A landsat series satellite transparency remote sensing inversion and consistency correction method and system

Abstract

The present application relates to the field of environmental science and monitoring technology, in particular to a Landsat series satellite transparency remote sensing inversion and consistency correction method and system, the method comprises the following steps: obtaining Landsat series satellite reflectivity image and carrying out water body range identification, combining the lake shoreline range to construct the water body mask; the pretreated Landsat series satellite reflectivity image is obtained by pretreating water body mask; based on high-quality optical deep water pixel set and measured transparency data, a transparency remote sensing inversion model suitable for multiple sensors is constructed; the cross-sensor correction model is established through the initial transparency estimation value of the overlapping period. The present application solves the technical problems of insufficient stability of water body identification, lack of comparability of model cross-sensor results and discontinuity of long-time series data sequence in the field of lake transparency remote sensing estimation, and provides a reliable and unified, high-precision standardized technical solution for long-term and continuous monitoring of lake transparency.

Description

Technical Field

[0001] This invention relates to the technical field of environmental science and monitoring technology, specifically to a method and system for Landsat series satellite transparency remote sensing inversion and consistency correction. Background Technology

[0002] Lake transparency (SDD) is an important indicator characterizing the optical properties and ecological environment of water bodies, comprehensively reflecting changes in optical components such as suspended solids concentration, chromatic substances content, and algal abundance. Driven by both global climate change and human activities, lake water environments are undergoing significant changes, making long-term monitoring and assessment of transparency a crucial requirement for lake ecological health management and water quality assurance.

[0003] The Landsat satellite series (including Landsat 5 TM, Landsat 7 ETM+, and Landsat 8 / 9 OLI) has acquired continuous, globally covering, medium spatial resolution (30m) multispectral imagery since the 1980s, serving as a crucial remote sensing data source for long-term lake monitoring. However, due to differences in sensor type, spectral response function, imaging characteristics, and radiometric calibration, systematic biases exist in reflectance levels across different generations of Landsat data, directly impacting the continuity and comparability of transparency inversion results. Furthermore, high-altitude lakes, small to medium-sized lakes, and complex aquatic environments affected by ice, snow, or suspended particles place higher demands on water extraction and transparency inversion from Landsat data.

[0004] While existing research has proposed some remote sensing inversion methods for lake transparency, they generally suffer from the following shortcomings: 1) Insufficient water body extraction accuracy: Complex terrain background, seasonal changes in water bodies, and image noise cause errors in water body masks based on single-scene images, making it difficult to meet the long-term monitoring needs of large-scale lakes. 2) Lack of a unified multi-sensor consistency correction system: There are non-negligible systematic errors in the inversion of transparency of Landsat data from different generations, which leads to unstable transparency change trends and incomparability of cross-sensor results in long-term series analysis; 3) Limited applicability and stability: Some models rely on single scene or regional data for construction, which cannot take into account lakes with different optical types, and the cross-regional and cross-time period inversion accuracy is insufficient.

[0005] Therefore, there is an urgent need to build a technical system that can simultaneously solve the problems of water body stability identification, multi-source Landsat data consistency correction, and high-precision transparency inversion, so as to achieve long-term, stable and comparable monitoring of lake transparency on a global scale. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for Landsat series satellite transparency remote sensing inversion and consistency correction. It specifically addresses three core technical challenges in the field of lake transparency remote sensing estimation: insufficient stability of water body identification, lack of comparability of model results across sensors, and discontinuity of long-term time-series data. It provides a reliable, unified, and highly accurate standardized technical solution for long-term and continuous monitoring of lake transparency at global and regional scales.

[0007] To achieve the above objectives, a Landsat series satellite transparency remote sensing inversion and consistency correction method is provided, comprising the following steps: acquiring Landsat series satellite reflectance images and identifying water body extents to determine long-term stable permanent lake water bodies, and constructing a water body mask in conjunction with the lake shoreline extent; preprocessing the water body mask to obtain preprocessed Landsat series satellite reflectance images, which is a high-quality optical deep-water pixel set; constructing a transparency remote sensing inversion model suitable for multiple sensors based on the high-quality optical deep-water pixel set and measured transparency data; inverting the reflectance images of each sensor using the transparency remote sensing inversion model to obtain the initial transparency estimate of the corresponding sensor; establishing a cross-sensor correction model through the initial transparency estimate of the overlap period and correcting the initial transparency estimate of a single satellite, performing interpolation compensation for anomalous years, and unifying the transparency estimates of all sensors to the same scale to form a consistent transparency sequence for a predetermined period.

[0008] A technical system was developed, which includes water body stability identification, multi-source Landsat data consistency correction, and high-precision transparency inversion, to achieve long-term, stable, and comparable monitoring of lake transparency on a global scale.

