Inshore water quality day-by-day kilometer-level inversion method, system, equipment and medium

Abstract

The invention relates to an inshore water quality day-by-day kilometer-level inversion method, system, device and medium, and belongs to the technical field of water quality monitoring, the method comprises the following steps: obtaining remote sensing image data, meteorological reanalysis data and ground water quality site measured data, preprocessing, carrying out space-time matching fusion, using the fusion data as a sample to carry out model training, and obtaining a model training model; obtaining a space model; a spatial model offshore water quality spatial distribution diagram; and constructing a CNN-BiLSTM mixed time sequence model, and training the CNN-BiLSTM mixed time sequence model by taking historical water quality data of a ground water quality station and the offshore water quality spatial distribution map as samples to obtain a prediction model. The method is suitable for red tide early warning, pollution tracing and marine ecological management, breaks through monthly monitoring limitation, and realizes day-by-day kilometer-level water quality dynamic tracking; remote sensing large-range coverage and site high-precision data are fused, and the problem of inshore environment heterogeneity is solved; the time-space double-model cascade architecture significantly improves the generalization ability under complex hydrological conditions.

Description

TECHNICAL FIELD

[0001] The present disclosure belongs to the technical field of water quality monitoring, and particularly relates to a near-shore water quality daily kilometer-level inversion method, system, device and medium. BACKGROUND

[0002] With the intensification of human activities, eutrophication, frequent red tides and sudden pollution events in near-shore waters have posed a serious threat to ecological safety and coastal economy. Traditional water quality monitoring relies on manual sampling and laboratory analysis, which is difficult to achieve large-scale high-frequency monitoring due to limitations of manpower, weather and cost. Indirect monitoring methods based on Kriging interpolation or semi-empirical models have poor spatial representativeness and low timeliness due to insufficient monitoring point density or strong regional empirical dependence, which cannot meet the requirements of dynamic management at the daily kilometer level.

[0003] In recent years, remote sensing and machine learning technologies have provided new ideas for water quality inversion. On the one hand, satellite data such as MODIS have the advantages of daily revisit and kilometer-level spatial resolution, which can capture the spatiotemporal heterogeneity of water color. On the other hand, models such as random forest and BiLSTM (Long Short-Term Memory Algorithm) can explore the nonlinear relationship between spectrum and water quality and the temporal evolution law. However, existing researches mostly focus on a single scale (such as monthly-scale inversion or station-level prediction), and do not effectively couple remote sensing spatial wide-area data with ground station high-precision data, resulting in insufficient model generalization ability and prediction timeliness in complex estuaries and near-shore areas with significant tidal influence.

[0004] In addition, near-shore water quality is driven by the coupling of meteorological factors (wind speed, radiation, etc.) and hydrological processes (runoff, tides), showing complex dynamic characteristics of superimposed "high-frequency mutation and seasonal trend". Traditional time series models (such as SARIMA) are difficult to capture multi-factor non-stationary disturbances, and single CNN or BiLSTM cannot consider spatial heterogeneity and temporal dependence. Therefore, it is necessary to build a cascade framework for multi-source data fusion of "remote sensing-weather-ground" in order to overcome these difficulties. It is necessary to provide a new near-shore water quality daily kilometer-level inversion method, system, device and medium with the properties of low cost, high precision, high efficiency and high coverage, to solve the above technical problems. SUMMARY

[0005] The purpose of the present disclosure is to provide a near-shore water quality daily kilometer-level inversion method, system, device and medium to solve the above problems.

[0006] The present disclosure achieves the above-mentioned purpose by the following technical solutions: A near-shore water quality daily kilometer-level inversion method, comprising the following steps: acquiring remote sensing image data, meteorological reanalysis data corresponding to the remote sensing image data at the corresponding time, and ground water quality station measured data, and performing preprocessing; spatially and temporally matching and fusing the preprocessed remote sensing image data, the meteorological reanalysis data and the ground water quality site measured data to obtain fused data; training a random forest regression model by taking the fused data as samples to obtain a spatial model; taking the remote sensing image data and the meteorological reanalysis data as inputs to perform per-pixel prediction on a target day, and outputting a per-day and per-kilometer precision near-shore water quality spatial distribution map; constructing a CNN-BiLSTM hybrid time series model, taking historical water quality data of the ground water quality site and the near-shore water quality spatial distribution map as samples to train the model, and obtaining a prediction model; inputting a to-be-predicted time point into the prediction model to obtain near-shore water quality prediction data.

