A method for retrieving transparency of inland water body based on MAPE-DR and improved QAA

Abstract

This invention provides a method for inland water transparency retrieval based on MAPE-DR and an improved QAA algorithm, belonging to the field of inland water transparency technology. The method includes: acquiring hyperspectral remote sensing data; processing the hyperspectral remote sensing data based on MAPE-DR; obtaining the absorption coefficient and backscattering coefficient from the processed hyperspectral remote sensing data based on the improved QAA algorithm; calculating the diffuse reflection attenuation coefficient based on the absorption coefficient and backscattering coefficient; and calculating the water transparency using the diffuse reflection attenuation coefficient. This invention can effectively improve data quality and ZSD retrieval accuracy, has broad application prospects, and provides a scientific basis for water environment monitoring and management.

Description

Technical Field

[0001] This invention relates to the field of inland water transparency technology, and in particular to an inversion method for inland water transparency based on MAPE-DR and improved QAA. Background Technology

[0002] Water transparency is an important water quality parameter in limnology and oceanography, closely related to the distribution of the light field beneath the water surface, and significant for aquatic biodiversity and productivity. Secchi disk depth (Zsd) is a crucial indicator of water transparency, typically measured by observing the depth at which a white or black-and-white transparent disk disappears into the water. Zsd directly reflects the concentration of suspended particulate matter, algae, and other pollutants in the water and is widely used in lake water quality monitoring. Changes in Zsd not only affect the health of lake and marine ecosystems but also have significant implications for environmental protection, water resource utilization, fisheries production, and tourism.

[0003] Traditional methods typically rely on large amounts of field sampling data to study Zsd distribution characteristics. However, this approach requires long-term monitoring at multiple sampling points, and the field sampling data is often spatially discontinuous, making it difficult to comprehensively reflect the Zsd distribution and dynamic changes of large-scale water bodies. Furthermore, traditional methods cannot accurately calculate the remote sensing reflectance of hyperspectral data. Therefore, it is essential to design an inland water transparency inversion method based on MAPE-DR and improved QAA. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide an inland water transparency inversion method based on MAPE-DR and improved QAA.

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

[0006] This invention provides a method for inverting inland water transparency based on MAPE-DR and improved QAA, comprising:

[0007] Acquire hyperspectral remote sensing data;

[0008] Processing hyperspectral remote sensing data based on MAPE-DR;

[0009] The improved QAA algorithm is used to obtain the absorption coefficient and backscattering coefficient from the processed hyperspectral remote sensing data. The diffuse reflection attenuation coefficient is calculated based on the absorption coefficient and backscattering coefficient, and the water transparency is calculated using the diffuse reflection attenuation coefficient.

[0010] Preferably, the hyperspectral remote sensing data is ZY1-02D ​​hyperspectral data, obtained from the Natural Resources Satellite Remote Sensing Cloud Service Platform.

[0011] Preferably, the hyperspectral remote sensing data is processed based on MAPE-DR, specifically as follows:

[0012] Atmospheric correction is performed on hyperspectral data to obtain surface reflectance data, and the surface reflectance data is converted into remote sensing reflectance based on remote sensing reflectance correction.

[0013] The remote sensing reflectance was denoised using LF, HF, MF, SGF and HA denoising methods respectively.

[0014] For each band, the MAPE values ​​of different combination methods and ASD equivalent spectra are compared, the best combination of methods is selected as the optimal band, and all the optimal bands are stacked to form a reconstructed image.

[0015] Preferably, the absorption coefficient and backscattering coefficient are obtained from the processed hyperspectral remote sensing data based on the improved QAA algorithm, specifically as follows:

[0016] An improvement upon QAAV6 and QAAL09, specifically in the following form:

[0017] r rs =R rs / (0.52+1.7R rs )

[0018]

[0019] In the formula, r rs R represents the remote sensing reflectance below the water surface. rs Represents the remote sensing reflectance above the water surface, u(λ) is r rs The function of a(λ0), where a(λ0) is the water absorption coefficient at the reference wavelength, and b bp (λ0) is the water backscattering coefficient at the reference wavelength, a w b represents the absorption coefficient of pure water. bw Let Y be the backscattering coefficient of pure water, and Y be empirical parameters, a(λ) and b(λ). bp (λ) represents the total absorption coefficient and backscattering coefficient for all wavelengths, respectively;

[0020] The QAAA optimization steps are as follows: taking advantage of the hyperspectral properties of ZY1-02D, λ0 is set to 713, 722, 730 and 739 respectively; λ2 is fixed at 748nm, and λ1 is varied between 400-800nm ​​to optimize the Y value.

[0021] Preferably, the diffuse reflection attenuation coefficient is calculated based on the absorption coefficient and the backscattering coefficient, and the water transparency is calculated using the diffuse reflection attenuation coefficient. The specific calculation method is as follows:

[0022]

[0023] In the formula, K d (λ) Diffuse reflection attenuation coefficient, This represents the minimum diffuse reflection attenuation coefficient of water within the transparent window range of 410-665nm. These are the remote sensing reflectances of the corresponding bands, a(λ) and b. bp (λ) represents the total absorption coefficient and backscattering coefficient of water, respectively, and b bw (λ) is the backscattering coefficient of pure water, θ s Let m0 be the solar zenith angle, m1 be 4.26, and m2 be 0.52.

