Tidal reach eutrophication inversion interpretation early warning system based on high-precision spectral method

Abstract

The application discloses a tidal river eutrophication inversion interpretation early warning system based on high-precision spectroscopy, relates to the technical field of water pollution monitoring, and comprises the following modules: a data acquisition module, which is used for acquiring spectral reflectance data and synchronous hydrological data; a data preprocessing module, which is used for constructing a preprocessing data set; a tidal coupling inversion interpretation module, which is used for screening eutrophication sensitive bands by using a continuous projection algorithm, constructing a tidal coupling inversion model, and generating an inversion result set; an early warning linkage module, which is used for constructing time series data, inputting the improved ModernTCN network, generating eutrophication trend prediction results, and performing early warning; a data transmission module, which is used for transmitting data based on multi-mode communication technology; and a data storage module, which is used for storing data by using a distributed database and supporting historical data query and statistical analysis. The application combines the tidal coupling inversion model and the improved ModernTCN network, and realizes accurate eutrophication inversion interpretation early warning.

Description

Technical Field

[0001] This invention relates to the field of water pollution monitoring technology, and in particular to a high-precision spectroscopic method-based eutrophication inversion interpretation and early warning system for tidal river sections. Background Technology

[0002] With the increasing demand for water environment management in tidal river sections, cross-regional water ecological protection, and eutrophication risk early warning, high-precision monitoring, inversion interpretation, and trend prediction technologies for the eutrophication status of tidal river sections have received widespread attention. Existing water pollution monitoring instruments and tidal river section water quality monitoring and early warning systems mainly rely on manual sampling and laboratory analysis, fixed monitoring station detection, or conventional spectral inversion methods for eutrophication identification. However, these methods generally suffer from the following problems in practical applications: Tidal river sections are affected by tidal fluctuations, water level fluctuations, flow velocity changes, and salinity intrusion, resulting in significant time-varying and nonlinear spectral responses. Traditional single-spectrum monitoring methods are insufficient to stably characterize eutrophication indicators such as chlorophyll a, total phosphorus, and total nitrogen, leading to inversion results that are easily affected by tidal disturbances and have insufficient accuracy. Fixed-point monitoring and mobile surveys have different collection frequencies and spatial coverage areas, making it difficult for existing data fusion processing methods to achieve unified alignment and continuous completion of spectral reflectance data and synchronous hydrological data, resulting in insufficient temporal continuity and weak spatial representativeness. For the complex non-stationary temporal changes in tidal river sections, traditional threshold alarms or conventional time-series prediction methods are poorly adapted to multi-scale tidal cycle changes and sudden abnormal fluctuations, making it difficult to take into account both long-term trend characteristics and short-term local change characteristics. This leads to lag in eutrophication level prediction, insufficient early warning accuracy, and affects the timeliness and reliability of early warning linkage.

[0003] Therefore, how to provide a high-precision spectral method-based eutrophication inversion interpretation and early warning system for tidal river sections is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a high-precision spectral method-based eutrophication inversion interpretation and early warning system for tidal river sections. This invention combines a tidal coupled inversion model with an improved ModernTCN network. By introducing a continuous projection algorithm to screen eutrophication-sensitive bands, and combining water level fluctuations, velocity gradients, and salinity values ​​to construct a tidal coupling factor, the spectral reflectance data and synchronous hydrological data are effectively coupled, improving the accuracy of the inversion calculation. Simultaneously, the improved ModernTCN network incorporates an exponentially progressive and periodically folded expansion rate structure and a phase equipotential expansion mechanism to enhance the stability and continuity of multi-scale time-series characteristics, effectively suppressing the non-stationary effects of tidal disturbances, thereby significantly improving the reliability of eutrophication trend prediction and early warning response capabilities.

[0005] According to an embodiment of the present invention, a high-precision spectroscopic method-based eutrophication inversion interpretation and early warning system for tidal river sections includes the following modules: The data acquisition module is used for fixed-point monitoring and mobile surveying of the target area of ​​the tidal river section to obtain spectral reflectance data and synchronous hydrological data of the tidal river section. The data preprocessing module is used to perform dark current correction, atmospheric correction and water scattering correction on spectral reflectance data, perform time series completion on synchronous hydrological data using cubic spline interpolation algorithm, and align the corrected spectral reflectance data with the completed synchronous hydrological data in time to construct a preprocessed dataset. The tidal coupling inversion and interpretation module is used to screen eutrophication sensitive bands based on the preprocessed dataset using the continuous projection algorithm, construct a training dataset, and use the random forest algorithm to introduce tidal coupling factors to construct a tidal coupling inversion model. It performs inversion calculations on chlorophyll a, total phosphorus, and total nitrogen, interprets the eutrophication level based on the inversion results, and generates an inversion result set. The early warning linkage module is used to construct time series data based on the inversion result set, input the improved ModernTCN network to predict eutrophication trends, generate eutrophication trend prediction results and issue early warnings. The improved ModernTCN network includes an input compression layer, a dilated convolutional layer, a temporal structure adjustment layer, and a prediction output layer. The temporal structure adjustment layer is equipped with a phase equipotential unfolding mechanism. The data transmission module is used to transmit spectral reflectance data, synchronous hydrological data, inversion result sets, and eutrophication trend prediction results based on multimode communication technology. The data storage module is used to store data using a distributed database and supports historical data querying and statistical analysis.

[0006] Optionally, the data acquisition module specifically comprises: Fixed-point monitoring equipment is deployed at key sections of the target area in the tidal river section. The key sections include the pollution source inflow area, the tidal influence area, and the drinking water source protection area. The fixed-point monitoring equipment collects spectral reflectance data and synchronous hydrological data. The synchronous hydrological data includes water level data, flow velocity data, and salinity data, forming fixed-point spectral reflectance sampling data and fixed-point synchronous hydrological sampling data. A drone equipped with an airborne hyperspectral device was used to conduct mobile surveys along the tidal river section, collecting spectral reflectance data and acquiring synchronous hydrological data at the corresponding locations, forming survey spectral reflectance sampling data and survey synchronous hydrological sampling data. The drone plans its patrol route according to the tidal cycle and collects data one hour before high tide, at the peak of high tide, at the peak of low tide, and one hour after low tide. The fixed-point spectral reflectance sampling data and the surveyed spectral reflectance sampling data are merged to obtain spectral reflectance data. The fixed-point synchronous hydrological sampling data and the surveyed synchronous hydrological sampling data are merged to obtain synchronous hydrological data.

[0007] Optionally, the dark current correction, atmospheric correction, and water scattering correction of the spectral reflectance data specifically include: Under conditions of no incident light, the dark current reference value corresponding to each band is obtained. The original sampled value of each band in the spectral reflectance data is subtracted from the dark current reference value of the corresponding band to obtain the dark current correction data. While acquiring spectral reflectance data, standard white board reflectance data is also acquired. The values ​​of each band in the dark current correction data are divided by the corresponding standard white board reflectance data, and the result is multiplied by the nominal reflectance of the standard white board to obtain atmospheric correction data. Atmospheric correction data corresponding to 1400nm and 1900nm in the near-infrared band are selected. The atmospheric correction data corresponding to 1400nm and 1900nm in the near-infrared band are added together and averaged to obtain the scattering reference value. The scattering reference value is multiplied by a preset scaling factor to obtain the scattering correction value. The scattering correction value is subtracted one by one from the atmospheric correction data corresponding to each band in the visible light band from 400nm to 700nm to obtain the corrected spectral reflectance data.