[0007] As a further optimization scheme of the present disclosure, the remote sensing image data is MODIS satellite visible light to near-infrared band remote sensing image data; the meteorological reanalysis data includes 2-meter air temperature, surface air pressure, sea level air pressure and 10-meter wind speed; and the ground water quality site measured data includes longitude and latitude, sampling time, sampling station name and key indicators, and the key indicators include chlorophyll a, dissolved oxygen, turbidity, total phosphorus, ammonia nitrogen and total nitrogen.

[0008] As a further optimization scheme of the present disclosure, the preprocessed remote sensing image data, meteorological reanalysis data and ground water quality site measured data are spatially and temporally matched and fused to obtain fused data, including: According to the longitude and latitude of the ground water quality site, the center point of the corresponding MODIS pixel is selected, and the preprocessed remote sensing image data, meteorological reanalysis data and ground water quality site measured data are spatially matched; From the MODIS daily transit time, an image closest to the sampling time of the ground water quality site is selected as a paired input, and the preprocessed remote sensing image data, meteorological reanalysis data and ground water quality site measured data are temporally matched; forming a remote sensing-water quality input-output sample pair, i.e., obtaining the fused data.

[0009] As a further optimization scheme of the present disclosure, the random forest regression model optimization adopts grid search to adjust hyperparameters, including the number of trees, the maximum tree depth, the minimum sample leaf node number and the number of split features.

[0010] As a further optimization scheme of the present disclosure, the following strategies are adopted for training the random forest regression model: multiple training subsets are generated by sampling with replacement; randomly selecting part of the features when each tree splits; The prediction results of the integrated multi-tree are averaged; Automatically handle multicollinearity and missing values.

[0011] As a further optimization of the present disclosure, the structure of the CNN-BiLSTM hybrid time series model comprises: A convolutional layer CNN is used to slide and extract the change features between adjacent time points, enhancing the response capability to the hourly fluctuations; A pooling layer is used to compress and denoise the extracted local features by using average pooling; A BiLSTM network is used to capture the forward and backward time dependencies in the sequence respectively by using a bidirectional structure; A Dropout layer is used to prevent overfitting and improve the generalization ability of the model in the variable conditions of the nearshore environment; A fully connected layer is responsible for integrating the high-dimensional time series features extracted by CNN and BiLSTM, and converting them into the input format of the prediction task; An output layer is used to generate the water quality prediction value sequence of future time points.

[0012] As a further optimization of the present disclosure, the CNN-BiLSTM hybrid time series model training uses a sliding window to generate samples, uses an Adam optimizer, and constructs 12 independent models per month to adapt to seasonal changes.

[0013] A nearshore water quality daily kilometer-level inversion system comprises: A data acquisition module is used to acquire remote sensing image data, meteorological reanalysis data corresponding to the remote sensing image data at the corresponding time, and ground water quality station measured data, and to preprocess them; A data fusion module is used to perform spatio-temporal matching and fusion of the preprocessed remote sensing image data, meteorological reanalysis data, and ground water quality station measured data to obtain fused data; A model training module is used to train a random forest regression model using the fused data as samples to obtain a spatial model; An inversion stage output module is used to input the remote sensing image data and meteorological reanalysis data to perform per-pixel prediction for a target day, and output a nearshore water quality spatial distribution map with daily and kilometer precision; A prediction model generation module is used to construct a CNN-BiLSTM hybrid time series model, train it using historical water quality data of ground water quality stations and the nearshore water quality spatial distribution map as samples, and obtain a prediction model; A water quality prediction module is used to input a to-be-predicted time point into the prediction model to obtain nearshore water quality prediction data.

