An Automated Fusion Method and System for Suspended Sediment Concentration from Multiple Ocean Color Satellites

Abstract

This invention discloses an automated fusion method and system for suspended sediment concentration from multi-source ocean color satellites, belonging to the field of marine remote sensing and geographic information processing technology. The method includes: S1, automatically identifying and reading the original satellite data file; S2, processing the quality labeling data to generate a comprehensive quality mask in one go, and marking invalid values ​​for the concentration data; S3, resampling all valid pixels onto a predefined regular geographic grid, calculating the cumulative concentration and pixel count of each grid cell to obtain the grid mean; S4, stacking multiple grid data from the same region and time window along the time axis, and performing spatiotemporal fusion using the median robust averaging method or the minimum value synthesis method; S5, automatically generating a standard raster file and a thematic map containing geographic features and embellishment information based on the fused data. This invention solves the problems of fragmented processing, low efficiency, and insufficient accuracy in existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of marine remote sensing and geographic information processing technology, and in particular relates to an automated fusion method and system for the concentration of suspended sediment from multi-source ocean color satellites. Background Technology

[0002] Suspended sediment concentration is an important indicator for assessing coastal erosion, estuarine deposition, water transparency, and the ecological environment. Satellite remote sensing technology, especially ocean color satellites such as Haiyang-1 and Sentinel-3, provides a core data source for large-scale, periodic monitoring of suspended sediment.

[0003] However, existing technologies have significant limitations when processing this data: Data formats are heterogeneous (HDF5, NetCDF, etc.), requiring manual writing of different scripts for reading; Quality control relies on pixel-by-pixel looping, which is inefficient and inconsistent in standards; Data from different satellites and at different times is difficult to automatically align to a unified spatial grid; Multi-temporal fusion methods are simple (such as direct averaging) but are easily affected by residual clouds and flare noise. The fragmented process from data to the final thematic map makes it difficult to meet the requirements of business systems for high timeliness, high accuracy, and unattended operation. Summary of the Invention

[0004] This invention provides an automated fusion method and system for the concentration of suspended sediment from multi-source ocean color satellites, which solves the problems of fragmented processing, insufficient accuracy, and low efficiency in existing technologies.

[0005] To achieve the above-mentioned objectives, the first objective of this invention is to provide an automated fusion method for the concentration of suspended sediment from multi-source ocean color satellites, comprising: S1. Automatically identify and read raw satellite data files of different formats, and extract suspended sediment concentration data, geographic coordinate data and quality label data; S2. Process the quality identification data based on vectorized bit operations to generate a comprehensive quality mask in one go, and mark the concentration data as invalid values ​​according to the mask; S3. Resample all valid pixels onto a predefined regular geographic grid, and use a vectorized accumulation method to calculate the concentration accumulation and pixel count of each grid cell in parallel to obtain the grid mean. S4. Stack multiple grid data from the same region and the same time window along the time axis, and perform spatiotemporal fusion using the median robust averaging method or the minimum value synthesis method to obtain the fused concentration data. S5. Automatically generate a standard grid file and a thematic map containing geographical features and finishing information based on the fused concentration data.

[0006] Preferably, S1 includes: For the HDF5 file of OceanSat-1 C / D satellite, read the TSM layer, longitude and latitude layer, and l2_flags layer; For the NetCDF data packet of Sentinel-3 satellite, read the TSM_NN variable, latitude / longitude variable, and WQSF variable, and convert the stored value to a scientific value V_science according to the following formula: V_science = scale_factor × V_storage + add_offset, where scale_factor is the scaling factor, add_offset is the offset; V_storage is the unknown quantity directly read from the file; Then obtain the actual concentration value through exponential operation: TSM = 10^{V_science}.

[0007] Preferably, S2 includes: Verify that the shapes of the quality flag array and the concentration data array are consistent; According to the preset list of invalid bits RemoveBits, perform the following operations on the quality flag array: Mask_i = (Flag & (1 << bit_i))!= 0, and reduce it to the total mask: Mask_total = np.bitwise_or.reduce([Mask_1,..., Mask_n]); Mark the pixels with Mask_total being true, the original concentration being a fixed invalid value, or the concentration being negative as NaN.