[0008] Optionally, the step of performing time series completion on the synchronized hydrological data using a cubic spline interpolation algorithm specifically involves: The water level data, flow velocity data, and salinity data in the synchronous hydrological data were arranged in the order of collection time to construct water level time series, flow velocity time series, and salinity time series respectively. In the water level time series, flow velocity time series and salinity time series, missing time points are identified. For any missing time point, two adjacent sampled time points before and after the missing time point are selected and recorded as the previous time point and the next time point, respectively. The corresponding synchronous hydrological data values ​​are recorded as the previous value and the next value, respectively. A piecewise cubic function is constructed between the previous time point and the next time point. The piecewise cubic function is determined by four coefficients. The calculation method of the piecewise cubic function is as follows: subtract the previous time point from the current time point to obtain the time difference value, multiply the cube of the time difference value by the first coefficient, add the square of the time difference value by the second coefficient, add the time difference value by the third coefficient, and add the fourth coefficient. The four coefficients are solved by setting four constraints: the function value of the piecewise cubic function at the previous time point is equal to the previous value, the function value of the piecewise cubic function at the next time point is equal to the next value, the first derivative of the piecewise cubic function at the previous time point is equal to the slope of the previous time interval, and the first derivative of the piecewise cubic function at the next time point is equal to the slope of the next time interval. The slope is obtained by dividing the difference between the values ​​of two adjacent sampled time points by the time difference. The second derivative is set to 0 at the start time point and 0 at the end time point of the time series. The coefficients of each piecewise cubic function are obtained by solving the constraints of each piecewise cubic function simultaneously. Substitute the time value corresponding to the missing time point into the piecewise cubic function to calculate the complete value, fill the complete value into the corresponding missing time point position, and obtain the completed synchronous hydrological data.

[0009] Optionally, the step of using a continuous projection algorithm to screen eutrophication-sensitive bands based on the preprocessed dataset specifically involves: The preprocessed dataset is centrally corrected spectral reflectance data is expanded according to bands and matched with water level data, flow velocity data and salinity data in the completed synchronous hydrological data according to the same timestamp to form a data sample arranged in chronological order. The continuous projection algorithm is applied to the corrected spectral reflectance data in the data sample. The band with the smallest wavelength is selected as the initial band and added to the selected bands. The remaining bands are selected as candidate bands. For each candidate band, extract the numerical sequence of the candidate band in all samples and the numerical sequence of the selected band in all samples. Multiply each value in the numerical sequence of the candidate band with the corresponding value in the numerical sequence of the selected band point by point, and accumulate all the product results. Divide the accumulated result by the square of each value in the numerical sequence of the selected band and sum the results to obtain the projection coefficient. Multiply each value in the selected band numerical sequence by the projection coefficient to obtain the fitted numerical sequence of the candidate band on the selected band; Subtract the corresponding value in the fitted numerical sequence from each value in the candidate band numerical sequence to obtain the residual sequence. Square each value in the residual sequence and sum them up. Take the square root of the summation result to obtain the projected distance corresponding to the candidate band. The projected distances corresponding to all candidate bands are compared. The candidate band with the largest projected distance is selected as the new band and added to the selected bands. At the same time, the new band is deleted from the candidate bands. After updating the selected bands, repeat the process of adding new bands for the remaining candidate bands until the number of selected bands reaches the preset threshold. Eutrophication-sensitive bands were identified from the selected bands, including the 685nm, 705nm, 560nm, 420nm, 620nm, 1400nm, and 1900nm bands. The corrected spectral reflectance data corresponding to the 685nm and 705nm bands were identified as chlorophyll a-sensitive band data, the corrected spectral reflectance data corresponding to the 560nm band was identified as total phosphorus-sensitive band data, the corrected spectral reflectance data corresponding to the 420nm and 620nm bands was identified as total nitrogen-sensitive band data, and the corrected spectral reflectance data corresponding to the 1400nm and 1900nm bands was identified as salinity-sensitive band data.

[0010] Optionally, the construction of a training dataset and the introduction of a tidal coupling factor using a random forest algorithm to construct a tidal coupling inversion model, the inversion calculation of chlorophyll a, total phosphorus, and total nitrogen, the interpretation of eutrophication levels based on the inversion results, and the generation of an inversion result set are specifically as follows: The corrected spectral reflectance data corresponding to the 685nm, 705nm, 560nm, 420nm, 620nm, 1400nm and 1900nm bands are matched with the water level data, flow velocity data and salinity data in the completed synchronous hydrological data in chronological order to construct a training dataset. The training dataset is divided into a training set and a validation set, and multiple decision trees are constructed. The water level fluctuation is obtained by subtracting the value of the previous time point from the value of the water level data at the current time point, the flow velocity gradient is obtained by subtracting the value of the previous time point from the value of the flow velocity data at the current time point, and the salinity value at the current time point is used as the salinity value. The water level fluctuation, flow velocity gradient, and salinity value are used as tidal coupling factors. A tidal coupling inversion model is constructed based on the corrected spectral reflectance data, the coupled spectral data, and the tidal coupling factor. In the tidal coupling inversion model, during the node partitioning process of each decision tree, partitioning variables are selected from the corrected spectral reflectance data, coupled spectral data, water level fluctuation, flow velocity gradient and salinity value, and a partitioning threshold is set. Samples with values ​​less than the partitioning threshold are partitioned to one side of the sub-nodes, and samples with values ​​greater than or equal to the partitioning threshold are partitioned to the other side of the sub-nodes. The input data is calculated using a tidal coupling inversion model. The chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration output by all decision trees are summed and divided by the number of decision trees to obtain the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration. Based on the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration, the eutrophication level is obtained and combined with the timestamp to generate an inversion result set.

[0011] Optionally, the early warning linkage module specifically comprises: Time-series data is constructed based on the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration arranged in chronological order in the inversion result set, as well as the eutrophication level. The time-series data is then input into the input compression layer of the improved ModernTCN network. In the input compression layer, a one-dimensional convolution operation is performed on the time series data, and the convolution output is compressed through a linear transformation to obtain compressed time series features; The compressed temporal features are input into the dilated convolutional layer, which contains multiple convolutional blocks. Each convolutional block uses a dilation rate sequence that combines exponential progression and periodic folding to perform one-dimensional convolution operations. The dilation rate sequence increases in power of 2 to a preset upper limit and then decreases in reverse order according to the same sequence to form a closed sequence. The compressed temporal features are input into the first convolutional block to obtain the first layer output features. The first layer output features are added element-wise to the compressed temporal features and used as the input to the second convolutional block to obtain the second layer output features. For each subsequent convolutional block, the previous layer output features are added element-wise to the compressed temporal features and used as the input to the current layer convolutional block to obtain the corresponding layer output features. The output features of each convolutional block are concatenated to obtain multi-scale temporal features. The multi-scale temporal features are input into the temporal structure adjustment layer, and the multi-scale temporal features are processed through the phase equipotential unfolding mechanism to obtain the unfolded correction features. The folding correction features are input into the prediction output layer, and through one-dimensional convolution and fully connected operations, the eutrophication trend prediction results for future time steps are output, and an early warning is given based on the eutrophication trend prediction results.

