A method for monitoring suspended sediment in the Yellow River estuary and its adjacent sea area based on multi-source remote sensing data

Abstract

This invention relates to the field of water conservancy engineering technology, specifically to a method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data. The method includes the following steps: selecting complementary multi-source remote sensing data (Landsat 8 OLI, Sentinel-2A / B, GF-1 WFV, GOCI), and performing radiometric calibration, atmospheric correction, and geometric fine correction standardization preprocessing; using an algorithm combining weighted average and principal component analysis to fuse the multi-source data and extract sediment-sensitive characteristic information; constructing an improved semi-empirical radiative transfer model incorporating tidal current, wind speed, and hydrodynamic correction factors to invert suspended sediment concentration, and ensuring the reliability of the results through multi-dimensional verification. This invention significantly improves monitoring accuracy and spatiotemporal coverage, and is adaptable to routine, emergency, and long-term dynamic monitoring of the Yellow River Estuary, providing scientific support for estuary ecological protection and management decisions.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering technology, specifically to a method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data. Background Technology

[0002] Suspended sediment is one of the core parameters of the estuarine and coastal aquatic environment. Its distribution and dynamic changes directly affect estuarine geomorphological evolution, navigation safety, aquatic ecosystem health, and coastal engineering construction. As a typical high-sediment-carrying estuary in my country, the Yellow River Estuary experiences large concentration gradients and dramatic spatiotemporal changes in suspended sediment in its adjacent sea areas, placing extremely high demands on the accuracy, spatiotemporal coverage, and environmental adaptability of monitoring methods. Currently, traditional on-site sampling and monitoring methods suffer from limited coverage, poor timeliness, and high manpower and material costs, making it difficult to meet the needs of large-scale, high-frequency, and high-precision dynamic monitoring of suspended sediment in the Yellow River Estuary.

[0003] Among existing related technologies, CN119600469A discloses a method and system for monitoring suspended sediment based on remote sensing technology, which is a general monitoring solution in this field. The core content of this prior art is as follows: its monitoring system mainly consists of a concentration inversion module, a model building module, a data processing module, a remote sensing data acquisition module, an alarm module, and a control platform. The remote sensing data acquisition module integrates satellite remote sensing data receiving units and UAV remote sensing data acquisition units, enabling simultaneous acquisition of multi-source remote sensing data. The data processing module can perform radiometric calibration, atmospheric correction, geometric correction, and feature extraction operations on remote sensing images. The model building module can construct statistical or machine learning models corresponding to suspended sediment concentrations based on extracted reflectance and scattering characteristics. Finally, the concentration inversion module completes sediment concentration inversion, generates a concentration distribution map, and outputs the monitoring results. This prior art achieves large-scale monitoring of suspended sediment through remote sensing technology, to a certain extent compensating for the shortcomings of traditional on-site sampling methods.

[0004] However, the aforementioned comparative document still has significant technical deficiencies and cannot meet the monitoring needs of suspended sediment in the Yellow River Estuary and its adjacent sea areas. Specifically, these deficiencies are reflected in the following aspects: First, the comparative document only generally mentions the collection of satellite and UAV remote sensing data, without selecting multi-source satellite data sources with complementary spatial, temporal, and spectral resolutions based on the spatiotemporal variation characteristics of suspended sediment in the Yellow River Estuary. It also fails to formulate spatiotemporal matching rules and unified resolution standards for multi-source data, making it unable to simultaneously consider the spatial detail accuracy and temporal dynamics of monitoring, and thus unable to capture the rapid intraday changes and seasonal transport patterns of sediment in the Yellow River Estuary. Second, the comparative document only adopts a generalized remote sensing data preprocessing workflow, without optimizing and correcting schemes for the optical characteristics of the nearshore turbid waters of the Yellow River Estuary and the high humidity and high aerosol atmospheric environment of the estuary. This fails to effectively eliminate the interference of environmental factors on the true spectral signal of the water body, resulting in insufficient accuracy of the basic data for subsequent inversion. Third, the comparative document does not involve the design of fusion algorithms for multi-source remote sensing data, failing to effectively integrate the advantages of data from different sensors. The first point is that the method has several limitations. First, it fails to address the core issues of low inversion accuracy in high-concentration sediment areas and insufficient signal response in low-concentration areas of the Yellow River Estuary from a single data source. Second, it does not perform dimensionality reduction and noise reduction processing for suspended sediment characteristics, resulting in weak anti-interference and specificity of the extracted feature information. Third, the comparative document only generalizes to adopt statistical or machine learning models, without constructing a targeted semi-empirical radiative transfer inversion model for the optical characteristics of suspended sediment in the Yellow River Estuary. It also fails to incorporate key hydrodynamic environmental factors such as tidal currents and wind speeds in the Yellow River Estuary for model correction. The model has poor universality under the complex hydrodynamic conditions of the Yellow River Estuary and cannot meet the full-gradient concentration inversion requirements from high-concentration areas at the estuary to low-concentration areas offshore. Fourth, the verification process of the comparative document only sets up a general model performance evaluation process, without developing a multi-dimensional verification scheme for the Yellow River Estuary region. It lacks verification with ground-based measured data covering different seasons, tidal phases, and concentration gradients, and it does not set up a cross-validation process. As a result, it cannot guarantee the stability and reliability of the method in long-term, large-scale monitoring of the Yellow River Estuary.

[0005] Therefore, developing a high-precision suspended sediment monitoring method based on multi-source remote sensing data fusion for the Yellow River Estuary and its adjacent sea areas is of great practical significance for addressing the problems of insufficient data source targeting, lack of multi-source data fusion capability, poor adaptability of inversion models, and insufficient monitoring accuracy and stability in existing technologies. This method is crucial for ecological environmental protection of the Yellow River Estuary, coastal engineering construction, and estuary management decisions. Summary of the Invention