[0008] Preferably, S3 includes: Set the boundaries and step sizes of the target grid; Extract the valid pixels that are not NaN and their longitude and latitude (lon, lat) and concentration values val; Calculate the grid row and column indices of each valid pixel: col_idx = floor((lon - left_lon) / StepSize), row_idx = floor((top_lat - lat) / StepSize); Flatten the row and column indices into a one-dimensional index: flat_idx = row_idx × N_cols + col_idx; The bincount function is used to calculate the cumulative sum SumMatrix for each grid cell in parallel, with flat_idx as the index and val as the weight, while simultaneously calculating the count matrix. Calculate the grid mean GridMean: GridMean = SumMatrix / CountMatrix, assign NaN to the position where the count is 0.

[0009] Preferred median robust averaging methods include: For a time series V_series, calculate the median and root mean square deviation (RMS): RMS = sqrt(mean((V_series-Median)^2)); Set a coefficient k and filter values ​​that satisfy |V_series_i-Median|≤k×RMS; The average value of the filtered values ​​is used as the fusion result.

[0010] A second objective of this invention is to provide an automated fusion system for the concentration of suspended sediment from multi-source ocean color satellites, comprising: The data reading module automatically identifies and reads raw satellite data files in different formats, extracting suspended sediment concentration data, geographic coordinate data, and quality label data; The quality control module processes the quality identification data based on vectorized bit operations, generates a comprehensive quality mask in one go, and marks the concentration data as invalid values ​​according to the mask; The gridding module resamples all valid pixels onto a predefined regular geographic grid, and uses a vectorized accumulation method to calculate the concentration sum and pixel count of each grid cell in parallel, thereby obtaining the grid mean. The spatiotemporal fusion module stacks multiple grid data from the same region and the same time window along the time axis and performs spatiotemporal fusion using the median robust averaging method or the minimum value synthesis method to obtain fused concentration data. The product output module automatically generates standard raster files and thematic maps containing geographic features and finishing information based on the fused concentration data.

[0011] Preferably, the data reading module includes: The first analysis unit reads the GeophysicalData / TSM layer as suspended sediment concentration data, the NavigationData / Latitude and Longitude layers as geographic coordinate data, and the GeophysicalData / l2_flags layer as mass label data from the HDF5 file of Haiyang-1 C / D satellite. The second parsing unit reads the TSM_NN variable from the tsm_nn.nc file, the latitude and longitude variables from the geo_coordinates.nc file, and the WQSF variable from the wqsf.nc file as quality identification data for the NetCDF data packet of Sentinel-3 satellite; The second parsing unit converts the stored value into a scientific value V_science according to the scale factor scale_factor and offset add_offset in the metadata of the TSM_NN variable by the following formula: V_science = scale_factor × V_storage + add_offset, and further obtains the actual suspended sediment concentration value through the exponential operation TSM = 10^{V_science}.

[0012] Preferably, the quality control module includes: Checking the shape consistency between the quality identification array and the suspended sediment concentration data array; Performing a vectorized bit operation on the quality identification array according to the preset list of invalid quality identification bits RemoveBits: generating a boolean mask Mask_i = (Flag & (1 << bit_i))!= 0 for each bit bit_i in the list, and generating a comprehensive quality mask Mask_total = np.bitwise_or.reduce([Mask_1, Mask_2,..., Mask_n]) through a bitwise OR reduction operation; Marking as NaN the pixels with the comprehensive quality mask being true, the preset fixed invalid values in the suspended sediment concentration data, and the pixel values with negative concentrations.

[0013] Preferably, the gridification module includes: Receiving the target grid boundary Bound and grid step size StepSize set by the user; Extracting the valid pixels that are not NaN from the suspended sediment concentration data, and obtaining the longitude lon_j, latitude lat_j and concentration value val_j corresponding to each valid pixel; Calculating the row index and column index of each valid pixel in the target grid according to the following formula: col_idx_j = floor((lon_j - Bound.left_lon) / StepSize), row_idx_j = floor((Bound.top_lat - lat_j) / StepSize); Convert the row index and column index into a flattened one-dimensional index flat_idx_j = row_idx_j × N_cols + col_idx_j, where N_cols is the number of grid columns; Using the bincount function of the NumPy library, with the one-dimensional index flat_idx_j as the index and the concentration value val_j as the weight, the concentration accumulation SumMatrix of each grid cell is calculated in parallel, and the pixel count CountMatrix of each grid cell is calculated at the same time. Calculate the mean value GridMean for each grid cell: GridMean = SumMatrix / CountMatrix, and assign NaN to the grid cells with a count of zero.