[0009] Preferably, the Landsat series satellite reflectance imagery includes: TM imagery, ETM+ imagery, and OLI imagery. Specifically, the TM (Thematic Mapper) sensor is mounted on Landsat 4 / 5 satellites, the ETM+ (Enhanced Thematic Mapper Plus) sensor is mounted on Landsat 7 satellites, and the OLI (Operational Land Imager) sensor is mounted on Landsat 8 / 9 satellites, all of which can provide multispectral reflectance data.

[0010] The identification process for water body range identification is as follows: The frequency raster data of water bodies within a specified range and the vector map layer of the lake shoreline in the study area are imported into the same spatial reference frame; the frequency value of water bodies in each pixel of the raster is read and marked: pixels with a frequency value greater than or equal to a preset frequency are identified as long-term stable permanent lake water bodies and marked as candidate permanent water body pixels, while those with a frequency value less than or equal to a preset frequency are marked as non-permanent water body pixels; the candidate permanent water body pixels are spatially overlaid with the lake shoreline vector map layer, and candidate permanent water body pixels located inside or in contact with the shoreline boundary are retained, while candidate permanent water body pixels that are clearly located outside the shoreline and not connected to any lake are removed.

[0011] By introducing a dataset of water body frequency occurrences within a specified range and combining it with the lake shoreline vector boundary to conduct spatial overlay analysis and precise screening, high-quality and reliable input data support is provided for water transparency inversion.

[0012] Preferably, for candidate permanent water body pixels that cross the shoreline boundary, the pixel center point of the candidate permanent water body pixel is determined, and the pixel center point is used as the basis for classification: if the pixel center point is located within the shoreline, the pixel is determined to be a lake water body and is retained; otherwise, it is removed.

[0013] It can effectively eliminate interference from pseudo-water bodies such as wetlands, rivers, and mixed land and water pixels.

[0014] Preferably, the preprocessing method for the water mask includes at least one of the following: estimating the beam attenuation coefficient based on inherent optical characteristics, combining the beam attenuation coefficient with water depth data to obtain an optical water depth index; using the optical water depth index to mask out pixels in optically shallow water; performing an inward erosion operation on the water mask to retain pixels with pure water properties, while simultaneously filtering the connected domain area of ​​continuous water bodies to remove tiny patches with an area smaller than a predetermined number of pixels. This effectively reduces the influence of optically shallow water areas.

[0015] Preferably, the beam attenuation coefficient To utilize the absorption coefficient and particle backscattering coefficient The estimation is performed using the following formula: ,in, The particle backscattering ratio; wavelength The backscattering coefficient of pure water at that location; The backscattering ratio of pure water; the optical water depth index To be based on the beam attenuation coefficient The calculation is performed using water depth data, and the formula is as follows: ,in, This is water depth data.

[0016] It provides high-quality and reliable input data support for water transparency inversion.

[0017] Preferably, the construction process of the multi-sensor transparency remote sensing inversion model is as follows: Based on Landsat series satellite reflectance images, two types of input features are constructed for water transparency inversion; the two types of input features include: band ratio features and inherent optical characteristic parameter features; a high-quality matching dataset is constructed based on the two types of input features and the measured water transparency data; a random partitioning strategy is adopted for each sensor, and the variables in the input features are standardized and normalized, and the optimal hyperparameters are determined by combining grid search and iteration to maximize the accuracy.

[0018] It effectively enhances the robustness and cross-regional adaptability of the model, and significantly improves the estimation accuracy of water transparency.

[0019] Preferably, the process of constructing the band ratio features is as follows: extracting visible light, near-infrared and short-wave infrared reflectance image data from Landsat series satellite images, and constructing band ratio features: Blue / Green, Blue / Red, Blue / NIR, Green / Red, Green / NIR, Red / NIR; Blue represents the blue band of Landsat satellite, Green represents the green band of Landsat satellite, and Red represents the red band of Landsat satellite.

[0020] Preferably, the process of creating the inherent optical characteristic parameter features is as follows: based on remote sensing reflectance image data, calculate the band ratio characteristic wavelength set. Corresponding wavelength within Backscattering coefficient and the overall absorption coefficient The backscattering coefficient The calculation formula is as follows: ,in, wavelength Remote sensing reflectance at the location, ; wavelength Remote sensing reflectance at the location; wavelength Particle backscattering coefficient at the location; wavelength The formula for calculating the total absorption coefficient at the location is as follows: ,in, wavelength The backscattering coefficient of pure water at that location; wavelength The particle backscattering coefficient at that location.

[0021] It achieves high-precision measurement of transparency inversion, cross-sensor consistency calibration, and long-term continuous output, providing high-quality and reliable data support for the analysis of lake optical environment evolution and ecological risk assessment.