[0006] The purpose of this invention is to provide a method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data, so as to solve the problems mentioned in the background art, such as insufficient monitoring accuracy, limited spatiotemporal coverage, weak multi-source data fusion capability, and poor adaptability of inversion models.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data includes the following steps: S1 Multi-source Remote Sensing Data Acquisition: For the monitoring area of ​​the Yellow River Estuary and its adjacent sea areas, remote sensing data from Landsat8OLI, Sentinel-2A / B, GF-1WFV and GOCI satellite sensors were selected as multi-source data sources to ensure that the spatiotemporal overlap of each data source meets the fusion requirements. The spatial resolution of all data was unified to 10m and the temporal resolution was controlled within 24h. S2 data preprocessing: The acquired multi-source remote sensing data are sequentially subjected to radiometric calibration, atmospheric correction and geometric fine correction to obtain a standardized remote sensing reflectance dataset; S3 Multi-Source Data Fusion: After completing the spatiotemporal registration of the preprocessed dataset, the weighted average method is used for preliminary fusion, and then principal component analysis (PCA) is used for dimensionality reduction and noise reduction to obtain a fused dataset that is strongly correlated with suspended sediment information. S4 suspended sediment concentration inversion: Based on the fused dataset, an improved semi-empirical radiative transfer model is used to invert the suspended sediment concentration (SSC). The core formula of the model is: ; In the formula, , , The apparent reflectance of remote sensing at the 655nm, 560nm, and 865nm wavelengths are respectively. , , , These are the undetermined coefficients of the model; S5 Result Verification: The accuracy of the inversion results was verified by using measured suspended sediment concentration data from ground stations in the Yellow River Estuary and adjacent sea areas.

[0008] Preferably, in step S1, the selection of data from each satellite sensor satisfies the spatiotemporal complementarity constraint, and the constraint formula is as follows: ; In the formula, , These represent the maximum and minimum imaging times for multi-source data, respectively. The spatial overlap rate of multi-source data within the monitoring area is calculated; simultaneously, Landsat8OLI and GF-1WFV data are resampled to a spatial resolution of 10m using bicubic convolution to match the spatial resolution of Sentinel-2A / B data.

[0009] Preferably, in step S2, the radiometric calibration uses sensor calibration coefficients to construct a linear conversion model, converting the image DN value into on-board radiance. The conversion formula is as follows: ; In the formula, The on-board radiance is in band λ. This refers to the band calibration gain coefficient. The band calibration offset coefficient is used; atmospheric correction uses the FLAASH model to eliminate atmospheric scattering and absorption effects, and geometric fine correction is based on ground control point GCPs and DEM data using second-order polynomial fitting. The geometric error of the corrected image is controlled within 0.5 pixels.

[0010] Preferably, in step S3, the weighting coefficients of the preliminary fusion using the weighted average method are determined by the Pearson correlation coefficient between each sensor band and the suspended sediment concentration. The weighting calculation formula is as follows: ; In the formula, The weighting coefficients for the λ-th band of the i-th sensor are... is the Pearson correlation coefficient between the reflectivity of this band and the measured suspended sediment concentration, where n is the number of sensors and m is the effective band number of the corresponding sensor.

[0011] Preferably, step S3, principal component analysis (PCA) dimensionality reduction and noise reduction, specifically includes the following steps: S31 Covariance Matrix Construction and Normalization: The weighted fused multi-band dataset is normalized in terms of band dimension to eliminate the numerical magnitude difference between different sensor bands and construct the covariance matrix of the normalized dataset. S32 Eigenvalue Calculation and Sorting: Calculate the eigenvalues ​​of the covariance matrix and the corresponding orthogonal eigenvectors, and sort the eigenvectors in descending order of eigenvalues. The magnitude of the eigenvalues ​​is positively correlated with the contribution of suspended sediment information. S33 Effective Principal Component Screening and Fusion Dataset Reconstruction: Calculate the cumulative contribution rate of the sorted feature values, select the top k principal components with a cumulative contribution rate ≥ 95%, remove noise and redundant information corresponding to the remaining principal components, and reconstruct the final fusion dataset based on the selected principal components.

[0012] Preferably, in step S4, the undetermined coefficients of the model , , , The objective function for fitting the measured data and the corresponding pixel reflectance using the least squares method is: ; In the formula, For the model inversion concentration value of the j-th station, Let N be the field-measured concentration value at the j-th station, and N be the number of measured stations used in the modeling.

[0013] Preferably, in step S4, an environmental correction factor is introduced into the improved semi-empirical radiative transfer model, and the corrected model formula is as follows: ; In the formula, V represents the tidal current velocity at the monitoring time, and U represents the wind speed at a height of 10m above the sea surface at the monitoring time. , The correction coefficients for tidal current speed and wind speed are determined by multiple linear regression fitting.

[0014] Preferably, step S5, which uses ground station measurement data for accuracy verification, specifically includes the following steps: S51 Validation Sample Set Construction: Select station-measured data covering the Yellow River Estuary and its adjacent sea areas, including river estuaries, nearshore tidal flats, and offshore shallow sea areas. The samples cover spring, summer, autumn, and winter, all tidal phases (high tide, low tide, low tide, and low tide), and all gradients of suspended sediment concentration (low, medium, and high). The number of valid samples is no less than 120. S52 Spatiotemporal Matching Processing: The sampling time and latitude and longitude coordinates of each measured station are accurately matched with the imaging time and pixel spatial position of the corresponding remote sensing image to ensure the spatiotemporal consistency between the measured data and the inversion results. The spatiotemporal deviation of the matching is controlled within ±1 hour and ±1 pixel. S53 accuracy index calculation: Based on the matched inversion results and measured data, the coefficient of determination is calculated. The three core accuracy indicators, mean absolute percentage error (MAPE), root mean square error (RMSE), etc., are used to quantitatively verify the inversion results.

[0015] Preferably, step S5, based on the verification of ground-based measured data, also includes a cross-validation step, as follows: S54 Single Sensor Independent Inversion Comparison: Select remote sensing data from Sentinel-2A / B, Landsat8OLI, and GF-1WFV single sensors respectively, and use the same inversion model as in step S4 to complete the independent inversion of suspended sediment concentration to obtain the single sensor inversion comparison dataset. S55 Multi-Source Consistency Comparison: The inversion results of the single-sensor inversion comparison dataset and the data fused by this method are compared pixel by pixel to verify the applicability of the method among different sensors. S56 Leave-one-out cross-validation: The leave-one-out method is used to verify the robustness of the inversion model. In each round, one measured station data is reserved as the validation set, and all remaining samples are used as the modeling set. Multiple rounds of iterative fitting and validation are completed, and the average accuracy index of multiple rounds of validation is statistically analyzed to confirm the stability and universality of the model.

