Long-time series monitoring method and system for sediment concentration of suspended sediment based on multi-reach scale

Abstract

The application discloses a kind of long-time sequence monitoring method and system of suspended load sediment concentration based on multiple river section scales, and relates to river monitoring technical field.The method includes the following steps: obtaining long-time sequence multispectral remote sensing image data and environmental data of the river to be detected, and pre-processing the multispectral remote sensing image data;According to the pre-defined river boundary, divide the river section, and extract the water body pixels of each river section by water surface extraction method;Using median extraction strategy to calculate the band reflectance and band index of each river section;Using machine learning algorithm to predict the suspended load sediment concentration of band reflectance, band index and environmental data.The present application carries out long-term monitoring of sediment concentration of different river section scales based on remote sensing data, and predicts by combining machine learning algorithm, which can accurately grasp the subtle dynamic changes of river sediment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of river monitoring, and particularly relates to a long-time series monitoring method and system for suspended sediment concentration based on multi-reach scale. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] Suspended sediment has characteristics such as wide particle size distribution, dynamic change, sedimentation characteristics and ecological impact. It contains particles of different sizes, usually from fine sediment to larger sand particles, which are affected by factors such as water flow speed, riverbed morphology and weather conditions, and show temporal and spatial dynamic changes. Different types of suspended sediment have different deposition rates in the water flow, which in turn affect the evolution of the riverbed and the ecological environment. In addition, suspended sediment concentration has an important influence on the habitat of aquatic organisms, light transmittance and water quality, and is an important indicator for assessing the ecological health of water bodies.

[0004] Traditional methods for monitoring the suspended sediment concentration of rivers usually rely on ground sampling and laboratory analysis, which not only requires a large amount of manpower and material resources, but also involves a long time for sample collection and analysis. In view of the low efficiency of traditional monitoring methods and the defects of large time and space errors, remote sensing technology has been increasingly concerned and applied in the process of monitoring the suspended sediment concentration of rivers due to its advantages of large range, high frequency and non-contact. However, current monitoring of suspended sediment concentration in rivers based on remote sensing technology is usually limited to a single river, making it difficult to achieve large-scale (such as watershed scale) monitoring.

[0005] Moreover, current technology is mostly limited to short-term or coarse-scale monitoring when monitoring the suspended sediment concentration of river reaches, lacking long-time series and fine-scale data support. This makes it impossible to fully grasp the change trend of suspended sediment at different time and spatial scales, especially the subtle changes under the influence of rapid climate change and human activities.

[0006] Most current remote sensing inversion models mainly rely on spectral band reflectance or simple band ratio index, but the change of suspended sediment concentration is affected by many complex factors, including upstream sediment, water flow conditions, surface erosion, vegetation coverage and precipitation. A single input feature cannot fully capture these complex relationships, resulting in insufficient model accuracy and robustness, especially when facing extreme events (such as floods) or river reaches affected by human activities (such as dam construction and land use changes).

[0007] Traditional river monitoring often operates on a basin-by-basin or hydrological station-by-station basis, making it difficult to capture subtle changes in suspended sediment concentration within rivers at the segmental scale. Furthermore, while satellite remote sensing data has accumulated decades of observational data, a unified, efficient, and operational framework is lacking to utilize this massive amount of data for continuous dynamic monitoring and trend analysis of suspended sediment concentration across the entire river network over decades. This makes it difficult to comprehensively grasp the subtle trends in suspended sediment concentration at different temporal and spatial scales, especially regarding regional responses to rapid climate change and anthropogenic influences.

[0008] In summary, how to achieve long-term, fine-grained dynamic monitoring of suspended sediment concentration across multiple river segments has become a problem that needs to be solved by existing technologies. Summary of the Invention

[0009] To address the shortcomings of existing technologies, the purpose of this invention is to provide a long-term monitoring method and system for suspended sediment concentration based on multiple river segment scales. This method uses remote sensing data to perform fine-grained long-term monitoring of sediment concentration at different river segment scales, and combines machine learning algorithms for prediction, thereby accurately grasping the subtle dynamic changes in river sediment.