[0022] Preferably, the process of establishing the cross-sensor correction model is as follows: based on the observation data of the two satellites during the same overlapping period, the median value of the initial transparency estimate of each sensor is taken, and then the average value of the two is calculated; the initial transparency estimate of the single satellite is calibrated for deviation through the correction equation.

[0023] Achieve scale uniformity and temporal continuity in transparency estimation using multiple sensors.

[0024] On the other hand, a system for Landsat series satellite transparency remote sensing inversion and consistency correction is also provided, for implementing the aforementioned Landsat series satellite transparency remote sensing inversion and consistency correction method, including: The first module is configured to acquire Landsat series satellite reflectance images and identify water bodies to determine long-term stable permanent lake bodies, and construct a water body mask based on the lake shoreline range. The second module is configured to preprocess the water body mask to obtain preprocessed Landsat series satellite reflectance images, which is a high-quality optical deep-water pixel set. The third module is configured to construct a transparency remote sensing inversion model suitable for multiple sensors based on the high-quality optical deep-water pixel set and measured transparency data; use the transparency remote sensing inversion model to invert the reflectance images of each sensor to obtain the initial transparency estimate of the corresponding sensor. The fourth module is configured to establish a cross-sensor correction model through the initial transparency estimate of the overlap period and correct the initial transparency estimate of a single satellite, perform interpolation compensation for anomalous years, unify the transparency estimates of all sensors to the same scale, and form a consistent transparency sequence for a predetermined period.

[0025] It provides a reliable, unified, and highly accurate technical approach for long-term monitoring of lake transparency globally and regionally.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This application innovatively introduces a global water body occurrence frequency dataset and combines it with the lake shoreline vector boundary to carry out spatial overlay analysis and precise screening, which can effectively eliminate false water body interference areas such as wetlands, rivers, and mixed shore-land pixels; at the same time, by implementing inward erosion treatment on the water body mask and screening the connected domain area threshold, the spatial stability and pixel purity of the lake water body mask are greatly improved, the land proximity effect and noise interference are significantly reduced, and high-quality and reliable input data support is provided for water transparency inversion; 2. By estimating key optical parameters through an optical model and combining them with water depth data, shallow water pixels can be accurately identified and eliminated, avoiding transparency estimation errors caused by bottom reflection. This method improves the physical meaning and stability of transparency inversion, and is particularly suitable for plateau lakes, shallow lakes, and various types of optically sensitive water bodies. 3. Based on the multi-source reflectance and measured transparency data of Landsat TM, ETM+, and OLI, a unified training dataset is constructed, and multi-dimensional features such as spectral ratio, absorption coefficient, and backscattering coefficient are introduced. Combined with machine learning models and spatial cross-validation strategies, the inversion model has stronger robustness and cross-regional adaptability, and the transparency estimation accuracy is significantly improved. 4. A cross-sensor correction equation was constructed using the median mean method over overlapping observation periods, and this equation was extended to single-satellite periods. Simultaneously, a neighboring-year fusion correction strategy was employed for years with discontinuous observations to achieve scale uniformity and temporal continuity in the transparency estimates from multiple sensors. This ultimately resulted in a highly consistent transparency sequence product. 5. This application takes into account different lake optical types, geographical environments, and interannual variations, and is applicable to complex water environments such as lakes, reservoirs, and high-altitude, high-turbidity water bodies. The generated transparency products have high spatial resolution and strong cross-sensor consistency, and can directly serve water quality assessment, ecological change analysis, and lake environmental monitoring. Furthermore, the technical system constructed in this application completely solves the key problems in remote sensing estimation of lake transparency: "difficulty in stable identification of water bodies—incomparability of models across sensors—discontinuity of long-term sequences," providing a reliable, unified, and high-precision technical route for long-term monitoring of global and regional lake transparency. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a diagram showing the construction and verification of the sensor transparency model corresponding to Landsat TM, ETM+, and OLI in this invention. Figure 3 This is a spatiotemporal cross-validation diagram of the transparency models of the transducers corresponding to Landsat TM, ETM+, and OLI in this invention; Figure 4 This is a consistency correction diagram of the sensor transparency time series results corresponding to Landsat TM, ETM+, and OLI in this invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. The embodiments described below 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.

[0029] Example 1 like Figure 1 As shown, this invention provides a Landsat series satellite transparency remote sensing inversion and consistency correction method, which is used to solve the technical system of water body stability identification, multi-source Landsat data consistency correction, and high-precision transparency inversion, so as to achieve long-term, stable and comparable monitoring of lake transparency on a global scale, including the following steps: This method acquires reflectance images from Landsat satellites and identifies water bodies to determine long-term stable, permanent lakes. A water body mask is then constructed based on the lake shoreline extent. In this embodiment, the Landsat satellites continuously provide global surface remote sensing data. Water body extents are extracted from TM, ETM+, and OLI reflectance images from the Landsat satellites using a global water body occurrence frequency dataset. Specifically, the TM (Thematic Mapper) sensor is mounted on Landsat 4 / 5 satellites, the ETM+ (Enhanced Thematic Mapper Plus) sensor on Landsat 7 satellites, and the OLI (Operational Land Imager) sensor on Landsat 8 / 9 satellites, all of which provide multispectral reflectance data. The water body occurrence frequency is used to determine permanent lake bodies, and a water body mask is constructed based on the lake shoreline extent.