[0016] As a preferred option, after step S4 completes the inversion of suspended sediment concentration, it also includes a step of standardizing the output of monitoring results, as follows: S41 concentration distribution thematic map generation: Based on the pixel-by-pixel suspended sediment concentration data obtained by inversion, a raster thematic map of the spatial distribution of suspended sediment concentration in the Yellow River Estuary and its adjacent sea areas with a spatial resolution of 10m consistent with the fused dataset is generated. S42 Concentration Gradient Zoning Statistics: The monitoring area is divided into four gradient zones: low concentration, medium-low concentration, medium-high concentration, and high concentration, based on preset suspended sediment concentration grading thresholds. The spatial distribution area, proportion, and average concentration of each zone are statistically analyzed. S43 Dynamic Feature Extraction and Results Archiving: Calculate the spatial centroid coordinates of suspended sediment accumulation areas in different time series monitoring results, extract the dynamic change characteristics of sediment transport, integrate thematic maps, regional statistical data, dynamic feature parameters and accuracy verification results to form a standardized monitoring results archive.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention addresses the spatiotemporal distribution and variation characteristics of suspended sediment in the Yellow River Estuary and its adjacent sea areas. It completes targeted screening and standardized preprocessing of multiple remote sensing data sources. By setting clear spatiotemporal complementarity constraints, it integrates the technical advantages of satellite sensors with different spatial, temporal, and spectral resolutions, overcoming the technical deficiency of existing technologies where a single data source cannot simultaneously ensure both spatial detail accuracy and temporal dynamics. Simultaneously, considering the optical characteristics of the turbid waters near the Yellow River Estuary and the complex atmospheric environment of the estuary, it optimizes the atmospheric correction and geometric fine correction processes, effectively eliminating the interference of environmental factors on the true spectral signal of the water body. Based on this, a multi-source data fusion algorithm combining weighted average method and principal component analysis is constructed. This achieves complementary advantages of different sensors in sensitive band information of suspended sediment, and effectively suppresses the interference of noise and redundant information through dimensionality reduction processing. This significantly improves the quality of remote sensing basic data and the extraction efficiency of suspended sediment characteristic information, providing stable and reliable data support for subsequent concentration inversion.

[0018] 2. This invention addresses the optical response characteristics of suspended sediment in the Yellow River Estuary by constructing an improved semi-empirical radiative transfer inversion model. Through the combination of red and green light band ratios and the coupling design of reflectivity in the near-infrared sensitive band, the model's signal response capability across the entire concentration gradient of suspended sediment is enhanced. This solves the technical problems of insufficient inversion accuracy in high-concentration sediment areas and weak signal response in low-concentration areas in existing general inversion models. Simultaneously, key hydrodynamic environmental factors such as tidal current velocity and sea surface wind speed in the Yellow River Estuary are introduced as model correction terms, improving the model's universality under complex estuarine dynamic conditions. Accurate fitting of model parameters is achieved through the least squares method, realizing accurate inversion of suspended sediment across the entire range and concentration gradient from the Yellow River estuary to adjacent offshore waters. Compared to existing general monitoring methods, the coefficient of determination of the inversion results is significantly improved, and the mean absolute percentage error and root mean square error are effectively controlled, demonstrating outstanding substantive features and significant technological advancements.

[0019] 3. This invention constructs a complete and replicable monitoring method system, encompassing data acquisition, preprocessing, multi-source fusion, concentration inversion, and multi-dimensional accuracy verification. Each operational step has a clear process, quantifiable parameters, and repeatable methods. It can stably adapt to the routine periodic monitoring of suspended sediment in the Yellow River Estuary and adjacent sea areas, emergency monitoring of water and sediment regulation during the flood season, and long-term interannual dynamic monitoring needs. This solves the technical defects of existing technologies, such as non-standardized monitoring processes, insufficient stability, and difficulty in meeting operational continuous monitoring requirements. Furthermore, the core technical solution of this invention has good scenario scalability. Its multi-source remote sensing data fusion method and targeted inversion model construction approach can be extended to suspended sediment monitoring in other estuary and coastal areas, providing a standardized technical path for remote sensing monitoring of nearshore marine water environment, possessing industrial applicability and clear engineering application value. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are explained in detail together with the embodiments of the invention, but do not constitute a limitation thereof.

[0021] Figure 1 This invention presents a flowchart of the Yellow River Estuary suspended sediment monitoring method based on multi-source remote sensing data. Figure 2 This is a flowchart of the data acquisition process of the present invention; Figure 3 This is a flowchart of the preprocessing process of the present invention; Figure 4 This is a flowchart of the multi-source data fusion process of the present invention; Figure 5 This is a flowchart illustrating the core closed-loop process of inversion, verification, and result output in this invention. Figure 6This is a flowchart of the multi-inversion model and verification method of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figures 1-6 As shown, this invention addresses the operational needs of monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas. It integrates the complementary spatiotemporal spectral advantages of multi-source satellite remote sensing data. Through standardized data preprocessing, multi-source data fusion combining weighted averaging and principal component analysis, an improved semi-empirical radiative transfer inversion model, and a multi-dimensional accuracy verification process, it achieves high-precision, wide-coverage, and high-timeliness dynamic monitoring of suspended sediment in the Yellow River Estuary. The complete implementation process of this invention is as follows: Multi-source remote sensing data acquisition: The monitoring area of ​​this invention is the Yellow River Estuary and its adjacent sea areas, geographically covering the sea area within 30km of the current Yellow River estuary and delta. The selected multi-source data sources include Landsat 8 OLI, Sentinel-2A / B, GF-1 WFV, and GOCI satellite sensor data. Data selection must meet spatiotemporal complementarity constraints, the constraints being as follows: ; ; in The maximum imaging time for multi-source data. For the minimum imaging time of multi-source data, This represents the spatial overlap rate of multi-source data within the monitoring area.

[0024] All data underwent spatial resolution unification. Landsat 8 OLI's 30m resolution data and GF-1 WFV's 16m resolution data were resampled to 10m using bicubic convolution to match the native 10m spatial resolution of the Sentinel-2A / B data. Simultaneously, the temporal resolution of all data was kept within 24 hours to capture the diurnal dynamic changes of suspended sediment. During data screening, image products with cloud cover below 10% were prioritized, while invalid data with excessive cloud coverage or severe stripe noise were removed.

[0025] Multi-source remote sensing data preprocessing: For all the acquired raw remote sensing data, three core preprocessing operations are performed in sequence: radiometric calibration, atmospheric correction, and geometric fine correction, to obtain a standardized remote sensing reflectance dataset.

[0026] Radiometric calibration uses the official calibration coefficients published by each sensor to construct a linear transformation model, converting the raw image DN values ​​into on-board radiance. The conversion formula is as follows: ; in For the corresponding band The star's radiance, This is the calibration gain coefficient for this band. This is the calibration offset coefficient for this band. This represents the original grayscale value of the image pixel.