[0010] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0011] The first aspect of this invention provides a long-term monitoring method for suspended sediment concentration based on multiple river segment scales, comprising the following steps:

[0012] Acquire long-term multispectral remote sensing image data and environmental data of the river to be detected, and preprocess the multispectral remote sensing image data;

[0013] The river segments are divided according to predefined river boundaries, and the water pixels of each river segment are extracted using a water surface extraction method.

[0014] The median extraction strategy was used to calculate the band reflectance and band index for each river segment;

[0015] Machine learning algorithms were used to predict suspended sediment content based on band reflectance, band index, and environmental data.

[0016] Furthermore, the specific steps for preprocessing multispectral remote sensing image data are as follows:

[0017] Radiometric calibration of multispectral remote sensing image data;

[0018] Atmospheric correction of multispectral remote sensing image data;

[0019] Geometric correction and registration of multispectral remote sensing image data;

[0020] Background removal is performed on multispectral remote sensing image data.

[0021] Furthermore, the specific steps for dividing the river into segments based on predefined river boundaries and extracting water pixels for each segment using a water surface extraction method are as follows:

[0022] Draw the river centerline and create a buffer zone based on the river centerline;

[0023] Water pixels for each river segment are extracted using a water surface extraction method, and noise in the water pixels within the buffer zone is removed. Initial noise removal is achieved by verifying the consistency of multiple water surface extraction methods, and further noise removal is achieved by determining the intersection of water pixels with the river centerline.

[0024] Furthermore, the specific steps for calculating the band reflectance and band index for each river segment using the median extraction strategy are as follows:

[0025] Water reflectance is extracted from water pixels within the buffer, and the median of reflectance data for each band is calculated.

[0026] The corresponding band index is calculated based on the median reflectance of each band.

[0027] Furthermore, the specific steps for predicting suspended sediment concentration using machine learning algorithms based on band reflectance, band index, and environmental data are as follows:

[0028] Build multiple machine learning models;

[0029] Each machine learning model is trained and validated using a dataset consisting of band reflectivity, band index, and environmental data, and the trained machine learning models are evaluated.

[0030] Based on the evaluation results, the optimal machine learning model is selected to predict suspended sediment concentration using band reflectance, band index, and environmental data.

[0031] A second aspect of the present invention provides a long-term monitoring system for suspended sediment concentration based on multiple river segment scales, comprising:

[0032] The data acquisition module is configured to acquire long-term multispectral remote sensing image data and environmental data of the river to be detected, and to preprocess the multispectral remote sensing image data.

[0033] The first data processing module is configured to divide the river into segments according to predefined river boundaries and extract water pixels for each river segment using a water surface extraction method.

[0034] The second data processing module is configured to use a median extraction strategy to calculate the band reflectance and band index of each river segment;

[0035] The machine learning module is configured to use machine learning algorithms to predict suspended sediment content from band reflectance, band index, and environmental data.

[0036] Furthermore, the data acquisition module also includes a preprocessing module, configured as follows:

[0037] Radiometric calibration of multispectral remote sensing image data;

[0038] Atmospheric correction of multispectral remote sensing image data;

[0039] Geometric correction and registration of multispectral remote sensing image data;

[0040] Background removal is performed on multispectral remote sensing image data.

[0041] Furthermore, the first data processing module is also configured as follows:

[0042] Draw the river centerline and create a buffer zone based on the river centerline;

[0043] Water pixels for each river segment are extracted using a water surface extraction method, and noise in the water pixels within the buffer zone is removed. Initial noise removal is achieved by verifying the consistency of multiple water surface extraction methods, and further noise removal is achieved by determining the intersection of water pixels with the river centerline.

[0044] Furthermore, the second data processing module is also configured as follows:

[0045] Water reflectance is extracted from water pixels within the buffer, and the median of reflectance data for each band is calculated.

[0046] The corresponding band index is calculated based on the median reflectance of each band.

[0047] Furthermore, the machine learning module is also configured as follows:

[0048] Build multiple machine learning models;

[0049] Each machine learning model is trained and validated using a dataset consisting of band reflectivity, band index, and environmental data, and the trained machine learning models are evaluated.

