatory variables, and using the response amplitude and recovery rate of the ecological health index as response variables, a generalized additive model is constructed. By analyzing the inflection points of the fitted curve, the critical rainfall intensity value at which the response amplitude of the ecological health index undergoes a step change and the critical rainfall interval value at which the recovery rate changes are identified. The critical rainfall intensity value and the critical rainfall interval value are used as the risk threshold.

[0015] Preferably, the method further includes a reverse constraint verification step: taking the top two rainstorm characteristic parameters with the highest contribution identified by the SHAP method as the dominant driving factors, backtracking to verify whether the sampling density of the original monitoring data corresponding to the dominant driving factors in the rainstorm event meets the effective analysis density requirement. When the sampling density is lower than the effective analysis density requirement, the data acquisition frequency of the monitoring parameters corresponding to the dominant driving factors is increased to twice the original frequency, and the increased acquisition frequency is synchronized to the acquisition configuration of the multidimensional monitoring data.

[0016] This invention provides a method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events. It has the following beneficial effects: By constructing a multi-dimensional monitoring data acquisition mechanism for the entire process of rainstorms, the acquisition frequency is dynamically adjusted with the moment of the rainstorm event as the time reference point, and differentiated sampling densities are adopted in different stages of rainstorms. By establishing a fusion and transformation path from monitoring data to ecological health index sequences, information from the three dimensions of hydrology, water quality and biological response is integrated into a unified quantitative representation.

[0017] Through back-substitution optimization and multi-stage iteration mechanisms, the mutation detection results obtained from the initial computation are back-substituted into the weight allocation stage of index synthesis. By correcting the weight coefficients of each indicator, the updated ecological health index sequence can effectively characterize the actual response sensitivity of each indicator during rainstorm disturbances. By setting qualified thresholds for location deviation, trend consistency, and recovery identifiability, the detection result is only considered a qualified output when all three thresholds are met simultaneously; otherwise, multi-stage iterations are triggered until the result stabilizes or the iteration limit is reached. This mechanism enables the identification process of the ecological health evolution trajectory to possess self-verification, self-optimization, and self-convergence capabilities, and the output mutation point locations and recovery trajectories exhibit stability and interpretability.

[0018] Using a generalized additive model, a nonlinear quantitative relationship between rainstorm characteristic parameters and ecological health response is established, and risk thresholds for system state abrupt changes are identified through inflection point analysis. Based on this, a random forest model and the SHAP method are employed to analyze the contribution and impact patterns of each rainstorm characteristic parameter on ecological health evolution, achieving a joint output of evolution trajectory, risk thresholds, and driving attribution information.

[0019] By employing a reverse constraint verification step, the dominant driving factors identified in the driving analysis phase are traced back to the monitoring data acquisition phase. This verifies the sampling density of the monitoring data corresponding to these factors during rainstorm events. When the sampling density is insufficient, the acquisition frequency of the corresponding monitoring parameters is automatically increased, allowing the acquisition strategy to adaptively adjust based on the analysis results. This forms a complete closed loop from monitoring to analysis and back to monitoring. This solves the problem in traditional solutions that cannot dynamically track and identify the evolution patterns of the entire process of urban river ecological health under high-frequency rainstorm disturbances. It can provide dynamic tracking and diagnostic support for ecological risk early warning and resilience management of urban rivers under rainstorm events. Attached Figure Description

[0020] Figure 1 This is a closed-loop flowchart of the method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events, as described in this invention. Figure 2 This is a schematic diagram of the multi-order iteration and reverse constraint verification feedback loop of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 and Figure 2 This invention provides a method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events, the implementation of which is as follows: First, for the target city's rivers, monitoring points are set up at key sections of the main stream, inlets of important tributaries, major sewage outlets, and ecologically sensitive areas. Hydrological monitoring equipment and in-situ water quality sensors are simultaneously deployed at each monitoring point, and biological sampling sections are established. When a weather forecast indicates a rainstorm event, the actual time of the rainstorm is used as the zero point for full-process monitoring. If a weather forecast indicates a rainstorm event but it does not actually occur, the monitoring system does not initiate the rainstorm-intensified data collection process, but continues to operate at the regular data collection frequency and continuously stores data, waiting for the next weather forecast to trigger it. If a rainstorm actually occurs but the forecast does not cover it in time, the monitoring system will use the actual start time of the rainfall as the zero point. After confirming the rainfall, it will immediately switch to encrypted acquisition mode and extract conventional frequency monitoring data from the stored conventional frequency monitoring data at several time points before the rainfall as background reference values. Background data will be collected before the rainstorm, encrypted acquisition will be carried out during the rainstorm, and continuous acquisition will be carried out after the rainstorm until all monitoring parameters return to the baseline level. The collected data includes hydrological dynamic data such as water level, flow velocity, and flow rate; water environmental quality data such as turbidity, pH value, and conductivity; and water ecological response data such as species composition, diversity index, and EPT species ratio of macrobenthic animals.

[0023] When collecting multi-dimensional monitoring data for the entire process of a rainstorm, the collection frequency is dynamically calibrated based on the actual time of the rainstorm event, rather than arranging collection tasks according to a fixed calendar cycle.

[0024] When meteorological forecasts indicate that a rainstorm may occur in the target watershed, the monitoring system enters a standby state. At this time, each monitoring device operates at the normal acquisition frequency to record the background environmental conditions before the rainstorm. The hydrological dynamic data, water environment quality data, and water ecological response data collected during this stage together constitute the background reference value before the rainstorm.