[0027] Atmospheric correction was performed using the FLAASH model. By simulating the transmission process of solar radiation in the atmosphere, the interference of atmospheric molecular scattering, aerosol absorption and scattering on the water body radiation signal was eliminated, the true spectral characteristics of water body radiation were restored, and the surface remote sensing apparent reflectance of each band was obtained.

[0028] Geometric fine correction is based on high-precision ground control points and digital elevation model (DEM) data. A second-order polynomial fitting method is used to complete the geometric fine correction of the image, eliminating geometric distortion caused by sensor attitude changes, Earth curvature, and terrain undulations. The geometric error of the corrected image is controlled within 0.5 pixels, ensuring accurate spatial matching of multi-source data.

[0029] Multi-source remote sensing data fusion: For the preprocessed multi-source dataset, spatiotemporal registration is first completed, then preliminary fusion is completed by weighted average method, and finally dimensionality reduction and noise reduction are completed by principal component analysis (PCA) to obtain a fused dataset that is strongly correlated with suspended sediment information.

[0030] Spatiotemporal registration is based on the 10m resolution image of Sentinel-2A / B, and pixel-level registration is completed on the preprocessed images of other sensors. After registration, the pixel offset of each image does not exceed 0.3 pixels, while ensuring that the imaging time difference of all data meets the constraint of within 24 hours, thus completing the matching of the time dimension.

[0031] The weighted average method was used to initially fuse the data by determining the weighting coefficients for each sensor band based on the Pearson correlation coefficients between each sensor band and the suspended sediment concentration. The weighting calculation formula is as follows: ; in For the first The first sensor Weighting coefficients for each band, The Pearson correlation coefficient between apparent reflectance and measured suspended sediment concentration in this band. The total number of sensors participating in the fusion. To determine the effective number of bands for sensors sensitive to suspended sediment, weighted fusion was performed on the registered multi-source, multi-band data based on weighting coefficients to obtain a preliminary fused multi-band dataset.

[0032] Principal Component Analysis (PCA) for dimensionality reduction and denoising consists of three steps. The first step is covariance matrix construction and normalization. For the weighted and fused multi-band dataset, minimum-maximum normalization of the band dimensions is performed to eliminate numerical magnitude differences between different sensor bands. A covariance matrix is ​​then constructed based on the normalized dataset. The second step is eigenvalue calculation and sorting. The eigenvalues ​​and corresponding orthogonal eigenvectors of the covariance matrix are calculated, and the eigenvectors are sorted in descending order of eigenvalues. The magnitude of the eigenvalues ​​is positively correlated with their contribution to suspended sediment information in the band data. The third step is effective principal component selection and fused dataset reconstruction. The cumulative contribution rate of the sorted eigenvalues ​​is calculated, and the top eigenvalues ​​with a cumulative contribution rate greater than or equal to 95% are selected. One principal component is selected, and noise and redundant information corresponding to the remaining principal components are removed. Based on the selected principal components, the data is reconstructed to obtain the final fused dataset.

[0033] Suspended sediment concentration inversion: Based on the fused dataset, an improved semi-empirical radiative transfer model was used to complete the inversion calculation of suspended sediment concentration in the Yellow River Estuary and its adjacent sea areas.

[0034] The core calculation formula of the model is ; in The suspended sediment concentration obtained from the inversion is... The apparent reflectance in the 655nm red light band for remote sensing. The apparent reflectance in the 560nm green light band is the remote sensing reflectance. The apparent reflectance for remote sensing in the 865nm near-infrared band. , , , These are the undetermined coefficients of the model.

[0035] The undetermined coefficients of the model were determined by least squares fitting, and the objective function of the fitting was: ; in For the first Concentration values ​​retrieved from the model at each station. For the first The actual measured concentration values ​​at each station. The total number of measured stations used for modeling.

[0036] To improve the model's versatility in complex hydrodynamic environments, an environmental correction factor is introduced to optimize the core model. The corrected model formula is as follows: ; in To monitor the power flow velocity at the corresponding point at a given time, To monitor the wind speed at a height of 10m above the sea surface at the corresponding location at that time, This is a correction factor for the current velocity. This is a correction factor for wind speed. and Determined by fitting using multiple linear regression.

[0037] Based on the inversion model with determined parameters, pixel-by-pixel calculations are performed on each effective pixel of the fused dataset to obtain raster data of suspended sediment concentration covering the entire monitoring area.

[0038] Verification of inversion results and output of results: First, the accuracy of the inversion results is verified by using ground station measurement data from the Yellow River Estuary and its adjacent sea areas, and then the standardized output of the monitoring results is completed.

[0039] Accuracy verification is divided into two parts: quantitative verification using ground-based measured data and cross-validation. The implementation process for quantitative verification using ground-based measured data is as follows: First, a verification sample set is constructed, selecting measured data from stations covering the Yellow River estuary, nearshore tidal flats, and offshore shallow sea areas. The samples cover all four seasons (spring, summer, autumn, and winter), all tidal phases (high tide, slack tide, and low tide), and all suspended sediment concentration ranges (low, medium, and high), with a minimum of 120 valid samples. Next, spatiotemporal matching processing is performed, precisely matching the sampling time and latitude / longitude coordinates of each measured station with the imaging time and pixel spatial location of the corresponding remote sensing image. This ensures the spatiotemporal consistency between the measured data and the inversion results, with the time deviation controlled within ±1 hour and the spatial deviation within ±1 pixel. Finally, accuracy indices are calculated based on the matched inversion results and measured data, calculating the coefficient of determination. Mean absolute percentage error Root mean square error Three core accuracy indicators were used to complete the quantitative verification of the inversion results.

[0040] The cross-validation process is as follows: First, independent inversion comparisons were performed using single sensors. Remote sensing data from Sentinel-2A / B, Landsat 8 OLI, and GF-1 WFV were selected, and the same inversion model was used to independently invert suspended sediment concentrations, resulting in single-sensor inversion comparison datasets. Next, multi-source consistency comparisons were performed. The inversion results of the single-sensor comparison datasets and the multi-source fused data were compared pixel-by-pixel to verify the applicability of the method across different sensors, focusing on spatial distribution trends, concentration gradient ranges, and extreme point locations. Finally, leave-one-out cross-validation was performed. The leave-one-out method was used to verify the robustness of the inversion model. In each round, one measured station's data was reserved as the validation set, and all remaining samples were used as the modeling set. Multiple rounds of iterative fitting and validation were completed, and the average accuracy index of the multiple rounds of validation was statistically analyzed to confirm the stability and universality of the model.