[0012] Optionally, the step of processing the multi-scale time series features through a phase equipotential unfolding mechanism to obtain unfolded corrected features specifically involves: The multi-scale temporal features are arranged in chronological order, and the multi-scale temporal features of the current time step and the multi-scale temporal features of adjacent time steps are extracted respectively. Subtracting the multi-scale temporal features of the previous time step from the multi-scale temporal features of the current time step yields the first difference sequence; subtracting the multi-scale temporal features of the current time step from the multi-scale temporal features of the next time step yields the second difference sequence. Add the corresponding elements of the first difference sequence and the second difference sequence, and then divide by 2 to obtain the symmetric difference sequence; The absolute value of each element in the symmetric difference sequence is calculated, and the sign consistency is judged with the first difference sequence at the corresponding position. When the corresponding element of the first difference sequence is positive, a positive value is taken, and when the corresponding element of the first difference sequence is negative, a negative value is taken, thus obtaining a phase-consistent sequence. At the current time step, extract the phase-consistent sequences corresponding to the previous time step, the current time step, and the next time step. Add the phase-consistent sequences corresponding to the previous time step, the current time step, and the next time step element by element to obtain the folded feature sequence corresponding to the current time step. Arrange the folded feature sequences obtained at each time step sequentially to obtain the unfolded feature sequence; The expanded feature sequence is added element-wise to the multi-scale temporal features to obtain the expanded correction features.

[0013] Optionally, the multi-mode communication technology specifically includes 4G / 5G, optical fiber, and BeiDou satellite.

[0014] Optionally, the data storage module specifically comprises: The spectral reflectance data, synchronous hydrological data, and the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level in the inversion result set are matched with timestamps and written to different storage nodes in the distributed database. In the distributed database, spectral reflectance data, synchronous hydrological data, and inversion result sets are stored in segments according to time order. Timestamp-based indexes are established for the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level in the spectral reflectance data, synchronous hydrological data, and inversion result sets, respectively. When querying historical data, based on the input time range, the system reads the spectral reflectance data, synchronous hydrological data, and predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level within the corresponding time period through the timestamp index. In the statistical analysis, the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration were accumulated in chronological order and the average value was calculated. The eutrophication level was also statistically analyzed to obtain the statistical results of historical data.

[0015] The beneficial effects of this invention are: This invention addresses the problems of unstable spectral response, low accuracy of eutrophication index inversion, and lag in trend prediction in tidal river sections under tidal influences through the collaborative construction of a tidal coupled inversion model and an improved ModernTCN network. It employs a continuous projection algorithm to screen eutrophication-sensitive bands and introduces water level variation, velocity gradient, and salinity as tidal coupling factors into a random forest algorithm to construct a tidal coupled inversion model. This couples spectral reflectance data with synchronous hydrological data, enabling high-precision inversion calculations of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration. In the trend prediction stage, the inversion result set is constructed as time-series data and input into an improved ModernTCN network. By introducing an exponentially progressive and periodically folded dilation rate sequence into the dilated convolutional layer, a closed structure of the temporal receptive field is formed. Simultaneously, a layer-by-layer injection method for compressing temporal features is introduced into multi-layer convolutional blocks to maintain the continuity of the original temporal information. Furthermore, a phase equipotential unfolding mechanism is set in the temporal structure adjustment layer to perform symmetric balancing and local enhancement processing on multi-scale temporal features, thereby effectively suppressing asymmetric fluctuations caused by tidal disturbances and improving feature stability and continuity. Ultimately, high-precision output of eutrophication trend prediction results is achieved, enhancing the system's adaptability to water quality changes under complex tidal environments and improving the accuracy, timeliness, and robustness of early warning results. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the eutrophication inversion and interpretation early warning system for tidal river sections based on high-precision spectroscopy proposed in this invention. Figure 2 This is a schematic diagram of the improved ModernTCN network structure for the eutrophication inversion interpretation and early warning system of tidal river sections based on high-precision spectroscopy proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figure 1 and Figure 2 The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy includes the following modules: The data acquisition module is used for fixed-point monitoring and mobile surveying of the target area of ​​the tidal river section to obtain spectral reflectance data and synchronous hydrological data of the tidal river section. The data preprocessing module is used to perform dark current correction, atmospheric correction and water scattering correction on spectral reflectance data, perform time series completion on synchronous hydrological data using cubic spline interpolation algorithm, and align the corrected spectral reflectance data with the completed synchronous hydrological data in time to construct a preprocessed dataset. The tidal coupling inversion and interpretation module is used to screen eutrophication sensitive bands based on the preprocessed dataset using the continuous projection algorithm, construct a training dataset, and use the random forest algorithm to introduce tidal coupling factors to construct a tidal coupling inversion model. It performs inversion calculations on chlorophyll a, total phosphorus, and total nitrogen, interprets the eutrophication level based on the inversion results, and generates an inversion result set. The early warning linkage module is used to construct time series data based on the inversion result set, input the improved ModernTCN network to predict eutrophication trends, generate eutrophication trend prediction results and issue early warnings. The improved ModernTCN network includes an input compression layer, a dilated convolutional layer, a temporal structure adjustment layer, and a prediction output layer. The temporal structure adjustment layer incorporates a phase equipotential unfolding mechanism. The data transmission module is used to transmit spectral reflectance data, synchronous hydrological data, inversion result sets, and eutrophication trend prediction results based on multimode communication technology. The data storage module is used to store data using a distributed database and supports historical data querying and statistical analysis.

[0019] In this embodiment, the data acquisition module specifically comprises: Fixed-point monitoring equipment is deployed at key sections of the target area in the tidal river section. The key sections include the pollution source inflow area, the tidal influence area, and the drinking water source protection area. The fixed-point monitoring equipment collects spectral reflectance data and synchronous hydrological data. The synchronous hydrological data includes water level data, flow velocity data, and salinity data, forming fixed-point spectral reflectance sampling data and fixed-point synchronous hydrological sampling data. A drone equipped with an airborne hyperspectral device was used to conduct mobile surveys along the tidal river section, collecting spectral reflectance data and acquiring synchronous hydrological data at the corresponding locations, forming survey spectral reflectance sampling data and survey synchronous hydrological sampling data. The drones plan their patrol routes according to the tidal cycle and collect data one hour before high tide, at the peak of high tide, at the peak of low tide, and one hour after low tide. The fixed-point spectral reflectance sampling data and the patrol-based spectral reflectance sampling data are merged to obtain spectral reflectance data. The fixed-point synchronous hydrological sampling data and the patrol-based synchronous hydrological sampling data are merged to obtain synchronous hydrological data. In the specific implementation process, fixed-point monitoring equipment and airborne hyperspectral equipment carried by UAVs are used to collect spectral reflectance sampling data. The fixed-point monitoring equipment integrates a high-resolution spectral sensor for water pollution monitoring, with a spectral range of 450-950nm, a spectral resolution of 10nm, and a data acquisition frequency of 10 minutes / time. The UAV carries a hyperspectral imager for mobile surveying, with a spectral range of 400-2500nm, a spectral resolution of 3nm, a spatial resolution of 0.5m, a flight altitude of 100m, and a data acquisition frequency of once per tidal cycle. By coordinating the acquisition of mid-to-high frequency continuous spectral reflectance sampling data from the fixed-point monitoring equipment with the broadband high-resolution spectral reflectance sampling data from the UAV surveying, the continuous variation characteristics of local areas and the spatial distribution characteristics of the entire river section are simultaneously covered. In subsequent data processing, the two types of spectral reflectance sampling data are processed accordingly within a common band range to achieve data alignment between different acquisition methods, thereby improving data utilization efficiency and enhancing the stability and spatial representativeness of eutrophication inversion results.