[0025] Once the monitoring system confirms the occurrence of a rainstorm event using rain gauge or weather radar data, it immediately switches the acquisition frequency of all in-situ monitoring equipment to an encrypted state, using this confirmation moment as the zero point. In this state, the hydrological monitoring equipment continuously collects data on water level, flow velocity, and flow rate at intervals higher than normal. Similarly, the in-situ water quality sensors continuously detect turbidity, pH, and conductivity at these encrypted intervals until the main rainfall event ends and river levels begin to recede. This encrypted acquisition phase aims to obtain continuous monitoring records during the period of drastic changes in river hydrological conditions and the pulse input of pollutants during the rainstorm impact.

[0026] After the main rainfall event of the rainstorm ends and river levels stabilize, the monitoring system continues to collect data, but the collection frequency switches from an encrypted state to a recovery monitoring frequency. This recovery monitoring frequency is lower than the encrypted collection frequency during the rainstorm but higher than the regular collection frequency before the rainstorm. During this phase, each monitoring device continues to operate and record various parameters. The monitoring system compares the currently collected parameter values ​​hourly with the baseline reference values ​​collected before the rainstorm to determine whether each parameter has recovered to a level close to the baseline reference value. This recovery monitoring phase continues until each monitoring parameter stabilizes within the baseline level range for multiple consecutive time points. At this point, it is determined that the ecosystem has completed the basic recovery after this round of rainstorm disturbance. The monitoring system then stops the specialized collection process for this rainstorm event and switches the collection frequency back to the regular frequency, awaiting the triggering of the next rainstorm event. Through the above-mentioned collection frequency control method, which uses the time of the rainstorm event as the zero point for dynamic calibration, the collection frequency before the rainstorm is lower than the collection frequency during the rainstorm, and the collection after the rainstorm continues until each parameter recovers to the baseline level.

[0027] Regarding the acquisition of hydrological dynamic data, water level monitoring equipment and flow velocity monitoring equipment are deployed at each monitoring section of the target river. Water level monitoring equipment is fixedly installed on vertical water gauge piles or bridge piers on both sides of the riverbank, with its sensing end extending below the water surface to continuously sense water level changes and output the water level data as an electrical signal. Flow velocity monitoring equipment is installed at the bottom of the river channel or suspended at a predetermined depth in the water body. It measures the flow velocity of the water body by emitting sound waves or electromagnetic waves and receiving the echoes, outputting the flow velocity data. Based on the known geometric parameters of the river channel at the monitoring section, the cross-sectional area is determined according to the water level data, and the product of the cross-sectional area and the flow velocity data is taken as the flow rate data for that section.

[0028] For the acquisition of water environmental quality data, in-situ water quality sensors are deployed at a predetermined depth below the water surface at each monitoring section. These in-situ water quality sensors integrate a turbidity detection unit, a pH detection unit, and a conductivity detection unit. The turbidity detection unit outputs turbidity data by emitting light signals and measuring changes in the intensity of transmitted or scattered light; the pH detection unit outputs pH data based on the potential difference between the reference electrode and the measuring electrode; and the conductivity detection unit outputs conductivity data by applying voltage to the water body and measuring the current response between the electrodes. These detection units operate continuously at set acquisition intervals, converting the detection results into digital signals and transmitting them in real time to the data collection terminal.

[0029] Regarding the collection of aquatic ecological response data, macrobenthic animal sampling was conducted at each biological monitoring section according to the preset sampling time nodes. During sampling, the sediment sampler was submerged to the river bottom and closed to capture bottom sediment samples. The collected bottom sediment samples were placed in a mesh sieve and repeatedly rinsed with on-site river water to remove fine particles from the bottom sediment through the mesh. Visible organisms remaining on the sieve were picked out one by one and placed in sample bottles. Fixative was added to the sample bottles to preserve the biological samples. In the laboratory, the preserved biological samples were observed under a stereomicroscope. Based on the morphological characteristics of the organisms, each individual was identified to the genus or species level taxonomic unit, and the number of individuals in each taxonomic unit was counted. Based on the identification and counting results, the Shannon diversity index, species richness index, and the ratio of the number of EPT taxonomic units (Mayflies, Pterygoptera, and Trichoptera) to the total number of taxonomic units were calculated. The Shannon diversity index is calculated by multiplying the proportion of individuals in each taxonomic unit to the total number of individuals and taking the negative value of the product of their logarithmic values. The EPT taxonomic unit proportion is calculated by the ratio of the sum of the number of individuals in the three taxonomic units of Ephemeroptera, Pterygoptera and Trichoptera to the total number of individuals collected.

[0030] Secondly, after aggregating all collected monitoring data, the candidate evaluation indicators were screened for dimensionality reduction. Specifically, the data matrix was standardized, and the data structure was tested for suitability for principal component analysis. After confirmation of suitability, the maximum variance rotation method was used to extract principal components, retaining indicators with loading values ​​greater than a set threshold as the core evaluation indicator set. For the core evaluation indicator set, weights were assigned based on the dispersion of each indicator value; indicators with greater dispersion had higher weights. The weighted indicators were then synthesized according to three dimensions: hydrological integrity, physicochemical integrity, and biological integrity, to obtain the ecological health index corresponding to each monitoring time point, forming a continuous time series.

[0031] Next, the initial computational analysis was performed on the aforementioned ecological health index time series. For mutation point detection, a non-parametric test method was used to scan the time series point by point, constructing a statistical sequence and identifying the time corresponding to its maximum value as a candidate mutation point. Simultaneously, the significance test index corresponding to this candidate mutation point was calculated. When the index is less than or equal to a set threshold, the point is determined to be a statistically significant mutation point and marked as the starting point of a drastic response. For trend decomposition, a seasonal trend decomposition method based on local weighted regression was used to decompose the time series into trend components, seasonal components, and residual components. Preliminary fitting parameters for the long-term evolution of the ecological health state were extracted from the trend components. After the initial computation, a first intermediate result containing candidate mutation point location information and preliminary fitting parameters for the trend components was obtained.