[0024] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0025] This invention provides a method for inland water transparency retrieval based on MAPE-DR and an improved QAA algorithm. The method includes: acquiring hyperspectral remote sensing data; processing the hyperspectral remote sensing data using MAPE-DR; obtaining the absorption coefficient and backscattering coefficient from the processed hyperspectral remote sensing data using the improved QAA algorithm; calculating the diffuse reflection attenuation coefficient based on the absorption coefficient and backscattering coefficient; and calculating the water transparency using the diffuse reflection attenuation coefficient. This invention can effectively improve data quality and ZSD retrieval accuracy, has broad application prospects, and provides a scientific basis for water environment monitoring and management. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram showing the overview of the study area and the locations of the sampling points;

[0028] Figure 2 Schematic diagram of different atmospheric correction effects;

[0029] Figure 3a This is a schematic diagram of the first type of remote sensing reflectance accuracy assessment.

[0030] Figure 3b This is a schematic diagram of the second method for assessing the accuracy of remote sensing reflectance.

[0031] Figure 3c This is a schematic diagram of the third type of remote sensing reflectance accuracy assessment.

[0032] Figure 4 This is a schematic diagram illustrating the performance evaluation of the denoising algorithm.

[0033] Figure 5 This is a schematic diagram illustrating the band accuracy assessment of the MAPE-DR method.

[0034] Figure 6 This is a schematic diagram illustrating the point accuracy assessment using the MAPE-DR method.

[0035] Figure 7 A schematic diagram illustrating the accuracy evaluation of different versions of the QAA algorithm;

[0036] Figure 8 A schematic diagram of QAA722 accuracy evaluation for ZY1-02 image data;

[0037] Figure 9 A schematic diagram of the spatial distribution of Zsd in Dushan Lake;

[0038] Figure 10 A schematic diagram of the spatial distribution of Zsd in the water body of Weishan Lake;

[0039] Figure 11 This is a flowchart of a method provided in an embodiment of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] The purpose of this invention is to provide an inland water transparency inversion method based on MAPE-DR and improved QAA, which can effectively improve data quality and Zsd inversion accuracy, has broad application prospects, and provides a scientific basis for water environment monitoring and management.

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] This invention takes the South Four Lakes of Shandong Province as its research background and describes the method described in this invention in detail.

[0044] First, let's introduce the South Four Lakes: The South Four Lakes (37°27′-35°20′N, 116°34′-117°21′E) are located in southwestern Shandong Province, China. Specifically... Figure 1 As shown, the South Four Lakes is the largest freshwater lake in Shandong Province, with a total area of ​​approximately 1266 km². 2It is formed by connecting Nanyang Lake, Dushan Lake, Zhaoyang Lake and Weishan Lake from northwest to southeast. It is 125km long from north to south and 6-25km wide from east to west. The average water depth is 1.5m, the maximum water depth during the flood season is 2.5-3m, and the maximum flood control capacity is 5.4 billion m3. It is an important regulating and storage lake in the eastern route of the South-to-North Water Diversion Project, and also an important biodiversity protection area.

[0045] This invention conducted water transparency and spectral measurements in the South Four Lakes on March 31 and April 1, 2023. Sampling points were set in the relatively open Dushan Lake and Weishan Lake areas of the South Four Lakes, as shown in the figures below. Figure 1 As shown in the figure. Water transparency was measured using a Seidon disk, and a total of 82 transparency samples were collected. The transparency data are statistically analyzed as follows. Figure 1 As shown in the bar chart, the maximum transparency of the 82 sample points was 1.10m, the minimum was 0.16m, and the average was approximately 0.50m. Following NASA's water spectral measurement specifications, the "above-surface measurement method" was used for water spectral measurements. The measuring instrument was a Fieldspec 4 high-resolution spectroradiometer (spectral range 350-2500nm, 1nm interval) manufactured by ASD Corporation. A total of 46 water spectra were collected simultaneously with the transparency measurements. The weather was favorable with no wind or clouds during the sampling period. Ten spectra were measured for each item, and the average was taken to ensure measurement accuracy. The water surface remote sensing reflectance was calculated using the following formula:

[0046]

[0047] In the formula, R rs (λ) represents the water surface remote sensing reflectance, and r represents the sky light reflectance (Fresnel reflectance) at the water-air interface. Based on existing research, the value of r in this invention is set to 0.025, and ρ p For the reflectance of a standard whiteboard, π is taken as 3.14, E s The mean value of the whiteboard was used to calculate the measured R value of the water surface. rs After (λ), the measured R is expressed by the following formula. rs (λ) is converted to the satellite equivalent band:

[0048]

[0049] In the formula, RSR i (λ) represents the satellite's spectral response function (SRF), λ max and λ min R represents the wavelength range of SRF in each band. rs (λ) represents the measured remote sensing reflectance of the water surface, R rs_ (λ iThe SRF of ZY1-02D ​​is assumed to be a Gaussian function, which is often used for SRF simulation of hyperspectral satellites. The specific method is shown in the following formula:

[0050]

[0051] In the formula, λ i and Δλ i These represent the center wavelength and full-width at half-maximum (FWHM) of the i-th band, respectively. The FWHM of the AHSI sensor of the ZY1-02D ​​satellite in the visible near-infrared and short-wave infrared bands are 8.67 nm and 16.26 nm, respectively.

