Water quality dynamic monitoring method based on multi-scale remote sensing image space-time difference cooperation

Abstract

The invention relates to a multi-scale remote sensing image spatial-temporal difference cooperative water quality dynamic monitoring method, and belongs to the technical field of water quality monitoring. The method comprises the steps that a multi-scale time sequence remote sensing image is acquired, a key monitoring domain is delimited through pollution risk and function partition coupling, and time-space registration and spectrum calibration are completed in combination with hydrological parameters; the method comprises the following steps: directionally extracting water quality parameter correlation difference characteristics, quantifying hierarchical characteristic correlation intensity and filtering non-pollution interference signals; a feature-oriented inversion framework is built, a pollution diffusion boundary is delimited, a pollution source is traced, and a water quality parameter space-time dynamic distribution diagram is generated through multi-feature adaptation fusion; verifying the adaptive deviation of the region and the local dimension, dynamically correcting the weight of the model, and constructing a two-factor early warning rule to form a whole-process monitoring link. According to the method, multi-scale image space-time difference characteristics are fully utilized, the accuracy and dynamic response capability of water quality monitoring are improved, and reliable support is provided for pollution source tracing and risk early warning.

Description

Technical Field

[0001] This invention belongs to the field of water quality monitoring technology, specifically relating to a method for dynamic water quality monitoring based on the spatiotemporal differences of multi-scale remote sensing images. Background Technology

[0002] Dynamic water quality monitoring is a core technical link in ensuring the ecological balance of aquatic areas, maintaining drinking water safety, and preventing water pollution incidents. Its monitoring accuracy and response efficiency are directly related to the scientific nature and timeliness of water environment management decisions. As water environment problems become increasingly complex, traditional water quality monitoring methods have gradually revealed significant limitations and are unable to meet the needs of large-scale, high-frequency, and high-precision monitoring.

[0003] Traditional water quality monitoring relies heavily on on-site sampling and laboratory analysis. This method requires manual on-site deployment of sampling points and collection of water samples for offline testing. This not only consumes significant manpower and time but also limits the coverage of sampling points, reflecting only the water quality at local locations and failing to achieve continuous dynamic monitoring of large areas of water. Especially for wide-area water bodies such as river basins and lakes, on-site sampling struggles to capture spatial distribution differences and temporal trends in water quality parameters, easily missing key nodes in the pollution diffusion process, leading to delays in pollution source tracing and untimely risk warnings.

[0004] To overcome the limitations of on-site sampling, remote sensing technology, with its advantages of large-scale, multi-temporal, and non-contact operation, is gradually being applied to water quality monitoring. However, existing remote sensing monitoring methods still have significant shortcomings in areas such as multi-scale image collaborative utilization, feature extraction and interference elimination, inversion result fusion, and early warning mechanism construction. At the multi-scale image processing level, remote sensing images of different resolutions differ in spectral information, spatial resolution, and temporal intervals. Existing methods lack effective spatiotemporal registration and spectral unification mechanisms, which can easily lead to distortion of water quality parameter inversion results due to image spatiotemporal deviations. At the same time, the optical properties of water bodies are affected by substances such as chlorophyll and suspended matter, and the differences in optical environments of different water bodies further exacerbate the difficulty of multi-scale image spectral calibration, making it difficult to establish a unified monitoring benchmark.

[0005] In the feature extraction and interference elimination stages, existing technologies mostly focus on the reflection characteristics of single-dimensional water quality parameters, failing to effectively distinguish between pollution-related differences and non-pollution-related differences caused by natural fluctuations. For example, natural factors such as water flow disturbances, water level changes, and seasonal changes can cause fluctuations in the spectral characteristics of water bodies. Such fluctuations are easily misjudged as pollution signals, interfering with the accuracy of monitoring results. Furthermore, there is a lack of correlation analysis mechanisms for different levels of difference characteristics, making it impossible to quantify the synergistic relationship between overall regional characteristics and local point characteristics, resulting in insufficient targeting and effectiveness of feature extraction.

[0006] Regarding the integration of inversion results and early warning mechanisms, existing methods mostly use simple weighting or splicing to integrate monitoring data from different sources. They fail to adapt and integrate the pollution diffusion characteristics of water quality parameters and the differences in functional zones of water areas, resulting in poor connection between regional and local scale inversion results and difficulty in generating accurate spatiotemporal dynamic distributions of water quality parameters. At the same time, early warning mechanisms often rely on single factors and do not consider the synergistic effect of spatial difference gradients and temporal change rates of water quality parameters, which can easily lead to delayed early warnings or false alarms, and cannot provide reliable support for the rapid response to water pollution incidents.

[0007] In summary, existing water quality monitoring technologies still have significant technical bottlenecks in areas such as multi-scale image collaboration, feature interference elimination, inversion fusion accuracy, and early warning response efficiency. There is an urgent need for a dynamic water quality monitoring method that can achieve spatiotemporal difference collaboration of multi-scale remote sensing images, accurately extract effective features, efficiently fuse inversion results, and construct a scientific early warning mechanism to meet the refined monitoring needs in complex water environments. Summary of the Invention

[0008] To address the aforementioned problems in the existing technology, this invention provides a method for dynamic water quality monitoring based on the spatiotemporal differences of multi-scale remote sensing images; The objective of this invention can be achieved through the following technical solutions: S1: Acquire multi-scale temporal remote sensing images of the monitored water area, and delineate key monitoring areas based on the spatial coupling of pollution risk level and water area functional zoning; select stable ground feature markers within the area as coordinate reference points, and perform spatiotemporal registration of images in combination with hydrological dynamic parameters; construct a unified spectral reference model based on water body optical characteristic parameters, and perform spectral feature calibration on the registered images; S2: For the images after benchmark unification, formulate difference feature capture rules according to the characteristics of key monitoring areas, and extract the band reflection differences and temporal variation characteristics corresponding to water quality parameters in a targeted manner; preset the scale feature transmission path, and quantify the correlation strength of difference features at different levels through feature contribution analysis; establish an interference feature filtering rule library, and remove non-pollution difference signals caused by natural fluctuations based on comparison with historical data from the same period. S3: Based on feature attributes, build a feature-guided inversion framework and divide it into regional feature layers and local feature layers; delineate pollution diffusion boundaries through feature-level analysis process and trace pollution sources by combining the distribution of feature anomalies; formulate multi-feature adaptation and fusion rules, and integrate inversion results from different levels to generate a spatiotemporal dynamic distribution map of water quality parameters. S4: Validate dimensions by dividing them into scale levels, calculate the adaptation deviation of the inversion results by region and local dimensions; preset a deviation closed-loop adjustment process, dynamically correct the feature weight parameters of the inversion model based on the deviation distribution; construct a dual-factor early warning rule of difference gradient and rate of change, and associate monitoring data to form a dynamic monitoring link throughout the entire process.