[0032] Subsequently, the first intermediate result is used as a correction term and substituted back into the indicator weight allocation and mutation point detection stages. Specifically, the substitution method is as follows: The candidate mutation point location information is extracted from the first intermediate result; the measured response amplitude of each individual indicator at the corresponding time point is extracted from the original multidimensional monitoring data; the response amplitude of each individual indicator at the mutation point is used as the measured sensitivity value; the assigned weights are then reversed, with positive weight increments assigned to indicators with higher response amplitudes, negative weight reductions assigned to indicators with lower response amplitudes, and the original weights of indicators with intermediate response amplitudes remaining unchanged; the corrected weights are then substituted back into the ecological health index synthesis process to obtain an updated ecological health index sequence; this updated sequence is then re-input into the mutation point detection stage to perform a second mutation point detection, yielding a second intermediate result containing the updated candidate mutation point locations and their significance indicators.

[0033] Next, the second intermediate result is compared with preset qualification thresholds. Qualification thresholds include a location deviation threshold, a trend consistency threshold, and a recovery identifiability threshold. The location deviation threshold is used to compare whether the time deviation of the mutation point position between the first and second intermediate results exceeds two monitoring time points; the trend consistency threshold is used to determine whether the trend direction of the updated ecological health index sequence and the original ecological health index sequence is consistent during non-mutation periods; the recovery identifiability threshold is used to determine whether a monotonic upward trend can be identified for more than three consecutive time points in the time series after the rainstorm event. When all three checks meet the preset threshold requirements, the second intermediate result is determined to be a qualified output result; when any check fails to meet the preset threshold requirements, multi-stage iteration is triggered, using the latest obtained candidate mutation point position as a correction term to re-execute the aforementioned back-substitution and secondary operations. After each iteration, the threshold comparison is performed again until all threshold conditions are met or the preset iteration limit is reached, and the final ecological health index sequence and its mutation point position that have passed the threshold checks are output.

[0034] Furthermore, based on the ecological health index sequence and its mutation point locations output after threshold verification, a nonlinear quantitative relationship between rainstorm characteristic parameters and ecological health response is established. Rainfall intensity, rainfall duration, rainfall interval, and peak runoff are used as explanatory variables, while the response amplitude and recovery rate of the ecological health index are used as response variables. A generalized additive model is employed for fitting, with each explanatory variable entering the model as a smooth function to characterize its nonlinear relationship with the response variables. By analyzing the inflection points of the fitted curves, the critical rainfall intensity value at which the ecological health index response amplitude undergoes a step change, and the critical rainfall interval value at which the recovery rate changes, are identified. These critical values ​​are used as risk thresholds for sudden changes in system state.

[0035] Finally, a random forest model was used to simulate the mapping relationship between rainstorm characteristic parameters and the ecological health index. The key rainstorm characteristic parameters selected above were used as input features, and the ecological health index and its recovery rate were used as prediction targets. This pre-trained random forest regression model was applied. The random forest model was pre-trained, and its training data came from historical monitoring data of each rainstorm event in the target river and synchronous records of the ecological health index for the corresponding time periods. After training, the model was applied to the analysis of current monitoring data. After model training, the SHAP method was used to analyze the model prediction results, calculating the contribution value of each input feature at each sample point. Based on the positive or negative sign and magnitude of the contribution value, the positive or negative effects of each rainstorm characteristic parameter on the ecological health index were quantitatively distinguished. The ecological health index time series output after threshold verification, the identified mutation points and critical risk thresholds, and the analyzed contribution and influence patterns of key driving factors were integrated to form a tracking result containing evolution trajectory identification information, risk threshold warning information, and driving attribution information, which was then output. The number of decision trees in the random forest regression model was determined according to the sample size, and the number of split nodes in each tree was set to one-third of the number of input features. During model training, out-of-bag samples are used for internal validation, and root mean square error is used as the evaluation metric for model accuracy.

[0036] When integrating multidimensional monitoring data into a time series of ecological health indices, all monitoring data collected during the entire monitoring period are first organized according to the monitoring time points to form an original data matrix. Each row of this matrix corresponds to a monitoring time point, and each column corresponds to an evaluation index. The evaluation indexes cover water level, flow velocity and flow rate in hydrological dynamic data, turbidity, pH value and conductivity in water environmental quality data, and species composition, diversity index and EPT species proportion in water ecological response data.

[0037] The original data matrix is ​​standardized to convert the data in each indicator column into dimensionless standardized values, eliminating the impact of different dimensions on the comparability between indicators. After standardization, the data matrix is ​​tested for suitability using principal component analysis (PCA). The suitability for dimensionality reduction is determined by calculating the sufficiency of sampling measure and the sphericity test. Once the data is deemed suitable, the correlation coefficient matrix of the standardized data matrix is ​​calculated. The eigenvalues ​​and eigenvectors of this correlation coefficient matrix are then solved. The eigenvalues ​​are arranged in descending order, and the proportion of each eigenvalue to the sum of all eigenvalues ​​is calculated as the variance contribution rate. The variance contribution rates are accumulated sequentially in descending order of eigenvalues. The first few principal components whose accumulated values ​​reach a set threshold are retained. Each principal component is a linear combination of the original indicators, and the loading values ​​of each original indicator on that principal component represent its representativeness. For each retained principal component, the original indicators whose loading values ​​on any principal component are greater than a set threshold are retained, and these retained indicators are used as the core evaluation indicator set.