[0052] Figure 11 The method flowchart provided in the embodiments of the present invention is as follows: Figure 11 As shown, this invention provides a method for inverting inland water transparency based on MAPE-DR and improved QAA, comprising:

[0053] Step 100: Acquire hyperspectral remote sensing data;

[0054] Step 200: Process hyperspectral remote sensing data based on MAPE-DR;

[0055] Step 300: Based on the improved QAA algorithm, the absorption coefficient and backscattering coefficient are obtained from the processed hyperspectral remote sensing data. The diffuse reflection attenuation coefficient is calculated based on the absorption coefficient and backscattering coefficient, and the water transparency is calculated through the diffuse reflection attenuation coefficient.

[0056] In step 100, the hyperspectral remote sensing data is ZY1-02D ​​hyperspectral data, obtained from the Natural Resources Satellite Remote Sensing Cloud Service Platform. The ZY1-02D ​​data closest to the sampling time is selected, with Dushan Lake on March 27, 2023, and Weishan Lake on March 30, 2023. All images are cloud-free and of good quality. Radiometric calibration, orthorectified image cropping, and water body extraction are performed in ENVI software. For water body extraction, the Normalized Difference Water Index (NDWI) is first calculated, and then the improved histogram bimodal method (MHBM) is used to segment water bodies from other land cover types.

[0057] In step 200, the hyperspectral remote sensing data is processed based on MAPE-DR, specifically as follows:

[0058] Step 201: Perform atmospheric correction on the hyperspectral data to obtain surface reflectance data. Based on remote sensing reflectance correction, convert the surface reflectance data into remote sensing reflectance, specifically as follows:

[0059] The atmospheric correction process is described below: Atmospheric correction is used to remove interference from atmospheric absorption and scattering on water body signals, so as to effectively retrieve water quality parameters. This invention briefly introduces several atmospheric correction methods:

[0060] 1. Flaash is based on a radiative transfer model and accurately corrects atmospheric effects by inputting scene parameters (such as sensor height, solar angle, aerosol type, etc.) to handle complex atmospheric phenomena such as water vapor absorption and aerosol scattering.

[0061] 2. QUAC achieves rapid correction based on scene statistics, assuming a constant mean spectral reflectance, without requiring external atmospheric parameters;

[0062] 3. The 6S model is used to simulate atmospheric transport by satellites within the solar spectral range and is suitable for multispectral and hyperspectral data;

[0063] 4. Modtran provides a variety of standard atmospheric models for calibration by simulating the transmission process of electromagnetic radiation in the atmosphere;

[0064] This paper introduces the correction of remote sensing reflectance: Hyperspectral data after atmospheric correction is essentially surface reflectance data. In water color remote sensing, remote sensing reflectance is often used to estimate various water quality parameters. The different radiative transport processes of water vapor and land-atmosphere lead to significant differences between the two reflectances. Therefore, this invention refers to a remote sensing reflectance correction method developed by Wang et al. to convert surface reflectance into remote sensing reflectance. The specific method is as follows:

[0065]

[0066] In the formula, R represents the corrected remote sensing reflectance. rs (λ) represents the surface reflectance after processing from each data source, min(R) swir ) represents the minimum value of the shortwave infrared band, and π is taken as 3.14;

[0067] In this invention, the mean value of the visible light band minus the short-wave infrared band (1530-1630nm) is used for remote sensing reflectance correction.

[0068] The remote sensing reflectance is denoised using LF, HF, MF, SGF, and HA denoising methods, respectively:

[0069] Denoising algorithms are used to eliminate noise in spectral data and enhance signal quality. Low-pass filtering (LF) and high-pass filtering (HF) allow low-frequency and high-frequency signals to pass through, respectively, while weakening signals outside the set frequency range. Median filtering (MF) is a nonlinear filtering technique based on sorting statistics. It sorts pixels by a sliding template and takes the median to suppress noise. Savitzky-Golay filtering (SGF) is based on local polynomial least squares, which smooths the spectral curve and calculates derivatives of each order. It is suitable for removing spectral glitches caused by mechanical vibration. Harmonic analysis (HA) decomposes the spectral curve into multiple sine and cosine wave forms, and after reconstruction, it effectively removes high-frequency noise and achieves spectral smoothing.

