Red tide dynamics evolution monitoring method based on stationary satellite

Abstract

The invention discloses a red tide dynamics evolution monitoring method based on a stationary satellite, and relates to the field of red tide dynamics evolution monitoring, and the method comprises the steps: obtaining satellite multi-source data, carrying out the time-space standardization processing of the multi-source data, and generating a standardized time sequence image matrix; performing flow field inversion on the standardized time sequence image matrix by adopting a three-level adaptive processing strategy, and outputting a high-precision sea surface flow field vector; based on the high-precision sea surface flow field vector, calculating a flow field and a derivative of a gradient thereof by using a central difference scheme, and extracting a shear strain rate; and constructing a binary logic mask matrix of a 75m isobath, and performing AND operation based on the binary logic mask matrix, the shear strain rate and the flow velocity amplitude to obtain a final red tide dynamics risk index. The problems that in the prior art, the monitoring frequency is low (discontinuous), flow field dynamic information is difficult to obtain, and the near-shore early warning precision is insufficient are solved.

Description

Technical Field

[0001] This invention relates to the field of red tide dynamics evolution monitoring, and specifically to a red tide dynamics evolution monitoring method based on geostationary satellites. Background Technology

[0002] Red tides (harmful algal blooms) are a hazardous phenomenon that moves and spreads rapidly under the influence of ocean dynamics. Currently, monitoring of red tides mainly relies on remote sensing observations by conventional polar-orbiting optical satellites (such as MODIS and Landsat).

[0003] However, this mainstream existing technology has the following significant technical problems in practical applications:

[0004] "One image per day" makes real-time monitoring impossible: Conventional polar-orbiting satellites only pass over the same sea area 1-2 times a day (usually in the morning or afternoon). Due to the influence of tides on nearshore currents, the direction and speed of the current can change dramatically within a few hours. Existing "capture-the-moment" observations cannot capture the continuous drift trajectory of red tides throughout the day, resulting in a huge time blind spot.

[0005] The current technology can only "interpret images" and lacks flow field data: existing technologies mainly focus on "where" the red tide is (the static distribution of chlorophyll concentration), but cannot directly measure "how" the red tide moves (the velocity of the sea surface flow field). Due to the lack of high-frequency continuous images, traditional methods cannot directly calculate the speed and direction of seawater flow from images, making it very difficult to predict the future drift path of the red tide.

[0006] Early warning capabilities are weak in complex nearshore environments: Near coastlines, the terrain is complex and currents are swift. Existing technologies often struggle to distinguish between areas where water flow is stagnant due to topography (making red tides more likely to accumulate) and areas where water flows through rapidly. This lack of dynamic analysis leads to inaccurate assessments of red tide outbreak risks. Summary of the Invention

[0007] To address the aforementioned shortcomings in existing technologies, this invention provides a method for monitoring the dynamic evolution of red tides based on geostationary satellites, which solves the problems of low (discontinuous) monitoring frequency, difficulty in obtaining flow field dynamic information, and insufficient nearshore early warning accuracy in existing technologies.

[0008] To achieve the aforementioned objectives, the technical solution adopted by this invention is as follows: a method for monitoring the dynamic evolution of red tides based on geostationary satellites, comprising the following steps:

[0009] S1: Acquire satellite multi-source data and perform spatiotemporal standardization processing on the multi-source data to generate a standardized time-series image matrix;

[0010] S2: A three-level adaptive processing strategy is used to perform flow field inversion on the standardized time-series image matrix and output a high-precision sea surface flow field vector.

[0011] S3: Based on high-precision sea surface current field vectors, the derivatives of the current field and its gradient are calculated using the central difference scheme, and the shear strain rate is extracted;

[0012] S4: Construct a binary logic mask matrix for the 75m isobath, and perform an AND operation based on the binary logic mask matrix, shear strain rate, and flow velocity amplitude to obtain the final red tide dynamic risk index, thereby realizing the monitoring of red tide dynamic evolution based on geostationary satellites.

[0013] Furthermore, step S1 includes the following sub-steps:

[0014] S11: Acquire multi-source satellite data, and perform regional stitching and resampling on the multi-source satellite data to form a time-series image matrix;

[0015] S12: Based on the time series image matrix, perform quality control and hole filling to generate a standardized time series image matrix.

[0016] Furthermore, the specific logic of the three-level adaptive processing strategy in S2 is as follows:

[0017] First-level strategy: Use the direct cross-correlation algorithm to calculate all image windows, count the proportion of effective vectors. If the proportion of effective vectors is ≥40%, accept the direct cross-correlation result and enter the sub-pixel fitting stage; otherwise, enter the second-level strategy.