[0030] Preprocessing the water mask yields preprocessed Landsat satellite reflectance images, which are high-quality optical deep-water pixel sets. This preprocessing removes pixels from optically shallow water areas and small lake areas with insufficient effective pixels.

[0031] Based on the high-quality optical deep-water pixel set and measured transparency data, a transparency remote sensing inversion model suitable for multiple sensors is constructed. This model is then used to invert the reflectance images of each sensor, obtaining the initial transparency estimates for that sensor. By constructing and evaluating the transparency inversion model, a consistent transparency inversion method applicable to different Landsat sensors is established, providing a foundation for subsequent unified multi-sensor products.

[0032] A cross-sensor calibration model was established using the initial transparency estimates during the overlap period, and the initial transparency estimates of individual satellites were corrected. Interpolation compensation was performed for anomalous years, unifying the transparency estimates of all sensors to the same scale, forming a consistent transparency sequence for a predetermined period. In years with observational anomalies or significantly reduced observation frequency, correction was performed by averaging the calibration results of adjacent years with the original value of that year to restore the sequence continuity. Systematic calibration and scale unification of the transparency estimates of the TM, ETM+, and OLI sensors ensured the temporal consistency of the multi-sensor transparency (SDD) inversion results from the Landsat series satellites.

[0033] Furthermore, the process for identifying the water body range is as follows: The frequency raster data of water bodies within a specified range and the vector map layer of lake shorelines in the study area are imported into the same spatial reference frame. In this embodiment, the specified range refers to the global range, that is, the frequency raster data of water bodies within the global range (spatial resolution 30m) and the vector map layer of lake shorelines in the study area are imported into the same spatial reference frame, and the frequency value of water bodies (0-100%) of each pixel in the raster is read.

[0034] The frequency of water bodies in each pixel of the raster is read and marked: pixels with a frequency greater than or equal to a preset frequency are identified as long-term stable permanent lake water bodies and marked as candidate permanent water body pixels. In this embodiment, the preset frequency is 75%, that is, pixels with a frequency ≥ 75% are identified as long-term stable permanent lake water bodies and marked as candidate permanent water body pixels, and pixels with a frequency < 75% are marked as non-permanent water body pixels. Non-permanent water body pixels can also be temporary water body pixels. A water body mask is constructed in combination with the lake shoreline range. That is, the water body mask is a binary image used to identify water body areas in remote sensing image processing. The algorithm marks water body pixels (such as lakes and rivers) as 1 and temporary water bodies (such as land and clouds) as 0 for subsequent quantitative analysis.

[0035] Subsequently, the candidate permanent water body pixels are spatially overlaid with the lake shoreline vector layer for analysis. Candidate pixels located inside or in contact with the shoreline boundary are retained, while isolated pixels that are clearly located outside the shoreline and not connected to any lake are removed to avoid misclassification of wetland, river, or noise pixels.

[0036] Furthermore, for candidate permanent water body pixels that cross the shoreline boundary, the pixel center point of the candidate permanent water body pixel is determined, and the pixel center point is used as the basis for classification: if the pixel center point is located within the shoreline, the pixel is identified as a lake water body and retained; otherwise, it is discarded. To further reduce the proximity effect caused by shore-land mixed pixels, an inward erosion operation of one pixel (30 m) is performed on the initially obtained permanent water body mask to remove shore-land mixed pixels at the mask edge and eliminate interference from the proximity effect, ensuring that the finally retained pixels are pure water body attribute pixels. At the same time, the connected domain area is used to screen continuous water body pixels. Continuous water body pixels refer to the set of water body pixels that meet the water body identification conditions in the raster space and are spatially connected by pixel edges or corners. The area of ​​these connected domains is screened to remove small patches with an area of ​​less than 4 pixels, thereby obtaining the effective water body pixels for each lake.

[0037] Furthermore, the preprocessing method for the water body mask includes: estimating the beam attenuation coefficient based on inherent optical properties, combining the beam attenuation coefficient with water depth data to obtain an optical water depth index, and using the optical water depth index to mask out pixels identified as optically shallow water. When identifying water bodies using the acquired Landsat series satellite reflectance images, in order to eliminate the interference of optically shallow water areas on remote sensing reflectance, i.e., when light penetrates to the bottom of the water, causing the observed reflectance to include contributions from both the water column and the bottom sediment, we use the following method to remove pixels identified as optically shallow water. To weaken the influence of optically shallow water areas, based on inherent optical properties, i.e., optical properties determined only by the composition of the water body itself and not changing with illumination conditions or observation geometry (solar altitude angle, sensor observation angle), we calculate the beam attenuation coefficient of the band and combine it with water depth data to obtain an optical water depth index, and use the optical water depth index to mask out pixels identified as optically shallow water.