[0070] For each band, the MAPE values ​​of different combination methods and ASD equivalent spectra are compared. The best combination of methods is selected as the optimal band, and all optimal bands are stacked to form a reconstructed image. Specifically:

[0071] In the processing of hyperspectral remote sensing data of inland water bodies, ensuring the accuracy of remote sensing reflectance and reducing errors to the greatest extent remains a challenge. To overcome this problem, this invention proposes a data processing method based on MAPE optimization and data reconstruction (MAPE-DR). Using MAPE-DR to process hyperspectral data of inland water bodies can ensure that the error between each band and the ASD equivalent spectrum is minimized. For each band, the MAPE values ​​of different algorithm combinations and the ASD equivalent spectrum are compared, and the best algorithm combination is selected as the optimal band. All optimal bands are stacked to form a reconstructed image. In simple terms, the hyperspectral remote sensing data is divided into multiple bands. For each band, atmospheric correction is performed based on four algorithms, resulting in four results. These four results are then denoised using five algorithms, resulting in a total of twenty results. The MAPE values ​​of different algorithm combinations and the ASD equivalent spectrum are compared, and the best algorithm combination is selected as the optimal band. All optimal bands are stacked to form a reconstructed image.

[0072] In step 300, the absorption coefficient and backscattering coefficient are obtained from the processed hyperspectral remote sensing data based on the improved QAA algorithm, specifically as follows:

[0073] An improvement upon QAAV6 and QAAL09, specifically in the following form:

[0074] r rs =R rs / (0.52+1.7R rs )

[0075]

[0076] In the formula, r rs R represents the remote sensing reflectance below the water surface. rs Represents the remote sensing reflectance above the water surface, u(λ) is r rs The function of a(λ0), where a(λ0) is the water absorption coefficient at the reference wavelength, and b bp (λ0) is the water backscattering coefficient at the reference wavelength, a w b represents the absorption coefficient of pure water. bw Let Y be the backscattering coefficient of pure water, and Y be empirical parameters, a(λ) and b(λ). bp (λ) represents the total absorption coefficient and backscattering coefficient for all wavelengths, respectively;

[0077] The QAAA optimization steps are as follows: taking advantage of the hyperspectral properties of ZY1-02D, λ0 is set to 713, 722, 730 and 739 respectively; λ2 is fixed at 748nm, and λ1 is varied between 400-800nm ​​to optimize the Y value.

[0078] The diffuse reflection attenuation coefficient is calculated based on the absorption coefficient and backscattering coefficient. The water transparency is then calculated using this diffuse reflection attenuation coefficient. The specific calculation method is as follows:

[0079]

[0080] In the formula, K d (λ) Diffuse reflection attenuation coefficient, This represents the minimum diffuse reflection attenuation coefficient of water within the transparent window range of 410-665nm. These are the remote sensing reflectances of the corresponding bands, a(λ) and b. bp (λ) represents the total absorption coefficient and backscattering coefficient of water, respectively, and b bw (λ) is the backscattering coefficient of pure water, θ s Let m0 be the solar zenith angle, m1 be 4.26, and m2 be 0.52.

[0081] Combining the two steps in step 300 above, explain in detail the origin of step 300:

[0082] The semi-analytical model based on the radiative transfer model and underwater visibility theory is robust in retrieving water transparency. The core principle is to calculate the diffuse reflection attenuation coefficient using the absorption coefficient and backscattering coefficient obtained from QAA, and then calculate the water transparency using the diffuse reflection attenuation coefficient. The specific formula is as follows:

[0083]

[0084] In the formula, The diffuse reflection attenuation coefficient represents the water body's transparency window in the 410-665nm range. These are the remote sensing reflectances of the corresponding bands, a(λ) and b. bp (λ) represents the total absorption coefficient and backscattering coefficient of water, respectively, and b bw (λ) is the backscattering coefficient of pure water. According to the results, this method does not require adjustment of each parameter. Therefore, it is particularly important to obtain the applicable water absorption coefficient and backscattering coefficient through the QAA algorithm. Currently, the commonly used QAA versions are QAAv5, QAAv6, QAAL09 and QAAM14. The calculation details of QAAV5 are shown in Table 1.

[0085] Table 1. Calculation steps of the QAAV5 algorithm

[0086]

[0087]

[0088] In Table 1, r rs R represents the remote sensing reflectance below the water surface. rs Represents the remote sensing reflectance above the water surface, u(λ) is r rs The function of a(λ0), where a(λ0) is the water absorption coefficient at the reference wavelength, and b bp (λ0) is the water backscattering coefficient at the reference wavelength, a w b represents the absorption coefficient of pure water. bw Let Y be the backscattering coefficient of pure water, and Y be empirical parameters, a(λ) and b(λ). bp (λ) represents the total absorption coefficient and backscattering coefficient for all wavelengths, respectively. QAAV6 is the step 2 in QAAV5 where Rrs is evaluated. If Rrs(670) > 0.0015sr -1 If the reference wavelength is changed to 670nm, the absorption coefficient is calculated using the following formula:

[0089]

[0090] QAAL09 is an image with the reference wavelength shifted to 710 nm and the backscattering coefficient of pure water ignored. The backscattering coefficient of the reference wavelength and the empirical parameter Y are calculated by the following formulas:

[0091]

[0092] QAAM14 primarily improves the calculation of the absorption coefficient at the reference wavelength, as shown in the following formula:

[0093]

[0094] This invention, based on QAAL09 and using the idea of ​​iterative optimization, proposes a Zsd inversion model applicable to the South Four Lakes region. The specific optimization involves the following two steps:

[0095] 1. Taking advantage of the high spectral resolution of ZY1-02D, shifting the reference wavelength to a longer band can improve the accuracy of QAA in inland water bodies. λ0 is set to 713, 722, 730 and 739 respectively.