[0038] For each indicator in the core evaluation indicator set, the entropy weight method is used to allocate weights. Specifically, the standardized values ​​of each indicator at each monitoring time point are normalized according to the monitoring time point, so that the proportion of the value at each monitoring time point to the sum of the values ​​at all time points of that indicator is taken as the weight value of that time point for that indicator; the information entropy of each indicator is calculated based on this proportion value. The smaller the information entropy, the greater the variability of the indicator between monitoring time points, i.e., the greater the amount of information carried by the indicator; the entropy weight of each indicator is calculated based on the information entropy, with indicators having lower information entropy receiving higher weights, and the sum of the weights of all indicators is normalized to one.

[0039] The core evaluation indicators are categorized into three dimensions based on their respective monitoring dimensions: hydrological integrity, physicochemical integrity, and biological integrity. The hydrological integrity dimension includes water level, flow velocity, and discharge; the physicochemical integrity dimension includes turbidity, pH, and conductivity; and the biological integrity dimension includes species composition, diversity index, and EPT species proportion. Within each dimension, the standardized values ​​of each indicator belonging to that dimension are multiplied by their corresponding entropy weights and summed to obtain the dimension score. Then, the scores of the three dimensions are multiplied by their respective dimension weights and summed to obtain the ecological health index for that monitoring point. The dimension weights are determined based on the proportion of the hydrological integrity, physicochemical integrity, and biological integrity dimensions in the total entropy weights of the entire core evaluation indicator set. The ecological health indices calculated for each monitoring point are arranged chronologically to form the time series of the ecological health indices.

[0040] After obtaining the time series of the ecological health index, the first calculation is performed on the time series. This first calculation includes two parallel sub-processes: a mutation point detection sub-process and a trend decomposition sub-process.

[0041] In the mutation point detection sub-process, the Pettitt nonparametric test method is used to scan the ecological health index time series point by point. Specifically, each time point in the time series is sequentially taken as a hypothetical mutation location, and the relationship between the data values ​​in the two subsequences before and after that time point is statistically analyzed. For a hypothetical mutation location, the data values ​​before and after that location are compared successively, and the difference between the number of times the former is less than the latter and the number of times the former is greater than the latter is counted. This difference is used as the statistic for that hypothetical location. After traversing all time points to obtain the statistic sequence, the time point corresponding to the maximum value is selected as the candidate mutation point location. This process is repeated to traverse all time points, obtaining a statistic sequence corresponding to each time point. The time point corresponding to the maximum statistic in this statistic sequence is selected as the candidate mutation point location. After determining the candidate mutation point location, a significance index is calculated based on the value of the maximum statistic and its distribution in the entire statistic sequence. The smaller the value of this significance index, the less likely the difference between the two subsequences before and after the candidate mutation point location is to be caused by random fluctuations. When the significance index is less than or equal to the preset threshold value, the candidate mutation point is determined to be a statistically significant mutation point and marked as the starting point of a violent response; the threshold value is 0.05, which is the α level value commonly used in significance testing.

[0042] In the trend decomposition sub-process, a seasonal trend decomposition method based on local weighted regression is used to decompose the ecological health index time series. This decomposition method treats the time series as the sum of three parts: trend component, seasonal component, and residual component. Specifically, the seasonal component is first removed from the original time series. The series after removing the seasonal component is then smoothed using local weighted regression to obtain a preliminary estimate of the trend component. During this smoothing process, for each time point to be estimated in the trend component, a time window is extracted centered on that time point. Weights are assigned to each data point falling within this window; data points closer to the center time point receive higher weights, and data points farther from the center time point receive lower weights. When the distance between a data point and the center time point exceeds the window boundary, the weight drops to zero. Weighted regression fitting is then performed on the data points within the window according to the above weighting method to obtain the trend estimate at the center time point. The window is moved point by point, traversing all time points of the entire time series to obtain a complete trend component estimate. After subtracting the obtained trend component from the original time series, the remaining part is averaged according to the seasonal cycle to obtain the seasonal component. After subtracting both the trend component and the seasonal component from the original time series, the remaining portion is used as the residual component. Preliminary fitting parameters characterizing the long-term evolution of ecological health status are extracted from the trend component, including the slope variation characteristics and numerical range of the trend component.

[0043] The period length of the seasonal component is determined based on the actual time span of the collected data. If the collected data covers multiple alternating cycles of wet and dry seasons, the seasonal component is extracted with a one-year period as the complete cycle; if the collected data only covers a single flood season, only the trend component and residual component are extracted, and the seasonal component is not extracted.

[0044] The candidate mutation point locations and their corresponding significance indices output by the mutation point detection sub-process, together with the preliminary fitting parameters of the trend components output by the trend decomposition sub-process, constitute the first intermediate result.

[0045] After obtaining the first intermediate result, the process proceeds to the back-substitution optimization stage. The core of this back-substitution stage lies in using the information provided by the candidate mutation point locations identified in the initial calculation to reverse-correct the index weight allocation in the ecological health index synthesis process, and then re-execute the comprehensive index synthesis and mutation point detection with the corrected weights, so that the detection results and the inherent response characteristics of the data mutually corroborate each other.

[0046] In practice, the location information of candidate mutation points is first extracted from the first intermediate results. These candidate mutation points correspond to a specific monitoring time point in the time series, representing a statistically significant change in the ecological health index at that time point. This specific monitoring time point is used as a backtracking index to return to the original multidimensional monitoring data, extracting the specific values ​​of all individual monitoring indicators at that time point, including but not limited to the measured values ​​of water level, flow velocity, flow rate, turbidity, pH value, conductivity, species composition, diversity index, and EPT species proportion. Simultaneously, the values ​​of the same set of individual indicators from several time points before and after this time point are also extracted to construct a reference benchmark for the change range of each indicator at the mutation point.