[0014] An electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to execute the program stored in the memory to implement the daily kilometer-level inversion method for near-shore water quality.

[0015] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for daily kilometer-level inversion of nearshore water quality.

[0016] The beneficial effects of this disclosure are as follows: This disclosure is applicable to red tide early warning, pollution source tracing, and marine ecological management. It breaks through the limitations of monthly monitoring and achieves daily kilometer-level dynamic tracking of water quality. It integrates remote sensing with large-scale coverage and high-precision station data to solve the problem of near-shore environmental heterogeneity. The spatiotemporal dual-model cascade architecture significantly improves the generalization ability under complex hydrological conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method in an embodiment of this disclosure; Figure 2 This is a system structure block diagram in an embodiment of this disclosure; Figure 3 This is a block diagram of the device structure in an embodiment of this disclosure. Detailed Implementation

[0018] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0019] like Figure 1 As shown, a daily kilometer-level inversion method for nearshore water quality includes the following steps: Acquire remote sensing image data, meteorological reanalysis data corresponding to the time of the remote sensing image data, and measured data from ground water quality stations, and perform preprocessing. The preprocessed remote sensing image data, the meteorological reanalysis data, and the measured data from the ground water quality station are spatiotemporally matched and fused to obtain fused data. The fused data is used as samples to train the random forest regression model, resulting in a spatial model. Using the remote sensing image data and the meteorological reanalysis data as input, pixel-by-pixel prediction is performed for the target day, and the output is a spatial distribution map of near-shore water quality with daily and kilometer-by-kilometer precision. A CNN-BiLSTM hybrid time series model is constructed, and historical water quality data of ground water quality stations and the near-shore water quality spatial distribution map are used as samples to train the model, so as to obtain a prediction model; The near-shore water quality prediction data is obtained by inputting the to-be-predicted time point into the prediction model.

[0020] In the embodiment, the following are specifically included: The spatial inversion stage: the nonlinear relationship between the water quality parameters (such as total nitrogen, total phosphorus, and salt content) and the spectral reflectivity is established by using MODIS satellite data (containing visible light to near-infrared bands) through a random forest regression model to generate a daily kilometer-level water quality spatial distribution map. Model optimization uses grid search to adjust key parameters (such as the number of trees and the number of split features), and cloud removal, band synthesis, and outlier removal preprocessing are performed in view of the complexity of the near-shore water body.

[0021] The main content of the spatial inversion stage is as follows: MODIS remote sensing image data is selected, which has the advantages of daily revisit and 1-kilometer spatial resolution, can capture the water color changes of the near-shore sea area at a high frequency, and is suitable for constructing a daily and kilometer-level water quality parameter inversion model, especially for the near-shore environment with strong spatiotemporal dynamics.

[0022] The 8th to 16th bands of the MODIS remote sensing image data are used as input features, which cover multiple frequency bands from visible light to near-infrared, and their main physical characteristics are as follows: (1) Band 8 (0.41-0.45 μm, blue light): sensitive to chlorophyll absorption, can reflect the change of algal concentration; (2) Band 9 (0.45-0.49 μm): representative band of organic pollutants and transparency changes in water body; (3) Band 10 (0.49-0.53 μm, green light): can reflect the content of total suspended solids, dissolved oxygen, etc.; (4) Band 11 (0.53-0.57 μm): can identify phytoplankton abundance; (5) Band 12 (0.63-0.67 μm, red light): sensitive to changes in turbidity, total phosphorus, total nitrogen, etc.; (6) Band 13 (0.85-0.88 μm): near-infrared band, reflecting the change of water profile, which is helpful for identifying the shoreline area; (7) Bands 14-16: mainly used for atmospheric correction, which is helpful for improving the stability of the band when combined modeling.

[0023] The MODIS remote sensing image data is preprocessed, including cloud mask removal (QA band judgment), image boundary cropping, projection conversion (WGS84 unified coordinate system), to ensure data cleanliness, consistency and matchability, laying a foundation for subsequent modeling.

