Water quality parameter inversion method for shallow water area based on substrate spectrum deduction in dry season

Abstract

The application discloses a water quality parameter inversion method for shallow water area based on substrate spectrum deduction in dry season, and relates to the technical field of remote sensing water environment monitoring. In view of the problem that the reflection of riverbed substrate in shallow water area penetrates the water surface and mixes into the water body spectrum, and causes the water quality parameters such as chlorophyll a to be abnormally high, the first multi-spectral image of the target river section in the dry season is obtained as the substrate background, and the second multi-spectral image in the ecological water supplement flood season is spatially aligned and the shallow water area is extracted; the double-pass attenuation coefficient matrix of the water body to the riverbed substrate spectrum is calculated according to the water depth; the substrate reflectivity is multiplied by the attenuation coefficient to obtain the substrate interference component penetrating the water surface; the component is pixel by pixel deducted from the image in the flood season to obtain the pure water body spectrum, which is input into an inversion model to calculate the water quality parameters. The application utilizes the phase difference between the flood season and the dry season, and physically deducts the exposed riverbed image as the substrate background, so that the high value in the shallow water area inversion is eliminated without radiation transmission modeling.

Description

Technical Field

[0001] This invention relates to the field of remote sensing water environment monitoring technology, and in particular to a method for inverting water quality parameters in shallow water areas based on the spectral subtraction of bottom sediment during the dry season. Background Technology

[0002] Using satellite or airborne multispectral remote sensing to retrieve water quality parameters from rivers and lakes has become an important means of obtaining water quality indicators such as chlorophyll a concentration and turbidity over a large area and quickly. The basic principle is that the water spectrum received by the sensor is related to the concentration of components such as suspended solids and chlorophyll in the water. By establishing an inversion model between the spectrum and water quality parameters, the spatial distribution of water quality parameters can be retrieved from remote sensing images.

[0003] When conducting ecological surveys and water quality assessments of rivers undergoing ecological restoration, many rivers undergoing restoration (such as the Hutuo River) have large areas of shallow water. When the water is clear and shallow, sunlight can penetrate the water surface and reach the riverbed substrate (such as gravel, sediment, or benthic algae). The spectral energy reflected by the substrate then penetrates the water surface again and is received by sensors.

[0004] Existing water quality inversion models typically assume that all signals received by sensors originate from suspended matter or chlorophyll within the water body. This leads to the misinterpretation of strong reflection signals from the riverbed sediment as excessively high concentrations of pollutants in the water. Consequently, the inverted water quality parameters (such as chlorophyll a concentration) show abnormally high values ​​in shallow water areas. For the same river, the inversion is normal in deep water areas, but the chlorophyll a concentration in shallow, clear water areas is much higher than the actual concentration, creating a false impression of "high pollution" that does not match reality and interfering with water quality assessment.

[0005] To eliminate the influence of sediment, existing technologies often employ complex water optical radiation transfer models, expressing shallow water reflectivity as a function of water depth, bottom albedo, and inherent optical properties of the water body, and then solving for and subtracting the contribution of sediment. These methods require the measurement of numerous inherent optical parameters of the water body (such as absorption coefficient and backscattering coefficient) or the classification of sediment types, resulting in high implementation costs. Furthermore, the parameters vary over time and across different river sections, making them difficult to widely apply in practical engineering projects.

[0006] Ultimately, existing methods treat "removing the influence of the riverbed sediment" as an inverse problem that needs to be solved within the framework of radiative transfer and with the help of a large number of measured optical quantities. However, they ignore a readily available observation condition for such ecologically recovering rivers: they have distinct wet and dry seasons. During the dry season, the riverbed is completely exposed. At this time, remote sensing images directly record the spectrum of the riverbed sediment itself, and the sediment background can be obtained without any modeling.