[0020] In this embodiment, dark current correction, atmospheric correction, and water scattering correction are performed on the spectral reflectance data, specifically as follows: Under conditions of no incident light, the dark current reference value corresponding to each band is obtained. The original sampled value of each band in the spectral reflectance data is subtracted from the dark current reference value of the corresponding band to obtain the dark current correction data. While acquiring spectral reflectance data, standard white board reflectance data is also acquired. The values ​​of each band in the dark current correction data are divided by the corresponding standard white board reflectance data, and the result is multiplied by the nominal reflectance of the standard white board to obtain atmospheric correction data. Atmospheric correction data corresponding to 1400nm and 1900nm in the near-infrared band are selected. The atmospheric correction data corresponding to 1400nm and 1900nm in the near-infrared band are added together and averaged to obtain the scattering reference value. The scattering reference value is multiplied by a preset scaling factor to obtain the scattering correction value. The scattering correction value is subtracted one by one from the atmospheric correction data corresponding to each band in the visible light band from 400nm to 700nm to obtain the corrected spectral reflectance data. In the specific implementation process, the dark current reference value is obtained by continuous sampling under completely shaded conditions of the acquisition equipment, with the sampling number set to 50 times. The 50 sampling results are averaged band by band to obtain a stable dark current reference value for each band. The standard white board reflectance data is obtained by spectral sampling of the standard white board at the same acquisition location and time conditions. The stable range of the standard white board reflectance is 0.98±0.01. The sampling height is consistent with the water body sampling height, both set at 1.5m above the water surface. The nominal reflectance of the standard white board adopts a pre-calibrated value, uniformly set to 0.99 in the 400nm to 700nm band and uniformly set to 0.98 in the 700nm to 2500nm band. During the dark current correction process, the original sampling value of each band is subtracted point by point from the corresponding band dark current reference value to form dark current correction data. In the atmospheric calibration... During the process, the dark current correction data for each band is divided point by point by the corresponding standard whiteboard reflectance data, and then multiplied by the nominal reflectance of the corresponding band to form atmospheric correction data. In the water scattering correction process, the atmospheric correction data of the 1400nm and 1900nm bands are summed and divided by 2 to obtain the scattering reference value. The scattering reference value is then multiplied by a preset scaling factor, which is set to 0.8. The resulting scattering correction values ​​are subtracted point by point from the atmospheric correction data of each band in the range of 400nm to 700nm to obtain the corrected spectral reflectance data. Dark current correction removes the background noise of the equipment, atmospheric correction eliminates the influence of illumination changes, and water scattering correction weakens the spectral shift caused by suspended particles, making the spectral reflectance data more stable and consistent, improving the comparability of data from different times and spaces, and thus improving the accuracy and reliability of eutrophication inversion calculation.

[0021] In this embodiment, time series completion of synchronous hydrological data is performed using a cubic spline interpolation algorithm, specifically as follows: The water level data, flow velocity data, and salinity data in the synchronous hydrological data were arranged in the order of collection time to construct water level time series, flow velocity time series, and salinity time series respectively. In the water level time series, flow velocity time series and salinity time series, missing time points are identified. For any missing time point, two adjacent sampled time points before and after the missing time point are selected and recorded as the previous time point and the next time point, respectively. The corresponding synchronous hydrological data values ​​are recorded as the previous value and the next value, respectively. Construct a piecewise cubic function between the previous time point and the next time point. The piecewise cubic function is determined by four coefficients. The calculation method of the piecewise cubic function is as follows: subtract the previous time point from the current time point to obtain the time difference, multiply the cube of the time difference by the first coefficient, add the square of the time difference by the second coefficient, add the time difference by the third coefficient, and add the fourth coefficient. The four coefficients are solved by setting four constraints: the function value of the piecewise cubic function at the previous time point is equal to the previous value, the function value of the piecewise cubic function at the next time point is equal to the next value, the first derivative of the piecewise cubic function at the previous time point is equal to the slope of the previous time interval, and the first derivative of the piecewise cubic function at the next time point is equal to the slope of the next time interval. The slope is obtained by dividing the difference between the values ​​of two adjacent sampled time points by the time difference. The second derivative is set to 0 at the start time point and 0 at the end time point of the time series. The coefficients of each piecewise cubic function are obtained by solving the constraints of each piecewise cubic function simultaneously. Substitute the time value corresponding to the missing time point into the piecewise cubic function to calculate the complete value, fill the complete value into the corresponding missing time point position to obtain the completed synchronous hydrological data. This invention reconstructs water level, flow velocity, and salinity time series using a unified time axis, ensuring consistent sampling intervals across various synchronous hydrological data and avoiding data misalignment caused by uneven sampling intervals. It employs piecewise cubic functions for interpolation of missing time points, maintaining continuity in function values, first-order rates of change, and second-order trends, thus guaranteeing the smoothness and physical consistency of the completed synchronous hydrological data over time. By imposing a constraint that the second derivative is zero at the start and end time points, the invention stabilizes the trends at the time series boundaries, preventing abrupt boundary changes. By calculating the slope based on the numerical and temporal differences between adjacent time points, the completed values ​​reflect the actual rate of hydrological change, improving the approximation of the true change process. The completed synchronous hydrological data obtained using the cubic spline interpolation algorithm significantly outperforms linear interpolation methods in terms of continuity, stability, and trend consistency, effectively reducing the impact of missing data on the tidal coupling inversion model calculation results and improving overall inversion accuracy and system reliability.

[0022] In this embodiment, a continuous projection algorithm is used to screen eutrophication-sensitive bands based on the preprocessed dataset, specifically as follows: The preprocessed dataset is centrally corrected spectral reflectance data is expanded according to bands and matched with water level data, flow velocity data and salinity data in the completed synchronous hydrological data according to the same timestamp to form a data sample arranged in chronological order. The continuous projection algorithm is applied to the corrected spectral reflectance data in the data sample. The band with the smallest wavelength is selected as the initial band and added to the selected bands. The remaining bands are selected as candidate bands. For each candidate band, extract the numerical sequence of the candidate band in all samples and the numerical sequence of the selected band in all samples. Multiply each value in the numerical sequence of the candidate band with the corresponding value in the numerical sequence of the selected band point by point, and accumulate all the product results. Divide the accumulated result by the square of each value in the numerical sequence of the selected band and sum the results to obtain the projection coefficient. Multiply each value in the selected band numerical sequence by the projection coefficient to obtain the fitted numerical sequence of the candidate band on the selected band; Subtract the corresponding value in the fitted numerical sequence from each value in the candidate band numerical sequence to obtain the residual sequence. Square each value in the residual sequence and sum them up. Take the square root of the summation result to obtain the projected distance corresponding to the candidate band. The projected distances corresponding to all candidate bands are compared. The candidate band with the largest projected distance is selected as the new band and added to the selected bands. At the same time, the new band is deleted from the candidate bands. After updating the selected bands, repeat the process of adding new bands for the remaining candidate bands until the number of selected bands reaches the preset threshold. Eutrophication-sensitive bands were identified from the selected bands, including the 685nm, 705nm, 560nm, 420nm, 620nm, 1400nm, and 1900nm bands. The corrected spectral reflectance data corresponding to the 685nm and 705nm bands were identified as chlorophyll a-sensitive band data, the corrected spectral reflectance data corresponding to the 560nm band was identified as total phosphorus-sensitive band data, the corrected spectral reflectance data corresponding to the 420nm and 620nm bands was identified as total nitrogen-sensitive band data, and the corrected spectral reflectance data corresponding to the 1400nm and 1900nm bands was identified as salinity-sensitive band data. In the specific implementation process, the corrected spectral reflectance data was discretized according to 1nm wavelength intervals, forming a spectral sequence of 2101 bands in the range of 400nm to 2500nm, and time-aligned with the completed synchronous hydrological data at a uniform time interval; no less than 1000 sets of valid sample data were collected, each set of samples including spectral reflectance data and the corresponding measured values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration; during the execution of the continuous projection algorithm, the threshold for the number of selected bands was set to 20 bands, and the band corresponding to the wavelength of 400nm was used as the initial band; when calculating the projection coefficient, all sample data were used in the calculation to ensure the accuracy of the projection distance calculation. To improve stability, among the selected bands, 685nm, 705nm, 560nm, 420nm, 620nm, 1400nm, and 1900nm were further identified as eutrophication-sensitive bands. The 685nm and 705nm bands correspond to the chlorophyll a absorption peak region, the 560nm band corresponds to the total phosphorus reflectance peak region, the 420nm and 620nm bands correspond to the total nitrogen characteristic variation region, and the 1400nm and 1900nm bands correspond to the salinity-sensitive variation region. This ensures that the selected sensitive bands possess both index response capability and environmental coupling capability under controlled quantity, thereby improving the representativeness and stability of the input data of the tidal coupling inversion model.