[0024] Meteorological reanalysis data (ERA5) corresponding to the time of MODIS remote sensing image data is obtained, including 2-meter air temperature, surface air pressure, sea level air pressure, 10-meter wind speed (U component and V component), etc., which are used to reflect the environmental background of the coastal area, such as climate, air pressure and hydrodynamic force, to enhance the physical meaning of the model. Meteorological data parameters are used to represent the actual nearshore environment.

[0025] Ground water quality station (buoy) measured data is obtained, including latitude and longitude, sampling time, sampling station name and key indicators: chlorophyll a, dissolved oxygen (DO), turbidity, total phosphorus, ammonia nitrogen, total nitrogen, etc. This data is the label variable in supervised learning, supporting model training.

[0026] MODIS remote sensing image data, meteorological reanalysis data and ground water quality station (buoy) measured data are spatio-temporally matched: Spatial matching: according to the latitude and longitude of the station, the corresponding MODIS pixel center point is selected; Time matching: from the daily transit time of MODIS, the image closest to the sampling time of the station is selected as the paired input.

[0027] Forming remote sensing-water quality input-output sample pairs for supervised learning.

[0028] MODIS remote sensing image data and meteorological data are used as input features, and ground water quality station (buoy) measured data is used as labels to construct the input structure of the machine learning model. Remote sensing features can reflect the optical properties of water and changes in water color, and meteorological data provides the environmental disturbance background.

[0029] All samples are subjected to outlier removal before modeling: MODIS remote sensing image data is determined to be abnormal if the Z-score is greater than ±3σ; ground water quality station (buoy) measured data is identified and removed using the box plot method (IQR) to identify and remove null values and extreme values.

[0030] Random forest regression model (Random Forest Regressor) is used for modeling. Its core is to build multiple weak decision trees and output predicted values through ensemble averaging, which is suitable for high-dimensional nonlinear feature regression tasks.

[0031] The following strategies are used for random forest model training: (1) With replacement sampling to generate multiple training subsets; (2) Randomly select part of features when splitting each tree, to enhance diversity; (3) Average the prediction results of multiple trees to improve stability; (4) Automatically handle multicollinearity and missing values, suitable for modeling of remote sensing multi-source data.

[0032] Before model training, the training set is divided into training set 80%, validation set 10%, and test set 10%, which are used for model fitting, parameter tuning, and final accuracy evaluation.

[0033] Use grid search (GridSearchCV) to tune parameters and optimize key hyperparameters: (1) Number of trees (n_estimators): [100, 200, 300]; (2) Maximum tree depth (max_depth): [10, 20, None]; (3) Minimum sample leaf node size (min_samples_leaf): [1, 2, 4]; (4) Number of split features (max_features): ['sqrt', 'log2'].

[0034] After model training, use R 2 , RMSE, MSE, and other accuracy indicators to evaluate the test set, with the following mathematical definitions: Mean Squared Error (MSE): ; The average of the square of the difference between the predicted value and the true value, the smaller the better.

[0035] Root Mean Squared Error (RMSE): ; The square root of MSE, the unit is consistent with the original data.

[0036] Coefficient of Determination (R 2 ): ; Where, is the number of samples, i.e., the number of observation data or samples used to calculate the error; is the actual value (true value) of the i-th sample; is the predicted value of the i-th sample. Measures the model's ability to explain variance, the closer to 1 indicates better fitting.

[0037] After training, the spatial model is obtained, and the complete remote sensing and meteorological input layers are used to predict the target day by pixel, and the output is the daily and kilometer-precision spatial distribution map of near-shore water quality.

[0038] The output results support saving in GeoTIFF raster format, adapting to the integration of marine environmental monitoring and GIS platforms, and supporting subsequent analysis requirements such as visualization, time series tracking, and hotspot identification.

[0039] The model training covers typical offshore areas (such as the Pearl River Estuary), and the verification generalization performance is excellent.

[0040] The spatial inversion stage breaks through the traditional monthly mapping limitations and supports daily water quality dynamic tracking.

[0041] The final output results are combined with the time series module and can be used as input for the CNN-BiLSTM model to build a complete "spatio-temporal double prediction system".