[0007] Therefore, a method for inverting water quality parameters in shallow water areas is needed. This method can utilize the temporal differences between wet and dry seasons, directly using images of the exposed riverbed during the dry season as the background spectrum of the sediment. Before inverting water quality during the wet season, the contribution of the sediment is physically subtracted according to the water depth, thereby eliminating the interference of the sediment on water quality inversion without relying on complex radiative transfer modeling. Summary of the Invention

[0008] To address the issues raised in the background art, such as the mixing of water spectral data with the reflection of riverbed sediment in shallow water areas, leading to abnormally high water quality inversion values, and the fact that existing radiative transfer sediment correction methods require extensive optical measurements and are difficult to implement, this invention provides a method for inverting water quality parameters in shallow water areas based on sediment spectral subtraction during the dry season. This method uses images of exposed riverbeds during the dry season as the sediment background and subtracts sediment interference by physical subtraction based on water depth attenuation.

[0009] The method described in this invention includes the following steps.

[0010] S1, acquire the first multispectral image of the target river section under the condition of low water flow and exposed riverbed, and the second multispectral image of the target river section under the condition of ecological water replenishment and high water flow.

[0011] S2, spatially align the first multispectral image with the second multispectral image, and extract the shallow water area from the second multispectral image. The shallow water area is obtained by calculating the water body index from the second multispectral image, extracting the water body, and then removing deep water areas whose depth exceeds a preset bottom-seeing depth threshold; the preset bottom-seeing depth threshold is determined based on the water transparency: if the measured transparency Z_SD is known, then h_clear = 1.5 · Z_SD; if there is no measured data, an empirical value of 1.5 to 3.0 meters can be used.

[0012] S3. Based on the water depth distribution in the shallow water area of ​​the second multispectral image, calculate the attenuation coefficient matrix of the water body on the spectrum of the riverbed bottom. The attenuation coefficient of each pixel in the shallow water area in the attenuation coefficient matrix is ​​calculated according to the formula α = e -2·K_d·h The calculation is performed, where h is the water depth and K_d is the diffuse attenuation coefficient of the water body; the diffuse attenuation coefficient K_d is obtained by any of the following methods: calculation based on an empirical formula of synchronously measured turbidity, inversion using a semi-analytical model of image band ratios, or taking 0.2–0.5 m when the water body is clear. -1 Typical values. The water depth distribution is obtained by water depth remote sensing inversion or measured water depth interpolation, and the inversion accuracy should be controlled within an average absolute error of less than 0.3 meters.

[0013] S4. Multiply the riverbed reflectance of each pixel in the first multispectral image corresponding to the shallow water area with the attenuation coefficient matrix to obtain the spectral component of the sediment interference that penetrates the water surface during the high-water season. For areas where the riverbed undergoes scouring and sedimentation changes between the dry and high-water seasons, a mask is generated by detecting the spectral differences between the two images (e.g., the difference in reflectance in the red or near-infrared bands exceeds a threshold of 0.05–0.15) or by manual interpretation. The masked area is excluded from sediment spectral subtraction, that is, the original high-water season reflectance value of the pixels within the masked area remains unchanged, and sediment subtraction is not performed.

[0014] S5. In the shallow water area of ​​the second multispectral image, the corresponding bottom sediment interference spectral components are subtracted pixel by pixel to obtain a pure water spectral image. This pure water spectral image is then input into a pre-established inversion model to calculate water quality parameters. The inversion model is an empirical regression model or machine learning model that takes the band reflectance or band ratio of the pure water spectral image as input and chlorophyll a concentration or turbidity as output. This model is trained using synchronously collected measured water quality data or adopts a validated regional model.

[0015] The present invention also provides a shallow water quality parameter inversion system based on bottom sediment spectral subtraction during the dry season, including an image acquisition module, an alignment extraction module, an attenuation calculation module, a bottom sediment component calculation module, and a subtraction inversion module, which are respectively used to perform the above steps.