[0050] Based on the evaluation results, the optimal machine learning model is selected to predict suspended sediment concentration using band reflectance, band index, and environmental data.

[0051] The above one or more technical solutions have the following beneficial effects:

[0052] This invention discloses a long-term monitoring method and system for suspended sediment concentration at multiple river segment scales. By constructing an innovative framework of multimodal machine learning models, this invention achieves dynamic monitoring of suspended sediment concentration at the river segment scale. The invention employs a "multimodal input feature fusion" strategy. In addition to traditional Landsat satellite spectral band surface reflectance and various water-related band indices, it innovatively integrates multivariate environmental predictive factors related to suspended sediment sources and environmental conditions as input. Furthermore, this invention utilizes the precisely depicted river boundaries from the Global River Width Database (GRWL), and based on this, defines river segments that maximize the representation of dynamic river characteristics through simplified vector version grouping and merging. By preprocessing and extracting pixels from Landsat imagery based on a double-buffered river width, this invention can acquire high-quality annual average remote sensing data for each river segment. This invention deeply integrates machine learning and remote sensing technologies, greatly reducing the reliance on expensive and time-consuming ground sampling. By automating the processing of massive amounts of remote sensing data, it significantly reduces monitoring costs and, for the first time, achieves dynamic monitoring of suspended sediment concentration at the river section scale over a large area and over a long time period, thus overcoming the limitations of traditional monitoring in terms of cost and efficiency.

[0053] To address the challenge of long-term, multi-segment scale monitoring in existing technologies, this invention leverages the Landsat satellite's mission duration of over forty years and its fine spatial resolution, combined with the Global Rivers Database (GRWL) for precise depiction of river boundaries. Utilizing the powerful processing capabilities of remote sensing cloud platforms such as Google Earth Engine (GEE), this invention can acquire and process multi-source, multispectral remote sensing image data, and, by incorporating various machine learning algorithms, conduct continuous dynamic monitoring of suspended sediment concentration for each defined river segment over nearly forty years (1984-2023).

[0054] Traditional single-input features limit the accuracy and robustness of suspended sediment concentration (SSC) inversion models. To overcome this limitation, this invention integrates multimodal data as model input. The machine learning model can learn the nonlinear, multicollinear, and heteroscedastic relationships between SSC and various complex driving factors, thereby significantly improving the accuracy of SSC inversion and the model's generalization ability.

[0055] This invention preprocesses and extracts pixels from Landsat imagery based on a double-buffered approach for river width, enabling the acquisition of high-quality annual average remote sensing data for each river segment. This strategy allows the invention to utilize nearly half a century of Landsat satellite archives, providing nearly forty years of data on suspended sediment concentration variations at the river segment scale, from 1984 to 2023, filling the gap in spatiotemporal resolution of traditional monitoring. This refined, long-term monitoring capability allows the invention to reveal long-term trends and seasonal variations in suspended sediment concentration at regional and even global scales, thereby greatly enhancing the understanding and prediction of river sediment dynamics.

[0056] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart of long-term monitoring of suspended sediment concentration based on multiple river segment scales in Embodiment 1 of the present invention;

[0059] Figure 2 This is a comparison chart of the inversion accuracy of various machine learning models in Embodiment 1 of the present invention. Detailed Implementation

[0060] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0061] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0062] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0063] Example 1:

[0064] Embodiment 1 of the present invention provides a long-term monitoring method for suspended sediment concentration based on multiple river segment scales, such as... Figure 1 As shown, it includes the following steps:

[0065] S1: Acquire long-term multispectral remote sensing image data and environmental data of the river to be detected, and preprocess the multispectral remote sensing image data.

[0066] S11: Acquire long-term multispectral remote sensing image data and environmental data of the river to be detected.