[0047] Based on this, for each individual indicator, the magnitude of its change at the candidate mutation point relative to the pre-mutation reference value is calculated. The specific calculation method for this magnitude is as follows: the average value of the indicator at multiple consecutive time points before the candidate mutation point is used as the pre-mutation reference value; the measured value at the candidate mutation point is used as the mutation value; and the degree of deviation of the mutation value from the pre-mutation reference value is used as the response magnitude of the indicator at the candidate mutation point. This process is repeated for all individual indicators to obtain a sequence of response magnitudes for each individual indicator at the candidate mutation point. The response magnitude is defined as a comprehensive representation of the direction and degree of deviation of the mutation value from the pre-mutation reference value. When an increase in the indicator value represents ecological degradation, the response magnitude is represented by a positive value (the difference between the mutation value and the reference value); when a decrease in the indicator value represents ecological degradation, the response magnitude is represented by a positive value (the difference between the reference value and the mutation value). Before participating in weight correction, the response magnitudes of each indicator are converted into dimensionless relative values.

[0048] The response magnitude of each individual indicator is used as a representation of its measured sensitivity. Indicators with higher response magnitudes indicate that they undergo significant changes at the moment of system abrupt change caused by rainstorm disturbances, meaning they have high measured sensitivity to rainstorm disturbances. Conversely, indicators with lower response magnitudes indicate that their changes are insignificant at the moment of abrupt change, indicating low measured sensitivity to rainstorm disturbances. Based on the ranking of the response magnitudes of the indicators, the weights of the indicators initially assigned using the entropy weight method are reverse-corrected. The correction rules are as follows: indicators with higher response magnitudes are given a positive weight increment to increase their contribution to the composite index synthesis; indicators with lower response magnitudes are given a negative weight reduction to decrease their contribution to the composite index synthesis; indicators with intermediate response magnitudes retain their original weights without adjustment. The sum of the corrected weights of all indicators is normalized to ensure that the dimensions and numerical range of the composite index remain unchanged before and after correction.

[0049] The updated weights obtained after the above reverse correction are then substituted back into the synthesis process of the ecological health index. Following the same method as the previous synthesis, the standardized values ​​of each individual indicator at each monitoring time point are multiplied by their corresponding updated weights and summed to obtain the new ecological health index values ​​at each monitoring time point. The new values ​​at each time point are then arranged in chronological order to form the updated ecological health index sequence.

[0050] The updated ecological health index sequence is re-input into the mutation point detection stage. Using the same Pettitt nonparametric test method as in the initial calculation, the updated sequence is scanned point-by-point again to construct a new statistical sequence. The time point corresponding to the new maximum statistical value is identified as the candidate mutation point location, and the update significance index corresponding to this candidate mutation point is calculated. After all the above back-substitution operations, a second intermediate result containing the updated ecological health index sequence, the candidate mutation point locations, and their update significance index is obtained, completing the entire closed loop from the initial detection result back-substitution to index synthesis and secondary detection.

[0051] After obtaining the second intermediate result, it is compared with a preset acceptable threshold to determine whether the back-substitution optimization and secondary calculation have produced sufficiently stable detection results. The acceptable threshold consists of three components: the positional deviation threshold, the trend consistency threshold, and the recovery identifiability threshold.

[0052] The positional deviation threshold is used to evaluate the consistency of mutation point locations in two independent detection results. The time interval between the two mutation point locations is calculated by comparing the monitoring time point number corresponding to the candidate mutation point in the first intermediate result with the monitoring time point number corresponding to the updated candidate mutation point in the second intermediate result. When this time interval is less than or equal to the interval between two monitoring time points, the second intermediate result is considered to have passed the positional deviation threshold verification; when the time interval exceeds the interval between two monitoring time points, the second intermediate result is considered to have failed the positional deviation threshold verification. The basic basis for this verification is that, under the premise that the core data information has not changed substantially, weight correction should not cause a significant drift in the mutation point location. If the drift is too large, it indicates that the back-substitution optimization process has introduced unacceptable signal distortion.

[0053] The trend consistency threshold is used to assess whether the overall evolution trend of the ecological health index remains stable before and after weight correction. The original ecological health index sequence corresponding to the first intermediate result is compared with the updated ecological health index sequence corresponding to the second intermediate result within the same interval. Data from non-mutation periods, excluding the time points of candidate mutations and several adjacent time points, are selected from both sequences. The direction of change for both sequences within these non-mutation periods is calculated, and the upward or downward trends of the two sequences in each corresponding interval are compared to see if they are consistent. When the two sequences show the same direction of change in each corresponding interval of the non-mutation period, the second intermediate result is deemed to have passed the trend consistency threshold verification; when there are instances where the two sequences show opposite directions of change in any corresponding interval, the second intermediate result is deemed to have failed the trend consistency threshold verification. The basic basis for this verification is that the purpose of weight correction is to strengthen the contribution of indicators sensitive to mutations, rather than to change the overall evolution direction of ecological health. If the trend direction reverses after correction, it indicates overcorrection.