[0014] A third objective of this invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned automated fusion method for the concentration of suspended sediment from multi-source ocean color satellites.

[0015] The advantages and positive effects of this application are: This invention enables end-to-end unattended processing from raw data to thematic maps. In particular, through vectorized quality control and bincount-based gridding, it avoids pixel-by-pixel looping, improving processing speed by more than an order of magnitude compared to traditional point-by-point methods, thus meeting the needs of high-frequency production in business applications.

[0016] This invention achieves high compatibility and consistency of multi-source data, unifies the parsing of two mainstream ocean color satellite data, Haiyang-1 C / D and Sentinel-3, and aligns them to the same spatial reference through standard gridding, enabling multi-source and multi-temporal data to be processed collaboratively, and the product has good spatiotemporal consistency.

[0017] This invention employs a dual approach: hierarchical quality labeling vectorization removal and median robust averaging fusion. The former accurately filters out interference from clouds, flares, land, and other sources; the latter effectively suppresses occasional outliers in the time series, resulting in a fusion result that is more physically robust than direct averaging or median averaging.

[0018] The scalability established by this invention allows users to configure core parameters such as grid step size, fusion algorithm type (robust average / minimum synthesis), quality control bit list, and synthesis cycle. It can adapt to business needs of different resolutions and regions without modifying the underlying code, and has extremely high reusability and platform potential. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a preferred embodiment of the present invention. Detailed Implementation

[0021] 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.

[0022] Please see Figure 1 An automated fusion method for the concentration of suspended sediment from multiple ocean color satellites mainly includes: S1. Automatically identify and read raw satellite data files of different formats, and extract suspended sediment concentration data, geographic coordinate data and quality label data; in the computer system, the quality label data is stored in the form of an array, named the quality label array; S2. Process the quality identification data based on vectorized bit operations to generate a comprehensive quality mask in one go, and mark the concentration data as invalid values ​​according to the mask; S3. Resample all valid pixels onto a predefined regular geographic grid, and use a vectorized accumulation method to calculate the concentration accumulation and pixel count of each grid cell in parallel to obtain the grid mean. S4. Stack multiple grid data from the same region and the same time window along the time axis, and perform spatiotemporal fusion using the median robust averaging method or the minimum value synthesis method to obtain the fused concentration data. S5. Based on the fused concentration data, automatically generate standard raster files and thematic maps containing geographic features and embellishment information.

[0023] To better understand the technical solution of the present invention, the above steps are described in detail below without limitation: S1 includes: automatically scanning the user-specified input directory to identify and extract raw data files in two main formats: HDF5 format (.h5) files from Ocean-1 C / D satellites and NetCDF data packets (.sen3 directory) from Sentinel-3.

[0024] S1 specifically includes: First, the system automatically scans the user-specified input directory to identify and extract raw data files in two main formats: HDF5 format (.h5) files from Ocean-1 C / D satellites and NetCDF data packets (.sen3 directory) from Sentinel-3 satellites. Then, based on the satellite type, the suspended sediment concentration data, latitude and longitude navigation data, and quality label data (Flag) are extracted respectively. Finally, for the Sentinel-3 data, based on the scale_factor and add_offset in its metadata, the conversion V_science = scale_factor × V_storage + add_offset was performed, and the actual concentration value was obtained through exponential calculation.

[0025] In one specific embodiment, the system acts as the execution entity, uniformly parsing data of different formats: For data from Haiyang-1C / D satellites: read the three core datasets in the HDF5 file: GeophysicalData / TSM (suspended sediment concentration data layer), NavigationData / Latitude and NavigationData / Longitude (latitude and longitude navigation layers), and GeophysicalData / l2_flags (mass labeling layer).

[0026] For Sentinel-3 data: Key data from multiple NetCDF files in the .sen3 directory were read, including the TSM_NN variable (logarithmic suspended sediment concentration) in the tsm_nn.nc file, the latitude and longitude variables (latitude and longitude) in the geo_coordinates.nc file, and the WQSF variable (mass identifier) ​​in the wqsf.nc file. The metadata (scale_factor and add_offset) of the TSM_NN variable was automatically identified, and data decompression and unit conversion were performed. The stored value V_storage was converted to the actual scientific value V_science using the following formula: V_science=scale_factor×V_storage+add_offset Here, `scale_factor` and `add_offset` are known quantities stored in the file attributes; `V_storage` is the unknown quantity (raw stored value) read directly from the file; and `V_science` is the scientific value to be calculated (i.e., log10(TSM)). Subsequently, the system uses the exponential operation `TSM_concentration = 10 × V_science` to restore the log10 concentration to the actual suspended sediment concentration value (unit: g / m³).