[0016] The beneficial effects of this invention are as follows: First, it utilizes the distinct wet and dry seasons of such rivers, directly using measured images of the exposed riverbed during the dry season as the background spectrum of the sediment, without the need for a radiative transfer model or actual measurements of the water's inherent optical quantities. This method is lightweight and easy to promote across a wide range of river sections. Second, it applies a two-way attenuation weighting to the sediment background reflection based on the water depth in shallow areas before subtracting it, ensuring that the subtracted sediment interference matches the actual sediment component penetrating the water surface during the wet season, avoiding over- or under-subtraction. Third, by subtracting sediment interference before inputting it into the inversion model, it eliminates abnormally high values ​​of water quality parameters caused by sediment reflection in shallow areas, making the inversion results in shallow and deep areas consistent and comparable. Fourth, it uses a masking method to remove areas where the riverbed changes during wet and dry seasons, ensuring the effectiveness of the sediment background. Attached Figure Description

[0017] Figure 1 This is a flowchart of the shallow water area water quality parameter inversion method described in this invention;

[0018] Figure 2 This is a schematic diagram of the shallow water area water quality parameter inversion system described in this invention;

[0019] Figure 3 This is a schematic diagram illustrating the cause of the higher water quality inversion due to the reflection of bottom sediment in shallow water areas penetrating the water surface and mixing into the water body spectrum.

[0020] Figure 4 This is a schematic diagram illustrating the principle of the present invention, which uses images of exposed riverbeds during the dry season as the baseline and subtracts interference from the substrate by two-way attenuation based on water depth.

[0021] Figure 5 This is a schematic diagram comparing the chlorophyll a inversion values ​​in shallow water areas before and after deducting sediment interference according to the present invention.

[0022] Figure 6 This is an end-to-end data flow walkthrough diagram of a specific embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. The specific water depths, attenuation coefficients, reflectivities, and waveband values ​​used in the following embodiments are illustrative examples and do not constitute a limitation on the scope of protection of this invention. Those skilled in the art can adjust the corresponding parameters according to the actual river section and the sensor; the adjusted implementation methods still fall within the scope of protection of this invention.

[0024] First, some terms involved in this invention will be explained. The multispectral image refers to a remote sensing image acquired by a satellite or airborne multispectral sensor, containing reflectance data across multiple spectral bands. The dry season refers to the period when a target river section dries up due to seasonal water shortage, and the riverbed is completely exposed. The ecological replenishment period refers to the period when the target river section is covered by water after ecological replenishment. The riverbed substrate refers to the surface layer of the riverbed, including sand, sediment, or benthic algae. The first multispectral image is an image of the exposed riverbed during the dry season; the reflectance of each pixel is the surface reflectance of the riverbed substrate, and therefore can be used as the background spectrum of the riverbed substrate. The shallow water area refers to a shallow area where sunlight can penetrate the water and reach the riverbed, causing the substrate to reflect the sunlight and receive it from the sensor. The water depth distribution refers to the water depth at each pixel in the shallow water area. The attenuation coefficient matrix refers to the matrix composed of the attenuation coefficients of the water body at each pixel in the shallow water area for the light reflected from the riverbed bottom. The reflected light from the riverbed bottom must first travel downwards through the water column to reach the riverbed, and then upwards through the water column to reach the water surface, hence it is a two-way attenuation. The diffuse attenuation coefficient refers to the degree of light attenuation per unit water depth, which increases with increasing water turbidity. The riverbed bottom interference spectral component refers to the portion of spectral energy reflected from the riverbed bottom after two-way attenuation by the water column, which penetrates the water surface during the high-water season and is received by the sensor. The pure water spectral image refers to the spectral image reflecting only the internal components of the water body after subtracting the riverbed bottom interference spectral component from the high-water season image. The inversion model refers to a model that takes the water body spectrum as input and outputs water quality parameters such as chlorophyll a concentration or turbidity. The diffuse attenuation coefficient K_d is calculated using the green band (center wavelength approximately 550 nm). If calculated using measured turbidity (NTU), the empirical formula K_d = 0.02 · Turb + 0.1 can be used (this coefficient can be adjusted according to the regional water body type). If retrieved from remote sensing images, a semi-analysis model using the blue-green band ratio can be used. The preset bottom depth threshold h_clear is set to 1.5 · Z_SD when measured transparency Z_SD is available; otherwise, it ranges from 1.5 to 3.0 meters depending on the river's clarity, with 2.0 meters being typical. Detection of scour and sedimentation changes: The red band reflectance of two images is calculated pixel-by-pixel. If the absolute value exceeds the preset threshold (e.g., 0.1, which can be adjusted within the range of 0.05 to 0.15 based on the image signal-to-noise ratio) and is located within the river channel, the pixel is considered to have significant scour and sedimentation, and a mask is generated for removal. Correction can also be achieved through manual visual interpretation.