[0054] The recoverable identifiable threshold is used to assess whether the ecological health index sequence after a rainstorm event contains identifiable recovery trajectory information. Starting from the end of the rainstorm event, the changes in the ecological health index values ​​at each subsequent point in the sequence are examined sequentially. Within a range of consecutive points after the rainstorm, it is determined whether there are at least three consecutive points where the ecological health index value at the later point is higher than that at the previous point, forming a continuous monotonic upward trend. When a monotonic upward trend of more than three consecutive points can be identified, the second intermediate result is deemed to have passed the recoverable identifiable threshold verification; otherwise, it is deemed to have failed the verification. The basic basis for this verification is that if the ecological health index shows no signs of recovery after the rainstorm, it indicates that the system may have entered a state of continuous degradation. In this case, the existing data no longer provides the conditions for further recovery trajectory modeling, and further iterative calculations are unnecessary.

[0055] When the second intermediate result simultaneously meets the above three threshold checks, the second intermediate result is used as a qualified output result and enters the subsequent modeling stage; when the second intermediate result fails to simultaneously meet the above three threshold checks, a multi-stage iteration mechanism is triggered.

[0056] When the second intermediate result fails to simultaneously satisfy any one of the three checks—position deviation threshold, trend consistency threshold, and recovery identifiability threshold—a multi-stage iteration mechanism is triggered.

[0057] In this multi-stage iterative mechanism, the latest obtained candidate mutation point position is used as the correction term for the new round of iteration. This candidate mutation point position originates from the output of the previous round's back-substitution and secondary computation; that is, the candidate mutation point position identified in the previous round after reverse weight correction, resynthesizing the ecological health index sequence, and re-performing mutation point detection. Using the candidate mutation point position obtained in this round as the correction term for the new round of iteration, the complete back-substitution and secondary computation process is re-executed: extracting the response amplitude of each indicator at the mutation point from the original monitoring data, reverse-correcting the entropy weights of each indicator based on the response amplitude ranking, re-substituting the corrected weights into the ecological health index to synthesize a new updated index sequence, and re-performing mutation point detection on the new updated sequence. Specifically, the monitoring time point corresponding to the updated candidate mutation point is extracted, and the measured values ​​of each single indicator at that time point are extracted from the original multidimensional monitoring data. The response amplitude of each indicator at the updated mutation point is calculated, and the entropy weight of each indicator is re-corrected based on the ranking of the response amplitude. The corrected weights are then substituted back into the synthesis process of the ecological health index to obtain the updated ecological health index sequence. The mutation point detection is then re-executed on the updated sequence to obtain the updated candidate mutation point positions.

[0058] After each round of the above iterative calculation, the updated candidate mutation point positions and corresponding updated ecological health index sequences output by that round are compared again with preset qualified thresholds, including position deviation threshold, trend consistency threshold, and recovery identifiability threshold. That is, position deviation verification, trend consistency verification, and recovery identifiability verification are re-executed. If the output of this round meets all three threshold requirements, the iteration is terminated, and the output of this round is taken as a qualified output result. If any one of the threshold requirements is still not met, the next round of iteration continues, using the updated candidate mutation point positions obtained in this round as the correction term for the new round, and repeating the complete process of back substitution, secondary calculation, and threshold comparison.

[0059] The aforementioned iterative process continues until the output of a certain iteration satisfies all three threshold conditions. Simultaneously, a preset upper limit for the number of iterations is implemented in the multi-stage iterative mechanism. When this upper limit is reached, the iteration is forcibly terminated even if the current output does not fully meet all threshold conditions, and the latest output updated ecological health index sequence and its corresponding candidate mutation point positions are taken as the current optimal output. This preset upper limit for the number of iterations is designed to prevent the iterative mechanism from falling into a non-convergent state, leading to unlimited consumption of computational resources, in cases where the data itself has a low signal-to-noise ratio or the ecological response process is excessively chaotic. Through this multi-stage iterative mechanism, continuous correction capabilities are achieved when a single iteration is insufficient to pass the threshold verification, enabling the detection results to adaptively approach a qualified state that meets the requirements of stability and interpretability.

[0060] After passing threshold verification and outputting the ecological health index sequence and its corresponding mutation point location, the next step is to establish the nonlinear quantitative relationship between rainstorm characteristic parameters and ecological health response.

[0061] First, the explanatory and response variables used in the modeling were determined. The explanatory variables were extracted from the monitoring records of the rainstorm event, including rainfall intensity, rainfall duration, rainfall interval, and peak runoff. Rainfall intensity was calculated based on the cumulative rainfall per unit time during the rainstorm event; rainfall duration was determined by the total duration of the rainstorm event from start to finish; rainfall interval was determined by the time difference between the start times of two adjacent rainstorm events; and peak runoff was determined by the maximum flow recorded at the monitoring section during the rainstorm event. The response variables were extracted from the ecological health index sequence that passed threshold validation, including response amplitude and recovery rate. The response amplitude was determined by the degree of decline in the ecological health index after the rainstorm event relative to the pre-rainstorm baseline level; and the recovery rate was determined by the change per unit time in the ecological health index as it recovered from its lowest point to a stable level after the rainstorm event.

[0062] After determining the explanatory and response variables, a generalized additive model is constructed. In this model, each explanatory variable is introduced as a smoothing function, rather than assuming a fixed linear functional relationship with the response variable. The model links the combination of smoothing functions of the explanatory variables to the expected value of the response variable through a link function. During model fitting, the smoothing functions of each explanatory variable are estimated from the data using locally weighted regression, without pre-defining the function form, thus characterizing various nonlinear relationships that may exist between the explanatory and response variables. The smoothing functions of each explanatory variable are estimated using penalized splines, and the penalty parameter is determined through cross-validation to achieve a balance between fitting accuracy and smoothness.