[0023] In this embodiment, a training dataset is constructed, and a tidal coupling factor is introduced using the random forest algorithm to build a tidal coupling inversion model. Chlorophyll a, total phosphorus, and total nitrogen are inverted and calculated. The eutrophication level is interpreted based on the inversion results, and an inversion result set is generated. Specifically: The corrected spectral reflectance data corresponding to the 685nm, 705nm, 560nm, 420nm, 620nm, 1400nm and 1900nm bands are matched with the water level data, flow velocity data and salinity data in the completed synchronous hydrological data in chronological order to construct a training dataset. The training dataset is divided into a training set and a validation set, and multiple decision trees are constructed. The water level fluctuation is obtained by subtracting the value of the previous time point from the value of the water level data at the current time point, the flow velocity gradient is obtained by subtracting the value of the previous time point from the value of the flow velocity data at the current time point, and the salinity value at the current time point is used as the salinity value. The water level fluctuation, flow velocity gradient, and salinity value are used as tidal coupling factors. A tidal coupling inversion model is constructed based on the corrected spectral reflectance data, the coupled spectral data, and the tidal coupling factor. In the tidal coupling inversion model, during the node partitioning process of each decision tree, partitioning variables are selected from the corrected spectral reflectance data, coupled spectral data, water level fluctuation, flow velocity gradient and salinity value, and a partitioning threshold is set. Samples with values ​​less than the partitioning threshold are partitioned to one side of the sub-nodes, and samples with values ​​greater than or equal to the partitioning threshold are partitioned to the other side of the sub-nodes. The input data is calculated using a tidal coupling inversion model. The chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration output by all decision trees are summed and divided by the number of decision trees to obtain the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration. Based on the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration, the eutrophication level is obtained and combined with the timestamp to generate an inversion result set. In the specific implementation process, the training dataset is divided into a training set and a validation set in a 7:3 ratio. A random forest model with 200 decision trees is constructed in the training set. During the construction of the tidal coupling factor, the water level fluctuation is obtained by subtracting the previous water level from the current water level data, the velocity gradient is obtained by subtracting the previous velocity from the current velocity data, and the salinity value is directly taken from the current salinity data. The corrected spectral reflectance data is multiplied point-by-point with the water level fluctuation, velocity gradient, and salinity value to form the coupled spectral data, which, along with the original spectral reflectance data and the tidal coupling factor, serves as input for decision tree node partitioning. During the inversion calculation, the outputs of all decision trees are accumulated and averaged to obtain the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration. Based on the "Surface Water Environmental Quality Standard" (GB... (3838-2002) divides chlorophyll a concentration into five levels: 10 μg / L, 20 μg / L, 40 μg / L, and 60 μg / L; total phosphorus concentration into five levels: 0.02 mg / L, 0.1 mg / L, 0.2 mg / L, and 0.4 mg / L; and total nitrogen concentration into five levels: 0.2 mg / L, 0.5 mg / L, 1.0 mg / L, and 2.0 mg / L. The corresponding levels of the three indicators are combined to determine the eutrophic, mesotrophic, slightly eutrophic, moderately eutrophic, and severely eutrophic levels, thus achieving quantitative classification and real-time interpretation of the inversion results.

[0024] In the node partitioning process of each decision tree, instead of directly selecting a single variable from all variables as the partitioning criterion, a subset of variables is first randomly selected from the complete variable set consisting of corrected spectral reflectance data, coupled spectral data, water level fluctuations, flow velocity gradients, and salinity values. The number of these subset variables is set to the integer value of the square root of the total number of variables, forming a candidate variable set. Then, each variable in the candidate variable set is partitioned separately. The values ​​of this variable in the current node sample are sorted from smallest to largest. The corresponding partitioning threshold is obtained by adding two adjacent values ​​and dividing by 2. Each threshold is used to divide the sample into two parts. The mean values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration in each sub-node are calculated. The mean value of each sample is subtracted from the mean value of its sub-node, squared, and summed to obtain the discrete value of the sub-node. The discrete values ​​of the left and right sub-nodes are summed to obtain the total discrete value. The variable with the smallest total discrete value and its corresponding threshold are selected from the candidate variable set as the optimal partitioning variable and optimal partitioning threshold for the current node. This completes the node partitioning, reduces the data differences within the partitioned sub-nodes, and improves the stability and accuracy of the model for water quality index inversion calculation.

[0025] In this embodiment, the early warning linkage module specifically includes: Time-series data were constructed based on the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level arranged in chronological order in the inversion results set, and the time-series data were input into the input compression layer of the improved ModernTCN network. In the input compression layer, one-dimensional convolution operation is performed on the time series data, and the convolution output is compressed through a linear transformation to obtain compressed time series features. The compressed temporal features are input into the dilated convolutional layer. The dilated convolutional layer sets up multiple convolutional blocks. Each convolutional block uses a dilation rate sequence that combines exponential progression and periodic folding to perform one-dimensional convolution operations. The dilation rate sequence increases in power of 2 to a preset upper limit and then decreases in reverse order according to the same sequence to form a closed sequence. In the specific implementation process, the number of convolutional blocks in the dilated convolutional layer corresponds to the length of the dilation rate sequence. The dilation rate sequence is set to 1, 2, 4, 8, 16, 8, 4, 2, 1 in an exponential progression and periodic foldback combination manner, corresponding to 9 convolutional blocks. That is, the dilation rate of each convolutional block is 1, 2, 4, 8, 16, 8, 4, 2, 1 in sequence. Each convolutional block performs one-dimensional convolution operation on the compressed temporal features in the above order, thereby forming a receptive field structure that first expands and then contracts. This allows the network to gradually expand the temporal perception range in the first half and gradually recover and strengthen local detail features in the second half, thereby realizing the continuous extraction of features at different time scales. The compressed temporal features are input into the first convolutional block to obtain the first layer output features. The first layer output features are then added element-wise with the compressed temporal features and used as the input to the second convolutional block to obtain the second layer output features. For each subsequent convolutional block, the previous layer output features are added element-wise with the compressed temporal features and used as the input to the current layer convolutional block to obtain the corresponding layer output features. The output features of each convolutional block are then concatenated to obtain the multi-scale temporal features. Multi-scale temporal features are input into the temporal structure adjustment layer, and the multi-scale temporal features are processed through the phase equipotential expansion mechanism to obtain the expanded correction features; The folded correction features are input into the prediction output layer. Through one-dimensional convolution and fully connected operations, the eutrophication trend prediction results for future time steps are output, and early warning is given based on the eutrophication trend prediction results. In the specific implementation process, the unfolded correction features are input into the prediction output layer in chronological order. First, a one-dimensional convolution operation is performed on the unfolded correction features, and the convolution kernel slides along the time dimension to extract continuous time change features, resulting in an intermediate feature sequence. Then, the intermediate feature sequence is input into a fully connected layer, and the features of each time step are linearly combined to obtain the output value of the corresponding time step. The output values ​​are arranged in chronological order to form the chlorophyll a concentration change trend, total phosphorus concentration change trend, and total nitrogen concentration change trend for several future time steps. Based on the change trend results corresponding to each time step, the eutrophication level change trend is determined, thereby obtaining the eutrophication trend prediction result within the future time range. This enables the prediction of eutrophication level changes 24 hours in advance, and sets three warning levels: blue, yellow, and red.