[0027] S2 specifically includes: Automatically perform quality control on the original suspended sediment concentration data read by S1 and the corresponding quality identification data. The core function of this step is to efficiently and accurately eliminate invalid pixels contaminated by factors such as clouds, flares, ice, land, and high solar zenith angles, ensuring the reliability of the subsequent analysis data. Specifically includes: 1) Data verification: Verify whether the shapes of the quality identification array Flag and the suspended sediment concentration data array Data are consistent.

[0028] 2) Quality identification mask generation: According to the preset list of invalid quality identification bits RemoveBits, which indicates which identification bits, when set to 1, represent unreliable pixel data (for example, the data of OceanSat-1 C / D satellites is set to [1, 3, 5, 7, 8, 9, 12, 14, 15, 21, 25]), perform vectorized bit operations on the Flag array. For each bit bit_i in the list, calculate the boolean mask Mask_i = (Flag & (1 << bit_i))!= 0. Subsequently, through a bitwise OR reduction operation, generate the final comprehensive quality mask Mask_total = np.bitwise_or.reduce([Mask_1, Mask_2,..., Mask_n]). This process operates on the entire array at once, completely replacing the traditional per-pixel loop judgment.

[0029] 3) Marking invalid values: Mark the pixel values with a comprehensive quality mask Mask_total of 1, the invalid values of the suspended sediment concentration data (-999 for the data of OceanSat-1 C / D satellites and FillValue for Sentinel-3 data), and the physical invalid conditions of the suspended sediment concentration data (Data < 0) as NaN.

[0030] S3 specifically includes: Resample the pixel data with quality-controlled irregular geographical coordinates (defined by the Longitude and Latitude arrays) to a standard regular geographical grid predefined by the user with a unified geographical range and a fixed step size. This step realizes the spatial alignment and standardization of multi-source and multi-scene heterogeneous observation data, laying a foundation for subsequent data fusion. Specifically includes: 1) Parameter definition: The user sets the boundaries of the target grid, namely the unified geographical range and the grid step size StepSize; 2) Spatial range truncation and extraction of valid data: Extract the non-NaN valid pixels from the suspended sediment concentration data Data, and obtain their corresponding longitude and latitude (lon, lat) and suspended sediment concentration value val.

[0031] 3) Grid Index Calculation: For each retained valid pixel (lon_j, lat_j), the system calculates its row index row_idx_j and column index col_idx_j in the target grid. The calculation formula is as follows: col_idx_j=floor((lon_j-Bound['left_lon']) / StepSize) row_idx_j=floor((Bound['top_lat']-lat_j) / StepSize) Floor() is the floor function.

[0032] 4) Vectorized Accumulation and Counting: The system converts the two-dimensional row and column index into a one-dimensional flat index: flat_idx_j = row_idx_j × N_cols + col_idx_j, where N_cols is the number of grid columns. The system uses the bincount function from the NumPy library to perform two vectorization operations: Using flat_idx as the index and val as the weight, calculate the cumulative sum of data in each grid cell, SumMatrix.

[0033] Using flat_idx as the index, calculate the pixel count CountMatrix for each grid cell.

[0034] 5) Grid mean calculation: The system calculates the mean of each grid cell based on the sum and count matrix: GridMean=SumMatrix / CountMatrix. For grid cells with a count of 0, the mean is assigned the value NaN.

[0035] S4 specifically includes: For multiple standard grid data points generated after S3 gridding within the same region and time period (e.g., one month), the system stacks them along the time dimension. The purpose of this step is to synthesize a suspended sediment concentration product that better represents the water condition within that time period, has more complete data coverage, and less noise. The fusion methods include the median robust averaging method and the minimum value synthesis method.

