River suspended sediment concentration remote sensing monitoring method based on data driving and machine learning

Abstract

The invention discloses a river suspended sediment concentration remote sensing monitoring method based on data driving and machine learning. According to the method, a typical river reach and upstream and downstream hydrological monitoring stations need to be determined, multi-year time sequence remote sensing image data of the typical river reach are collected, and the remote sensing image data are preprocessed through radiometric calibration, atmospheric correction and the like to obtain water surface reflectivity data; synchronously matching measured sediment content data of a hydrometric station, constructing a spectrum-sediment data set, and grading and dividing a training set and a test set according to low, medium and high concentrations; and training and establishing a suspended sediment inversion model by adopting a least square support vector machine algorithm, and after verification of a test set, inverting the suspended sediment concentration of the whole river reach and generating a spatial-temporal distribution thematic map. The problems that a traditional method is high in cost, low in efficiency and limited in coverage and a traditional remote sensing model is poor in applicability to inland rivers are effectively solved, and efficient, accurate and macroscopic remote sensing monitoring on the concentration of the suspended sediment in the river is achieved.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring technology, and in particular to a remote sensing monitoring method for river suspended sediment concentration based on data-driven and machine learning approaches. Background Technology

[0002] River suspended sediment concentration is a key indicator characterizing the intensity of soil erosion, river channel evolution trends, and the health of the aquatic ecosystem in a watershed. Traditional methods for measuring river suspended sediment concentration are mainly divided into direct measurement methods and indirect measurement methods. However, while direct measurement methods (such as field sampling) have high accuracy, they are limited by high cost, low efficiency, high risk, and difficulty in reflecting the spatiotemporal distribution of the entire watershed. Traditional remote sensing inversion methods (empirical, physical, and semi-empirical models) suffer from poor universality, complex mechanisms, or insufficient migration capacity when applied to inland rivers with complex optical characteristics and narrow water surfaces.

[0003] To address the aforementioned problems, this invention proposes a remote sensing monitoring method for river suspended sediment concentration based on data-driven and machine learning. This method leverages the high spatiotemporal and hyperspectral resolution of the Sentinel-2 satellite, combines synchronous hydrological cross-section measured data, and utilizes machine learning algorithms to establish a robust and accurate inversion model, providing a solution for monitoring river suspended sediment concentration. Summary of the Invention

[0004] The present invention aims to provide a remote sensing monitoring method for river suspended sediment concentration based on data-driven and machine learning, in order to solve the problems of high cost, low efficiency, limited coverage and poor applicability of traditional remote sensing models to inland rivers.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A remote sensing method for monitoring river suspended sediment concentration based on data-driven and machine learning includes the following steps:

[0007] S1. The river section with the largest annual variation in sediment content is identified as the research object for remote sensing inversion of suspended sediment concentration. Typical river sections are selected, and the hydrological monitoring stations with the closest upstream and downstream distances in the typical river section area are determined.

[0008] S2. Collect multi-year time series remote sensing image data of this typical river section, and preprocess these remote sensing image data to obtain surface reflectance data of the river water body.

[0009] S3. Based on the imaging time of the collected remote sensing images, select the sediment content data of the nearest hydrological monitoring station upstream and downstream of the typical river section at the time point closest to the imaging time of the remote sensing images.

[0010] S4. Based on the surface reflectance data of water bodies in typical river sections obtained in step S2 and the measured sediment concentration data of the nearest upstream and downstream hydrological stations obtained in step S3, construct a spectral sediment monitoring dataset for river sections.

[0011] S5. Take the river section spectral sediment monitoring dataset established in step S4, select a threshold based on the sediment monitoring data, and classify the dataset into low, medium and high sediment content. Each level is divided into training set and test set according to the proportion.

[0012] S6. The least squares support vector machine algorithm is used to train the river section spectral sediment monitoring training set constructed in step S5 to establish a machine learning-based river section suspended sediment inversion model, and the test set is used for verification.

[0013] S7. Finally, using the suspended sediment inversion model established in step S6, the suspended sediment concentration of the entire river section is calculated from the remote sensing data of typical river sections, and the spatiotemporal distribution thematic map of suspended sediment in the monitored river section is obtained.

[0014] Furthermore, the preprocessing steps in step S2 include radiometric calibration, image registration, cloud detection, atmospheric correction, and image cropping.

[0015] Furthermore, in step S2, data from the European Space Agency's Sentinel-2 satellite is selected as the primary data source. Sentinel-2 L1C-level data is used, which has undergone radiometric calibration and geometric correction. Cloud detection is performed through visual interpretation to filter out images that are greatly affected by clouds and cloud shadows.

[0016] Furthermore, in step S5, the determination of the grading threshold is achieved by statistically analyzing the measured data of suspended sediment concentration in typical river sections, selecting an appropriate threshold, ensuring that the amount of data for each level reaches at least 10, and dividing the data for each level into a 70% training set and a 30% test set to ensure that there are at least 2 data points in the test set for each level.