[0025] Figure 1 This is a flowchart of the shallow water area water quality parameter inversion method described in this invention. Figure 1 As shown, the method of the present invention includes the following steps.

[0026] Step S1 involves acquiring a first multispectral image of the target river section during the dry season when the riverbed is exposed, and a second multispectral image of the target river section during the ecological water replenishment and high-water season. This step is the foundation of this invention and will be described in detail below.

[0027] like Figure 3 As shown, the reason for the inflated water quality inversion in shallow water areas is that when the water is clear and shallow, sunlight penetrates the water surface and shines on the riverbed sediment. The spectral energy reflected by the sediment penetrates the water surface again and is received by the sensor, superimposed with the spectrum of the internal components of the water. Existing inversion models mistakenly treat this sediment reflection as a contribution from suspended matter or chlorophyll in the water, thus inverting the water quality parameters. To deduct the sediment contribution, the key is to first obtain the spectrum of the riverbed sediment itself, i.e., the sediment background. This invention does not use a radiative transfer model to infer the sediment, but instead utilizes the existing condition that such ecologically recovering rivers have distinct wet and dry seasons: during the dry season, the riverbed is completely exposed and without water cover. The reflectance recorded in the first multispectral image acquired at this time is the surface reflectance of the riverbed sediment, i.e., the sediment background spectrum, which can be obtained directly without any modeling. Then, a second multispectral image is acquired during the ecological replenishment period when the riverbed is covered by water, as the object to be inverted. As can be seen, this step transforms the inverse problem of "solving the bottom sediment problem" into the direct problem of "directly observing the bottom sediment during the dry season" by taking advantage of the temporal differences between the wet and dry seasons.

[0028] Step S2 involves spatially aligning the first multispectral image with the second multispectral image and extracting the shallow water area from the second multispectral image. Since the first and second multispectral images were acquired from the same river segment at different times, they must first be registered to the same coordinate system, ensuring that corresponding pixels correspond to the same location on the riverbed. Registration can be achieved based on ground control points or existing image registration algorithms. Subsequently, the shallow water area is extracted from the second multispectral image: first, the water body is extracted using a water body index; then, deep water areas exceeding a preset bottom-visibility threshold are removed. The remaining shallow water area, where sunlight can penetrate to the bottom and bottom reflections mix in, requires bottom subtraction. The preset bottom-visibility threshold is determined based on water transparency: if the measured transparency Z_SD is known, then h_clear = 1.5 · Z_SD; if no measured data is available, an empirical value of 2.0 meters can be used, with a larger threshold for clearer water. The registration and water extraction in this step are mature processes in the field of remote sensing and will not be elaborated upon here.

[0029] Step S3: Based on the water depth distribution in the shallow water area of ​​the second multispectral image, calculate the attenuation coefficient matrix of the water body on the riverbed bottom spectrum. This step is crucial for accurate bottom sediment subtraction; its principle is as follows... Figure 4As shown, the reflected light from the riverbed does not reach the sensor directly, but rather undergoes a two-way attenuation through the water column: sunlight first travels downwards through the water column to reach the riverbed, is reflected by the riverbed, and then travels upwards through the water column to reach the water surface. Both segments of the light path are attenuated by absorption and scattering by the water. The deeper and more turbid the water, the stronger the attenuation, and the weaker the riverbed component reaching the surface. Therefore, for each pixel in shallow water, this invention calculates its attenuation coefficient based on the water depth h and the diffuse attenuation coefficient K_d of that pixel, according to the two-way light path of the riverbed reflected light traveling downwards and upwards through the water column: Attenuation coefficient α = e -2·K_d·h The diffuse attenuation coefficient K_d can be obtained through methods including: calculation using an empirical formula based on synchronously measured turbidity, inversion using a semi-analytical model of image band ratios, or taking a value of 0.2–0.5 m when the water is clear. -1 Typical values. It is recommended that the water depth inversion accuracy be controlled to a mean absolute error of less than 0.3 meters. The attenuation coefficients of each pixel in the shallow water area constitute an attenuation coefficient matrix. The water depth distribution in the shallow water area can be obtained from water depth remote sensing inversion or from measured water depth interpolation. It can be seen that this step characterizes the proportion of sediment reflection attenuated when it passes through the water column to reach the water surface, providing a basis for subtracting sediment based on the actual penetration amount.