[0036] 1) Median Robust Averaging Method: The system first calculates the median of V_series, then calculates the root mean square (RMS) of the deviation of each value from the median: RMS = sqrt(mean((V_series-Median)^2)). Subsequently, the system selects all values ​​that satisfy |V_series_i-Median| ≤ ​​k × RMS (where k is a preset coefficient, such as 1.5), and finally calculates the average of these "robust" values ​​as the fusion value for that grid location. This method effectively removes the interference of random outliers in the time series.

[0037] 2) Minimum value synthesis method: The system directly takes the minimum value in V_series as the fusion value, aiming to minimize the impact of residual pollution such as clouds.

[0038] S5 specifically includes: Based on the fused mesh data generated by S4, two standardized products are automatically generated.

[0039] 1) Standard Georeferenced Raster File: The system generates a standard GeoTIFF format file by combining the data matrix, along with its precise geographic transformation parameters (affine transformations calculated from Bound and StepSize) and coordinate system information.

[0040] 2) Thematic Maps: Based on the dynamic range of the data, color scales and classifications are set, basic geographic elements such as coastlines and administrative divisions are overlaid, and finishing elements such as map titles, legends, scales, data sources, cartographic units, and dates are added. The final output is a high-resolution image file (such as PNG format). This step achieves unattended output of "data to map".

[0041] An automated fusion system for measuring suspended sediment concentration from multiple ocean color satellites includes: The data reading module is used to execute S1; The quality control module is used to execute S2; The mesh module is used to execute authority S3; The spatiotemporal fusion module is used to execute S4; The product output module is used to execute S5.

[0042] The data reading module includes: The first parsing unit is used to read the GeophysicalData / TSM layer as suspended sediment concentration data, the NavigationData / Latitude and Longitude layers as geographic coordinate data, and the GeophysicalData / l2_flags layer as mass label data from the HDF5 file of Haiyang-1 C / D satellite. The second parsing unit is used to read the TSM_NN variable from the tsm_nn.nc file, the latitude and longitude variables from the geo_coordinates.nc file, and the WQSF variable from the wqsf.nc file as quality identification data for the NetCDF data packets of the Sentinel-3 satellite; The second parsing unit is further configured to convert the stored value into a scientific value according to the scale_factor and add_offset in the metadata of the TSM_NN variable according to the following formula: V_science = scale_factor × V_storage + add_offset, and further obtain the actual suspended sediment concentration value (unit: g / m³) through the exponential operation TSM = 10^{V_science}.

[0043] The quality control module is specifically configured to: Verify the shape consistency between the quality identification array and the suspended sediment concentration data array; Perform a vectorized bit operation on the quality identification array according to the preset list of invalid quality identification bits RemoveBits: generate a Boolean mask Mask_i = (Flag & (1 << bit_i))!= 0 for each bit bit_i in the list, and generate a comprehensive quality mask Mask_total = np.bitwise_or.reduce([Mask_1, Mask_2,..., Mask_n]) through a bitwise OR reduction operation; Mark as NaN those with the comprehensive quality mask being true, the preset fixed invalid values in the suspended sediment concentration data, and the pixel values with negative concentrations.

[0044] The preset list of invalid quality identification bits in the quality control module is set to 1, 3, 5, 7, 8, 9, 12, 14, 15, 21, 25 for OceanSat-1 C / D satellite data, and is used to remove clouds, land, flares, high solar zenith angles, and ice pixels.

[0045] The gridification module is specifically configured to: Receive the target grid boundary Bound and grid step size StepSize set by the user; Extract the valid pixels that are not NaN in the suspended sediment concentration data, and obtain the longitude lon_j, latitude lat_j, and concentration value val_j corresponding to each valid pixel; Calculate the row index and column index of each valid pixel in the target grid according to the following formula: col_idx_j = floor((lon_j - Bound.left_lon) / StepSize), row_idx_j = floor((Bound.top_lat - lat_j) / StepSize); Convert the row index and column index into a flattened one-dimensional index flat_idx_j = row_idx_j × N_cols + col_idx_j, where N_cols is the number of grid columns; Using the bincount function of the NumPy library, with the one-dimensional index flat_idx_j as the index and the concentration value val_j as the weight, the concentration accumulation SumMatrix of each grid cell is calculated in parallel, and the pixel count CountMatrix of each grid cell is calculated at the same time. Calculate the mean value of each grid cell: GridMean = SumMatrix / CountMatrix, and assign the value of the grid cell with a count of zero to NaN.