[0017] The principle and beneficial effects of this technical solution: This invention addresses the problem of effectively monitoring suspended sediment concentration in inland river waters on a macroscopic scale. It fully utilizes existing high temporal and spatial resolution satellite remote sensing data and historical sediment monitoring data. By processing data from the European Space Agency's Sentinel-2 L1C satellite through atmospheric correction and cloud detection, it extracts the surface reflectance spectral information of turbid inland river waters. This data is then paired with synchronized sediment monitoring data from hydrological stations to construct a spectral sediment dataset based on Sentinel-2 data. Using the least squares support vector machine algorithm, a machine learning model is constructed to correlate remote sensing water color spectra with the average sediment concentration at river cross-sections. This enables effective remote sensing inversion of sediment concentration along the river channel, providing significant scientific value for research on river water and sediment heterogeneity analysis, scouring, and sedimentation. Compared to existing technologies, this invention will fundamentally change the current situation where water color remote sensing technology is widely used in oceans and lakes, but limited in remote sensing research of suspended sediment in inland river waters. It overcomes the shortcomings of traditional hydrological sediment monitoring, which is limited to monitoring cross-sections, and can reflect the spatiotemporal distribution and changes of suspended sediment in rivers on a macroscopic scale. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention;

[0019] Figure 2 Schematic diagram of the Wuhan-Jiujiang section of the Yangtze River in this embodiment of the invention.

[0020] Figure 3 This is a schematic diagram illustrating the principle of remote sensing water color inversion for suspended sediment in rivers using the method of this invention.

[0021] Figure 4 This is a scatter plot of the predicted and measured data from the entire data model in the method of this invention;

[0022] Figure 5 This is a scatter plot of the model predictions and measured data from the training set and test set in the method of this invention;

[0023] Figure 6 This is a spatial distribution map of suspended sediment in the Wuhan Hankou to Jiujiang section of the Yangtze River during the 2018 flood season, obtained using the method of this invention.

[0024] Figure 7 This invention provides a spatial distribution map of suspended sediment in the Wuhan Hankou to Jiujiang section of the Yangtze River during the 2020 flood season. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments:

[0026] like Figure 1 The diagram shows a flowchart of a remote sensing monitoring method for river suspended sediment concentration based on data-driven and machine learning methods according to the present invention. A detailed description of specific embodiments follows:

[0027] Step 1: Determine the selected study river section. In this embodiment, the section from Hankou, Wuhan to Jiujiang is selected as the study river section. Figure 2 As shown, the typical river section with the largest annual variation in suspended sediment content was selected as the research object for remote sensing inversion of suspended sediment concentration. Based on the region where the typical river section is located, the hydrological monitoring station with the closest upstream and downstream distance of the typical river section was determined.

[0028] Step 2: In this embodiment, data from the European Space Agency's Sentinel-2 satellite is selected as the primary data source. Multi-year time-series remote sensing imagery of typical river sections is collected. This embodiment uses Sentinel-2 L1C-level data, which has undergone radiometric calibration and geometric correction. This step primarily involves visual cloud detection to filter out images heavily affected by clouds and cloud shadows. After preliminary cloud detection, the images are processed using Acolite atmospheric correction software provided by the Royal Institute of Natural Sciences in Belgium. This software automatically identifies dark pixels, estimates the atmospheric dark spectrum for each band using the spectrum of the dark pixels, and estimates the atmospheric path radiation for each band. This data is then subtracted from the total signal to obtain the true reflectance signal of the surface water, completing the atmospheric correction of the turbid inland river water and obtaining the surface reflectance data of the river water. Figure 3 The diagram shows the principle of remote sensing water color inversion for suspended sediment in rivers. The method of this invention utilizes this principle, namely, the principle that changes in the concentration of suspended sediment will cause regular changes in the color (spectral characteristics) of the water body. This color change is observed through satellite sensors, and the concentration of sediment is calculated and displayed through a mathematical model.

[0029] Step 3: Collect and organize sediment monitoring data from upstream and downstream hydrological monitoring stations within this typical river section. In this embodiment, the hydrological monitoring stations used are Hankou Station and Jiujiang Station. Based on the monitoring section coordinates of Hankou Station and Jiujiang Station, draw polygon vectors, crop the river water surface reflectance data obtained in Step 2, and perform mean calculation on the extracted water surface reflectance for each band to obtain the spectral average reflectance value of the section. The average value is mainly used to eliminate random noise errors in remote sensing images. Based on the imaging time of the remote sensing images, extract the sediment data from the collected and organized sediment monitoring data at the time closest to the imaging time of the remote sensing images.

[0030] Step 4: Based on the sediment data and average spectral reflectance values ​​of the cross section obtained in Step 3 at the time closest to the remote sensing image imaging time, construct a river cross section spectral sediment dataset.