[0030] Step S4: Multiply the riverbed reflectance of each pixel in the first multispectral image corresponding to the shallow water area with the attenuation coefficient matrix to obtain the sediment interference spectral component that penetrates the water surface during the high-water season. The reflectance of each pixel in the first multispectral image is the reflectance of the exposed riverbed during the dry season, which is the total reflection of the sediment when there is no water cover; while during the high-water season, this part of the reflection has to be attenuated twice by the water column before reaching the water surface. Therefore, multiplying the sediment background reflectance with the attenuation coefficient obtained in S3 pixel by pixel and band by band yields the sediment interference spectral component that actually penetrates the water surface and mixes into the water spectrum during the high-water season. Furthermore, since the riverbed may undergo erosion and deposition changes between the dry and high-water seasons, the exposed period image no longer represents the sediment during the high-water season. This invention generates a mask for areas where erosion and deposition changes occur. Specifically, the reflectance difference between two images in the red or near-infrared bands can be used for detection: if the absolute difference in reflectance of a pixel exceeds a threshold (typically 0.05–0.15), it is identified as a region of scouring and silting, and a binary mask is generated. The masked area is then excluded from bottom sediment subtraction, meaning the original high-water-season reflectance value of the pixels within the masked area remains unchanged, and bottom sediment subtraction is not performed. Thus, this step converts the "dry season total reflectance" into a "high-water-season penetration component" based on water depth, obtaining the amount of bottom sediment interference to be subtracted from the high-water-season image.

[0031] Step S5: In the shallow water area of ​​the second multispectral image, the corresponding bottom sediment interference spectral component is subtracted pixel by pixel to obtain a pure water spectral image. This pure water spectral image is then input into a pre-established inversion model to calculate water quality parameters. The inversion model can be an empirical regression model (such as multiple linear regression or logarithmic regression) or a machine learning model (such as random forest) that takes the band reflectance or band ratio of the pure water spectrum as input and chlorophyll a concentration or turbidity as output. This model needs to be trained using synchronously collected measured water quality data, or a validated regional model can be used. During training, it is recommended that the cross-validation determination coefficient R² > 0.7, or other statistical significance indicators can be used. For each pixel in the shallow water area, in each spectral band, the bottom sediment interference spectral component obtained in S4 is subtracted from the reflectance during the high-water season. The remaining image is the pure water spectrum that only reflects the internal components of the water body. Considering errors or the possibility of negative values ​​after subtraction of individual pixels, negative pixels are truncated to zero. Finally, the spectral image of the purified water body after removing the sediment was input into the pre-established inversion model to calculate water quality parameters such as chlorophyll a concentration and turbidity. Since the sediment interference has been removed, the shallow water area no longer shows abnormally high values ​​of water quality parameters caused by sediment reflection, and the inversion results of the shallow water area and the deep water area are consistent and comparable.