[0063] After model fitting, inflection point analysis is performed on the smoothed function curves of each explanatory variable. The specific implementation of inflection point analysis is as follows: The first and second derivatives of the smoothed function curve for each explanatory variable are calculated. Points where the first derivative changes from positive to negative or vice versa are identified as inflection points indicating a reversal in the curve's direction. Points where the second derivative is zero and the signs of the second derivatives on both sides change are identified as inflection points indicating changes in the curve's curvature. Among all inflection points, the focus is on the locations where the response variable undergoes a step or abrupt change with the explanatory variables. For the explanatory variable of rainfall intensity, the critical position on its smoothed function curve where the response amplitude of the ecological health index undergoes a step change is identified, and the rainfall intensity value corresponding to this position is taken as the critical rainfall intensity value. For the explanatory variable of rainfall interval, the critical position on its smoothed function curve where the recovery rate of the ecological health index changes significantly is identified, and the rainfall interval value corresponding to this position is taken as the critical rainfall interval value.

[0064] The aforementioned critical rainfall intensity and critical rainfall interval values ​​are used as risk thresholds for abrupt changes in the system's state. These risk thresholds have the following implications: when the rainfall intensity of an actual rainstorm event exceeds the critical rainfall intensity value, it indicates that the event may trigger a step-like deterioration in the ecosystem's health; when the rainfall interval between two adjacent rainstorm events is less than the critical rainfall interval value, it indicates that the ecosystem may suffer the next shock before the previous disturbance has fully recovered, and the cumulative effect may lead to continuous degradation of the system's state. These risk thresholds are used to determine the risk level and trigger early warnings for rainstorm events in subsequent real-time monitoring.

[0065] After analyzing the contribution of various rainfall characteristic parameters to ecological health evolution using the random forest model and the SHAP method, the process moves to the reverse constraint verification stage. This stage uses the driving analysis results to reverse verify the adequacy of the monitoring data collection stage, and adaptively adjusts the collection strategy based on the verification results.

[0066] First, from the contribution analysis results of each rainstorm characteristic parameter output by the SHAP method, the top two rainstorm characteristic parameters in terms of contribution are extracted, and these two parameters are taken as the dominant driving factors. This contribution ranking is determined based on the degree of influence of each rainstorm characteristic parameter on the prediction target in the random forest model. The top two parameters represent the two factors with the strongest impact on ecological health evolution among all rainstorm characteristic parameters included in the analysis. For example, it may be rainfall intensity and rainfall interval, or rainfall intensity and runoff peak, depending on the analysis results of the actual monitoring data.

[0067] After identifying the dominant driving factor, the original monitoring data collection records of the entire rainstorm event are traced back to extract the original monitoring data items directly corresponding to the dominant driving factor. If the dominant driving factor is rainfall intensity or rainfall duration, the rain gauge records for the corresponding time period in the meteorological monitoring data are traced back; if the dominant driving factor is rainfall interval, the timestamp records of multiple rainstorm events are traced back; if the dominant driving factor is peak runoff, the flow records for the corresponding time period in the hydrological monitoring data are traced back. The actual number of sampling points for this original monitoring data item in each time period of the entire rainstorm event is counted, and the actual sampling density is calculated by comparing the actual number of sampling points with the total duration of the entire rainstorm event. At the same time, based on the data attribute characteristics of the dominant driving factor and the minimum data volume requirements for this data item in subsequent modeling analysis, the effective analysis density requirement is determined. The effective analysis density requirement is determined based on the data change rate of the dominant driving factor. The entire rainstorm event is divided into three stages: before the rainstorm, during the rainstorm, and after the rainstorm. The effective analysis density requirements for each stage are as follows: during the rainstorm stage, there must be at least two sampling points before and after the peak of the driving factor; during the recovery stage, there must be at least one sampling point at each inflection point of the ecological health index trend. The minimum sampling density for each stage is determined based on a combination of the total duration of the stage and the required number of sampling points. When the actual sampling density is lower than this minimum sampling density, it is considered insufficient.

[0068] The calculated actual sampling density is compared with the preset effective analysis density requirement. When the actual sampling density reaches or exceeds the effective analysis density requirement, it indicates that the current monitoring scheme's data collection sufficiency for the dominant driving factor meets the needs of subsequent analysis, and no adjustment is required; the original collection configuration can be maintained. When the actual sampling density is lower than the effective analysis density requirement, it indicates that the current monitoring scheme's data collection density for the dominant driving factor is insufficient to support subsequent detailed analysis, and the collection strategy needs to be modified.

[0069] If adjustments are needed, the data acquisition frequency for the monitoring parameters corresponding to the dominant driving factor will be doubled. This frequency increase is based on the baseline frequency in the original acquisition configuration and does not change the acquisition frequency of the monitoring parameters corresponding to other non-dominant driving factors. After the frequency increase, the original monitoring data items corresponding to the dominant driving factor will receive twice the number of sampling points per unit time, thereby compensating for the insufficient sampling density of the original data.

[0070] The increased acquisition frequency is synchronized to the acquisition configuration of the multidimensional monitoring data, serving as the updated acquisition parameters for subsequent heavy rain events. This synchronization ensures that the frequency increase becomes a persistent configuration update. During subsequent heavy rain events, the monitoring system automatically performs data acquisition corresponding to the dominant driving factor at the updated acquisition frequency, rather than requiring manual readjustment each time. In this way, the output of the driving analysis phase inversely constrains the configuration parameters of the data acquisition phase, creating a complete closed loop for the entire ecological health dynamic evolution tracking method, from acquisition to analysis and then back to acquisition configuration.

[0071] Through the implementation of the above steps, a complete tracking process of urban river ecological health driven by rainstorm events was achieved, from data collection, index construction, mutation identification, iterative optimization, threshold determination to driver analysis.