[0067] In one specific implementation, this embodiment acquires satellite remote sensing data via sensors. It primarily acquires and utilizes Landsat series multispectral remote sensing imagery, including Landsat-5 Thematic Mapper™, Landsat-7 Enhanced Thematic Mapper Plus (ETM+), and Landsat-8 Operational Land Imager (OLI) images. The remote sensing data covers the period from 1984 to 2023, providing fine-scale spatial resolution (30 meters) and periodic revisits (16 days). Its rich visible, near-infrared, and shortwave infrared information is crucial for identifying water body characteristics and analyzing sediment content. To avoid inconsistencies that may arise from different sensors, this embodiment prioritizes and primarily uses Landsat-5 Thematic Mapper™ or Landsat-7 Enhanced Thematic Mapper Plus (ETM+) data when developing models based on ground data. For large-scale applications, the compatibility of Landsat-8 Operational Land Imager (OLI) / Thermal Infrared Sensor (TIRS) is considered, or independent models are used for processing, ensuring the consistency and reliability of long-term time-series data.

[0068] To better predict suspended sediment load in rivers, this embodiment additionally acquires a series of environmental data closely related to the sources and environmental conditions of suspended sediment load, enhancing the model's interpretability and prediction accuracy. These data primarily originate from the ECMWF Reanalysis v5 (ERA5) Land product. Environmental data include high vegetation leaf area index (h_LAI), low vegetation mean leaf area index (l_LAI), daily average temperature, wind speed, net solar radiation at the surface, inland water evaporation, and daily precipitation. The ECMWF Reanalysis v5 (ERA5) Land product data is widely recognized for its high reliability and accuracy, and its time span is also from 1984 to 2023. The introduction of these multimodal input features enables the machine learning model to capture the influence of complex environmental drivers beyond the optical properties of water bodies, such as non-point source material input caused by extreme rainfall.

[0069] S12: Preprocess multispectral remote sensing image data.

[0070] S121: Radiometric calibration of multispectral remote sensing image data.

[0071] Specifically, the digital quantization (DN) values ​​of the original multispectral remote sensing image data from the sensor are converted into physically meaningful radiance or reflectance values, eliminating errors caused by the sensor itself.

[0072] S122: Perform atmospheric correction on multispectral remote sensing image data.

[0073] Specifically, the LaSRC atmospheric correction algorithm is used to eliminate the influence of atmospheric molecules and aerosols on the reflectivity of ground objects, converting the reflectivity of the top atmospheric layer into the true reflectivity of the ground surface.

[0074] Atmospheric corrections for Landsat-5 Thematic Mapper™ and Landsat-7 Enhanced Thematic Mapper Plus (ETM+) were performed by the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS).

[0075] S123: Perform geometric correction and registration on multispectral remote sensing image data.

[0076] Specifically, it corrects the geometric distortion of multispectral remote sensing image data and ensures that images acquired at different times and by different sensors are spatially precisely aligned for time-series analysis.

[0077] S124: Remove background from multispectral remote sensing image data.

[0078] Specifically, background elements such as clouds, cloud shadows, and snowmelt are identified and removed. Multispectral remote sensing image data is used to identify and mask pixels covered by clouds, cloud shadows, snow, or ice, avoiding interference with water feature extraction.

[0079] S2: Divide the river into segments according to predefined river boundaries, and extract water pixels for each segment using a water surface extraction method. In this embodiment, a division scale of 20km is set, dividing the river into 20km long segments on average.

[0080] S21: Draw the river centerline and create a buffer zone based on the river centerline.

[0081] This embodiment uses the Global River Width Dataset (GRWL) to define river boundaries. The GRWL database uses Landsat data to accurately depict the river centerline. This invention creates a buffer zone twice the river width based on the GRWL river centerline, and processes multispectral remote sensing image data within this buffer zone to further ensure that the extracted pixels represent the true water body characteristics of the river section.

[0082] S22: Extract water pixels for each river segment using a water surface extraction method, and remove noise from the water pixels in the buffer.

[0083] S221: Extract water pixels for each river segment using a water surface extraction method to determine the water body range.

[0084] Specifically, for each defined river segment, this embodiment integrates multiple advanced water surface extraction methods for water body pixel extraction, including Dynamic Surface Water Range (DSWE), Modified Normalized Difference Water Index (MNDWI), Arid Zone Water Detection Rule (ARWDR), and Automatic Water Extraction Index (AWEI).