[0009] As a preferred embodiment of the present invention, the method for delineating key monitoring areas based on the spatial coupling of pollution risk level and water functional zoning is as follows: By combining the frequency of historical pollution events, the spread range of pollutants, and the residual concentration of pollutants in the monitored waters, a multi-dimensional pollution risk assessment system is constructed to determine the pollution risk level of different areas. Based on the ecological function positioning of the waters, the functional types of the waters are distinguished, and the monitoring priority of each functional zone is clarified. A spatial weight coupling model between risk level and functional zone is established to correlate risk level parameters with functional zone priority. At the same time, the hydrological connectivity characteristics of the waters are superimposed to screen out key areas under the synergistic effect of risk-function-connectivity as key monitoring areas.

[0010] Specifically, the method for performing spatiotemporal registration of images by combining hydrological dynamic parameters is as follows: The core hydrological dynamic parameters are selected from the water flow velocity, water level change range, and water turbulence intensity of the monitored water area, and are associated and matched with the coordinate attributes of the land feature markers. A mapping relationship between hydrological parameters and image coordinate deviation is constructed. By training with historical hydrological-image data from the same period, the coordinate correction thresholds corresponding to different parameter change ranges are determined. The current hydrological dynamic parameters are accessed in real time, and the iteration step size of the registration mechanism is dynamically adjusted by comparing the mapping relationship. This corrects the spatial offset of the image caused by water flow and water level fluctuations, and at the same time, the time reference of different time series images is calibrated to form an image spatiotemporal registration adapted to hydrological dynamics.

[0011] Specifically, the method for constructing a unified spectral benchmark model based on the optical property parameters of water bodies is as follows: Chlorophyll concentration, suspended solids content, and yellow substance absorption coefficient were selected as core optical characteristic parameters. Differentiated weights were assigned to these parameters based on the type of monitored water area, and normalization was performed. Spectral data from historically unpolluted water areas were introduced as a baseline to establish a nonlinear mapping function between optical characteristic parameters and spectral reflectance deviation. The mapping function was iteratively optimized using measured spectral data to determine the matching relationship between parameter weights and spectral correction coefficients, thus forming a unified spectral benchmark model adaptable to different water quality conditions.

[0012] Specifically, the method for formulating the differential feature capture rules is as follows: This study distinguishes between two core objectives: band reflectance differences and temporal variation characteristics corresponding to water quality parameters. It clarifies that band reflectance differences focus on spatial spectral value variations, while temporal variation characteristics focus on parameter fluctuation trends over time. Based on the hierarchical characteristics of multi-scale images, it sets regional feature extraction priorities for coarse-scale images and local point feature extraction priorities for fine-scale images, thus defining feature capture ranges at different scales. It pre-sets feature validity screening conditions, combining the sensitive band range of water quality parameters to limit the difference threshold of band reflectance differences and the duration of continuous fluctuations in temporal variation characteristics. Simultaneously, it correlates with historical baseline features from the same period, eliminating non-water quality-related features that deviate from the baseline, thus forming multi-dimensional difference feature capture rules.

[0013] Specifically, the method for quantifying the correlation strength of different levels of features through feature contribution analysis is as follows: Distinguishing between regional and local layer differences, the spectral response values ​​and time-series change rates of these two types of features are extracted as analytical dimensions. The information dependence between features is calculated based on mutual information entropy, and the linear and nonlinear correlations between features are analyzed in conjunction with grey relational analysis. A feature contribution evaluation matrix is ​​constructed, and different dimensions are weighted according to the pollution sensitivity of the monitored water area. The contribution coefficients of each level of features to the inversion of water quality parameters are obtained through matrix operations. The correlation strength is quantified based on the difference in contribution coefficients, and feature combinations with correlation strengths higher than a preset threshold are selected to form a hierarchical feature correlation map.

[0014] Specifically, the method for establishing the interference feature filtering rule base is as follows: Remote sensing images and measured water quality data from the same historical period of the monitored water area are collected to construct a historical baseline dataset. The natural fluctuation range and time series features of water quality parameters within this dataset are extracted and used as baseline feature thresholds. The images and water quality data of the current period are then compared with the baseline dataset. The feature deviation values ​​of the two datasets are evaluated through time series trend fitting to distinguish between non-pollution difference signals caused by natural fluctuations and pollution-related feature signals. Based on the feature deviation values, a filtering threshold is set to generate rules for the identification and removal of non-pollution difference signals. The effectiveness of the rules is verified by periodically integrating newly added data from the same period. The baseline feature thresholds and filtering thresholds are then adjusted to form a dynamically optimized interference feature filtering rule library.

[0015] Specifically, the method for building the feature-guided inversion framework is as follows: After interference filtering, the differential features are hierarchically divided into regional and local layers. Based on the feature contribution analysis results, feature association clusters directly related to the target water quality parameters are selected, and the mapping relationship between each feature and the target water quality parameters is clarified. Multi-scale inversion sub-models are constructed. The regional layer model focuses on integrating large-scale spectral mean features, while the local layer model focuses on temporal abrupt changes in point locations. Sub-model synergy is achieved through feature weight allocation. A feature-model dynamic matching mechanism is established. The selection threshold of the core feature set is adjusted based on real-time water quality parameter fluctuations. The sub-model parameters are periodically calibrated based on measured data, forming a water quality parameter inversion framework with feature-driven as the core.

[0016] Specifically, the method for defining the pollution diffusion boundary through the feature-level analysis process is as follows: Based on the spatial distribution differences between regional and local layer features, the critical regions of pollution feature gradient changes are identified; spatial clustering mechanisms are used to analyze the density of feature anomalies, and combined with water flow direction and velocity parameters, a spatiotemporal evolution model of pollution diffusion is constructed; the main trend boundary of diffusion is determined by the regional layer features, the boundary morphology is refined by the local layer features, and the continuous pollution zone and temporary fluctuation zone are distinguished based on the stability of temporal features, so as to dynamically delineate the pollution diffusion boundary.