[0032] Figure 2 This is a schematic diagram of the shallow water area water quality parameter inversion system described in this invention. Figure 2 As shown, the system of the present invention includes an image acquisition module, an alignment and extraction module, an attenuation calculation module, a sediment component calculation module, and a subtraction and inversion module connected in sequence. Each module is a corresponding functional unit for implementing the above functions; this division is illustrative and represents only one logical functional division. The image acquisition module is configured to acquire a first multispectral image during the dry season and a second multispectral image during the wet season. The alignment and extraction module is configured to spatially align the two images and extract the shallow water area. The attenuation calculation module is configured to calculate a two-way attenuation coefficient matrix based on water depth. The sediment component calculation module is configured to generate a mask for the area where the riverbed undergoes erosion and deposition changes between the dry and wet seasons, and multiply the riverbed sediment reflectance of each pixel in the first multispectral image corresponding to the shallow water area with the attenuation coefficient matrix to obtain the sediment interference spectral component penetrating the water surface during the wet season. The subtraction inversion module is configured to subtract the corresponding bottom sediment interference spectral components pixel-by-pixel in the shallow water area of ​​the second multispectral image, excluding the masked area. For the masked area, the original high-water-season reflectance value remains unchanged, and no bottom sediment subtraction is performed, resulting in a pure water spectral image. This image is then input into a pre-established inversion model to calculate water quality parameters. The modules in this system are used to execute the above method and achieve the same effect; the similarities will not be elaborated further.

[0033] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the above-described method for inverting water quality parameters in shallow water areas. The present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for inverting water quality parameters in shallow water areas. The electronic device may be a remote sensing data processing server or workstation, and the computer-readable storage medium may be a disk, optical disk, or solid-state storage device; the present invention is not limited thereto.

[0034] Example 1

[0035] The following uses a section of the Hutuo River as an example to illustrate the implementation process of the method of this invention. Figure 6 Two multispectral images were obtained for this river section: the first multispectral image was taken during the dry season when the riverbed was completely exposed, and the second multispectral image was taken during the ecological water replenishment period. Both images contain blue, green, red, and near-infrared bands.

[0036] like Figure 6 As shown, the first step corresponds to steps S1 and S2. The first multispectral image of the dry season and the second multispectral image of the wet season are registered to the same coordinate system. Water bodies are extracted from the wet season image using water body indices, and deep water areas exceeding the bottom-seeing depth threshold of 2 meters are removed to obtain the shallow water area range. In this embodiment, the synchronously measured transparency Z_SD = 1.3 meters is used, and h_clear = 1.5 · Z_SD = 1.95 meters, rounded down to 2.0 meters, is used as the bottom-seeing depth threshold.

[0037] The second step corresponds to step S3. Obtain the water depth distribution in the shallow water area. This embodiment uses a measured interpolation method: 15 depth sounding points are set up in the shallow water area, and a water depth grid is generated using Kriging interpolation. The data is then resampled to image resolution, and the average absolute error after interpolation is 0.2 meters. For a specific pixel in the shallow water area with a depth of 0.8 meters, the diffuse attenuation coefficient K_d is obtained: the turbidity of the water at this point is simultaneously collected at 20 NTU, and the empirical formula K_d = 0.02 · Turb + 0.1 = 0.5 m is used. -1 (This formula has been calibrated for this river section, R² = 0.85). The attenuation coefficient of this pixel, calculated using a two-way optical path, is e. -2·0.5·0.8 = e -0.8 ≈0.449. This calculation is performed for each pixel in the shallow water area to obtain the attenuation coefficient matrix.

[0038] The third step corresponds to step S4. The bottom reflectance of this pixel in the green band of the exposed riverbed image during the dry season is 0.15. Multiplying this by the attenuation coefficient of 0.449, the bottom interference spectral component of the green band penetrating the water surface during the wet season is 0.067. For areas with changes in scour and sedimentation: comparing the red band reflectance of the two images, it was found that the difference in most areas was less than 0.08, indicating no significant scour or sedimentation, therefore no mask was generated. If the difference in a certain area exceeded 0.1, it was manually checked and removed.