[0096] 2. Fix λ2 at 748, and vary λ1 between 400-800nm ​​to optimize the Y value.

[0097] Based on specific data and experimental results obtained from this invention, the following explanation is provided:

[0098] 1. Accuracy assessment of atmospheric correction and remote sensing reflectance correction

[0099] The ZY1-02D ​​data were processed using the Flaash, QUAC, 6S, and Modtran algorithms to obtain surface reflectance data, as shown in the figure. Figure 2 As shown;

[0100] This invention uses the 400-900nm band (bands 2 to 59 of ZY1-02D) for subsequent transparency inversion. Negative Flaash values ​​appear in some bands due to over- or under-correction of atmospheric effects in these bands. QUAC exhibits more pronounced water spectral characteristics in the 404-687nm range (Band 2-35), with reflectance ranging from 0.025 to 0.08 sr. -1 The reflectance fluctuates between 696nm (Band 36) and 696nm, with the reflectance approaching 0, indicating that the algorithm performs poorly in these bands. The 6S algorithm exhibits abnormal reflectance in the first few bands, but the spectral characteristics of the water body are also relatively clear, with reflectance ranging from 0.02 to 0.08 sr. -1 The reflectivity fluctuates between 0.02 and 0.08 sr. Different locations in the Modtran algorithm show significant variations in reflectivity in the short wavelength band, with the overall reflectivity ranging from 0.02 to 0.08 sr. -1 The reflectance fluctuates between these values. The red dashed line in the figure represents the average reflectance value from 1530 to 1630 nm. It can be observed that the red dashed lines for all four algorithms are basically below the visible and near-infrared bands (404-894 nm). Therefore, we subtract the average reflectance value from the 1530-1630 nm range from the visible and near-infrared bands, and then divide by π to convert the surface reflectance to remote sensing reflectance. The result is as follows. Figures 3a-3c As shown;

[0101] Figure 3a The results show that the ZY1-02D ​​data, after remote sensing reflectance correction, better matches the ASD equivalent spectrum. Figure 3cThe MRE and RMSE before and after correction are shown. Flaash exhibits negative values ​​in multiple bands, resulting in poor correction performance. However, after 500nm, the MRE before correction is 64%-94%, and the RMSE is 0.02-0.04 sr⁻¹; after correction, the MRE is 22%-73%, and the RMSE is 0.002-0.005, with an average MRE improvement of approximately 37%. QUAC has a large number of zero reflectance values ​​in bands after 696nm, so the correction effect is only evaluated in the 404-687nm range. Before correction, the MRE is 50%-82%, and the RMSE is 0.01-0.04; after correction, the MRE is 12%-61%, and the RMSE is [missing value]. The average MRE was improved by approximately 41% with a correction of 0.002-0.006 sr-1. Before correction using the 6S algorithm, the MRE was 71%-94% and the RMSE was 0.02-0.05 sr-1. After correction, the MRE was 11%-76% and the RMSE was 0.002-0.008 sr-1. The MRE was below 50% in the 404-730nm range, with the MRE below 20% in the 490-662nm range. The average MRE improvement was 45%. There were a few outliers after Modtran correction, but the overall correction effect was significant. The MRE was below 50% in the 499-722nm range and below 30% in the 524-585nm range.

[0102] By selecting ten bands and two ratios for ZY1-02D, the accuracy of different atmospheric correction algorithms in each band was systematically evaluated. The results are as follows: Figure 3b As shown, the MAPE of the 6S algorithm at 447nm is 37%, while Modtran has the worst accuracy (71%). At 490nm, the 6S algorithm performs best (MAPE = 26%), with accuracy improvements of 16%, 6%, and 14% compared to Flaash, QUAC, and Modtran, respectively. At 550nm, the accuracy of QUAC and 6S is similar (12% and 13%), while Flaash has the lowest accuracy (23%). At 620nm, the MAPE of 6S is 15%, indicating that the 6S algorithm has the best atmospheric correction effect on the water body of Nansi Lake in the range of 447-620nm. However, in the range of 670-748nm, the accuracy of 6S decreases while the accuracy of Flaash increases, and the difference between the two increases with the longer the wavelength.

[0103] QUAC performs moderately well at short wavelengths, but its accuracy deteriorates at longer wavelengths. Modtran's accuracy drops to 151% at 739nm and 174% at 748nm. For the two ratios (490 / 560 and 709 / 665), the ratio form can improve the accuracy of a single band. For example, QUAC's MAPE is 32% at 490nm, but the MAPE drops to 17% at 490 / 550nm. Modtran's MAPE is 36% at 705nm, while the MAPE at 705 / 670nm is only 3%, effectively offsetting the common errors of different bands and reducing the influence of atmospheric scattering and absorption.