[0085] S222: Remove noise from water pixels within the buffer.

[0086] S2221: Initial noise removal was achieved through consistency verification of multiple water surface extraction methods.

[0087] A sample is considered valid only when all these water surface extraction methods consistently identify the sampled point data (or potential water body pixels) as water body pixels. This largely eliminates the interference of incorrect or inaccurate sampled point coordinate records and land background noise on water body feature extraction.

[0088] S2222: Further noise removal is achieved by determining the intersection of water pixels with the river centerline.

[0089] Specifically, the system determines whether the identified water body intersects with the GRWL river centerline, and further removes small, non-intersecting water bodies such as lake pixels to ensure that the analysis is performed only on the river.

[0090] S3: The median extraction strategy is used to calculate the band reflectance and band index for each river segment.

[0091] S31: Extract water reflectance from water pixels in the buffer and calculate the median of reflectance data for each band.

[0092] Specifically, to minimize the uncertainty caused by outliers (such as optical reflections from the bottom of shallow water or incompletely identified thin clouds), this embodiment employs a median extraction strategy for the band reflectance of each river segment. This strategy, based on the water body range precisely extracted within the buffer zone in the previous step, effectively filters out extreme values ​​by calculating the median of the reflectance data for each band within that range. The median represents the stability and reliability of the data, thus more accurately representing the true reflectance of the river segment.

[0093] S32: Calculate the corresponding band index based on the median reflectance of each band.

[0094] Specifically, a series of band indices closely related to suspended sediment content are calculated based on the extracted reflectance of each band, serving as input features for the machine learning model. These indices include, but are not limited to, the Normalized Difference Turbidity Index (NDTI), the Normalized Difference Water Index (NDWI), and ratio indices such as red / green, red / blue, and near-infrared / red, thereby significantly enhancing the sediment's spectral response and improving the model's sensitivity. Overall, the river segment-scale processing strategy in this embodiment greatly expands the studyable scope (e.g., watershed scale) while ensuring the representativeness of the extracted river feature information. This allows for monitoring of suspended sediment content in rivers at a wider range (e.g., watershed scale) and a finer scale (river segment scale) without increasing computational complexity.

[0095] S4: Utilize machine learning algorithms to predict suspended sediment content based on band reflectivity, band index, and environmental data.

[0096] S41: Build multiple machine learning models.

[0097] Specifically, this invention employs multiple machine learning algorithms and comparative analysis to achieve accurate prediction of suspended sediment concentration. Specifically, it uses Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), and Deep Neural Networks (DNN) to construct multiple machine learning models. The DNN model is implemented using the TensorFlow (version 2.14.0) open-source machine learning framework. Its structure includes an input layer, multiple hidden layers, and an output layer, employing a Multilayer Perceptron (MLP) architecture and using the ReLU activation function. The number of hidden layers and neurons is optimized through repeated cross-validation and hyperparameter tuning to ensure optimal prediction performance. All other machine learning models, including Random Forest, XGBoost, and SVR, are implemented using the scikit-learn (version 1.3.1) package in the Python 3.8 environment. Random Forest improves stability by aggregating prediction results from a large number of bootstrap sampling decision trees. XGBoost sequentially trains weak predictors to fit the residuals and introduces a regularization term to prevent overfitting. SVR maps the data to a high-dimensional space using a kernel function to find the optimal hyperplane for regression.

[0098] S42: Train and validate each machine learning model using a dataset consisting of band reflectivity, band index, and environmental data, and evaluate the trained machine learning model.

[0099] Specifically, to comprehensively evaluate the performance and generalization ability of these models, this embodiment divides the dataset into training, testing, and independent validation sets in a strict 7:2:1 ratio. The independent validation set is not involved in model training or calibration and is used solely for the final generalization ability evaluation. Furthermore, this embodiment implements 10-fold cross-validation to further optimize the model's hyperparameters. The hyperparameter adjustments for each model are shown in Table 1, and the potential impact of dataset partitioning changes on model performance is evaluated to ensure the robustness and reliability of the models.