[0072] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events, characterized in that, Includes the following steps: S1. Collect multi-dimensional monitoring data of the entire process of the rainstorm event. The multi-dimensional monitoring data includes hydrological dynamic data, water environment quality data and water ecological response data. The entire process includes three periods: before the rainstorm, during the rainstorm and after the rainstorm. S2. Integrate the multidimensional monitoring data into a time series of the ecological health index; S3. Perform the first operation on the time series to obtain a first intermediate result containing candidate mutation point location information and preliminary fitting parameters of trend components; S4. The first intermediate result is used as a correction term and substituted back into the weight allocation stage of the ecological health index and the mutation point detection stage of the first calculation to perform a second calculation and obtain the second intermediate result. S5. Compare the second intermediate result with the preset qualified threshold. When the second intermediate result meets the qualified threshold, output the second intermediate result. When the second intermediate result does not meet the qualified threshold, trigger multi-stage iteration and repeat the back substitution and the second operation until the qualified threshold is met or the preset iteration cycle limit is reached. S6. Based on the second intermediate result that meets the qualified threshold, establish a nonlinear quantitative relationship between rainstorm characteristic parameters and ecological health response, and identify the risk threshold of system state change. S7. Based on the random forest model and SHAP method, analyze the contribution and impact pattern of each rainstorm characteristic parameter on the evolution of ecological health, and output the tracking results including evolution trajectory, risk threshold and driving attribution.

2. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 1, characterized in that: The collection frequencies for the three periods before, during, and after the rainstorm are dynamically calibrated with the occurrence time of the rainstorm event as the zero point. The collection frequency before the rainstorm is lower than that during the rainstorm, and the collection after the rainstorm continues until the parameters of the multidimensional monitoring data recover to the baseline level.

3. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 2, characterized in that: The hydrological dynamic data includes water level, flow velocity and flow rate; the water environment quality data includes turbidity, pH value and conductivity; and the water ecological response data includes species composition, diversity index and EPT species ratio of macrobenthic animals.

4. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 3, characterized in that: The method for integrating the multidimensional monitoring data into the time series of the ecological health index is as follows: Principal component analysis is used to reduce the dimensionality of the candidate evaluation indicators in the multidimensional monitoring data, and indicators with loading values ​​greater than a set threshold are retained to form a core evaluation indicator set. The entropy weight method is used to assign weights to the core evaluation indicator set, and the weighted core evaluation indicators are synthesized according to the three dimensions of hydrological integrity, physicochemical integrity and biological integrity to obtain the ecological health index corresponding to each monitoring time point.

5. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 4, characterized in that: The initial calculation includes a mutation point detection subprocess and a trend decomposition subprocess. The mutation point detection subprocess uses the Pettitt nonparametric test method to scan the time series point by point and identify the time corresponding to the maximum value of the statistic as a candidate mutation point. The trend decomposition subprocess uses a seasonal term trend decomposition method based on local weighted regression to decompose the time series into trend components, seasonal components and residual components, and extracts preliminary fitting parameters from the trend components.

6. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 5, characterized in that: The back-substitution method is as follows: extract the candidate mutation point location information from the first intermediate result, backtrack to the multidimensional monitoring data to extract the response amplitude of each single indicator at the corresponding time, use the response amplitude of each single indicator at the candidate mutation point as the measured sensitivity value, reverse correct the weight assigned by the entropy weight method, resubstitute the corrected weights into the synthesis process of the ecological health index to obtain an updated ecological health index sequence, re-input the updated ecological health index sequence into the mutation point detection stage, perform secondary mutation point detection, and obtain the second intermediate result.

7. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 6, characterized in that: The qualified thresholds include a location deviation threshold, a trend consistency threshold, and a recovery identifiable threshold. The location deviation threshold requires that the time deviation between the mutation point location in the first intermediate result and the second intermediate result does not exceed two monitoring time points. The trend consistency threshold requires that the updated ecological health index sequence and the original ecological health index sequence have the same trend direction during non-mutation periods. The recovery identifiable threshold requires that a monotonic upward trend can be identified in the time series after the rainstorm event for more than three consecutive time points.

8. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 7, characterized in that: In each iteration of the multi-stage iteration, the back substitution and the second operation are re-executed with the latest obtained candidate mutation point position as the correction term. After each iteration, the comparison with the qualified threshold is re-executed until all threshold conditions are met or the preset iteration cycle limit is reached.

9. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 8, characterized in that: The method for establishing the nonlinear quantitative relationship between rainstorm characteristic parameters and ecological health response is as follows: using rainfall intensity, rainfall duration, rainfall interval, and peak runoff as explanatory variables, and using the response amplitude and recovery rate of the ecological health index as response variables, a generalized additive model is constructed. By analyzing the inflection points of the fitted curve, the critical rainfall intensity value at which the response amplitude of the ecological health index undergoes a step change and the critical rainfall interval value at which the recovery rate changes are identified. The critical rainfall intensity value and the critical rainfall interval value are used as the risk threshold.

10. The method for tracking the dynamic evolution of urban river ecological health driven by high-frequency rainstorm events according to claim 9, characterized in that: It also includes a reverse constraint verification step: taking the top two rainstorm characteristic parameters with the highest contribution identified by the SHAP method as the dominant driving factors, backtracking to verify whether the sampling density of the original monitoring data corresponding to the dominant driving factors in the rainstorm event meets the effective analysis density requirement. When the sampling density is lower than the effective analysis density requirement, the data acquisition frequency of the monitoring parameters corresponding to the dominant driving factors is increased to twice the original frequency, and the increased acquisition frequency is synchronized to the acquisition configuration of the multidimensional monitoring data.

Images