[0017] Specifically, the method for formulating multi-feature adaptation and fusion rules is as follows: Based on the pollution diffusion range and impact degree of water quality parameters, a hierarchical weight allocation mechanism is established. The regional layer inversion results are allocated and fused based on spatial coverage, and the local layer inversion results are allocated based on parameter sensitivity. Hydrological dynamic parameters and historical data from the same period are introduced as verification benchmarks to calibrate the deviations in the spatial dimension caused by water flow disturbances and the fluctuations in the temporal dimension caused by monitoring intervals, and to remove abnormal inversion values ​​that deviate from the benchmarks. The temporal dimension results are stitched together according to the image time interval, and the spatial dimension data is integrated by matching the grid division of key monitoring areas, and finally a dynamic distribution map containing parameter concentration gradients and spatiotemporal evolution trends is generated.

[0018] Specifically, the method for calculating the adaptation bias of the inversion results by region and local dimension is as follows: The regional dimension uses the spatial grid of key monitoring areas as the unit, selects the mean and spatial distribution trend of the regional layer inversion results as the benchmark value, and counts the deviation of the inversion value within the grid from the benchmark value. The local dimension selects the inversion data of the points corresponding to the characteristic anomalies, compares them with the benchmark value of the regional grid to which the point belongs, and determines the single-point deviation by combining the sensitivity threshold of water quality parameters. The overall fit is calculated by combining the two types of deviations, and the proportion of local points with deviations exceeding the threshold within the regional grid and the cumulative fluctuation of local point deviations are counted to form a quantifiable fit deviation distribution.

[0019] Specifically, the method for constructing the dual-factor early warning rule based on differential gradient and rate of change is as follows: The differential gradient factor is calculated based on the spatial difference between regional and local inversion results. It takes the absolute value of the difference in water quality parameters between adjacent grids at the same time and sets the gradient threshold according to the functional zoning of water areas. The rate of change factor counts the temporal change of water quality parameters per unit time and determines the rate threshold by combining the historical fluctuation range of the same period. The dual-factor linkage triggering condition is set. When the differential gradient exceeds the threshold and the coverage ratio of monitoring points that exceed the threshold reaches the preset value, and the magnitude of the single factor exceeding the threshold reaches the preset ratio, the corresponding level of warning is triggered, and the warning response link is generated by associating real-time monitoring data.

[0020] The beneficial effects of this invention are as follows: (1) By setting up a spatial coupling mechanism for delineating key monitoring areas based on pollution risk level and water functional zoning, combining image spatiotemporal registration mechanism with hydrological dynamic parameters, and a unified spectral benchmark model based on water body optical characteristic parameters, and by using an interference feature filtering rule library to remove non-pollution difference signals caused by natural fluctuations, the problem of spatiotemporal deviation of multi-scale images, inconsistent spectral benchmarks and misjudgment of interference signals in traditional remote sensing monitoring is effectively solved, providing a characteristic data basis for subsequent water quality parameter inversion.

[0021] (2) By setting up regional and local feature layer collaborative inversion under the feature-guided inversion framework, multi-feature adaptation and fusion rules adapted to the diffusion characteristics of water pollution, and dual-factor early warning rules of difference gradient and change rate, and by combining the deviation closed-loop adjustment process to dynamically correct the feature weights of the inversion model, the generation of the spatiotemporal dynamic distribution of water quality parameters is realized, and the problems of poor connection of inversion results and single early warning mechanism in traditional monitoring, such as early warning lag and false alarms, are solved.

[0022] (3) By setting up a hierarchical feature transmission path for multi-scale images, a feature contribution quantification analysis mechanism, and a full-process dynamic optimization closed loop, the advantages of the large-scale coverage of coarse-scale images and the local accurate representation capabilities of fine-scale images are fully integrated. This solves the problem that monitoring accuracy and coverage are difficult to balance due to the fragmentation of multi-scale data and unclear feature correlation in traditional monitoring. It can not only comprehensively control the overall water quality of a wide area of ​​water, but also capture pollution anomalies at local points. At the same time, through dynamic calibration of parameters throughout the process, the monitoring method can be adapted to different types of water areas and complex hydrological environments. Attached Figure Description

[0023] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0024] Figure 1 This is a flowchart illustrating a method for dynamic water quality monitoring based on multi-scale remote sensing image spatiotemporal differences according to the present invention. Figure 2 This is an architecture diagram of a multi-scale remote sensing image spatiotemporal difference collaborative water quality dynamic monitoring method according to the present invention; Detailed Implementation

[0025] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0026] Please see Figure 1-2 A method for dynamic monitoring of water quality based on spatiotemporal differences in multi-scale remote sensing images, comprising: S1: Acquire multi-scale temporal remote sensing images of the monitored water area, and delineate key monitoring areas based on the spatial coupling of pollution risk level and water area functional zoning; select stable ground feature markers within the area as coordinate reference points, and perform spatiotemporal registration of images in combination with hydrological dynamic parameters; construct a unified spectral reference model based on water body optical characteristic parameters, and perform spectral feature calibration on the registered images; S2: For the images after benchmark unification, formulate difference feature capture rules according to the characteristics of key monitoring areas, and extract the band reflection differences and temporal variation characteristics corresponding to water quality parameters in a targeted manner; preset the scale feature transmission path, and quantify the correlation strength of difference features at different levels through feature contribution analysis; establish an interference feature filtering rule library, and remove non-pollution difference signals caused by natural fluctuations based on comparison with historical data from the same period. S3: Based on feature attributes, build a feature-guided inversion framework and divide it into regional feature layers and local feature layers; delineate pollution diffusion boundaries through feature-level analysis process and trace pollution sources by combining the distribution of feature anomalies; formulate multi-feature adaptation and fusion rules, and integrate inversion results from different levels to generate a spatiotemporal dynamic distribution map of water quality parameters. S4: Validate dimensions by dividing them into scale levels, calculate the adaptation deviation of the inversion results by region and local dimensions; preset a deviation closed-loop adjustment process, dynamically correct the feature weight parameters of the inversion model based on the deviation distribution; construct a dual-factor early warning rule of difference gradient and rate of change, and associate monitoring data to form a dynamic monitoring link throughout the entire process.