[0042] Time series optimization stage: 1. Integrate hourly water quality observation data from automated stations to build a CNN-BiLSTM hybrid time series model: 2. Convolutional layer (CNN) extracts local spatio-temporal features; 3. Bidirectional LSTM layer captures long-term dependencies; 4. Input the spatial inversion results to output daily water quality dynamic prediction for the next 24 hours.

[0043] 5. Collaborative verification: evaluate accuracy through root mean square error (RMSE) and coefficient of determination (R 2 ) in offshore areas such as the Pearl River Estuary, with total nitrogen / total phosphorus inversion R 2 above 0.85 and prediction error within 48 hours ≤15%.

[0044] The main technical content of the time series optimization stage is as follows: Use hourly water quality parameters (such as total nitrogen, total phosphorus, and salinity) from automated water quality observation stations (sea surface buoys) to build a high-time-resolution time series sample set as the deep learning model training data source.

[0045] For the characteristics of nearshore water quality being strongly influenced by tides, runoff, human activities, wind waves, and changing rapidly with short-term and sudden characteristics, build an adaptive CNN-BiLSTM time series prediction model to enhance the modeling ability of nearshore dynamic water quality processes.

[0046] The CNN substructure is used to extract local mutation features of water quality time series, such as pollutants brought by rainwater runoff, sudden pollution events, and short-term high-frequency fluctuations caused by high tide and low tide. The model scans the water quality time series through a sliding small window (one-dimensional convolution kernel) to mine local patterns such as change intensity and fluctuation frequency at the hourly scale, enhancing the response ability to short-term water quality disturbances.

[0047] BiLSTM substructure is used to capture the long-term dynamic evolution characteristics of nearshore water quality, such as seasonal climate, long-term hydrological driving, background nutrient load changes and other factors affecting the trend changes. Through the bidirectional LSTM unit, the model analyzes the information in the past and future directions at the same time, forming a deep memory of the temporal context, thus accurately modeling complex temporal dependencies and providing structural support for future prediction.

[0048] The model input not only includes the historical water quality sequence of the automated site, but also integrates the current daily 1 km nearshore water quality spatial distribution map obtained by spatial inversion of the random forest regression model, which serves as the model's current state condition input, providing an accurate initial basis for future dynamic prediction.

[0049] The convolutional layer (CNN) is set to use 64 one-dimensional filters with a kernel size of 2 to extract the change characteristics between adjacent time points and enhance the response capability to hourly fluctuations.

[0050] The pooling layer uses average pooling to compress and denoise the extracted local features, reducing information redundancy while maintaining the change trend.

[0051] The BiLSTM network adopts a bidirectional structure to capture the forward and backward time dependencies in the sequence, enhancing the prediction accuracy of future trends.

[0052] The Dropout layer is set after the fully connected layer to prevent overfitting and improve the model's generalization ability in the variable conditions of the nearshore environment.

[0053] The fully connected layer is responsible for integrating the high-dimensional time series features extracted by CNN and BiLSTM and converting them into the input format for the prediction task.

[0054] The output layer is set as a single neuron structure to generate the water quality prediction value sequence for each time point in the future, with units in hours.

[0055] The model predicts the change trend of nearshore water quality parameters (such as total nitrogen, total phosphorus, etc.) in the next 24 to 48 hours.

[0056] During the model training process, the Adam optimizer is used with a learning rate of 1e-4 and a training round of 300 rounds.

[0057] The sample input generates time series samples using a sliding window method with a window length of 8 hours and a step size of 1 hour.

[0058] The data division ratio is: the training set accounts for 80%, the validation set accounts for 10%, and the test set accounts for 10%, so as to ensure the stability of model training and the representativeness of evaluation.

[0059] The model performance evaluation adopts R 2 , mean square error (MSE) and root mean square error (RMSE) and other indicators. The test in the typical nearshore area of the Pearl River Estuary shows that the prediction error is ≤15% within 48 hours, and the prediction accuracy reaches the practical level.

[0060] An independent CNN-BiLSTM time series model is trained for each water quality parameter, and 12 groups of models are constructed monthly to adapt to the change mode of nearshore water quality in different seasons.