[0026] In this embodiment, the multi-scale time series features are processed through a phase equipotential unfolding mechanism to obtain unfolded and corrected features, specifically: The multi-scale temporal features are arranged in chronological order, and the multi-scale temporal features of the current time step and the multi-scale temporal features of adjacent time steps are extracted respectively. Subtracting the multi-scale temporal features of the previous time step from the multi-scale temporal features of the current time step yields the first difference sequence; subtracting the multi-scale temporal features of the current time step from the multi-scale temporal features of the next time step yields the second difference sequence. Add the corresponding elements of the first difference sequence and the second difference sequence, and then divide by 2 to obtain the symmetric difference sequence; Perform absolute value operations on each element in the symmetric difference sequence and check the sign consistency with the first difference sequence at the corresponding position. If the corresponding element in the first difference sequence is positive, take the positive value; if the corresponding element in the first difference sequence is negative, take the negative value. This yields the phase-consistent sequence. At the current time step, extract the phase-consistent sequences corresponding to the previous time step, the current time step, and the next time step. Add the phase-consistent sequences corresponding to the previous time step, the current time step, and the next time step element by element to obtain the folded feature sequence corresponding to the current time step. Arrange the folded feature sequences obtained at each time step sequentially to obtain the unfolded feature sequence; The expanded feature sequence is added element-wise to the multi-scale temporal features to obtain the expanded correction features; This invention processes multi-scale time series features through a phase equipotential unfolding mechanism, balancing and enhancing changes between adjacent time steps while maintaining the overall trend of the original time series. By symmetrically calculating and averaging the differences between consecutive time steps, the current time step is numerically constrained by the changes of both the previous and subsequent time steps, resulting in a relatively balanced variation amplitude within the local time series and preventing unilateral abrupt changes from shifting the overall trend. Simultaneously, sign consistency processing maintains the continuity of the time series' direction of change, avoiding distortion of trend judgment due to local reversals. During the folding process, features from the previous, current, and subsequent time steps are superimposed, concentrating information within the local time range and improving the sensitivity of features to short-period changes. During the unfolding process, the folded features are rearranged chronologically, evenly distributing the enhanced information across all time steps. This enhances the overall feature expressive power without altering the time structure, effectively suppressing high-frequency fluctuations caused by tidal disturbances, improving the stability and continuity of multi-scale time series features, and ultimately enhancing the accuracy and robustness of eutrophication trend prediction results.

[0027] The improved ModernTCN network in this invention still follows the basic idea of ​​the traditional ModernTCN network in terms of overall structure, which is to construct a temporal feature extraction framework based on one-dimensional convolution and dilated convolution. That is, the input temporal data is extracted layer by layer through convolution operation, and the temporal receptive field is expanded by dilated convolution, so as to realize the modeling of long-term dependencies. On this basis, the network structure is improved for the tidal river section scenario. Introducing an exponentially increasing and periodically folding dilation rate sequence into the dilated convolutional layer transforms the convolutional receptive field from unidirectional expansion to a closed structure that expands and then contracts, thereby simultaneously enhancing the expressive power of long-term trend features and short-term local fluctuation features within the same network layer. A layer-by-layer injection method based on compressed temporal features is introduced between multiple convolutional blocks, allowing each convolutional block to retain original temporal information while extracting higher-order features, avoiding feature shift and information attenuation in deep networks. An equipotential unfolding mechanism is introduced into the temporal structure adjustment layer, balancing the asymmetric fluctuations caused by tidal changes in the time series by symmetrically processing adjacent time step differences and folding and unfolding features within local time ranges, and enhancing the information expressive power within local time windows. Through these improvements, the improved ModernTCN network, while maintaining its original efficient temporal modeling capabilities, further enhances its adaptability to complex periodic changes in tidal river sections, reduces the interference of abnormal fluctuations on prediction results, and thus significantly improves the stability, accuracy, and robustness of eutrophication trend prediction.

[0028] In this embodiment, multi-mode communication technologies specifically include 4G / 5G, optical fiber, and BeiDou satellite; In practical implementation, 4G / 5G communication networks are used for real-time data uploading between fixed-point monitoring equipment and mobile survey equipment, with transmission latency controlled within 50ms. Fiber optic communication links are used for high-capacity data transmission between the monitoring center and the data processing server, with a transmission bandwidth of over 1Gbps to meet the stable transmission requirements of high-frequency spectral reflectance data and synchronous hydrological data. BeiDou satellite communication links are used for supplementary data transmission in areas with insufficient 4G / 5G network coverage, ensuring the continuity and integrity of monitoring data in remote areas. Through the combined application of these three communication methods, reliable data transmission under different environmental conditions is achieved, improving the overall transmission stability of the system and avoiding data loss due to communication interruptions, thereby ensuring the continuity and accuracy of the eutrophication inversion interpretation and early warning process.

[0029] In this embodiment, the data storage module specifically comprises: The spectral reflectance data, synchronous hydrological data, and the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level in the inversion result set are matched with timestamps and written to different storage nodes in the distributed database. In the distributed database, spectral reflectance data, synchronous hydrological data, and inversion result sets are stored in segments according to time order. Time-stamp-based indexes are established for the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level in the spectral reflectance data, synchronous hydrological data, and inversion result sets. When querying historical data, based on the input time range, the system reads the spectral reflectance data, synchronous hydrological data, and predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level within the corresponding time period through the timestamp index. In the statistical analysis, the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration were accumulated in chronological order and the average value was calculated. The eutrophication level was also statistically analyzed to obtain the statistical results of historical data.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a eutrophication monitoring and early warning scenario in a tidal river section flowing into the sea in a coastal city. The upstream area is significantly affected by urban non-point source pollution and tributary inflows, while the downstream area is significantly affected by tidal cycles and salinity backwater effects. Long-standing problems include long manual sampling cycles, insufficient representativeness of fixed stations, increased inversion errors due to tidal changes, and delayed early warnings. The project operated continuously for 6 months, covering both the high-water and low-water seasons. Fixed-point monitoring equipment was deployed in the pollution source inflow area, the tidal influence area, and the drinking water source protection area. Unmanned aerial vehicles (UAVs) were used for mobile surveys one hour before high tide, at the peak of high tide, at the peak of low tide, and one hour after low tide. The fixed-point monitoring equipment collected spectral reflectance data and synchronous hydrological data every 10 minutes, and the UAVs conducted one survey per tidal cycle.

[0031] The collected spectral reflectance data were first corrected for dark current, atmosphere, and water scattering. Simultaneous hydrological data was completed using cubic spline interpolation and then aligned with the spectral reflectance data to construct a preprocessed dataset. Based on this, a continuous projection algorithm was used to screen eutrophication-sensitive bands, identifying the 685nm, 705nm, 560nm, 420nm, 620nm, 1400nm, and 1900nm bands as key inputs. Furthermore, water level fluctuations, velocity gradients, and salinity values ​​were introduced as tidal coupling factors into a random forest model to construct a tidal coupling inversion model. A grid search determined the number of decision trees to be 200 and the maximum depth to be 15. The inversion result set was then input into an improved ModernTCN network. Trend prediction was performed using an exponential progression and periodic foldback combination of expansion rate sequences and a phase equipotential expansion mechanism. The eutrophication trend prediction results for the next 24 hours were output and pushed to the management platform according to blue, yellow, and red three-level warning rules.