[0027] As a preferred embodiment of the present invention, the method for delineating key monitoring areas based on the spatial coupling of pollution risk level and water functional zoning is as follows: By combining the frequency of historical pollution events, the spread range of pollutants, and the residual concentration of pollutants in the monitored waters, a multi-dimensional pollution risk assessment system is constructed to determine the pollution risk level of different areas. Based on the ecological function positioning of the waters, the functional types of the waters are distinguished, and the monitoring priority of each functional zone is clarified. A spatial weight coupling model between risk level and functional zone is established to correlate risk level parameters with functional zone priority. At the same time, the hydrological connectivity characteristics of the waters are superimposed to screen out key areas under the synergistic effect of risk-function-connectivity as key monitoring areas.

[0028] Specifically, the method for performing spatiotemporal registration of images by combining hydrological dynamic parameters is as follows: The core hydrological dynamic parameters are selected from the water flow velocity, water level change range, and water turbulence intensity of the monitored water area, and are associated and matched with the coordinate attributes of the land feature markers. A mapping relationship between hydrological parameters and image coordinate deviation is constructed. By training with historical hydrological-image data from the same period, the coordinate correction thresholds corresponding to different parameter change ranges are determined. The current hydrological dynamic parameters are accessed in real time, and the iteration step size of the registration mechanism is dynamically adjusted by comparing the mapping relationship. This corrects the spatial offset of the image caused by water flow and water level fluctuations, and at the same time, the time reference of different time series images is calibrated to form an image spatiotemporal registration adapted to hydrological dynamics.

[0029] In this embodiment, several stably distributed ground feature markers within the monitored water area are selected as coordinate references. These markers include man-made structures such as bridges and fixed docks. Through monitoring over a period of time, multiple sets of valid paired data are collected at each ground feature marker.

[0030] Each set of data includes the following synchronous measurements: hydrological dynamic parameters: water flow velocity V (unit: m / s) was measured using an acoustic Doppler current profiler; water level change ΔH (unit: m) was measured using a water level gauge; and water turbulence intensity I (dimensionless) was calculated based on turbulence intensity. Actual coordinate offset: the actual coordinates of ground feature markers were measured in the field using differential GPS, and the pixel coordinates of the corresponding points were extracted from remote sensing images. The offsets of the two in the X direction (Δx_actual, pixel) and Y direction (Δy_actual, pixel) were calculated.

[0031] Based on multiple sets of data, a mapping model was established using the multiple linear regression method. The regression equations were obtained by solving the least squares method: Δx_predicted=aV+bΔH+cI+d, Δy_predicted=eV+fΔH+gI+h, where Δx_predicted and Δy_predicted represent the X and Y coordinate offsets predicted by the model (unit: pixels), and a, b, c, d, e, f, g, and h are coefficients obtained through regression analysis.

[0032] To verify the reliability of the model, the coefficient of determination R was calculated. 2 First, calculate the average value of the actual offset in the X direction: Δx_mean = (ΣΔx_actual) / N; then calculate the total sum of squares SST_x = Σ(Δx_actual - Δx_mean). 2 This reflects the total degree of variation in the data itself; then, the regression sum of squares, SSR_x = Σ(Δx_predicted - Δx_mean), is calculated. 2 This reflects the variance that the model can explain; then, the sum of squared residuals SSE_x = Σ(Δx_actual - Δx_predicted) is calculated. 2 This reflects the variation that the model cannot explain; finally, the coefficient of determination R is calculated. 2 _x = SSR_x / SST_x = -(SSE_x / SST_x). The hold-out method is used to further verify the model's generalization ability. The data is randomly divided into training and test sets. The regression coefficients are recalculated using only the training set data to build the model, and R is calculated on the test set. 2 This demonstrates that the model has good predictive power and generalization ability.

[0033] Based on the established mapping model, parameter variation intervals are divided and corresponding correction thresholds are determined. Flow velocity V is divided into multiple intervals, with corresponding X-direction coordinate correction thresholds of different ranges; water level change ΔH is divided into multiple intervals, with corresponding Y-direction coordinate correction thresholds of different ranges. This hierarchical threshold system enables precise control of coordinate correction amounts under different hydrological conditions.

[0034] Real-time access to current hydrological monitoring data is used. When flow velocity V, water level change ΔH, and turbulence intensity I are detected, the theoretical offset is calculated according to the mapping model: Δx_predicted = aV + bΔH + cI + d, Δy_predicted = eV + fΔH + gI + h. The iteration step size adjustment coefficient α is determined based on the parameter intervals. The registration algorithm iterates at a step size of α times the standard step size. After several iterations, the coordinate corrections converge to Δx_final and Δy_final. The final residuals meet the registration accuracy requirements.

[0035] A dynamic compensation mechanism was established between image acquisition time and hydrological parameter monitoring time. When a sudden change in hydrological parameters is detected (such as the rate of change of flow velocity dV / dt exceeding a threshold), the time compensation amount is automatically calculated: Δt = k1×|dV / dt| + k2×|dΔH / dt|, with the maximum compensation time set not to exceed a specific value. This mechanism effectively eliminates the problem of inconsistent time references in time-series images caused by sudden changes in hydrological parameters, ensuring the time synchronization of multi-temporal data.

[0036] Specifically, the method for constructing a unified spectral benchmark model based on the optical property parameters of water bodies is as follows: Chlorophyll concentration, suspended solids content, and yellow substance absorption coefficient were selected as core optical characteristic parameters. Differentiated weights were assigned to these parameters based on the type of monitored water area, and normalization was performed. Spectral data from historically unpolluted water areas were introduced as a baseline to establish a nonlinear mapping function between optical characteristic parameters and spectral reflectance deviation. The mapping function was iteratively optimized using measured spectral data to determine the matching relationship between parameter weights and spectral correction coefficients, thus forming a unified spectral benchmark model adaptable to different water quality conditions.

[0037] Specifically, the method for formulating the differential feature capture rules is as follows: This study distinguishes between two core objectives: band reflectance differences and temporal variation characteristics corresponding to water quality parameters. It clarifies that band reflectance differences focus on spatial spectral value variations, while temporal variation characteristics focus on parameter fluctuation trends over time. Based on the hierarchical characteristics of multi-scale images, it sets regional feature extraction priorities for coarse-scale images and local point feature extraction priorities for fine-scale images, thus defining feature capture ranges at different scales. It pre-sets feature validity screening conditions, combining the sensitive band range of water quality parameters to limit the difference threshold of band reflectance differences and the duration of continuous fluctuations in temporal variation characteristics. Simultaneously, it correlates with historical baseline features from the same period, eliminating non-water quality-related features that deviate from the baseline, thus forming multi-dimensional difference feature capture rules.