[0061] The final output of the model is the nearshore water quality prediction data sequence for the next 48 hours, which can be used for red tide early warning, dynamic monitoring, pollution diffusion modeling and other actual marine environmental management scenarios.

[0062] As shown in Figure 2 , the embodiment of the present disclosure provides a nearshore water quality daily kilometer-level inversion system, comprising: A data acquisition module is configured to acquire remote sensing image data, meteorological reanalysis data corresponding to the remote sensing image data at a moment, and ground water quality station measured data, and perform preprocessing; A data fusion module is configured to perform spatio-temporal matching and fusion of the preprocessed remote sensing image data, meteorological reanalysis data and ground water quality station measured data to obtain fused data; A model training module is configured to use the fused data as a sample to train a random forest regression model to obtain a spatial model; An inversion stage output module is configured to use the remote sensing image data and the meteorological reanalysis data as input to perform per-pixel prediction for a target day, and output a nearshore water quality spatial distribution map with daily and kilometer precision; A prediction model generation module is configured to construct a CNN-BiLSTM hybrid time series model, use historical water quality data of a ground water quality station and the nearshore water quality spatial distribution map as samples to train the model, and obtain a prediction model; A water quality prediction module is configured to input a to-be-predicted time point into the prediction model to obtain nearshore water quality prediction data.

[0063] The implementation process of the functions and roles of each module in the above system is specifically described in the implementation process of the corresponding steps in the above method, which will not be repeated here.

[0064] Referring to Figure 3The electronic device provided by the embodiment of the present disclosure comprises a processor 1110, a communication interface 1120, a memory 1130 and a communication bus 1140, wherein the processor 1110, the communication interface 1120 and the memory 1130 complete mutual communication through the communication bus 1140. The memory 1130 is used for storing a computer program. The processor 1110 is used for executing the program stored on the memory 1130, and realizes the offshore water quality daily kilometer-level inversion method. The communication bus 1140 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0065] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0066] The memory 1130 can comprise a Random Access Memory (RAM) and can also comprise a non-volatile memory, for example at least one disk memory. Optionally, the memory 1130 can also be at least one storage device located away from the aforementioned processor 1110.

[0067] The processor 1110 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; can also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0068] The embodiment of the present disclosure further provides a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the offshore water quality daily kilometer-level inversion method.

[0069] The computer readable storage medium can be included in the device / apparatus described in the above embodiments; or can exist separately and not be assembled into the device / apparatus. The above computer readable storage medium carries one or more programs, when the one or more programs are executed, the offshore water quality daily kilometer-level retrieval method according to the embodiments of the present disclosure is implemented.

[0070] According to the embodiments of the present disclosure, the computer readable storage medium can be a non-volatile computer readable storage medium, which can include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in conjunction with an instruction execution system, device or apparatus.

[0071] The above embodiments only express several implementation manners of the present disclosure, which are described in a more specific and detailed manner, but should not be understood as a limitation on the scope of the patent of the present disclosure. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present disclosure, which are all within the protection scope of the present disclosure.

Claims

1. A method for daily kilometer-level inversion of nearshore water quality, characterized in that, Includes the following steps: Acquire remote sensing image data, meteorological reanalysis data corresponding to the time of the remote sensing image data, and measured data from ground water quality stations, and perform preprocessing. The preprocessed remote sensing image data, the meteorological reanalysis data, and the measured data from the ground water quality station are spatiotemporally matched and fused to obtain fused data. The fused data is used as samples to train the random forest regression model, resulting in a spatial model. Using the remote sensing image data and the meteorological reanalysis data as input, pixel-by-pixel prediction is performed for the target day, and the output is a spatial distribution map of near-shore water quality with daily and kilometer-by-kilometer precision. A CNN-BiLSTM hybrid time series model was constructed, and historical water quality data from ground water quality stations and the spatial distribution map of near-shore water quality were used as samples to train the model to obtain a prediction model. Input the predicted time point into the prediction model to obtain nearshore water quality prediction data.