[0046] The spatiotemporal fusion module includes: The median robust averaging unit stacks multiple grid data points within the same grid region and time window along the time axis into a sequence V_series. It calculates the median and root mean square (RMS) of this sequence: sqrt(mean((V_series-Median)²)). Values ​​satisfying |V_series_i-Median| ≤ ​​k × RMS, where k is a preset coefficient, are selected, and their average is used as the fusion result. The minimum value synthesis unit directly takes the minimum value in the sequence V_series as the fusion result; The spatiotemporal fusion module selectively enables the median robust average unit or the minimum value synthesis unit according to the fusion mode specified by the user.

[0047] The preset coefficient k in the median robust average unit is 1.5.

[0048] The product output module includes: The raster file generation unit is used to generate a standard GeoTIFF format file based on the fused mesh data, combined with the affine transformation parameters calculated from the target mesh boundary and step size, and the preset coordinate system information. The thematic map generation unit is used to automatically classify and color the data according to the dynamic range of the fused grid data, overlay preset coastlines and basic geographic elements of administrative divisions, and add map title, legend, scale, data source, cartographic unit and date finishing elements to finally generate a high-resolution image file.

[0049] The system also includes a parameter configuration module, which allows users to set at least one of the following parameters: target grid step size, spatiotemporal fusion method type, quality control invalid bit list, and synthesis cycle; the data reading module, quality control module, meshing module, and spatiotemporal fusion module adapt to the processing requirements of different regions, different resolutions, and different synthesis cycles according to the settings of the parameter configuration module.

[0050] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An automated fusion method of multi-source ocean color satellite suspended sediment concentration, characterized in that, It includes: S1. Automatically identify and read raw satellite data files in different formats, and extract suspended sediment concentration data, geographic coordinate data, and quality flag data; S2. Process the quality flag data based on vectorized bit operations to generate a comprehensive quality mask at once, and mark invalid values for the concentration data according to the mask; S3. Resample all valid pixels onto a predefined regular geographic grid, and use the vectorized accumulation method to calculate the concentration sum and pixel count of each grid cell in parallel, and then obtain the grid mean value; S4. Stack multiple grid data within the same region and the same time window along the time axis, and perform spatio-temporal fusion using the median robust averaging method or the minimum composite method to obtain the fused concentration data; S5. Automatically generate a standard raster file and a thematic map containing geographic features and finishing information based on the fused concentration data.

2. The automated fusion method of multi-source ocean color satellite suspended sediment concentration according to claim 1, characterized in that, S1 includes: For the HDF5 file of OceanSat-1C / D satellite, read the TSM layer, longitude and latitude layer, and l2_flags layer; For the NetCDF data packet of Sentinel-3 satellite, read the TSM_NN variable, latitude / longitude variable, and WQSF variable, and convert the stored value to the scientific value V_science according to the following formula: V_science = scale_factor × V_storage + add_offset, where, scale_factor is the scaling factor, add_offset is the offset; V_storage is the unknown quantity directly read from the file; Then obtain the actual concentration value through exponential operation: TSM = 10^{V_science}.

3. The automated fusion method of multi-source ocean color satellite suspended sediment concentration according to claim 1, wherein S2 It includes: Verify that the shapes of the quality flag array and the concentration data array are consistent; Execute on the quality flag array according to the preset invalid bit list RemoveBits: Mask_i = (Flag & (1 << bit_i))!= 0, and reduce it to the total mask: Mask_total = np.bitwise_or.reduce([Mask_1,..., Mask_n]); 4. The automated fusion method of multi-source ocean color satellite suspended sediment concentration according to claim 1, characterized in that, Mark the pixels where Mask_total is true, the original concentration is a fixed invalid value, or the concentration is negative as NaN. S3 includes: Set the boundaries and step sizes of the target grid; Extract the valid pixels that are not NaN and their longitude, latitude (lon, lat), and concentration value val; Calculate the grid row and column indices of each valid pixel: col_idx = floor((lon - left_lon) / StepSize), row_idx = floor((top_lat - lat) / StepSize); Flatten the row and column indices into a one-dimensional index: flat_idx = row_idx × N_cols + col_idx; Use the bincount function to calculate the sum matrix SumMatrix of each grid in parallel with flat_idx as the index and val as the weight, and calculate the count matrix CountMatrix at the same time. Calculate the grid mean GridMean: GridMean = SumMatrix / CountMatrix, assign NaN to the position where the count is 0.