[0038] Specifically, the method for quantifying the correlation strength of different levels of features through feature contribution analysis is as follows: Distinguishing between regional and local layer differences, the spectral response values ​​and time-series change rates of these two types of features are extracted as analytical dimensions. The information dependence between features is calculated based on mutual information entropy, and the linear and nonlinear correlations between features are analyzed in conjunction with grey relational analysis. A feature contribution evaluation matrix is ​​constructed, and different dimensions are weighted according to the pollution sensitivity of the monitored water area. The contribution coefficients of each level of features to the inversion of water quality parameters are obtained through matrix operations. The correlation strength is quantified based on the difference in contribution coefficients, and feature combinations with correlation strengths higher than a preset threshold are selected to form a hierarchical feature correlation map.

[0039] Specifically, the method for establishing the interference feature filtering rule base is as follows: Remote sensing images and measured water quality data from the same historical period of the monitored water area are collected to construct a historical baseline dataset. The natural fluctuation range and time series features of water quality parameters within this dataset are extracted and used as baseline feature thresholds. The images and water quality data of the current period are then compared with the baseline dataset. The feature deviation values ​​of the two datasets are evaluated through time series trend fitting to distinguish between non-pollution difference signals caused by natural fluctuations and pollution-related feature signals. Based on the feature deviation values, a filtering threshold is set to generate rules for the identification and removal of non-pollution difference signals. The effectiveness of the rules is verified by periodically integrating newly added data from the same period. The baseline feature thresholds and filtering thresholds are then adjusted to form a dynamically optimized interference feature filtering rule library.

[0040] In this embodiment, remote sensing image data (including spectral reflectance sequences) and synchronously measured water quality data (chlorophyll a concentration and suspended matter concentration) for the same quarter (e.g., the second quarter) of each year over the past five years in the watershed are selected. These data are then organized into a historical benchmark dataset for the same period according to time series. For the chlorophyll a concentration in the dataset, its mean μ1 and standard deviation σ1 at each monitoring point are calculated. The range [μ1-2σ1, μ1+2σ1] (approximately 95% of the sample data in a normal distribution will fall within the range of mean ± 2 times the standard deviation; combined with the retrospective analysis of measured data from historical pollution-free periods in the watershed, this range can completely cover the fluctuations in chlorophyll a concentration caused by natural factors such as water temperature changes, light intensity, and slight disturbances in water flow) is set as the benchmark range for the natural concentration fluctuation of chlorophyll a. For the suspended matter concentration, the mean rate of change ν1 and the upper limit of the fluctuation amplitude θ1 of its adjacent time periods are calculated through a sliding window. The time series characteristics with a rate of change ≤ ν1 and an amplitude ≤ θ1 are used as the benchmark characteristics for the natural fluctuation of suspended matter. The two together constitute the benchmark characteristic threshold.

[0041] After acquiring remote sensing images (at the same resolution and in the same band) and measured data for the current second quarter, the current chlorophyll a concentration sequence and suspended matter change rate sequence are extracted. Then, time-series trend fitting is performed with the benchmark dataset. For chlorophyll a, the least squares method is used to fit the deviation between the current concentration curve and the benchmark mean concentration curve, obtaining the characteristic deviation value ΔChl (the difference between the current concentration and the benchmark mean). For suspended matter, the ratio of the current change rate to the benchmark mean change rate ΔSus (the current change rate ν1) is calculated. Historical data verification shows that natural fluctuations (such as water temperature changes and slight water flow disturbances) typically result in ΔChl ≤ σ1 and ΔSus typically ≤ α1, while pollution-related signals (such as wastewater discharge) will cause ΔChl > σ1 or ΔSus > α1. This allows for the differentiation between non-pollution difference signals and pollution characteristic signals.

[0042] Based on the aforementioned characteristic deviation values, filter thresholds are set: the filter threshold for chlorophyll a is set to σ1, and the filter threshold for suspended matter is set to α1. A rule is generated: when ΔChl≤σ1, it is determined to be a non-polluting difference signal such as water temperature fluctuation and is removed; when ΔSus≤α1, it is determined to be a non-polluting difference signal caused by natural water flow disturbance and is removed. Subsequently, at the end of each quarter, newly added concurrent remote sensing images and measured data are integrated into the baseline dataset. The effectiveness of the rule is verified by the proportion of false positives (denoted as β) among the non-polluting signals removed from the rule base that are actually polluting signals. If β>5% (a preset acceptable upper limit for false positives), the baseline characteristic thresholds are recalculated (e.g., the baseline range for chlorophyll a is adjusted to [μ1-2.5σ1,μ1+2.5σ1]), and the filter thresholds are simultaneously updated to 1.25σ1 (chlorophyll a) and 1.08α1 (suspended matter). This achieves dynamic optimization of the rule base, ensuring the filtering of non-polluting interference signals.

[0043] Specifically, the method for building the feature-guided inversion framework is as follows: After interference filtering, the differential features are hierarchically divided into regional and local layers. Based on the feature contribution analysis results, feature association clusters directly related to the target water quality parameters are selected, and the mapping relationship between each feature and the target water quality parameters is clarified. Multi-scale inversion sub-models are constructed. The regional layer model focuses on integrating large-scale spectral mean features, while the local layer model focuses on temporal abrupt changes in point locations. Sub-model synergy is achieved through feature weight allocation. A feature-model dynamic matching mechanism is established. The selection threshold of the core feature set is adjusted based on real-time water quality parameter fluctuations. The sub-model parameters are periodically calibrated based on measured data, forming a water quality parameter inversion framework with feature-driven as the core.

[0044] Specifically, the method for defining the pollution diffusion boundary through the feature-level analysis process is as follows: Based on the spatial distribution differences between regional and local layer features, the critical regions of pollution feature gradient changes are identified; spatial clustering mechanisms are used to analyze the density of feature anomalies, and combined with water flow direction and velocity parameters, a spatiotemporal evolution model of pollution diffusion is constructed; the main trend boundary of diffusion is determined by the regional layer features, the boundary morphology is refined by the local layer features, and the continuous pollution zone and temporary fluctuation zone are distinguished based on the stability of temporal features, so as to dynamically delineate the pollution diffusion boundary.