[0039] Step 4, corresponding to step S5. The reflectance of this pixel in the green band of the high-water season image is 0.12. Subtracting the spectral component of sediment interference (0.067), the reflectance of the pure water spectrum in the green band is 0.053. This subtraction is repeated for each band to obtain the pure water spectrum for this pixel. Inversion model establishment: 30 sample points were collected in the deep water area (sediment not visible), and the chlorophyll a concentration was simultaneously measured (laboratory analysis). Using the green / red band ratio after sediment subtraction as the independent variable, a logarithmic regression model was established: Chl-a = 2.5 · ln(G / R) + 3.2, with cross-validation R² = 0.81. The pure water spectrum image after sediment subtraction was input into the inversion model, and the inverted chlorophyll a concentration of approximately 5 mg / m³ was obtained for this pixel, consistent with the measured value for this river section. Figure 5 As shown, for comparison, if the substrate is not deducted and the reflectance during the wet season is directly inverted at 0.12, the inverted chlorophyll a concentration is about 15 mg / m³ because the green band reflection is raised by the substrate, which is much higher than the measured value.

[0040] To illustrate the effectiveness of this invention, the inventors selected one hundred sampling pixels in the shallow water area of ​​this river section and compared them with synchronously measured data. When directly inverting without subtracting the sediment, the average relative error between the inverted chlorophyll a values ​​and the measured values ​​in the shallow water area was approximately 120%, generally too high. Using this invention to subtract the sediment before inversion, the average relative error decreased to approximately 18%, essentially eliminating the abnormally high values ​​in the shallow water area. The above data may vary in different river sections and with different sensors; this is merely an illustrative example.

[0041] Comparative Example 1 illustrates the effect of sediment subtraction. Pixels from the shallow water area were used without sediment subtraction, and the image from the high-water season was directly inverted. Because the shallow water is crystal clear, the green band reflectance of 0.12 includes 0.067% sediment reflectance, almost half of which is present. The inversion model mistakenly treats this sediment reflectance as a contribution from chlorophyll in the water, inverting a chlorophyll a concentration of approximately 15 mg / m³, while the actual concentration is only about 5 mg / m³, creating a false impression of high pollution. Because sediment was not subtracted, the sediment reflectance in the shallow water area was incorrectly included in the water quality parameters, which is why existing methods invert shallow water areas with higher values. Therefore, the absence of sediment subtraction causes the water quality parameters in shallow water areas to be inflated by sediment reflectance.

[0042] Comparative Example 2 illustrates the effect of depth-based two-way attenuation weighted subtraction. The aforementioned pixels were taken without weighting by the attenuation coefficient; the dry season bottom reflectance of 0.15 was directly subtracted from the wet season reflectance of 0.12. Since the actual component of bottom reflection reaching the water surface after two-way attenuation by the water column is only 0.067, not 0.15, directly subtracting 0.15 would result in oversubtraction. The subtraction result is -0.03, truncated to zero, and the pure water spectrum is incorrectly flattened to zero, failing to retrieve meaningful water quality parameters. Therefore, directly subtracting the exposed period reflectance without depth-based two-way attenuation weighting leads to oversubtraction, demonstrating that calculating the attenuation coefficient based on water depth and weighting the subtraction is indispensable to this invention.

[0043] The method described in this embodiment is applicable to river sections with the following characteristics:

[0044] (1) There is a clear dry season with a large area of ​​exposed riverbed (no water accumulation or only a thin water layer).

[0045] (2) During the high-water season, the water is relatively clear, allowing the reflection from the bottom sediment to penetrate the water surface and be received by the sensor (usually requiring a diffuse attenuation coefficient K_d < 1.0 m). -1 (or water depth h < 2.5 m).

[0046] (3) The interval between the high water period and the low water period shall not exceed one year, and the riverbed shall not be subject to large-scale artificial modification or severe erosion and siltation.

[0047] If the water is highly turbid during the high-water season (e.g., high sediment content), the bottom sediment reflection is negligible, and there is no need to subtract the bottom sediment. If the riverbed is covered by vegetation during the dry season, it is not suitable to directly use the first multispectral image as the bottom sediment background. Those skilled in the art should determine the applicability based on the actual conditions of the target river section before implementation.