[0038] Furthermore, beam attenuation coefficient To utilize the absorption coefficient and particle backscattering coefficient The estimation is performed using the following formula: , in, The particle backscattering ratio; wavelength The backscattering coefficient of pure water at that location; This represents the backscattering ratio of pure water.

[0039] The optical depth index To be based on the beam attenuation coefficient The calculation is performed using water depth data, and the formula is as follows: , in, This is water depth data.

[0040] In this embodiment, the wavelength is taken as... At that time, beam attenuation coefficient To utilize the absorption coefficient and particle backscattering coefficient Perform estimation and calculate wavelength Beam attenuation coefficient The process is as follows: , Among them, the backscattering ratio of pure water Set to 0.5; Particle backscattering ratio Set it to 0.015.

[0041] Then, based on the beam attenuation coefficient and water depth data calculation Optical water depth index The process is as follows: .

[0042] in, c (547) is the beam attenuation coefficient (unit: m⁻¹) for the 547 nm band (green light). The depth is represented by the water depth (in meters). Therefore, the optical depth index OD(547) is dimensionless, reflecting the overall attenuation of 547 nm light by the water. A smaller OD(547) indicates limited light attenuation before reaching the bottom, suggesting that bottom reflection may significantly affect remote sensing reflectivity. This paper defines pixels with an optical depth index less than 20 as shallow optical water areas and performs masking to remove pixels in these areas, reducing the interference of bottom reflection on the water's optical parameter inversion results.

[0043] By estimating key optical parameters using an optical model and combining this with water depth data, shallow optical pixels can be accurately identified and eliminated, avoiding transparency estimation errors caused by bottom reflection. This method improves the physical meaning and stability of transparency inversion, and is particularly applicable to plateau lakes, shallow lakes, and various types of optically sensitive water bodies.

[0044] Wherein, absorption coefficient With particle backscattering coefficient The calculation process is as follows: , , , in, wavelength The pure water absorption coefficient; For intermediate process variables; wavelength The remote sensing reflectance; wavelength Remote sensing reflectance.

[0045] Based on the calculated absorption coefficient Calculate the particle backscattering coefficient The calculation process is as follows: , , in, wavelength The backscattering coefficient of pure water; For intermediate process variables; wavelength The remote sensing reflectance.

[0046] In this embodiment, the absorption coefficient is used. and particle backscattering coefficient For example, the calculation process is as follows: , , In the formula, 469 nm and 547 nm correspond to the blue and green light bands of the Landsat satellite, respectively. These are the remote sensing reflectance ratios for the blue and green light bands, respectively.

[0047] , Based on the calculated absorption coefficient Calculate the particle backscattering coefficient The process is as follows: , .

[0048] Furthermore, the process of constructing the multi-sensor transparency remote sensing inversion model is as follows: Based on Landsat series satellite reflectance images, two types of input features are constructed for water transparency inversion; the two types of input features include: band ratio features and inherent optical property parameter features; 1) The process of constructing the band ratio feature is as follows: extract the reflectance image data of visible light, near infrared and shortwave infrared bands from the images of the Landsat series satellites, and construct the band ratio feature. The Landsat series satellites include Landsat 5, Landsat 7 and Landsat 8. There are six ratio features, namely: Blue / Green, Blue / Red, Blue / NIR, Green / Red, Green / NIR and Red / NIR. Among them, Blue represents the blue band of Landsat satellites, Green represents the green band of Landsat satellites, and Red represents the red band of Landsat satellites.

[0049] 2) The process of creating inherent optical characteristic parameters is as follows: Calculate the band ratio characteristic wavelength set based on remote sensing reflectance image data. Corresponding wavelength within Backscattering coefficient and the overall absorption coefficient This forms a feature set of inherent optical characteristic parameters, including the backscattering coefficient. The calculation formula is as follows: , in, wavelength Remote sensing reflectance at the location, ; wavelength Remote sensing reflectance at that location.

[0050] wavelength Total absorption coefficient at the location The calculation formula is as follows: , in, wavelength The backscattering coefficient of pure water at that location; wavelength The particle backscattering coefficient at that location.

[0051] Based on remote sensing reflectance The converted value is an intermediate process quantity, and its calculation formula is as follows: , in, wavelength Remote sensing reflectance at that location.