[0045] Specifically, the method for formulating multi-feature adaptation and fusion rules is as follows: Based on the pollution diffusion range and impact degree of water quality parameters, a hierarchical weight allocation mechanism is established. The regional layer inversion results are allocated and fused based on spatial coverage, and the local layer inversion results are allocated based on parameter sensitivity. Hydrological dynamic parameters and historical data from the same period are introduced as verification benchmarks to calibrate the deviations in the spatial dimension caused by water flow disturbances and the fluctuations in the temporal dimension caused by monitoring intervals, and to remove abnormal inversion values ​​that deviate from the benchmarks. The temporal dimension results are stitched together according to the image time interval, and the spatial dimension data is integrated by matching the grid division of key monitoring areas, and finally a dynamic distribution containing parameter concentration gradients and spatiotemporal evolution trends is generated.

[0046] In this embodiment, the monitoring of chlorophyll a concentration in a lake is taken as an example. There is obvious pollution diffusion in this water area, and it is necessary to obtain its spatiotemporal dynamic distribution through multi-scale remote sensing images.

[0047] The multi-feature adaptation and fusion rules are implemented through the following operation process: First, a hierarchical fusion strategy is established, and the fusion ratio is dynamically allocated according to the pollution diffusion range and impact degree. The regional layer inversion results are allocated a ratio α based on the spatial coverage index C_region, and the local layer inversion results are allocated a ratio β based on the parameter sensitivity index S_local. The weights are dynamically adjusted through the constraint condition α+β=1 to ensure the organic unity of large-scale trends and local details.

[0048] Next, a set of hydrological dynamic parameters consisting of flow velocity and flow direction, along with a historical benchmark dataset from the same period, were introduced as verification benchmarks. Spatial smoothing based on a flow disturbance correction factor was performed on the regional layer results, and time series alignment and missing value imputation were performed on the local layer results. At the same time, an outlier threshold δ was set, and outlier inversion values ​​that deviated from the historical fluctuation range [L_min,L_max]±δ were removed by scanning. The spatial neighborhood mean replacement method was used to complete the data correction.

[0049] Then, a spatiotemporal integration operation is performed, in which the images are integrated according to their acquisition time sequence T1, T2, ..., T. n The inversion results are sequentially arranged, and cubic spline interpolation is used to generate a continuous time series followed by smoothing and denoising. Spatially, the regional layer results are resampled to a unified grid G1, G2, ..., G m The local layer results are interpolated to the corresponding grid using the inverse distance weighting method, and finally the data fusion is completed according to the formula Final_Value=α×Value_region+β×Value_local.

[0050] Subsequently, a dynamic distribution map containing physical meaning is generated. A concentration gradient field is constructed by calculating the concentration difference between grids. A spatiotemporal evolution trend vector is established by combining the change direction and rate of different time phases. Color gradient and arrow symbols are used to represent the concentration gradient and diffusion trend, respectively, forming a comprehensive distribution map that can intuitively reflect the spatiotemporal evolution law of water quality parameters.

[0051] Finally, by comparing the implementation effect with the field monitoring data, the dynamic distribution map generated by this method showed excellent performance in terms of spatial morphology matching, gradient change accuracy and trend prediction reliability. The entire process, through a clear parameter system, verification mechanism and fusion steps, achieved the operability and repeatability of multi-feature adaptation and fusion rules.

[0052] Specifically, the method for calculating the adaptation bias of the inversion results by region and local dimension is as follows: The regional dimension uses the spatial grid of key monitoring areas as the unit, selects the mean and spatial distribution trend of the regional layer inversion results as the benchmark value, and counts the deviation of the inversion value within the grid from the benchmark value. The local dimension selects the inversion data of the points corresponding to the characteristic anomalies, compares them with the benchmark value of the regional grid to which the point belongs, and determines the single-point deviation by combining the sensitivity threshold of water quality parameters. The overall fit is calculated by combining the two types of deviations, and the proportion of local points with deviations exceeding the threshold within the regional grid and the cumulative fluctuation of local point deviations are counted to form a quantifiable fit deviation distribution.

[0053] Specifically, the method for constructing the dual-factor early warning rule based on differential gradient and rate of change is as follows: The differential gradient factor is calculated based on the spatial difference between regional and local inversion results. It takes the absolute value of the difference in water quality parameters between adjacent grids at the same time and sets the gradient threshold according to the functional zoning of water areas. The rate of change factor counts the temporal change of water quality parameters per unit time and determines the rate threshold by combining the historical fluctuation range of the same period. The dual-factor linkage triggering condition is set. When the differential gradient exceeds the threshold and the coverage ratio of monitoring points that exceed the threshold reaches the preset value, and the magnitude of the single factor exceeding the threshold reaches the preset ratio, the corresponding level of warning is triggered, and the warning response link is generated by associating real-time monitoring data.

[0054] In this embodiment, a cross-regional watershed is taken as the monitoring object. The watershed is divided into three functional zones: drinking water source protection area, agricultural irrigation area, and landscape and recreation area. The specific implementation of the dual-factor early warning rule based on difference gradient and rate of change is as follows: First, the monitoring area is divided into grids. The entire watershed is divided into continuous grid units of 1km×1km. Each grid serves as an independent monitoring node. Regional layer inversion water quality parameters (such as COD concentration) and local layer inversion water quality parameters (such as COD concentration at key points within the grid) are retrieved simultaneously from each grid. When calculating the differential gradient factor, for monitoring data at the same time, the absolute value of the difference in water quality parameters between each grid and its four adjacent grids is extracted, i.e., ΔGi (where i is the grid number). Gradient thresholds are set based on the protection level and pollution tolerance of each functional zone: Drinking water source protection areas are sensitive to pollution, and based on the maximum natural gradient fluctuation value of their historical 20-year pollution-free period, a gradient threshold Gth1=μG1-σG1 (μG1 is the historical natural gradient mean of the area, and σG1 is the standard deviation) is set to ensure that even slight gradient anomalies trigger early warnings; the threshold Gth2=μG2 (μG2 is the historical natural gradient mean of the area) is set for agricultural irrigation areas; and the threshold Gth3=μG3+σG3 is set for landscape and recreation areas. This setting is based on the pollution prevention and control priorities of different functional zones and has been verified by retrospective analysis of historical pollution events, which can accurately match the early warning needs of each area.