2. The method for daily kilometer-level inversion of nearshore water quality according to claim 1, characterized in that, The remote sensing image data is visible to near-infrared remote sensing image data from the MODIS satellite; the meteorological reanalysis data includes 2-meter air temperature, surface air pressure, sea level air pressure, and 10-meter wind speed; the measured data from the ground water quality stations include latitude and longitude, sampling time, sampling station name, and key indicators, including chlorophyll a, dissolved oxygen, turbidity, total phosphorus, ammonia nitrogen, and total nitrogen.

3. The method for daily kilometer-level inversion of nearshore water quality according to claim 1, characterized in that, The preprocessed remote sensing image data, the meteorological reanalysis data, and the measured data from the ground water quality stations are spatiotemporally matched and fused to obtain fused data, including: Based on the latitude and longitude of the ground water quality stations, the corresponding MODIS pixel center point is selected, and the preprocessed remote sensing image data, the meteorological reanalysis data, and the measured data of the ground water quality stations are spatially matched. The image closest to the sampling time of the ground water quality station is selected from the daily MODIS transit time as the pairing input, and the preprocessed remote sensing image data, the meteorological reanalysis data and the measured data of the ground water quality station are time-matched. By forming input-output sample pairs of remote sensing and water quality data, fused data is obtained.

4. The method for daily kilometer-level inversion of nearshore water quality according to claim 1, characterized in that, The random forest regression model is optimized using grid search to adjust hyperparameters, including the number of trees, maximum tree depth, minimum number of sample leaf nodes, and number of split features.

5. The method for daily kilometer-level inversion of nearshore water quality according to claim 1, characterized in that, The random forest regression model is trained using the following strategy: Multiple training subsets are generated by sampling with replacement. Each tree randomly selects some features when splitting; The prediction results from multiple trees are averaged. Automatically handles multicollinearity and missing values.

6. The method for daily kilometer-level inversion of nearshore water quality according to claim 1, characterized in that, The structure of the CNN-BiLSTM hybrid temporal model includes: Convolutional layers (CNNs) are used to slide and extract changing features between adjacent time points, enhancing the ability to respond to hourly fluctuations. The pooling layer uses average pooling to compress and reduce noise in the extracted local features; The BiLSTM network employs a bidirectional structure to capture the forward and backward time dependencies in the sequence, respectively. Dropout layers prevent overfitting and improve the model's generalization ability under varying near-shore environmental conditions. The fully connected layer is responsible for integrating the high-dimensional temporal features extracted by CNN and BiLSTM and converting them into the input format for the prediction task; The output layer is used to generate a sequence of predicted water quality values ​​for future time points.

7. The method for daily kilometer-level inversion of nearshore water quality according to claim 1, characterized in that, The CNN-BiLSTM hybrid temporal model is trained using a sliding window to generate samples and the Adam optimizer is used to build 12 independent models per month to adapt to seasonal changes.

8. A daily kilometer-level inversion system for nearshore water quality, characterized in that, include: The data acquisition module is used to acquire remote sensing image data, meteorological reanalysis data corresponding to the time of the remote sensing image data, and measured data from ground water quality stations, and to perform preprocessing. The data fusion module is used to perform spatiotemporal matching and fusion of the preprocessed remote sensing image data, the meteorological reanalysis data, and the measured data from the ground water quality station to obtain fused data; The model training module is used to train the random forest regression model using the fused data as samples to obtain a spatial model; The inversion stage output module is used to take the remote sensing image data and the meteorological reanalysis data as input, perform pixel-by-pixel prediction for the target day, and output a near-shore water quality spatial distribution map with daily and kilometer-by-kilometer accuracy. The prediction model generation module is used to construct a CNN-BiLSTM hybrid time series model. It uses historical water quality data from ground water quality stations and the near-shore water quality spatial distribution map as samples to train the model and obtain the prediction model. The water quality prediction module is used to input the time point to be predicted into the prediction model to obtain nearshore water quality prediction data.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor for executing a program stored in a memory to implement the daily kilometer-level inversion method for nearshore water quality as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the daily kilometer-level inversion method for nearshore water quality as described in any one of claims 1-7.

Images