[0104] 2. Denoising Algorithm Accuracy Evaluation

[0105] The atmospheric correction results from Flaash, QUAC, 6S, and Modtran were denoised using LF, HF, MF, SGF, and HA methods, respectively. The SNR of all data was calculated using the local variance method. The results are as follows: Figure 4 As shown;

[0106] Figure 4The results show that the denoising algorithms have different effects on different atmospheric correction algorithms. The SNR of the original Flaash data shows a relatively stable upward trend between wavelengths of 400-500nm and 680-750nm, with the SNR reaching its peak at 748nm. As the wavelength increases, the signal quality after denoising by the Flaash algorithm improves. The effects of MF, HA, and SGF are similar, especially between wavelengths of 500-700nm, where the SNR curves of these methods are generally higher than those of the original data, indicating that they have a certain enhancement effect on the signal. HF has the worst effect, with the SNR remaining at a very low level across the entire band, indicating that while HF removes low-frequency noise, it also weakens the signal strength. LF shows a unique trend, maintaining a high SNR level, indicating that while LF removes high-frequency noise, it also preserves the effective signal. The SNR of QUAC is between 400-600nm. It outperforms the Flaash algorithm and has no negative values, but the SNR drops slightly in the 600-700nm range before suddenly increasing. Different denoising algorithms show significant differences in their performance on QUA. Overall, LF has the best denoising effect, HF the worst, and MF is better than SGF. The SNR curve of 6S shows large fluctuations between 400-900nm, reaching a peak near 740nm, but the drop after the peak is more significant than that of Flaash and QUAC. LF and MF show good denoising effects, indicating that these two methods have significant advantages in signal enhancement. SGF is relatively smooth, but its SNR is slightly lower than MF and LF, and it almost overlaps with the original data in some bands. Modtran's SNR performance is similar to Flaash. LF and MF significantly improve the SNR of the original data. HA and SGF have advantages in some bands and generally overlap with the original data. HF is not effective.

[0107] 3. Reconstructing Data Accuracy Assessment

[0108] The conclusions above show that different atmospheric correction algorithms have different effects on different bands in ZY1-02D ​​data, and denoising algorithms can significantly improve the signal-to-noise ratio of hyperspectral data. To improve the correction accuracy of hyperspectral imagery in inland water bodies, 10 bands required for the subsequent QAA algorithm were selected, and the MAPE-DR method proposed in this invention was used to process ZY1-02D ​​hyperspectral data. The accuracy of the reconstructed imagery was compared with that of traditional methods. Figure 5 As shown;

[0109] Figure 5It can be seen that the MAPE of the reconstructed data is the lowest globally. The MAPE of the reconstructed data at 447nm (Band7) can be reduced by up to 35%. The QUAC effect is the best at 490nm (Band12). The accuracy of the reconstructed data at 490nm can be improved by 3%-19%, and the accuracy of the reconstructed data at 620nm (Band27) can be improved by about 8%. In the 670–722 nm band (Band 27–39), QUAC, 6S, and Modtran all performed worse than Flaash. In these bands, the MAPE of the reconstructed data ranged from 23% to 39%, indicating that Flaash is more suitable for processing the red light band of the Nansi Lake. At 739 nm (Band 41), QUAC and Flaash showed little difference, but at 748 nm (Band 42), the accuracy of the reconstructed data improved by 25%, 13%, 76%, and 104% respectively compared to other methods. These results show that this method is suitable for ZY1-02D ​​data and has good transferability. In bands where certain algorithms perform poorly, the advantages of other algorithms can be used to compensate for their shortcomings. The reconstruction method significantly improves the matching accuracy between remote sensing data and the true values. Furthermore, four points were randomly selected from both the upper and lower lakes of Nansi Lake, and the MAPE and RMSE of the reconstructed data and other methods were evaluated based on the point locations. The results are as follows: Figure 6 As shown;

[0110] Figure 6 The results of the MAPE radar chart show that the MAPE of the reconstructed data is the lowest among all algorithms, especially at Point 3 and Point 8. The MAPE of the reconstructed data at Point 3 is 28%, while the MAPE of other methods is 46%, 70%, 98%, and 73%, respectively. The MAPE of the reconstructed data at Point 8 is 48%, while the MAPE of other methods is as high as 92%, 63%, 136%, and 166%. Although the MAPE of Flaash at Point 2 is close to that of the reconstructed data, its overall accuracy is still not as stable as that of the reconstruction methods. This is also because the denoising algorithm further improves the data accuracy. A similar trend is evident in the RMSE radar chart. Flaash performs close to the reconstructed data at some measurement points, but the overall RMSE is still higher than that of the reconstructed data. QUAC and Modtran perform relatively poorly. The RMSE of the reconstructed data is the lowest at Point 4, and close to that at Point 6. However, QUAC and Modtran have higher RMSEs at this point. Modtran has the highest error values ​​at multiple points. The point-based accuracy assessment further illustrates the accuracy and stability of MAPE-DR in the processing of hyperspectral remote sensing data of inland water bodies.