[0100] Table 1 Hyperparameters of various machine learning algorithms

[0101]

[0102] For different machine learning algorithms, ten-fold cross-validation is used for dynamic adjustment during the parameter tuning process to ensure that each method exerts its maximum advantage. Evaluation is based on the coefficient of determination (R²). 2 The root mean square error (RMSE) is used for evaluation. 2The larger the value and the lower the RMSE, the more accurate and reliable the models are selected as the final inversion models. These models can automatically adapt to changes under different environmental conditions, significantly improving the accuracy and reliability of monitoring results, and have real-time update capabilities, enabling rapid response in constantly changing environments. According to specific evaluation rules, the inversion accuracy of each model is as follows: Figure 2 As shown in the figure, XGBoost is selected as the best inversion algorithm in this embodiment.

[0103] In summary, this invention, through the combination of remote sensing and machine learning technologies, enables dynamic monitoring of suspended sediment over long time periods and at the river section scale, providing a reliable and automated technical solution for river sediment monitoring.

[0104] S43: Based on the evaluation results, select the optimal machine learning model to predict suspended sediment concentration using band reflectance, band index, and environmental data.

[0105] In this embodiment, the machine learning model input features include multiple dimensions, including band reflectance, band index, and environmental data. Compared with traditional machine learning models, the multi-dimensional data input of this invention can enhance the model's generalization ability and accuracy. Furthermore, this invention uses the interpretable machine learning method SHAP (SHapley Additive ex Planations) to quantitatively evaluate the contribution of each input feature.

[0106] Example 2:

[0107] Embodiment 2 of the present invention provides a long-term monitoring system for suspended sediment concentration based on multiple river segment scales, comprising:

[0108] The data acquisition module is configured to acquire long-term multispectral remote sensing image data and environmental data of the river to be detected, and to preprocess the multispectral remote sensing image data.

[0109] The data acquisition module also includes a preprocessing module, which is configured as follows:

[0110] Radiometric calibration of multispectral remote sensing image data.

[0111] Atmospheric correction is performed on multispectral remote sensing image data.

[0112] Geometric correction and registration of multispectral remote sensing image data.

[0113] Background removal is performed on multispectral remote sensing image data.

[0114] The first data processing module is configured to divide the river into segments according to predefined river boundaries and extract water pixels for each segment using a water surface extraction method.

[0115] The first data processing module is also configured as follows:

[0116] Draw the river centerline and create a buffer zone based on the river centerline.

[0117] Water pixels for each river segment are extracted using a water surface extraction method, and noise in the water pixels within the buffer is removed.

[0118] The noise is initially removed by verifying the consistency of various water surface extraction methods, and then further removed by judging the intersection of water pixels with the river centerline.

[0119] The second data processing module is configured to use a median extraction strategy to calculate the band reflectance and band index for each river segment.

[0120] The second data processing module is also configured as follows:

[0121] Water reflectance is extracted from the water pixels within the buffer, and the median of the reflectance data for each band is calculated.

[0122] The corresponding band index is calculated based on the median reflectance of each band.

[0123] The machine learning module is configured to use machine learning algorithms to predict suspended sediment content from band reflectance, band index, and environmental data.

[0124] The machine learning module is also configured as follows:

[0125] Build multiple machine learning models.

[0126] Each machine learning model is trained and validated using a dataset consisting of band reflectivity, band index, and environmental data, and the trained machine learning models are evaluated.

[0127] Based on the evaluation results, the optimal machine learning model is selected to predict suspended sediment concentration using band reflectance, band index, and environmental data.

[0128] The steps and methods involved in the above embodiment two correspond to those in embodiment one. For specific implementation details, please refer to the relevant description section of embodiment one.

[0129] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0130] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0131] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A long-term monitoring method for suspended sediment concentration based on multiple river segment scales, characterized in that, Includes the following steps: Acquire long-term multispectral remote sensing image data and environmental data of the river to be detected, and preprocess the multispectral remote sensing image data; acquire a series of environmental data closely related to the source of suspended sediment concentration and environmental conditions; The river segments are divided according to predefined river boundaries, and the water pixels of each river segment are extracted using a water surface extraction method. The median extraction strategy was used to calculate the band reflectance and band index for each river segment. The specific steps for calculating the band reflectance and band index for each river segment using the median extraction strategy are as follows: Water reflectance is extracted from water pixels within the buffer, and the median of reflectance data for each band is calculated. Calculate the corresponding band index based on the median reflectance of each band; Machine learning algorithms were used to predict suspended sediment content based on band reflectivity, band index, and environmental data.