[0052] In this embodiment, wavelength The absorption coefficients are respectively taken as ( , , The backscattering coefficients are respectively taken as ( , , ). , These represent the blue, green, and red bands of the Landsat satellite, respectively. Absorption coefficient. and backscattering coefficient The calculation process has been given above. Based on this, the backscattering coefficient... , The calculation formula is as follows: , , When wavelength or Total absorption coefficient at the location The calculation formula is as follows: ,or , In the formula or For intermediate process quantities, the calculation formula is as follows: ,or .

[0053] A high-quality matching dataset is constructed based on the two types of input features and measured water transparency data. In this embodiment, a high-quality matching dataset is constructed based on the above two types of input parameters of Landsat 5, 7, and 8 and measured transparency data. The consistency and reliability of the data are improved by using a ±1 day time difference and a 3×3 pixel window low variability screening strategy.

[0054] like Figure 2 As shown, a random partitioning strategy is adopted for each sensor, and the variables in the input features are standardized and normalized. Grid search and iteration are combined to determine the optimal hyperparameters to maximize accuracy. In this embodiment, for each sensor, a random partitioning training / testing strategy is used 10 times (70% training, 30% testing), and all input variables are preprocessed with standardization and normalization to ensure consistent dimensions for different features. Furthermore, grid search and 100 rounds of training iterations are combined to determine the optimal model hyperparameters, as shown in Table 1, to maximize the accuracy of the inversion model.

[0055] Table 1: List of Optimal Hyperparameters like Figure 3As shown, to ensure the model's stability globally, a spatial cross-validation method based on geospatial clustering is employed to analyze the model's adaptability to different geographical environments. Model performance is comprehensively evaluated using mean relative error (MRE), root mean square error (RMSE), regression slope (β), and coefficient of determination (R²) to ensure the model has reliable generalization capabilities.

[0056] Furthermore, the process of establishing the cross-sensor calibration model is as follows: like Figure 4 As shown, based on observation data from the same overlapping period of two satellites, the median value of the initial transparency estimates from each sensor is taken, and then the average of the two values ​​is calculated. In this embodiment, during the overlap period of the two satellites, the median values ​​of the transparency estimates from TM and ETM+ are averaged to obtain the consistency benchmark value for that year, ensuring that the water transparency for that year can be consistently adjusted. Taking the median value can avoid SDD anomalies caused by image noise and local water anomalies (such as sudden changes in suspended matter) during single-sensor inversion, making the representative value of a single sensor more robust. Averaging the median value can balance the system bias of the two sensors, allowing the fusion result to take into account the characteristics of both TM and ETM+ data, avoiding the bias of a single sensor data from dominating the final result. The calculation method is as follows: , Where y represents the year; This is the calibrated transparency estimate; The median value of the transparency estimate obtained from TM; The median value of the transparency estimate obtained from ETM+; Based on this, a cross-sensor correction equation is established: , in, The estimated transparency value of the TM sensor after correction; This is the estimated transparency value of the TM sensor before calibration.

[0057] Taking 2000 as an example, the calculation method is as follows: , in, Transparency estimates were calibrated in 2000. : Median of transparency estimates obtained by TM in 2000; : Median of transparency estimates obtained by ETM+ in 2000.

[0058] The initial transparency estimates of the single satellite were calibrated for bias using a correction equation. For the transparency estimates obtained from the Landsat 5 single satellite between 1995 and 1999, the correction equation was extended to the single-satellite stage for bias calibration, systematically correcting the early Landsat 5 transparency estimates to ensure consistency across sensor time series.

[0059] For years with observed anomalies or significantly reduced observation frequency, correction is performed by averaging the calibration results of adjacent years with the original value of that year to restore the continuity of the sequence. The calculation method is as follows: , in, This is the estimated transparency value after calibration that year; The transparency estimate is based on the calibration from the previous year. This is an estimated transparency value after calibration for the following year; This is the median value of the transparency estimate obtained from the Landsat 7 ETM+ sensor back then; Taking 2012 as an example, the calculation method is as follows: , in, This is a calibrated estimate of transparency for 2012; This is a calibrated estimate of transparency for 2011; This is a calibrated estimate of transparency for 2013; This is the median value of the transparency estimate obtained from the Landsat 7 ETM+ sensor inversion in 2012.

[0060] Ultimately, the transparency estimates from all sensors were transformed to a uniform scale, resulting in a highly consistent long-term transparency dataset spanning 1995–2023.

[0061] A cross-sensor correction equation was constructed using the median mean method over overlapping observation periods, and then extended to single-star periods. Meanwhile, a neighboring year fusion correction strategy was adopted for years with discontinuous observations to achieve scale uniformity and temporal continuity of the transparency estimates from multiple sensors, ultimately forming a highly consistent transparency sequence product from 1995 to 2023.