[0055] When calculating the rate of change factor, the time unit is the interval of remote sensing image acquisition (e.g., once a day). The temporal change of water quality parameters in each grid cell is calculated as Vi = (parameter value at the current moment - parameter value at the previous moment) / time interval. Water quality parameter change rate data for the same period over the past ten years (e.g., June to August each year) are retrieved to construct a historical temporal fluctuation dataset. After verifying that the data conforms to a normal distribution using the KS test, the mean μV and standard deviation σV are calculated, and rate thresholds Vth are set: Vth1 = μV + 1.5σV for drinking water source protection areas, Vth2 = μV + 2σV for agricultural irrigation areas, and Vth3 = μV + 2.5σV for landscape and recreational areas. This setting is based on the fact that more than 90% of the natural fluctuation rate changes in historical data fall within the range of μV + 2σV. Changes exceeding this range are mostly related to pollution emissions. The adjustment of the standard deviation multiple for different functional areas is combined with their response sensitivity requirements to water quality changes.

[0056] The system sets two-factor linkage triggering conditions: First, it presets a monitoring point coverage ratio threshold P1 = 30% and a single-factor exceedance threshold P2 = 20%, and divides the system into two warning levels. When the number of monitoring grids within a functional zone where the difference gradient ΔGi > the corresponding gradient threshold Gth and the rate of change Vi > the corresponding rate threshold Vth accounts for ≥ P1 of the total number of grids in that zone, and simultaneously the exceedance magnitude of any factor (e.g., ΔGi - Gth) / Gth ≥ P2, a Level 1 warning is triggered. If only one of the following conditions is met: coverage ratio ≥ P1 or single-factor exceedance magnitude ≥ P2, a Level 2 warning is triggered. For example, if 35% of the grids in a drinking water source protection area simultaneously satisfy ΔGi > Gth1 and Vi > Vth1, and the difference gradient exceedance magnitude of some grids reaches 25%, a Level 1 warning is triggered. The system automatically associates real-time monitoring data for that area (e.g., grid coordinates, specific parameter values, hydrological dynamics) to generate a warning response link containing the location of suspected pollution areas, abnormal parameter trends, and emergency response suggestions, which is then simultaneously pushed to the watershed water environment management platform and relevant responsible units.

[0057] Every six months thereafter, based on newly added monitoring data and feedback from pollution incident handling, the gradient thresholds, rate thresholds, and proportional parameters in the triggering conditions of each functional zone will be recalibrated to ensure that the early warning rules are adapted to the dynamic changes in water quality in the basin.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for water quality dynamic monitoring based on multi-scale remote sensing image spatio-temporal difference coordination, characterized in that, The method comprises the following steps: S1: Obtain multi-scale time-series remote sensing images of the monitored water area, and determine the key monitoring area based on the spatial coupling of pollution risk level and water area function zoning; Select stablely distributed ground feature markers as coordinate reference points, and perform image space-time registration combined with hydrodynamic parameters; construct a spectral reference uniform model based on water body optical property parameters, and calibrate the spectral characteristics of the registered image; S2: According to the characteristics of the key monitoring area, the image after the reference unification is processed, the difference feature capture rule is formulated, and the corresponding band reflection difference and time series change characteristics of the water quality parameter are extracted; Pre-set scale feature conduction path, analyze and quantify the correlation strength of different levels of difference features through feature contribution degree; establish an interference feature filtering rule library, and exclude non-pollution difference signals caused by natural fluctuations based on historical synchronous data comparison; S3: Build a feature-oriented inversion framework based on feature attributes, divide regional feature layer and local feature layer; determine the pollution diffusion boundary through the feature level analysis process, trace the pollution source combined with the distribution of feature abnormal points; formulate multi-feature adaptive fusion rules, and integrate different level inversion results to generate a time-space dynamic distribution map of water quality parameters; S4: Verify the dimension through scale level division, calculate the adaptive deviation of the inversion result according to the regional and local dimensions; preset the deviation closed-loop adjustment process, dynamically correct the feature weight parameters of the inversion model based on the deviation distribution; construct a difference gradient and change rate double-factor early warning rule, and form a whole-process dynamic monitoring link by associating the monitoring data.

2. The method of claim 1, wherein, The key monitoring area is determined based on the spatial coupling of pollution risk level and water area function zoning, and the specific method is: Combined with the occurrence frequency of historical pollution events in the monitored water area, the pollution diffusion range and the residual concentration, a multi-dimensional pollution risk assessment system is constructed to determine the pollution risk level of different regions; Based on the ecological function positioning of the water area, the water area function type is distinguished, and the monitoring priority of each functional partition is clear; a spatial weight coupling model of risk level and functional zoning is established, the risk level parameters and functional zoning priority are associated and calculated, and the hydrological connectivity characteristics of the water area are superimposed to select the key regions under the synergistic action of risk-function-connectivity as the key monitoring area.

3. The method of claim 1, wherein, The specific method of combining hydrodynamic parameters to perform image space-time registration is: Select the flow velocity, water level variation amplitude and water body turbulence intensity of the monitored water area as the core hydrodynamic parameters, and associate and match the coordinate attributes of the ground feature markers; construct a mapping relationship between hydrological parameters and image coordinate deviation, determine the coordinate correction threshold corresponding to different parameter variation intervals through historical synchronous hydrological-image data training; real-time access to current hydrodynamic parameters, compare the mapping relationship to dynamically adjust the iteration step of the registration mechanism, correct the image space shift caused by water flow and water level fluctuation, and calibrate the time base of different time series images to form an image space-time registration adaptive to hydrodynamic.

4. The method of claim 1, wherein, The specific method of constructing a spectral reference uniform model based on water body optical property parameters is: The core optical characteristic parameters are screened as chlorophyll concentration, suspended matter content and yellow substance absorption coefficient, the parameter differentiation weight is given based on the monitored water type and the normalization processing is completed; the spectral data of the historical same period non-polluted water area is introduced as a baseline, and a nonlinear mapping function of the optical characteristic parameters and spectral reflectance deviation is established; The mapping function is iteratively optimized through the measured spectral data, the matching relationship between the parameter weight and the spectral correction coefficient is determined, and a spectral baseline unified model adaptable to different water quality conditions is formed.