[0041] The standardized output process for monitoring results involves several steps. First, a thematic map of concentration distribution is generated. Based on the pixel-by-pixel suspended sediment concentration data obtained through inversion, a raster thematic map of the spatial distribution of suspended sediment concentration in the Yellow River Estuary and its adjacent sea areas with a spatial resolution consistent with the fused dataset (10m resolution) is generated. Next, concentration gradient zoning statistics are completed. Preset suspended sediment concentration grading thresholds divide the monitoring area into four gradient zones: low concentration, low-medium concentration, medium-high concentration, and high concentration. The spatial distribution area, proportion, and average concentration of each zone are then statistically analyzed. Finally, dynamic feature extraction and result archiving are completed. The spatial centroid coordinates of suspended sediment accumulation areas in different time-series monitoring results are calculated, dynamic changes in sediment transport are extracted, and the thematic map, zoning statistics, dynamic feature parameters, and accuracy verification results are integrated to form a standardized monitoring results archive.

[0042] Example 1: Routine suspended sediment monitoring during the spring water level period in the Yellow River Estuary This routine monitoring of suspended sediment in the Yellow River Estuary during the spring low-water season took place on April 12, 2019. The monitoring area was the Yellow River estuary at its current Qingshuigou channel and the adjacent sea area within 30 km. During this period, the Yellow River's runoff was stable and the hydrodynamic conditions in the sea area were gentle, making it suitable for routine monitoring of suspended sediment distribution.

[0043] In the multi-source remote sensing data acquisition phase, the multi-source data sources selected for this implementation included Sentinel-2B imagery acquired on April 12, 2019; GF-1 WFV imagery acquired on April 12, 2019; GOCI hourly imagery acquired on April 12, 2019; and Landsat 8 OLI imagery acquired on April 13, 2019. The difference between the maximum and minimum imaging time for all data was 22 hours, which met the requirements. The spatial overlap rate within the monitoring area is 92%, which meets the constraints. Constraints. The 30m resolution data of Landsat 8 OLI and the 16m resolution data of GF-1 WFV were resampled to 10m using bicubic convolution and spatially matched with the 10m resolution data of Sentinel-2B. The cloud cover of all data was less than 8% and there was no obvious strip noise.

[0044] In the data preprocessing stage, radiometric calibration is first performed. Taking the green band of Sentinel-2B imagery as an example, the calibration gain coefficient of this band is... Scale offset coefficient The original of a certain pixel Substitute into the radiation calibration formula The on-stellar radiance of the pixel was calculated. After radiometric calibration calculations were completed for all bands, atmospheric correction was performed using the FLAASH model to obtain the apparent reflectance of each band. Subsequently, based on 18 high-precision ground control points and 30m resolution DEM data in the Yellow River Delta region, geometric fine correction was performed using a second-order polynomial. After correction, the geometric error of all images was controlled within 0.4 pixels.

[0045] In the multi-source data fusion stage, firstly, using Sentinel-2B imagery as a reference, pixel-level registration was completed for all preprocessed images, with the pixel offset after registration not exceeding 0.25 pixels. Subsequently, the Pearson correlation coefficients between the sensitive bands of each sensor and the measured suspended sediment concentrations were calculated, with the correlation coefficient for the Landsat 8 OLI 655nm red band being... Correlation coefficient of 560nm green light band Correlation coefficient in the 865nm near-infrared band The correlation coefficients for the corresponding bands of Sentinel-2B are 0.91, 0.78, and 0.84, respectively; the correlation coefficients for the corresponding bands of GF-1 WFV are 0.87, 0.75, and 0.80, respectively; and the correlation coefficients for the corresponding bands of GOCI are 0.85, 0.81, and 0.78, respectively. Summing the absolute values ​​of all correlation coefficients yields a total of 13.05. Taking the Sentinel-2B 655nm band as an example, its weighting coefficient... The weight coefficients of all bands were calculated using the same method, and a weighted average was used for preliminary fusion based on the weights. Subsequently, PCA dimensionality reduction was performed on the preliminary fused dataset, and the cumulative contribution rate of the first three principal components of the covariance matrix was found to be 96.2%, which meets the requirement of a cumulative contribution rate ≥ 95%. The first three principal components were selected to reconstruct the final fused dataset.

[0046] In the suspended sediment concentration inversion step, this implementation used measured data from 45 stations during the spring low-water period of the Yellow River Estuary to complete the model parameter fitting, and obtained the undetermined coefficients of the model by solving the least squares method. , , , Simultaneously, by combining tidal current and wind speed data during the monitoring period, an environmental correction coefficient was obtained through fitting. , Taking a certain pixel as an example, the pixel's... , , Monitoring the current velocity at any given time Wind speed First, substitute the values ​​into the core model to calculate the basic values: .

[0047] Substituting the environmental correction term, the correction factor is... The final inversion concentration of that pixel The same method was used to perform pixel-by-pixel calculations for all pixels, thus obtaining the suspended sediment concentration distribution data for the monitored area.

[0048] In the results verification and output phase, this verification used measured data from 42 stations in the Yellow River Estuary area during the same period as the verification set. After completing spatiotemporal matching, the determination coefficient of the inversion results was calculated. Mean absolute percentage error Root mean square error This meets the monitoring accuracy requirements. Simultaneously, single-sensor inversion comparison and leave-one-out cross-validation were completed, with the average value of the leave-one-out cross-validation... ,average The model exhibits good stability. A thematic map of suspended sediment concentration distribution at a resolution of 10m was generated, and statistical analysis of four concentration gradient zones was completed. The high-concentration area is mainly distributed near the estuary, covering an area of ​​128.6 km², with an average concentration of 215.3 mg / L. The centroid coordinates of the sediment accumulation areas were extracted, and the monitoring results were archived.

[0049] Example 2: Emergency Monitoring of High Sediment Concentrations in the Yellow River Estuary During the Flood Season This emergency monitoring of high sediment concentrations following the Yellow River's flood season water and sediment regulation was conducted on July 8, 2020, covering the Yellow River estuary and its adjacent 40km-range sea area. During this period, the Yellow River's water and sediment regulation led to a surge in sediment entering the sea, resulting in a large gradient and rapid dynamic changes in the concentration of suspended sediment in the sea area, necessitating high-frequency and high-precision emergency monitoring.