[0062] This method takes into account different lake optical types, geographical environments, and interannual variations, and is applicable to complex water environments such as lakes, reservoirs, and high-altitude, high-turbidity water bodies. The generated transparency products have high spatial resolution and strong cross-sensor consistency, and can directly serve water quality assessment, ecological change analysis, and lake environmental monitoring.

[0063] Example 2 The difference from Example 1 is that the pretreatment method for pretreating the water mask further includes: An inward erosion operation is performed on the water mask to retain pixels with pure water properties. Simultaneously, the connected domain area of ​​continuous water bodies is screened, removing tiny patches with an area smaller than a predetermined number of pixels. In this embodiment, to reduce the land proximity effect, the water mask is etched inward by one pixel. Lakes with an area less than four effective 30×30m pixels are also excluded from the analysis, thereby obtaining a high-quality optical deep-water pixel set that meets the transparency inversion requirements.

[0064] Example 3 A Landsat series satellite transparency remote sensing inversion and consistency correction system is also provided to implement the Landsat series satellite transparency remote sensing inversion and consistency correction method described in Example 1, including: The first module is configured to acquire Landsat series satellite reflectivity images and identify water bodies, determining long-term stable permanent lake water bodies, and constructing a water body mask based on the lake shoreline extent. By introducing a global water body occurrence frequency dataset and combining it with the water body mask constructed from the lake shoreline extent, the vector boundaries of the lake shoreline are spatially overlaid and filtered to effectively eliminate false water body areas such as wetlands, rivers, and mixed shoreline pixels.

[0065] The second module is configured to preprocess the water mask to obtain preprocessed Landsat series satellite reflectivity images, which are high-quality optical deep-water pixel sets. By implementing inward erosion processing and connected domain area threshold screening on the water mask, the spatial stability and pixel purity of the lake water mask are significantly improved, the land proximity effect and noise interference are significantly reduced, and high-quality, high-reliability input data support is provided for water transparency inversion.

[0066] The third module is configured to construct a transparency remote sensing inversion model suitable for multiple sensors based on the high-quality optical deep-water pixel set and measured transparency data. This model is then used to invert the reflectivity images of each sensor, obtaining the initial transparency estimate for that sensor. By estimating key optical parameters through the optical model and combining them with water depth data, optical shallow-water pixels can be accurately identified and eliminated, avoiding transparency estimation errors caused by bottom reflection. This method improves the physical meaning and stability of transparency inversion, and is particularly suitable for plateau lakes, shallow lakes, and various types of optically sensitive water bodies.

[0067] The fourth module is designed to establish a cross-sensor correction model using initial transparency estimates from the overlap period and to correct the initial transparency estimates for individual stars. It performs interpolation compensation for anomalous years, unifying the transparency estimates from all sensors to the same scale, thus forming a consistent transparency sequence for a predetermined period. A cross-sensor correction equation is constructed using the median mean method of the overlap observation period and extended to the individual star period. Simultaneously, a neighboring year fusion correction strategy is employed for years with discontinuous observations, achieving scale uniformity and temporal continuity in multi-sensor transparency estimates. The final result is a highly consistent transparency sequence product.

[0068] Embodiments of the present invention have been shown and described. It will be apparent to those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for Landsat series satellite transparency remote sensing inversion and consistency correction, characterized in that, Includes the following steps: Acquire Landsat series satellite reflectivity images and identify water body extent to determine long-term stable permanent lake water bodies, and construct water body masks in combination with lake shoreline extent; The water mask is preprocessed to obtain the preprocessed Landsat series satellite reflectivity image, which is a high-quality optical deep-water pixel set. Based on the high-quality optical deep-water pixel set and measured transparency data, a transparency remote sensing inversion model suitable for multiple sensors is constructed; the reflectance images of each sensor are inverted using the transparency remote sensing inversion model to obtain the initial transparency estimate of the corresponding sensor. A cross-sensor calibration model is established by using the initial transparency estimates during the overlap period and correcting the initial transparency estimates of individual stars. Interpolation compensation is performed for anomalous years, and the transparency estimates of all sensors are unified to the same scale to form a consistent transparency sequence for a predetermined period.

2. The Landsat series satellite transparency remote sensing inversion and consistency correction method according to claim 1, characterized in that, The Landsat series satellite reflectivity images include: TM images, ETM+ images, and OLI images; The identification process for the water body range is as follows: Import the raster data of water body frequency within the specified range and the vector map layer of lake shoreline in the study area into the same spatial reference frame; Read the water occurrence frequency value of each cell in the raster and mark it: cells with occurrence frequency values ​​greater than or equal to the preset frequency are identified as long-term stable permanent lake water bodies and marked as candidate permanent water body cells, and vice versa are marked as non-permanent water body cells. The candidate permanent water body pixels are spatially overlaid with the lake shoreline vector layer. Candidate permanent water body pixels located inside or in contact with the shoreline boundary are retained, while candidate permanent water body pixels that are clearly located outside the shoreline and not connected to any lake are removed.