5. The automated fusion method of multi-source ocean color satellite suspended sediment concentration according to claim 1, characterized in that, The median robust averaging method includes: For a time series V_series, calculate the median and root mean square deviation (RMS): RMS = sqrt(mean((V_series-Median)^2)); Set a coefficient k and filter values ​​that satisfy |V_series_i-Median|≤k×RMS; The average value of the filtered values ​​is used as the fusion result.

6. An automated fusion system for measuring suspended sediment concentration from multiple ocean color satellites, characterized in that, include: The data reading module automatically identifies and reads raw satellite data files in different formats, extracting suspended sediment concentration data, geographic coordinate data, and quality label data; The quality control module processes the quality identification data based on vectorized bit operations, generates a comprehensive quality mask in one go, and marks the concentration data as invalid values ​​according to the mask; The gridding module resamples all valid pixels onto a predefined regular geographic grid, and uses a vectorized accumulation method to calculate the concentration sum and pixel count of each grid cell in parallel, thereby obtaining the grid mean. The spatiotemporal fusion module stacks multiple grid data from the same region and the same time window along the time axis and performs spatiotemporal fusion using the median robust averaging method or the minimum value synthesis method to obtain fused concentration data. The product output module automatically generates standard raster files and thematic maps containing geographic features and finishing information based on the fused concentration data.

7. The automated fusion system for multi-source ocean color satellite suspended sediment concentration according to claim 6, characterized in that, The data reading module includes: The first analysis unit reads the GeophysicalData / TSM layer as suspended sediment concentration data, the NavigationData / Latitude and Longitude layers as geographic coordinate data, and the GeophysicalData / l2_flags layer as mass label data from the HDF5 file of Haiyang-1 C / D satellite. The second parsing unit reads the TSM_NN variable from the tsm_nn.nc file, the latitude and longitude variables from the geo_coordinates.nc file, and the WQSF variable from the wqsf.nc file as quality identification data for the NetCDF data packets of the Sentinel-3 satellite; The second parsing unit converts the stored value into a scientific value V_science according to the scaling factor (scale_factor) and offset (add_offset) in the metadata of the TSM_NN variable, using the following formula: V_science = scale_factor × V_storage + add_offset, and further obtains the actual suspended sediment concentration value through the exponential operation TSM = 10^{V_science}.

8. The automated fusion system for multi-source ocean color satellite suspended sediment concentration according to claim 6, characterized in that, The quality control module includes: Verify the shape consistency between the quality label array and the suspended sediment concentration data array; Perform a vectorized bit operation on the quality identification array according to the preset invalid quality identification bit list RemoveBits: generate a boolean mask Mask_i = (Flag & (1 << bit_i))!= 0 for each bit bit_i in the list, and generate a comprehensive quality mask Mask_total = np.bitwise_or.reduce([Mask_1, Mask_2,..., Mask_n]) through a bitwise OR reduction operation; Mark as NaN those with the comprehensive quality mask being true, the preset fixed invalid values in the suspended sediment concentration data, and the pixel values with negative concentrations.

9. The automated fusion system for multi-source ocean color satellite suspended sediment concentration according to claim 6, characterized in that, The gridification module includes: Receive the target grid boundary Bound and grid step size StepSize set by the user; Extract the valid pixels in the suspended sediment concentration data that are not NaN, and obtain the longitude lon_j, latitude lat_j, and concentration value val_j corresponding to each valid pixel; Calculate the row index and column index of each valid pixel in the target grid according to the following formula: col_idx_j = floor((lon_j - Bound.left_lon) / StepSize), row_idx_j = floor((Bound.top_lat - lat_j) / StepSize); Convert the row index and column index to a flattened one-dimensional index flat_idx_j = row_idx_j × N_cols + col_idx_j, where N_cols is the number of grid columns; Using the bincount function of the NumPy library, with the one-dimensional index flat_idx_j as the index and the concentration value val_j as the weight, calculate the concentration sum SumMatrix of each grid cell in parallel, and at the same time calculate the pixel count CountMatrix of each grid cell; Calculate the mean GridMean of each grid cell: GridMean = SumMatrix / CountMatrix, and assign the value of the grid cell with a count of zero as NaN.

10. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by a processor, it implements the automated fusion method for the suspended sediment concentration of multi-source ocean color satellites as described in any one of claims 1-5.

Images