[0050] In the multi-source remote sensing data acquisition phase, the multi-source data sources selected for this implementation included Sentinel-2A imagery acquired on July 8, 2020; GF-1 WFV imagery acquired on July 8, 2020; GOCI imagery acquired hourly on July 8, 2020; and Landsat 8 OLI imagery acquired on July 7, 2020. The difference between the maximum and minimum imaging time for all data was 23 hours, which meets the requirements. The spatial overlap rate within the monitoring area is 89%, which meets the constraints. Constraints were considered. Landsat 8 OLI's 30m resolution data and GF-1 WFV's 16m resolution data were resampled to 10m using bicubic convolution, and spatially matched with Sentinel-2A's 10m resolution data. All data showed cloud cover below 5% and no significant stripe noise. GOCI data provided high-frequency observations once per hour, meeting the temporal resolution requirements for emergency monitoring.

[0051] In the data preprocessing stage, radiometric calibration is first performed. Taking the Landsat 8 OLI red band as an example, the calibration gain coefficient of this band is... Scale offset coefficient The original of a certain pixel Substitute into the radiation calibration formula The on-stellar radiance of the pixel was calculated. After radiometric calibration of all bands, atmospheric correction was performed using the FLAASH model to eliminate the influence of high humidity during the flood season on the radiation signal, and the apparent reflectance of each band was obtained. Subsequently, geometric fine correction was performed based on 22 high-precision ground control points and DEM data, and the geometric error of the corrected image was controlled within 0.35 pixels.

[0052] In the multi-source data fusion stage, firstly, using Sentinel-2A imagery as a reference, pixel-level registration was completed for all preprocessed images, with the pixel offset after registration not exceeding 0.2 pixels. Subsequently, the Pearson correlation coefficients between the sensitive bands of each sensor and the measured data of high sediment concentration during the flood season were calculated, with the correlation coefficient for the Landsat 8 OLI 655nm red band being the most significant. Correlation coefficient of 560nm green light band Correlation coefficient in the 865nm near-infrared band The correlation coefficients for the corresponding bands of Sentinel-2A are 0.94, 0.74, and 0.89, respectively; the correlation coefficients for the corresponding bands of GF-1 WFV are 0.91, 0.71, and 0.86, respectively; and the correlation coefficients for the corresponding bands of GOCI are 0.90, 0.78, and 0.85, respectively. The sum of the absolute values ​​of all correlation coefficients is 13.81. Taking the Sentinel-2A 655nm band as an example, its weighting coefficient... After calculating the weights of all bands, a weighted average preliminary fusion was performed based on the weights. Subsequently, PCA dimensionality reduction was performed on the preliminary fused dataset, and the cumulative contribution rate of the first three principal components was found to be 97.1%, which meets the requirement of a cumulative contribution rate ≥ 95%. The first three principal components were selected to reconstruct the final fused dataset.

[0053] In the suspended sediment concentration inversion step, this implementation used measured high sediment concentration data from 52 stations in the Yellow River Estuary during the flood season to fit the model parameters, and obtained the undetermined coefficients of the model using the least squares method. , , , By combining tidal current and wind speed data during the monitoring period, an environmental correction coefficient was obtained through fitting. , Taking a high-density pixel near the estuary as an example, the pixel's... , , Monitoring the current velocity at any given time Wind speed First, calculate the base value: .

[0054] Substituting the environmental correction term, the correction factor is... The final inversion concentration of that pixel The same method was used to complete the pixel-by-pixel calculation of all pixels, and the intraday dynamic changes were inverted by combining the GOCI hourly data to obtain the suspended sediment concentration distribution data at different times.

[0055] In the results verification and output phase, this verification used measured data from 38 stations in the Yellow River Estuary area during the same period as the verification set, with high-concentration stations accounting for more than 60%. After completing spatiotemporal matching, the determination coefficient of the inversion results was calculated. Mean absolute percentage error Root mean square error The inversion accuracy in high-concentration regions is significantly better than that of traditional single-sensor methods. Simultaneously, single-sensor inversion comparison and leave-one-out cross-validation were performed, with the leave-one-out cross-validation average... The model exhibits good stability. A thematic map of hourly suspended sediment concentration distribution at 10m resolution was generated, and concentration gradient zoning statistics were completed. The high-concentration area reached 312.4 km², significantly larger than during the normal water period. Simultaneously, the diffusion direction and transport velocity of the sediment plume were extracted, generating an emergency monitoring report and providing data support for estuarine ecological management and navigation safety.

[0056] Example 3: Long-term interannual dynamic monitoring of suspended sediment in the Yellow River Estuary This study conducted long-term interannual dynamic monitoring of suspended sediment in the Yellow River Estuary, covering the period from 2016 to 2020. Images from three typical periods—normal water season, flood season, and dry season—were selected each year. The monitoring area covered the entire coastline of the Yellow River Delta and adjacent sea areas, aiming to reveal the interannual variation patterns and spatial distribution evolution characteristics of suspended sediment in the Yellow River Estuary.

[0057] In the multi-source remote sensing data acquisition phase, this implementation selected Landsat 8OLI, Sentinel-2A / B, GF-1 WFV, and GOCI imagery data from three typical time periods between 2016 and 2020, totaling 120 valid images. The multi-source data for each monitoring period met the requirements... Time constraints, and Spatial overlap constraints were implemented. All data underwent spatial resolution unification. Images with resolutions other than 10m were resampled to 10m using bicubic convolution to ensure spatial consistency of long-term time-series data. Cloud cover in all images was less than 10%, with no significant noise impact.

[0058] In the data preprocessing stage, all valid imagery from 2016 to 2020 underwent standardized processing using a unified preprocessing workflow. All imagery was radiometrically calibrated using the corresponding sensor's official calibration coefficients, atmospheric correction was performed using the FLAASH model, and geometric correction was completed using standardized ground control points and DEM data. After correction, the geometric error of all imagery was controlled within 0.5 pixels, ensuring the radiometric and geometric consistency of long-term data and eliminating systematic errors from different years and different sensors.

[0059] In the multi-source data fusion stage, a unified fusion process was used to process the preprocessed data for each monitoring period. First, spatiotemporal registration was performed using Sentinel-2 imagery as a reference, with the pixel offset not exceeding 0.3 pixels after registration. Based on the full-time measured dataset of the Yellow River Estuary, the Pearson correlation coefficients of the sensitive bands of each sensor were calculated, and a unified weighting formula was used to determine the weight coefficients for each band, completing the initial weighted average fusion. Subsequently, PCA dimensionality reduction was performed on the initial fused dataset for each period, and principal components with a cumulative contribution rate ≥95% were selected to reconstruct the fused dataset, ensuring the consistency and comparability of the methods used in long-term time-series fusion data.