3. The Landsat series satellite transparency remote sensing inversion and consistency correction method according to claim 2, characterized in that, For candidate permanent water body pixels that cross the shoreline boundary, determine the pixel center point of the candidate permanent water body pixel and use the pixel center point as the basis for classification: if the pixel center point is located within the shoreline, the pixel is determined to be a lake water body and is retained; otherwise, it is removed.

4. The Landsat series satellite transparency remote sensing inversion and consistency correction method according to claim 1, characterized in that, The pretreatment method for pretreating the water mask includes at least one of the following: The beam attenuation coefficient is estimated based on the inherent optical characteristics, and the beam attenuation coefficient is combined with water depth data to obtain an optical water depth index; the optical water depth index is used to mask out pixels in optical shallow water. An inward erosion operation is performed on the water mask to retain pixels with pure water properties. At the same time, the connected domain area of ​​continuous water is screened to remove tiny patches with an area smaller than a predetermined number of pixels.

5. The Landsat series satellite transparency remote sensing inversion and consistency correction method according to claim 4, characterized in that, The beam attenuation coefficient To utilize the absorption coefficient and particle backscattering coefficient The estimation is performed using the following formula: ， in, The particle backscattering ratio; wavelength The backscattering coefficient of pure water at that location; The backscattering ratio of pure water; The optical depth index To be based on the beam attenuation coefficient The calculation is performed using water depth data, and the formula is as follows: ， in, This is water depth data.

6. The Landsat series satellite transparency remote sensing inversion and consistency correction method according to claim 1, characterized in that, The process of constructing the multi-sensor transparency remote sensing inversion model is as follows: Based on Landsat series satellite reflectance images, two types of input features are constructed for water transparency inversion; the two types of input features include: band ratio features and inherent optical property parameter features; A high-quality matching dataset is constructed based on the two types of input features and measured water transparency data. A random partitioning strategy is adopted for each sensor, and the variables in the input features are standardized and normalized. The optimal hyperparameters are determined by combining grid search and iteration to maximize the improvement of accuracy.

7. The Landsat series satellite transparency remote sensing inversion and consistency correction method according to claim 6, characterized in that, The process of constructing the band ratio feature is as follows: Reflectance data in the visible, near-infrared, and short-wave infrared bands were extracted from images from Landsat satellites to construct band ratio characteristics: Blue / Green, Blue / Red, Blue / NIR, Green / Red, Green / NIR, and Red / NIR. Blue represents the blue band of Landsat satellites, Green represents the green band of Landsat satellites, and Red represents the red band of Landsat satellites.

8. The Landsat series satellite transparency remote sensing inversion and consistency correction method according to claim 6, characterized in that, The process for creating the inherent optical characteristic parameters is as follows: Calculate the band ratio characteristic wavelength set based on remote sensing reflectance image data. Corresponding wavelength within Backscattering coefficient and the overall absorption coefficient , The backscattering coefficient The calculation formula is as follows: ， in, wavelength Remote sensing reflectance at the location, ;; wavelength Remote sensing reflectance at the location; wavelength The particle backscattering coefficient at that location; wavelength Total absorption coefficient at the location The calculation formula is as follows: ， in, wavelength The backscattering coefficient of pure water at that location; wavelength The particle backscattering coefficient at that location.

9. The Landsat series satellite transparency remote sensing inversion and consistency correction method according to claim 1, characterized in that, The process of establishing the cross-sensor calibration model is as follows: Based on observation data from the same overlapping period of the two satellites, the median value of the initial transparency estimates of each sensor is taken, and then the average value of the two is calculated. The initial transparency estimate of the single star is calibrated for deviation using a correction equation.

10. A Landsat series satellite transparency remote sensing inversion and consistency correction system, used to implement the Landsat series satellite transparency remote sensing inversion and consistency correction method as described in any one of claims 1-9, characterized in that, include: The first module is set to acquire Landsat series satellite reflectivity images and identify water body extents to determine long-term stable permanent lake water bodies, and construct water body masks in combination with lake shoreline extents. The second module is configured to preprocess the water mask to obtain preprocessed Landsat series satellite reflectivity images, which are high-quality optical deep-water pixel sets. The third module is configured to construct a transparency remote sensing inversion model suitable for multiple sensors based on the high-quality optical deep-water pixel set and measured transparency data; and to use the transparency remote sensing inversion model to invert the reflectance images of each sensor to obtain the initial transparency estimate of the corresponding sensor. The fourth module is configured to establish a cross-sensor correction model based on the initial transparency estimates during the overlap period and correct the initial transparency estimates of individual stars, perform interpolation compensation for anomalous years, unify the transparency estimates of all sensors to the same scale, and form a consistent transparency sequence for a predetermined period.

Images