[0031] Step 5: Based on the river cross-section spectral sediment dataset established in Step 4, considering the characteristics of relatively few high values ​​and many low values ​​in the suspended sediment concentration data, select a threshold according to the histogram distribution of the measured sediment data, and classify the sediment data into low sediment content, medium sediment content and high sediment content. Randomly select 70% of each class to form the training set and 30% to form the validation set.

[0032] Step 6: Based on the river cross-section spectral sediment training set established in Step 5, a least squares support vector machine (LSVM) algorithm is used to construct an inversion model of remote sensing water color spectrum and river cross-section sediment content. Compared with the traditional support vector machine regression model, the least squares LSVM algorithm changes inequality constraints to equality constraints and replaces the insensitive loss function with the least squares loss function, transforming the optimization problem into solving a system of linear equations, avoiding complex quadratic programming solutions, and making it suitable for handling small sample datasets. The accuracy of the inversion model is evaluated using the river cross-section spectral sediment test set established in Step 5, and the model parameters are fine-tuned.

[0033] Step 7: Based on the remote sensing sediment concentration inversion model established in Step 6, perform model calculations on the remote sensing images of the entire river section and obtain the suspended sediment concentration distribution map of the entire river section.

[0034] like Figure 4 The figure shows a scatter plot of the calculated values ​​of the remote sensing sediment concentration inversion model established by the method of this invention and the measured values ​​of the entire dataset. The figure shows that the data points are closely distributed near the baseline, indicating that the calculated values ​​and measured values ​​are almost identical. The inversion model established by the method of this invention can effectively invert and display the true data situation, and the coefficient of determination (R²) is high. 2 The value of ) is 0.83349, indicating that its predictive ability is relatively strong.

[0035] like Figure 5 As shown, this is a comparison of scatter plots of the calculated values ​​of the training set and test set of the remote sensing sediment concentration inversion model established by this invention. The left side is the training set, and the right side is the test set. As can be seen from the figure, the two scatter plots are highly similar, indicating that the model has learned the pattern of the training data, has good generalization ability, and is a reliable and robust model.

[0036] like Figures 6-7 The figures show the spatial distribution maps of suspended sediment in the Hankou-Jiujiang section during the 2018 flood season and the 2020 flood season, obtained using the remote sensing sediment concentration inversion model established by the method of this invention.

[0037] The above descriptions are merely embodiments of the present invention, and common technical solutions or characteristics known in the schemes are not described in detail here. For those skilled in the art, various modifications and improvements can be made without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application shall be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A remote sensing monitoring method for river suspended sediment concentration based on data-driven and machine learning methods, characterized in that, Includes the following steps: S1. Select the typical river section with the largest annual variation in sediment content as the research object of remote sensing inversion of suspended sediment concentration, and determine the hydrological monitoring station with the closest upstream and downstream distance in the typical river section area; S2. Collect multi-year time series remote sensing image data of this typical river section, and preprocess these remote sensing image data to obtain surface reflectance data of the river water body. S3. Based on the imaging time of the collected remote sensing images, select the sediment content data of the nearest hydrological monitoring station upstream and downstream of the typical river section at the time point closest to the imaging time of the remote sensing images. S4. Based on the surface reflectance data of water bodies in typical river sections obtained in step S2 and the measured sediment concentration data of the nearest upstream and downstream hydrological stations obtained in step S3, construct a spectral sediment monitoring dataset for river sections. S5. Take the river section spectral sediment monitoring dataset established in step S4, select a threshold based on the sediment monitoring data, and classify the dataset into low, medium and high sediment content. Each level is divided into training set and test set according to the proportion. S6. The least squares support vector machine algorithm is used to train the river section spectral sediment monitoring training set constructed in step S5 to establish a machine learning-based river section suspended sediment inversion model, and the test set is used for verification. S7. Finally, using the suspended sediment inversion model established in step S6, the suspended sediment concentration of the entire river section is calculated from the remote sensing data of typical river sections, and the spatiotemporal distribution thematic map of suspended sediment in the monitored river section is obtained.

2. The remote sensing monitoring method for river suspended sediment concentration based on data-driven and machine learning according to claim 1, characterized in that: The preprocessing steps in step S2 include radiometric calibration, image registration, cloud detection, atmospheric correction, and image cropping.

3. The remote sensing monitoring method for river suspended sediment concentration based on data-driven and machine learning according to claim 1, characterized in that: In step S2, data from the European Space Agency's Sentinel-2 satellite is selected as the primary data source. Sentinel-2 L1C-level data is used, which has undergone radiometric calibration and geometric correction. Cloud detection is performed through visual interpretation to filter out images that are greatly affected by clouds and cloud shadows.

4. The remote sensing monitoring method for river suspended sediment concentration based on data-driven and machine learning according to claim 1, characterized in that: In step S5, the determination of the grading threshold is achieved by statistically analyzing the measured data of suspended sediment concentration in typical river sections, selecting an appropriate threshold, ensuring that the amount of data for each level reaches at least 10, and dividing the data for each level into a 70% training set and a 30% test set to ensure that there are at least 2 data points in the test set for each level.

Images