[0060] In the suspended sediment concentration inversion phase, this study used full-gradient measured data from 145 stations in the Yellow River Estuary from 2016 to 2020 to complete the parameter fitting of the long-term time-series general inversion model. The undetermined coefficients of the general model were obtained by solving the least squares method. , , , By combining hydrological and meteorological data from different time periods, seasonal environmental correction coefficients were obtained, with the coefficients for the normal water period being... , ;flood season , Dry season , Taking a certain pixel during the dry season of 2016 as an example, the pixel's... , , Monitoring the current velocity at any given time Wind speed First, calculate the base value: .

[0061] Substituting the dry season environmental correction term into the equation, the correction factor is: The final inversion concentration of that pixel Following a unified model and process, suspended sediment concentration inversion was completed for all time periods of images from 2016 to 2020, resulting in a long-term continuous concentration distribution dataset.

[0062] In the results verification and output phase, this verification used measured data from 145 stations from 2016 to 2020, divided into 5 verification sets according to year. After completing spatiotemporal matching, the average coefficient of determination of the long-term time series inversion results was calculated. Mean absolute percentage error Root mean square error The inversion accuracy remained stable across different years, meeting the requirements for long-term time-series monitoring. Simultaneously, single-sensor inversion comparisons and cross-validations were completed for each year, confirming the model's universality and stability in long-term time series. Finally, thematic atlases of suspended sediment concentration distribution at 10m resolution were generated for each year and time period, and concentration gradient zoning statistics for each year were completed. Analysis revealed the interannual variation trend of the area of ​​high-concentration suspended sediment areas in the Yellow River Estuary from 2016 to 2020, as well as the interannual shift characteristics of the centroid of sediment accumulation areas. This revealed the impact of water and sediment regulation activities on the distribution of suspended sediment in the Yellow River Estuary, forming long-term time-series monitoring and analysis results, and providing data support for research on the geomorphological evolution and ecological environmental protection of the Yellow River Delta.

[0063] This invention addresses the spatiotemporal distribution and variation characteristics of suspended sediment in the Yellow River Estuary and its adjacent sea areas. It completes targeted screening and standardized preprocessing of multiple remote sensing data sources. By setting clear spatiotemporal complementarity constraints, it integrates the technical advantages of satellite sensors with different spatial, temporal, and spectral resolutions, overcoming the technical deficiency of existing technologies where a single data source cannot simultaneously capture both spatial detail accuracy and temporal dynamics. Simultaneously, considering the optical characteristics of the turbid waters near the Yellow River Estuary and the complex atmospheric environment of the estuary, it optimizes the atmospheric correction and geometric fine correction processes, effectively eliminating the interference of environmental factors on the true spectral signals of the water body. Based on this, a multi-source data fusion algorithm combining weighted average and principal component analysis is constructed. This achieves complementary advantages of different sensors in sensitive bands of suspended sediment information, and effectively suppresses the interference of noise and redundant information through dimensionality reduction processing. This significantly improves the quality of remote sensing basic data and the extraction efficiency of suspended sediment characteristic information, providing stable and reliable data support for subsequent concentration inversion.

[0064] This invention addresses the optical response characteristics of suspended sediment in the Yellow River Estuary by constructing an improved semi-empirical radiative transfer inversion model. Through the combination of red and green light band ratios and the coupling design of reflectivity in the near-infrared sensitive band, the model's signal response capability across the entire concentration gradient of suspended sediment is enhanced. This solves the technical problems of insufficient inversion accuracy in high-concentration sediment areas and weak signal response in low-concentration areas in existing general inversion models. Simultaneously, key hydrodynamic environmental factors such as tidal current velocity and sea surface wind speed in the Yellow River Estuary are introduced as model correction terms, improving the model's universality under complex estuarine dynamic conditions. Accurate fitting of model parameters is achieved through the least squares method, realizing accurate inversion of suspended sediment across the entire range and concentration gradient from the Yellow River estuary to adjacent offshore waters. Compared to existing general monitoring methods, the coefficient of determination of the inversion results is significantly improved, and the mean absolute percentage error and root mean square error are effectively controlled, demonstrating outstanding substantive features and significant technological advancements.

[0065] This invention constructs a comprehensive and replicable monitoring methodology system encompassing data acquisition, preprocessing, multi-source fusion, concentration inversion, and multi-dimensional accuracy verification. Each operational step has a clearly defined process, quantifiable parameters, and repeatable methods. It can stably adapt to the routine periodic monitoring of suspended sediment in the Yellow River Estuary and adjacent sea areas, emergency monitoring of water and sediment regulation during the flood season, and long-term interannual dynamic monitoring needs. This addresses the technical shortcomings of existing technologies, such as non-standardized processes, insufficient stability, and difficulty in meeting operational continuous monitoring requirements. Furthermore, the core technical solution of this invention has excellent scenario scalability. Its multi-source remote sensing data fusion method and targeted inversion model construction approach can be extended to suspended sediment monitoring in other estuary and coastal areas, providing a standardized technical path for remote sensing monitoring of nearshore marine water environments. It possesses industrial applicability and clear engineering application value.

[0066] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data, characterized in that, Includes the following steps: S1 Multi-source Remote Sensing Data Acquisition: For the monitoring area of ​​the Yellow River Estuary and its adjacent sea areas, remote sensing data from Landsat8OLI, Sentinel-2A / B, GF-1WFV and GOCI satellite sensors were selected as multi-source data sources to ensure that the spatiotemporal overlap of each data source meets the fusion requirements. The spatial resolution of all data was unified to 10m and the temporal resolution was controlled within 24h. S2 data preprocessing: The acquired multi-source remote sensing data are sequentially subjected to radiometric calibration, atmospheric correction and geometric fine correction to obtain a standardized remote sensing reflectance dataset; S3 Multi-Source Data Fusion: After completing the spatiotemporal registration of the preprocessed dataset, the weighted average method is used for preliminary fusion, and then principal component analysis (PCA) is used for dimensionality reduction and noise reduction to obtain a fused dataset that is strongly correlated with suspended sediment information. S4 suspended sediment concentration inversion: Based on the fused dataset, an improved semi-empirical radiative transfer model is used to invert the suspended sediment concentration (SSC). The core formula of the model is: In the formula, , , The apparent reflectance of remote sensing at the 655nm, 560nm, and 865nm wavelengths are respectively. , , , These are the undetermined coefficients of the model; S5 Result Verification: The accuracy of the inversion results was verified by using measured suspended sediment concentration data from ground stations in the Yellow River Estuary and adjacent sea areas.

2. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 1, characterized in that, In step S1, the selection of data from each satellite sensor satisfies the spatiotemporal complementarity constraint, and the constraint formula is as follows: In the formula, , These represent the maximum and minimum imaging times for multi-source data, respectively. The spatial overlap rate of multi-source data within the monitoring area is calculated; simultaneously, Landsat8OLI and GF-1WFV data are resampled to a spatial resolution of 10m using bicubic convolution to match the spatial resolution of Sentinel-2A / B data.

3. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 1, characterized in that, In step S2, radiometric calibration uses sensor calibration coefficients to construct a linear conversion model, converting the image DN values ​​into on-board radiance. The conversion formula is as follows: In the formula, The on-board radiance is in band λ. This refers to the band calibration gain coefficient. The band calibration offset coefficient is used; atmospheric correction uses the FLAASH model to eliminate atmospheric scattering and absorption effects, and geometric fine correction is based on ground control point GCPs and DEM data using second-order polynomial fitting. The geometric error of the corrected image is controlled within 0.5 pixels.

4. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 1, characterized in that, In step S3, the weighting coefficients of the initial fusion using the weighted average method are determined by the Pearson correlation coefficient between each sensor band and the suspended sediment concentration. The weighting calculation formula is as follows: In the formula, The weighting coefficients for the λ-th band of the i-th sensor are... is the Pearson correlation coefficient between the reflectivity of this band and the measured suspended sediment concentration, where n is the number of sensors and m is the effective band number of the corresponding sensor.

5. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 4, characterized in that, In step S3, principal component analysis (PCA) dimensionality reduction and noise reduction specifically includes the following steps: S31 Covariance Matrix Construction and Normalization: The weighted fused multi-band dataset is normalized in terms of band dimension to eliminate the numerical magnitude difference between different sensor bands and construct the covariance matrix of the normalized dataset. S32 Eigenvalue Calculation and Sorting: Calculate the eigenvalues ​​of the covariance matrix and the corresponding orthogonal eigenvectors, and sort the eigenvectors in descending order of eigenvalues. The magnitude of the eigenvalues ​​is positively correlated with the contribution of suspended sediment information. S33 Effective Principal Component Screening and Fusion Dataset Reconstruction: Calculate the cumulative contribution rate of the sorted feature values, select the top k principal components with a cumulative contribution rate ≥ 95%, remove noise and redundant information corresponding to the remaining principal components, and reconstruct the final fusion dataset based on the selected principal components.

6. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 1, characterized in that, In step S4, the undetermined coefficients of the model , , , The objective function for fitting the measured data and the corresponding pixel reflectance using the least squares method is: In the formula, For the model inversion concentration value of the j-th station, Let N be the field-measured concentration value at the j-th station, and N be the number of measured stations used in the modeling.

7. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 6, characterized in that, In step S4, an environmental correction factor is introduced into the improved semi-empirical radiative transfer model. The corrected model formula is as follows: In the formula, V represents the tidal current velocity at the monitoring time, and U represents the wind speed at a height of 10m above the sea surface at the monitoring time. , The correction coefficients for tidal current speed and wind speed are determined by multiple linear regression fitting.

8. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 1, characterized in that, Step S5, which uses ground station measurement data for accuracy verification, specifically includes the following steps: S51 Validation Sample Set Construction: Select station-measured data covering the Yellow River Estuary and its adjacent sea areas, including river estuaries, nearshore tidal flats, and offshore shallow sea areas. The samples cover spring, summer, autumn, and winter, all tidal phases (high tide, low tide, low tide, and low tide), and all gradients of suspended sediment concentration (low, medium, and high). The number of valid samples is no less than 120. S52 Spatiotemporal Matching Processing: The sampling time and latitude and longitude coordinates of each measured station are accurately matched with the imaging time and pixel spatial position of the corresponding remote sensing image to ensure the spatiotemporal consistency between the measured data and the inversion results. The spatiotemporal deviation of the matching is controlled within ±1 hour and ±1 pixel. S53 accuracy index calculation: Based on the matched inversion results and measured data, the coefficient of determination is calculated. The three core accuracy indicators, mean absolute percentage error (MAPE), root mean square error (RMSE), etc., are used to quantitatively verify the inversion results.

9. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 8, characterized in that, Step S5, based on the verification of ground-based measured data, also includes a cross-validation step, as detailed below: S54 Single Sensor Independent Inversion Comparison: Select remote sensing data from Sentinel-2A / B, Landsat8OLI, and GF-1WFV single sensors respectively, and use the same inversion model as in step S4 to complete the independent inversion of suspended sediment concentration to obtain the single sensor inversion comparison dataset. S55 Multi-Source Consistency Comparison: The inversion results of the single-sensor inversion comparison dataset and the data fused by this method are compared pixel by pixel to verify the applicability of the method among different sensors. S56 Leave-one-out cross-validation: The leave-one-out method is used to verify the robustness of the inversion model. In each round, one measured station data is reserved as the validation set, and all remaining samples are used as the modeling set. Multiple rounds of iterative fitting and validation are completed, and the average accuracy index of multiple rounds of validation is statistically analyzed to confirm the stability and universality of the model.

10. The method for monitoring suspended sediment in the Yellow River Estuary and its adjacent sea areas based on multi-source remote sensing data according to claim 1, characterized in that, After step S4 completes the inversion of suspended sediment concentration, it also includes a step of standardizing the output of monitoring results, as follows: S41 concentration distribution thematic map generation: Based on the pixel-by-pixel suspended sediment concentration data obtained by inversion, a raster thematic map of the spatial distribution of suspended sediment concentration in the Yellow River Estuary and its adjacent sea areas with a spatial resolution of 10m consistent with the fused dataset is generated. S42 Concentration Gradient Zoning Statistics: The monitoring area is divided into four gradient zones: low concentration, medium-low concentration, medium-high concentration, and high concentration, based on preset suspended sediment concentration grading thresholds. The spatial distribution area, proportion, and average concentration of each zone are statistically analyzed. S43 Dynamic Feature Extraction and Results Archiving: Calculate the spatial centroid coordinates of suspended sediment accumulation areas in different time series monitoring results, extract the dynamic change characteristics of sediment transport, integrate thematic maps, regional statistical data, dynamic feature parameters and accuracy verification results to form a standardized monitoring results archive.

Images