[0111] This invention also provides a detailed process for developing a Zsd inversion model using in-situ data from Nanshan Lake, verifying the feasibility of the method. Based on the above description, a Zsd inversion model for the four lakes of Nanshan Lake is first developed using in-situ data. Algorithms QAAV5, QAAV6, QA-L09, and QAA-M14 are constructed respectively. Comparison shows that QAA-L09 is more suitable for the four lakes of Nanshan Lake in Shandong Province. While there are some empirical steps in the QAA process, this invention does not have measured absorption coefficient and backscattering coefficient data. Therefore, referring to the research of Feng et al., measured Zsd data is used to correct the algorithm results. The results show that using a power function to correct the QAA results in the four lakes of Nanshan Lake yields higher accuracy. The correction forms for different algorithms are shown in the following formulas:

[0112]

[0113]

[0114] In the formula, The results represent the measured Zsd data after power function correction for different algorithm results, where x is the original result of the Zsd algorithm;

[0115] Accuracy comparison of different versions of QAA algorithm Figure 7 As shown, the Zsd retrieved by QAA722 has better consistency with the field measurements, and its scatter distribution more closely matches the 1:1 straight line. This algorithm also produces higher R² values. 2 With lower RMSE (0.8539 and 0.0544), QAAV5 and QAAM14 showed larger errors between their estimated Zsd values ​​and measured values. They significantly overestimated Zsd in the medium concentration range (0.4-0.6 m), and QAAV5 significantly underestimated Zsd in the 0.6-0.8 m range. QAAL09 and QAAV6 also showed good inversion results. 2 The values ​​were 0.8020 and 0.7998, respectively. QAAL09's accuracy in the low-to-medium concentration Zsd range (0.2-0.5m) approached that of QAA722, but its overall accuracy was still significantly lower. Compared to QAAV5, QAAV6, QAAL09, and QAAM14, QAA722 showed lower accuracy in Zsd retrieval in the South Four Lakes. 2 The accuracy of Zsd inversion can be improved by approximately 37%, 7%, 6% and 12% respectively, and the RMSE can be reduced by 38%, 15%, 14% and 22% respectively. In summary, the QAA722 proposed in this study has effectively improved the existing QAA algorithm, significantly improved the inversion accuracy and model robustness of Zsd, and can provide a more adaptable solution for the high accuracy requirements of Zsd inversion of the South Four Lakes.

[0116] This invention provides an evaluation method for image inversion results, specifically:

[0117] Image spectral features of all sample points were extracted from Flaash, QUAC, 6S, Modtran, and reconstructed data, respectively, and the QAA722 algorithm was constructed. Since QUAC has almost zero reflectance in the bands beyond 696nm, the correction effect is unsatisfactory, leading to a large number of negative values ​​when using QAA. Therefore, the performance of QAA722 on the other four image types was evaluated only using MAPE, RMSE, AURE, and MAE. The results are as follows. Figure 8 As shown;

[0118] Figure 8 The results show that there are significant differences in prediction accuracy among different algorithms. The QAA722 algorithm based on reconstructed data performs best in all evaluation metrics, with an RMSE of only 0.1774, which is about 37%, 42%, and 37% lower than Flaash, 6S, and Modtran, respectively. The MAPE of QAA722 based on reconstructed data is 0.2328, which is 36%, 44%, and 41% higher than other algorithms, respectively, showing a strong relative accuracy advantage. AURE and MAE also show similar trends, with reconstructed data being about 55.1% and 48.4% lower than 6S, about 46% and 44% lower than Flaash, and about 51% and 44% lower than Modtran.

[0119] In summary, reconstructed data demonstrates significant advantages in ZSD inversion of the Nansi Lake water body, improving inversion accuracy and reducing errors. The 6S algorithm performs the worst, particularly on MAPE and AURE. Flaash and Modtran achieve approximately 10% improvement in inversion accuracy compared to 6S. The QAA722 algorithm based on reconstructed data is the optimal choice for ZSD inversion of the Nansi Lake water body in this analysis, exhibiting strong prediction accuracy and stability. For future research, especially in water body inversion under complex water quality and illumination conditions, the QAA722 algorithm is recommended as a priority. Applying the proposed QAA algorithm to the reconstructed data yields the spatial distribution map of the Nansi Lake water body. The image inversion results for Dushan Lake are shown below. Figure 9 As shown, the results for Weishan Lake are as follows: Figure 10 As shown;

[0120] Figure 9It can be seen that the Zsd of Dushan Lake exhibits a significant spatial distribution difference. The central area of ​​the lake has a relatively high Zsd, while the Zsd values ​​in the southern area are mostly concentrated above 0.9m, reflecting a relatively clear water condition. However, near the lake edge, the Zsd decreases significantly, especially in areas with more frequent human activities (such as the northwest and eastern coast), where the transparency value drops below 0.4m. The lake center is far from the direct impact of human activities, such as agricultural runoff, domestic sewage, and industrial wastewater discharge, so the water pollution level is relatively low, and the Zsd is higher. In addition, the greater water depth in the central area of ​​the lake and the faster settling speed of suspended solids also contribute to the increase in Zsd level. The lake edge area is more affected by human activities, with agricultural irrigation, aquaculture activities, and domestic sewage being directly discharged into the lake, exacerbating the input of nutrients and leading to eutrophication. At the same time, the shallower water depth at the lake edge, the disturbance of bottom sediment, and the growth of algae also have an adverse effect on Zsd.