[0032] Table 1. Comparison of Eutrophication Trend Prediction and Early Warning Performance As can be seen from the data in Table 1 above, there are significant differences in the accuracy of eutrophication trend prediction and the reliability of early warning among different methods. The system of this invention achieves a trend prediction accuracy of 92.8%, significantly higher than the 82.4% of the LSTM prediction method and the 86.9% of the traditional ModernTCN network method. This indicates that by introducing a tidal coupling inversion model and improving the ModernTCN network structure, the complex temporal variation characteristics of tidal river sections affected by tidal cycles can be more effectively characterized, thus improving the overall prediction capability. Regarding the false alarm rate, the system of this invention is controlled at 6.4%, which is more than half lower than the 13.6% of the LSTM prediction method and also significantly lower than the 10.8% of the traditional ModernTCN network method. This shows that after balancing the multi-scale temporal characteristics through the phase equipotential expansion mechanism, abnormal fluctuations caused by tidal disturbances can be effectively suppressed, reducing misjudgments. Regarding the false negative rate, the system of this invention is 4.1%, significantly lower than the 9.8% of the LSTM prediction method and the 7.6% of the traditional ModernTCN network method, indicating that the system of this invention is more sensitive and stable in identifying potential eutrophication risks. In summary, the system of this invention performs best in the three key indicators of trend prediction accuracy, false alarm rate, and missed alarm rate, demonstrating strong prediction reliability and practical value for early warning.

[0033] Meanwhile, in the early warning stage, the system of this invention can achieve an early warning response time of no more than 5 minutes, which significantly improves the response speed to sudden eutrophication risks and provides timely and effective data support for relevant departments to carry out emergency dispatch and pollution control. The system of this invention is superior to traditional methods in terms of accuracy, stability and response efficiency, which verifies the practicality and reliability of the system of this invention.

[0034] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A high-precision spectroscopic method-based eutrophication inversion interpretation and early warning system for tidal river sections, characterized in that, Includes the following modules: The data acquisition module is used for fixed-point monitoring and mobile surveying of the target area of ​​the tidal river section to obtain spectral reflectance data and synchronous hydrological data of the tidal river section. The data preprocessing module is used to perform dark current correction, atmospheric correction and water scattering correction on spectral reflectance data, perform time series completion on synchronous hydrological data using cubic spline interpolation algorithm, and align the corrected spectral reflectance data with the completed synchronous hydrological data in time to construct a preprocessed dataset. The tidal coupling inversion and interpretation module is used to screen eutrophication sensitive bands based on the preprocessed dataset using the continuous projection algorithm, construct a training dataset, and use the random forest algorithm to introduce tidal coupling factors to construct a tidal coupling inversion model. It performs inversion calculations on chlorophyll a, total phosphorus, and total nitrogen, interprets the eutrophication level based on the inversion results, and generates an inversion result set. The early warning linkage module is used to construct time series data based on the inversion result set, input the improved ModernTCN network to predict eutrophication trends, generate eutrophication trend prediction results and issue early warnings. The improved ModernTCN network includes an input compression layer, a dilated convolutional layer, a temporal structure adjustment layer, and a prediction output layer. The temporal structure adjustment layer is equipped with a phase equipotential unfolding mechanism. The data transmission module is used to transmit spectral reflectance data, synchronous hydrological data, inversion result sets, and eutrophication trend prediction results based on multimode communication technology. The data storage module is used to store data using a distributed database and supports historical data querying and statistical analysis.

2. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 1, characterized in that, The data acquisition module is specifically: Fixed-point monitoring equipment is deployed at key sections of the target area in the tidal river section. The key sections include the pollution source inflow area, the tidal influence area, and the drinking water source protection area. The fixed-point monitoring equipment collects spectral reflectance data and synchronous hydrological data. The synchronous hydrological data includes water level data, flow velocity data, and salinity data, forming fixed-point spectral reflectance sampling data and fixed-point synchronous hydrological sampling data. A drone equipped with an airborne hyperspectral device was used to conduct mobile surveys along the tidal river section, collecting spectral reflectance data and acquiring synchronous hydrological data at the corresponding locations, forming survey spectral reflectance sampling data and survey synchronous hydrological sampling data. The drone plans its patrol route according to the tidal cycle and collects data one hour before high tide, at the peak of high tide, at the peak of low tide, and one hour after low tide. The fixed-point spectral reflectance sampling data and the surveyed spectral reflectance sampling data are merged to obtain spectral reflectance data. The fixed-point synchronous hydrological sampling data and the surveyed synchronous hydrological sampling data are merged to obtain synchronous hydrological data.

3. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 1, characterized in that, The dark current correction, atmospheric correction, and water scattering correction of the spectral reflectance data are specifically performed as follows: Under conditions of no incident light, the dark current reference value corresponding to each band is obtained. The original sampled value of each band in the spectral reflectance data is subtracted from the dark current reference value of the corresponding band to obtain the dark current correction data. While acquiring spectral reflectance data, standard white board reflectance data is also acquired. The values ​​of each band in the dark current correction data are divided by the corresponding standard white board reflectance data, and the result is multiplied by the nominal reflectance of the standard white board to obtain atmospheric correction data. Atmospheric correction data corresponding to 1400nm and 1900nm in the near-infrared band are selected. The atmospheric correction data corresponding to 1400nm and 1900nm in the near-infrared band are added together and averaged to obtain the scattering reference value. The scattering reference value is multiplied by a preset scaling factor to obtain the scattering correction value. The scattering correction value is subtracted one by one from the atmospheric correction data corresponding to each band in the visible light band from 400nm to 700nm to obtain the corrected spectral reflectance data.

4. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 1, characterized in that, The time series completion of synchronous hydrological data using a cubic spline interpolation algorithm is specifically as follows: The water level data, flow velocity data, and salinity data in the synchronous hydrological data were arranged in the order of collection time to construct water level time series, flow velocity time series, and salinity time series respectively. In the water level time series, flow velocity time series and salinity time series, missing time points are identified. For any missing time point, two adjacent sampled time points before and after the missing time point are selected and recorded as the previous time point and the next time point, respectively. The corresponding synchronous hydrological data values ​​are recorded as the previous value and the next value, respectively. A piecewise cubic function is constructed between the previous time point and the next time point. The piecewise cubic function is determined by four coefficients. The calculation method of the piecewise cubic function is as follows: subtract the previous time point from the current time point to obtain the time difference value, multiply the cube of the time difference value by the first coefficient, add the square of the time difference value by the second coefficient, add the time difference value by the third coefficient, and add the fourth coefficient. The four coefficients are solved by setting four constraints: the function value of the piecewise cubic function at the previous time point is equal to the previous value, the function value of the piecewise cubic function at the next time point is equal to the next value, the first derivative of the piecewise cubic function at the previous time point is equal to the slope of the previous time interval, and the first derivative of the piecewise cubic function at the next time point is equal to the slope of the next time interval. The slope is obtained by dividing the difference between the values ​​of two adjacent sampled time points by the time difference. The second derivative is set to 0 at the start time point and 0 at the end time point of the time series. The coefficients of each piecewise cubic function are obtained by solving the constraints of each piecewise cubic function simultaneously. Substitute the time value corresponding to the missing time point into the piecewise cubic function to calculate the complete value, fill the complete value into the corresponding missing time point position, and obtain the completed synchronous hydrological data.

5. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 1, characterized in that, The method for filtering eutrophication-sensitive bands based on the preprocessed dataset using a continuous projection algorithm is as follows: The preprocessed dataset is centrally corrected spectral reflectance data is expanded according to bands and matched with water level data, flow velocity data and salinity data in the completed synchronous hydrological data according to the same timestamp to form a data sample arranged in chronological order. The continuous projection algorithm is applied to the corrected spectral reflectance data in the data sample. The band with the smallest wavelength is selected as the initial band and added to the selected bands. The remaining bands are selected as candidate bands. For each candidate band, extract the numerical sequence of the candidate band in all samples and the numerical sequence of the selected band in all samples. Multiply each value in the numerical sequence of the candidate band with the corresponding value in the numerical sequence of the selected band point by point, and accumulate all the product results. Divide the accumulated result by the square of each value in the numerical sequence of the selected band and sum the results to obtain the projection coefficient. Multiply each value in the selected band numerical sequence by the projection coefficient to obtain the fitted numerical sequence of the candidate band on the selected band; Subtract the corresponding value in the fitted numerical sequence from each value in the candidate band numerical sequence to obtain the residual sequence. Square each value in the residual sequence and sum them up. Take the square root of the summation result to obtain the projected distance corresponding to the candidate band. The projected distances corresponding to all candidate bands are compared. The candidate band with the largest projected distance is selected as the new band and added to the selected bands. At the same time, the new band is deleted from the candidate bands. After updating the selected bands, repeat the process of adding new bands for the remaining candidate bands until the number of selected bands reaches the preset threshold. Eutrophication-sensitive bands were identified from the selected bands, including the 685nm, 705nm, 560nm, 420nm, 620nm, 1400nm, and 1900nm bands. The corrected spectral reflectance data corresponding to the 685nm and 705nm bands were identified as chlorophyll a-sensitive band data, the corrected spectral reflectance data corresponding to the 560nm band was identified as total phosphorus-sensitive band data, the corrected spectral reflectance data corresponding to the 420nm and 620nm bands was identified as total nitrogen-sensitive band data, and the corrected spectral reflectance data corresponding to the 1400nm and 1900nm bands was identified as salinity-sensitive band data.

6. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 1, characterized in that, The process involves constructing a training dataset and using a random forest algorithm to introduce a tidal coupling factor to build a tidal coupling inversion model. Chlorophyll a, total phosphorus, and total nitrogen are then inverted and calculated. The eutrophication level is interpreted based on the inversion results, generating an inversion result set. Specifically: The corrected spectral reflectance data corresponding to the 685nm, 705nm, 560nm, 420nm, 620nm, 1400nm and 1900nm bands are matched with the water level data, flow velocity data and salinity data in the completed synchronous hydrological data in chronological order to construct a training dataset. The training dataset is divided into a training set and a validation set, and multiple decision trees are constructed. The water level fluctuation is obtained by subtracting the value of the previous time point from the value of the water level data at the current time point, the flow velocity gradient is obtained by subtracting the value of the previous time point from the value of the flow velocity data at the current time point, and the salinity value at the current time point is used as the salinity value. The water level fluctuation, flow velocity gradient, and salinity value are used as tidal coupling factors. A tidal coupling inversion model is constructed based on the corrected spectral reflectance data, the coupled spectral data, and the tidal coupling factor. In the tidal coupling inversion model, during the node partitioning process of each decision tree, partitioning variables are selected from the corrected spectral reflectance data, coupled spectral data, water level fluctuation, flow velocity gradient and salinity value, and a partitioning threshold is set. Samples with values ​​less than the partitioning threshold are partitioned to one side of the sub-nodes, and samples with values ​​greater than or equal to the partitioning threshold are partitioned to the other side of the sub-nodes. The input data is calculated using a tidal coupling inversion model. The chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration output by all decision trees are summed and divided by the number of decision trees to obtain the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration. Based on the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration, the eutrophication level is obtained and combined with the timestamp to generate an inversion result set.

7. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 1, characterized in that, The aforementioned early warning linkage module is specifically: Time-series data is constructed based on the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration arranged in chronological order in the inversion result set, as well as the eutrophication level. The time-series data is then input into the input compression layer of the improved ModernTCN network. In the input compression layer, a one-dimensional convolution operation is performed on the time series data, and the convolution output is compressed through a linear transformation to obtain compressed time series features; The compressed temporal features are input into the dilated convolutional layer, which contains multiple convolutional blocks. Each convolutional block uses a dilation rate sequence that combines exponential progression and periodic folding to perform one-dimensional convolution operations. The dilation rate sequence increases in power of 2 to a preset upper limit and then decreases in reverse order according to the same sequence to form a closed sequence. The compressed temporal features are input into the first convolutional block to obtain the first layer output features; The first layer output feature is added element-wise to the compressed temporal feature and used as the input to the second layer convolutional block to obtain the second layer output feature. For each subsequent convolutional block, the previous layer output feature is added element-wise to the compressed temporal feature and used as the input to the current layer convolutional block to obtain the corresponding layer output feature. The output features of each convolutional block are concatenated to obtain the multi-scale temporal feature. The multi-scale temporal features are input into the temporal structure adjustment layer, and the multi-scale temporal features are processed through the phase equipotential unfolding mechanism to obtain the unfolded correction features. The folding correction features are input into the prediction output layer, and through one-dimensional convolution and fully connected operations, the eutrophication trend prediction results for future time steps are output, and an early warning is given based on the eutrophication trend prediction results.

8. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 7, characterized in that, The process of processing the multi-scale time series features through a phase equipotential unfolding mechanism to obtain unfolded and corrected features is as follows: The multi-scale temporal features are arranged in chronological order, and the multi-scale temporal features of the current time step and the multi-scale temporal features of adjacent time steps are extracted respectively. Subtracting the multi-scale temporal features of the previous time step from the multi-scale temporal features of the current time step yields the first difference sequence; subtracting the multi-scale temporal features of the current time step from the multi-scale temporal features of the next time step yields the second difference sequence. Add the corresponding elements of the first difference sequence and the second difference sequence, and then divide by 2 to obtain the symmetric difference sequence; The absolute value of each element in the symmetric difference sequence is calculated, and the sign consistency is judged with the first difference sequence at the corresponding position. When the corresponding element of the first difference sequence is positive, a positive value is taken, and when the corresponding element of the first difference sequence is negative, a negative value is taken, thus obtaining a phase-consistent sequence. At the current time step, extract the phase-consistent sequences corresponding to the previous time step, the current time step, and the next time step. Add the phase-consistent sequences corresponding to the previous time step, the current time step, and the next time step element by element to obtain the folded feature sequence corresponding to the current time step. Arrange the folded feature sequences obtained at each time step sequentially to obtain the unfolded feature sequence; The expanded feature sequence is added element-wise to the multi-scale temporal features to obtain the expanded correction features.

9. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 1, characterized in that, The multi-mode communication technologies specifically include 4G / 5G, optical fiber, and BeiDou satellite.

10. The eutrophication inversion interpretation and early warning system for tidal river sections based on high-precision spectroscopy as described in claim 1, characterized in that, The data storage module is specifically: The spectral reflectance data, synchronous hydrological data, and the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level in the inversion result set are matched with timestamps and written to different storage nodes in the distributed database. In the distributed database, spectral reflectance data, synchronous hydrological data, and inversion result sets are stored in segments according to time order. Time-stamp-based indexes are established for the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level in the spectral reflectance data, synchronous hydrological data, and inversion result sets, respectively. When querying historical data, based on the input time range, the system reads the spectral reflectance data, synchronous hydrological data, and predicted values ​​of chlorophyll a concentration, total phosphorus concentration, total nitrogen concentration, and eutrophication level within the corresponding time period through the timestamp index. In the statistical analysis, the predicted values ​​of chlorophyll a concentration, total phosphorus concentration, and total nitrogen concentration were accumulated in chronological order and the average value was calculated. The eutrophication level was also statistically analyzed to obtain the statistical results of historical data.

Images