[0048] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for inverting water quality parameters in shallow water areas based on bottom sediment spectral subtraction during the dry season, characterized in that, include: S1, acquire the first multispectral image of the target river section under the condition of low water flow and exposed riverbed, and the second multispectral image of the target river section under the condition of ecological water replenishment and high water flow. S2, Spatially align the first multispectral image with the second multispectral image, and extract the shallow water area range in the second multispectral image; S3. Based on the water depth distribution in the shallow water area in the second multispectral image, calculate the attenuation coefficient matrix of the water body on the spectrum of the riverbed bottom. S4. Multiply the riverbed reflectance of each pixel in the first multispectral image corresponding to the shallow water area with the attenuation coefficient matrix to obtain the bottom sediment interference spectral component that penetrates the water surface during the high water season. S5. Subtract the corresponding bottom sediment interference spectral components pixel by pixel from the shallow water area of ​​the second multispectral image to obtain a pure water spectral image, and input the pure water spectral image into the pre-established inversion model to calculate water quality parameters.

2. The method according to claim 1, characterized in that, The spatial alignment is based on ground control points or an image registration algorithm to register the first multispectral image and the second multispectral image to the same coordinate system. The shallow water area is obtained by calculating the water body index from the second multispectral image to extract the water body, and then removing the deep water area where the water depth exceeds the preset bottom depth threshold.

3. The method according to claim 1, characterized in that, The attenuation coefficient of each pixel in the attenuation coefficient matrix for shallow water areas is determined by the diffuse attenuation coefficient of that pixel based on the water depth and the water volume, and by the exponential attenuation law of the two-way optical path reflected from the bottom substrate and traveling down and up through the water column. The formula for calculating the attenuation coefficient is α = e -2·K_d·h Where h is the water depth and K_d is the diffuse attenuation coefficient; The water depth distribution in the shallow water area in this second multispectral image is obtained by water depth remote sensing inversion or by interpolation of measured water depth.

4. The method according to claim 1, characterized in that, The reflectance of the riverbed substrate in each pixel of the first multispectral image is the surface reflectance of the exposed riverbed during the dry season. The method also includes generating a mask for the area where the riverbed changes due to erosion and sedimentation between the dry and wet seasons, excluding the masked area from the bottom sediment spectral subtraction, keeping the original wet season reflectance value of the pixels in the masked area unchanged, and not performing bottom sediment subtraction.

5. The method according to claim 1, characterized in that, The pixel-by-pixel subtraction refers to subtracting the corresponding bottom sediment interference spectral component from each pixel of the shallow water area in the second multispectral image in each spectral band. And the pixel values ​​that result in a negative value after subtraction are truncated to zero.

6. The method according to claim 1, characterized in that, The water quality parameters include at least one of chlorophyll a concentration and turbidity; The inversion model is an empirical regression model or a machine learning model that takes pure water spectral images as input and water quality parameters as output.

7. A system for retrieving water quality parameters in shallow water areas based on bottom sediment spectral subtraction during the dry season, characterized in that, include: The image acquisition module is used to acquire the first multispectral image of the target river section in the state of low water flow and exposed riverbed, and the second multispectral image of the target river section in the ecological water replenishment and high water period. The alignment and extraction module is used to spatially align the first multispectral image with the second multispectral image and extract the shallow water area range in the second multispectral image. The attenuation calculation module is used to calculate the attenuation coefficient matrix of the water body on the spectrum of the riverbed bottom based on the water depth distribution in the shallow water area of ​​the second multispectral image. The sediment component calculation module is used to generate a mask for the area where the riverbed changes due to scouring and deposition between the dry and wet seasons, and to multiply the riverbed sediment reflectance of each pixel in the first multispectral image corresponding to the shallow water area with the attenuation coefficient matrix to obtain the sediment interference spectral component that penetrates the water surface during the wet season. The subtraction inversion module is used to subtract the corresponding bottom sediment interference spectral components pixel by pixel in the shallow water area of ​​the second multispectral image, except for the masked area. For the masked area, the original high-water reflectance value remains unchanged and no bottom sediment subtraction is performed to obtain a pure water spectral image. This image is then input into a pre-established inversion model to calculate water quality parameters.

8. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 6.

Images