[0121] Figure 10 The results showed that the Zsd distribution in Weishan Lake was more complex and the overall level was lower. Zsd was mainly concentrated between 0.1 and 0.5 meters, with the Zsd in the central area generally lower than that of Dushan Lake, mostly between 0.35 and 0.55 meters. Along the shore, Zsd was mainly below 0.3 meters, especially in the southeastern coastal area. From a natural perspective, Weishan Lake has a larger area and weaker hydrodynamic conditions, making it difficult for suspended solids to settle, thus exacerbating the overall turbidity. From a human activity perspective, the shores of Weishan Lake are densely populated with extensive farmland and industrial areas, making agricultural irrigation drainage, domestic sewage, and industrial wastewater the main sources of pollution. In particular, agricultural activities in the northwest and urban wastewater in the southeast have a significant impact on lake water quality. The large input of nutrients leads to eutrophication, and the proliferation of algae further reduces Zsd. In addition, aquaculture activities and localized sand mining along the shores of Weishan Lake also exacerbate the deterioration of the lake's Zsd.

[0122] This invention summarizes and generalizes its findings. Taking the South Four Lakes of Shandong Province as an example, this invention uses the MAPE-DR method to process ZY1-02D ​​hyperspectral remote sensing data. The results show that the MAPE of MAPE-DR at 447nm in ZY1-02 can be reduced by up to 35%, the accuracy at 490nm can be improved by 3%-19%, the accuracy at 620nm can be improved by about 8%, and at 748nm, the accuracy of MAPE-DR can be improved by 25%, 13%, 76%, and 104% respectively compared with other methods. MAPE-DR is based on AS... MAPE of D-equivalent spectra, by selecting the optimal algorithm combination for each band, achieves a higher degree of matching between reconstructed data and real spectral data, and demonstrates robustness in both band and spatial positioning assessment. Based on the existing QAA algorithm, this paper develops a QAA722 algorithm suitable for Zsd inversion in the South Four Lakes by using iterative optimization to shift the reference wavelength to a longer band and improve the empirical parameter Y value. Compared with QAAV5, QAAV6, QAAL09, and QAAM14, QAA722 shows better performance in Zsd inversion of the South Four Lakes. 2 The improvements are approximately 37%, 7%, 6%, and 12%, respectively, while the RMSE is reduced by 38%, 15%, 14%, and 22%, respectively. In inland water Zsd inversion, this invention can effectively improve data quality and Zsd inversion accuracy, has broad application prospects, and provides a scientific basis for water environment monitoring and management.

[0123] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0124] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for inverting inland water transparency based on MAPE-DR and improved QAA, characterized in that, include: Acquire hyperspectral remote sensing data; The hyperspectral remote sensing data is processed based on MAPE-DR, specifically as follows: Atmospheric correction is performed on hyperspectral data to obtain surface reflectance data, and the surface reflectance data is converted into remote sensing reflectance based on remote sensing reflectance correction. The remote sensing reflectance was denoised using LF, HF, MF, SGF and HA denoising methods respectively. For each band, the MAPE values ​​of different combination methods and ASD equivalent spectra are compared, the best combination of methods is selected as the optimal band, and all the optimal bands are stacked to form a reconstructed image; Based on the improved QAA algorithm, the absorption coefficient and backscattering coefficient are obtained from the processed hyperspectral remote sensing data. The diffuse reflection attenuation coefficient is then calculated from these coefficients, and the water transparency is obtained from the diffuse reflection attenuation coefficient. Specifically: An improvement upon QAAV6 and QAAL09, specifically in the following form: r rs =R rs / (0.52+1.7R rs ) In the formula, r rs R represents the remote sensing reflectance below the water surface. rs Represents the remote sensing reflectance above the water surface, u(λ) is r rs The function of a(λ0), where a(λ0) is the water absorption coefficient at the reference wavelength, and b bp (λ0) is the water backscattering coefficient at the reference wavelength, a w b represents the absorption coefficient of pure water. bw Let Y be the backscattering coefficient of pure water, and Y be empirical parameters, a(λ) and b(λ). bp (λ) represents the total absorption coefficient and backscattering coefficient for all wavelengths, respectively; The QAAA optimization steps are as follows: taking advantage of the hyperspectral properties of ZY1-02D, λ0 is set to 713, 722, 730 and 739 respectively; λ2 is fixed at 748nm, and λ1 is varied between 400-800nm ​​to optimize the Y value.

2. The method according to claim 1, characterized in that, The hyperspectral remote sensing data is ZY1-02D ​​hyperspectral data, obtained from the Natural Resources Satellite Remote Sensing Cloud Service Platform.

3. The method according to claim 2, characterized in that, The diffuse reflection attenuation coefficient is calculated based on the absorption coefficient and backscattering coefficient. The water transparency is then calculated using this diffuse reflection attenuation coefficient. The specific calculation method is as follows: In the formula, K d (λ) Diffuse reflection attenuation coefficient, This represents the minimum diffuse reflection attenuation coefficient of water within the transparent window range of 410-665nm. These are the remote sensing reflectances of the corresponding bands, a(λ) and b. bp (λ) represents the total absorption coefficient and backscattering coefficient of water, respectively, and b bw (λ) is the backscattering coefficient of pure water, θ s Let m0 be the solar zenith angle, m1 be 4.26, and m2 be 0.52.

Images