2. The long-term monitoring method for suspended sediment concentration based on multi-segment scale as described in claim 1, characterized in that, The specific steps for preprocessing multispectral remote sensing image data are as follows: Radiometric calibration of multispectral remote sensing image data; Atmospheric correction of multispectral remote sensing image data; Geometric correction and registration of multispectral remote sensing image data; Background removal is performed on multispectral remote sensing image data.

3. The long-term monitoring method for suspended sediment concentration based on multi-segment scale as described in claim 1, characterized in that, The specific steps for dividing the river into segments based on predefined river boundaries and extracting water pixels for each segment using a water surface extraction method are as follows: Draw the river centerline and create a buffer zone based on the river centerline; Water pixels for each river segment are extracted using a water surface extraction method, and noise in the water pixels within the buffer zone is removed. The noise is initially removed by verifying the consistency of multiple water surface extraction methods, and then further removed by determining the intersection of the water pixels with the river centerline.

4. The long-term monitoring method for suspended sediment concentration based on multi-segment scale as described in claim 1, characterized in that, The specific steps for predicting suspended sediment concentration using machine learning algorithms based on band reflectance, band index, and environmental data are as follows: Build multiple machine learning models; Each machine learning model is trained and validated using a dataset consisting of band reflectivity, band index, and environmental data, and the trained machine learning models are evaluated. Based on the evaluation results, the optimal machine learning model is selected to predict suspended sediment concentration using band reflectance, band index, and environmental data.

5. A long-term monitoring system for suspended sediment concentration based on multiple river segment scales, characterized in that, include: The data acquisition module is configured to acquire long-term multispectral remote sensing image data and environmental data of the river to be detected, and to preprocess the multispectral remote sensing image data; and to acquire a series of environmental data closely related to the source of suspended sediment concentration and environmental conditions. The first data processing module is configured to divide the river into segments according to predefined river boundaries and extract water pixels for each river segment using a water surface extraction method. The second data processing module is configured to calculate the band reflectance and band index of each river segment using a median extraction strategy. The specific steps for calculating the band reflectance and band index of each river segment using the median extraction strategy are as follows: Water reflectance is extracted from water pixels within the buffer, and the median of reflectance data for each band is calculated. Calculate the corresponding band index based on the median reflectance of each band; The machine learning module is configured to use machine learning algorithms to predict suspended sediment content from band reflectance, band index, and environmental data.

6. The long-term monitoring system for suspended sediment concentration based on multiple river segment scales as described in claim 5, characterized in that, The data acquisition module also includes a preprocessing module, which is configured as follows: Radiometric calibration of multispectral remote sensing image data; Atmospheric correction of multispectral remote sensing image data; Geometric correction and registration of multispectral remote sensing image data; Background removal is performed on multispectral remote sensing image data.

7. The long-term monitoring system for suspended sediment concentration based on multiple river segment scales as described in claim 5, characterized in that, The first data processing module is also configured as follows: Draw the river centerline and create a buffer zone based on the river centerline; Water pixels for each river segment are extracted using a water surface extraction method, and noise in the water pixels within the buffer zone is removed. The noise is initially removed by verifying the consistency of multiple water surface extraction methods, and then further removed by determining the intersection of the water pixels with the river centerline.

8. The long-term monitoring system for suspended sediment concentration based on multiple river segment scales as described in claim 5, characterized in that, The machine learning module is also configured as follows: Build multiple machine learning models; Each machine learning model is trained and validated using a dataset consisting of band reflectivity, band index, and environmental data, and the trained machine learning models are evaluated. Based on the evaluation results, the optimal machine learning model is selected to predict suspended sediment concentration using band reflectance, band index, and environmental data.

Images