5. The method of claim 1, wherein, The specific method for formulating the difference characteristic capturing rule is: The two core targets of band reflectance difference and time sequence change characteristic corresponding to water quality parameters are distinguished, the spectral value anomaly focusing on the spatial dimension of the band reflectance difference and the parameter fluctuation trend focusing on the time dimension of the time sequence change characteristic are determined; based on the hierarchical characteristics of multi-scale images, the regional feature extraction priority is set for coarse scale images, and the local point feature extraction priority is set for fine scale images, and the feature capturing range of different scales is divided; The preset feature effectiveness screening condition is limited in combination with the sensitive band range of the water quality parameter, the difference value threshold of the band reflectance difference and the sustained fluctuation duration of the time sequence change characteristic, and meanwhile, the historical same period characteristic baseline is associated to eliminate the non-water quality related features deviating from the baseline, thereby forming a multi-dimensional difference characteristic capturing rule.

6. The method of claim 1, wherein, The specific method for quantifying the correlation strength of different hierarchical difference characteristics through feature contribution degree analysis is: The regional layer difference characteristic and the local layer difference characteristic are distinguished, the spectral response value and the time sequence change rate of the two types of characteristics are extracted as analysis dimensions; the information dependence degree between the characteristics is calculated based on mutual information entropy, the linear and nonlinear correlation between the characteristics is analyzed in combination with the grey correlation degree, the feature contribution degree evaluation matrix is constructed, the weight is given to different dimensions according to the pollution sensitivity of the monitored water area; the contribution coefficient of each hierarchical feature to the water quality parameter inversion is obtained through matrix operation, the correlation strength is quantified based on the contribution coefficient difference, the feature combination with the correlation strength higher than the preset threshold is screened out, and a hierarchical feature correlation atlas is formed.

7. The method of claim 1, wherein, The specific method for establishing the interference feature filtering rule library is: The historical same period remote sensing images and measured water quality data of the monitored water area are collected, a historical same period baseline data set is constructed, the natural fluctuation range and time sequence characteristics of the water quality parameters in the data set are extracted as baseline feature thresholds; Then, the images and water quality data of the current period are compared with the baseline data set, the feature deviation values of the two are evaluated through time sequence trend fitting, the non-pollution difference signals caused by natural fluctuation and the pollution related feature signals are distinguished; the filtering threshold is set based on the feature deviation values, the identification and elimination rule for the non-pollution difference signals is generated, the baseline feature threshold and the filtering threshold are adjusted based on the period integration of the newly added same period data, and a dynamically optimized interference feature filtering rule library is formed.

8. The method of claim 1, wherein, The specific method for building the feature-oriented inversion framework is: The interference filtered difference features are hierarchically divided according to regional layers and local layers, and feature contribution degree analysis results are combined to screen out feature association clusters directly associated with the target water quality parameter, so as to clearly define the mapping relationship between each feature and the target water quality parameter; multi-scale inversion sub-models are constructed, the regional layer model focuses on integrating large-scale spectral mean features, and the local layer model focuses on point time series mutation features, and feature weight distribution is used to realize the cooperation of sub-models; a feature-model dynamic matching mechanism is established, the selection threshold of the core feature set is adjusted based on real-time water quality parameter fluctuations, and sub-model parameters are calibrated regularly combined with measured data, forming a water quality parameter inversion framework with feature-driven as the core.

9. The method of claim 1, wherein, The pollution diffusion boundary is determined through the feature hierarchical analysis process, and the specific method is as follows: Based on the spatial distribution difference of regional layer and local layer features, the critical area of pollution feature gradient change is identified; a space-time evolution model of pollution diffusion is constructed by using the spatial clustering mechanism to analyze the density of feature abnormal points and combining the water flow direction and flow rate parameters; the main trend boundary of diffusion is determined by the regional layer features, the boundary form is refined by the local layer features, and the continuous pollution area and temporary fluctuation area are distinguished based on the stability of the time series features, so as to dynamically determine the pollution diffusion boundary.

10. The method of claim 1, wherein, The multi-feature adaptive fusion rule is formulated, and the specific method is as follows: Based on the pollution diffusion range and influence degree of water quality parameters, a hierarchical weight distribution mechanism is established, and the regional layer inversion results are distributed and fused based on the spatial coverage, and the local layer inversion results are distributed and fused based on the parameter sensitivity; hydrodynamic parameters and historical data of the same period are introduced as a verification benchmark to calibrate the deviation of the spatial dimension caused by water flow disturbance and the fluctuation caused by monitoring interval, and to eliminate abnormal inversion values deviating from the benchmark; the time dimension results are spliced according to the image time interval, the spatial dimension data are integrated by matching the key monitoring domain grid, and finally the dynamic distribution map containing the parameter concentration gradient and the space-time evolution trend is generated.

11. The method of claim 1, wherein, The adaptive deviation of the inversion result is calculated according to the regional and local dimensions, and the specific method is as follows: The regional dimension takes the spatial grid of the key monitoring domain as a unit, selects the mean value and spatial distribution trend of the regional layer inversion result as a benchmark value, and calculates the deviation amplitude of the inversion value in the grid from the benchmark value; the local dimension selects the point inversion data corresponding to the feature abnormal point, and compares it with the benchmark value of the regional grid to which the point belongs, and determines the single-point deviation amount combined with the sensitive threshold of the water quality parameter; the overall adaptation degree is calculated by comprehensively considering the two types of deviations, the proportion of local points with deviation exceeding the threshold in the regional grid is calculated, and the cumulative fluctuation amplitude of the local point deviation is calculated, to form a quantifiable adaptive deviation distribution.

12. The method of claim 1, wherein, The specific method for constructing the difference gradient and change rate double-factor early warning rule is as follows: The difference gradient factor is calculated based on the spatial difference value of regional and local inversion results, taking the absolute value of the difference of water quality parameters of adjacent grids at the same time, and setting a gradient threshold according to the water function partition; the change rate factor is calculated by the time series change amount of the water quality parameters in a unit time, and the rate threshold is determined by combining the historical fluctuation range; the double-factor linkage trigger condition is set, when the difference gradient exceeds the threshold value and the change rate exceeds the threshold value, the proportion of the monitoring points covered reaches the preset value, and the amplitude of the single factor exceeding the threshold value reaches the preset proportion, the corresponding grade warning is triggered, and the real-time monitoring data is associated to generate a warning response link.

Images