[0018] The second-level strategy is to use fast Fourier transform to calculate the cross-correlation function in the frequency domain and perform quality control based on the anomaly capture mechanism. If no anomaly is thrown during the execution process, the result is accepted and the process enters the sub-pixel fitting stage; if an anomaly occurs, the third-level strategy is entered.

[0019] The third-level strategy is to use neighborhood reference interpolation as a fallback solution, use the spatial continuity assumption to estimate the flow velocity, and then enter the sub-pixel fitting stage.

[0020] The subpixel fitting stage uses a two-dimensional Gaussian surface fitting method to locate the peak center based on the obtained integer pixel displacement peaks, and obtains the displacement amount with subpixel accuracy. A quality check is then performed to ensure that at least one valid vector exists, and finally, a high-precision sea surface current field vector is obtained.

[0021] Furthermore, the shear strain rate in S3 The calculation formula is:

[0022]

[0023] in, The flow velocity is in the horizontal direction. The vertical flow velocity. and These are the spatial coordinates in the horizontal and vertical directions, respectively. and Each is the current pixel The vertical velocity values ​​in the upper and lower neighborhoods. and Each is the current pixel The flow velocity values ​​in the left and right neighborhoods in the horizontal direction, and This represents the actual geographic spatial grid spacing.

[0024] Furthermore, the method for constructing the binary logic mask matrix in S4 is as follows:

[0025] If a region with a water depth of less than or equal to 75m is defined as a potential red tide outbreak zone, then the binary logic mask matrix is ​​1.

[0026] If the area with a water depth greater than 75m is defined as the non-red tide background area, then the binary logic mask matrix is ​​0.

[0027] Furthermore, the red tide dynamics risk index in S4 The calculation formula is:

[0028]

[0029] in, Shear strain rate For the velocity amplitude, It is a binary logic mask matrix for the 75m isobath.

[0030] The beneficial effects of this invention are:

[0031] (1) Overcoming the time blind zone and achieving hourly continuous dynamic monitoring. This invention utilizes the high-frequency observation capabilities of the GOCI-II geostationary satellite and combines it with multi-slot stitching technology to achieve hourly continuous tracking of the red tide drift and diffusion process. This enables the monitoring system to capture short-term drastic changes in flow direction and velocity caused by nearshore tides, filling the time blind zone of traditional monitoring. It overcomes the time resolution limitation of traditional polar-orbiting satellites (such as MODIS) that only capture one scene per day, and solves the problem of not being able to capture the rapid drift trajectory of red tides with the tides within a day.

[0032] (2) A unique three-level adaptive flow field inversion significantly improves inversion accuracy under complex sea conditions. This invention proposes a three-level adaptive strategy of "DCC (high robustness) + FFT (high efficiency) + Quick PIV (backup plan)". In low signal-to-noise ratio regions, the high robustness of the DCC algorithm ensures accurate matching; in normal regions, FFT significantly improves computational efficiency; in extreme cases, Quick PIV prevents data gaps; finally, through sub-pixel Gaussian fitting, high accuracy and high coverage (effective vector rate > 60%) of flow velocity inversion are achieved. This solves the problem that traditional single algorithms (such as simple FFT) are prone to failure or mismatch in cloud edges, low-texture or high-noise regions.

[0033] (3) Constructing a "dynamic-nutrient coupling" risk model significantly reduces the false alarm rate in the open ocean. This invention innovatively introduces the 75m isobath as a topographic constraint mask, combining dynamic indicators (shear strain rate × current velocity) with seabed topographic depth to forcibly filter out risk values ​​in open ocean areas with water depths greater than 75m. This conforms to the scientific law that red tide outbreaks require both "dynamic accumulation" and "nutrients," significantly improving the scientific validity and accuracy of nearshore red tide early warning. It solves the problem that existing methods, relying solely on dynamic parameters, misjudge eddies in deep-sea (nutrient-poor) areas as high-risk red tide zones.

[0034] (4) This invention represents a leap from "reading pictures and describing them" to "quantitative dynamic analysis." It not only outputs chlorophyll concentration maps but also retrieves high-precision sea surface current vectors (U, V) and shear strain rates (U, V). By calculating shear strain rate, frontal zones with intense water deformation can be identified, thereby accurately predicting high-risk sea areas where red tide algae are prone to accumulate, providing direct decision-making data support for disaster prevention and mitigation. This changes the previous situation where only the static distribution of red tides (chlorophyll concentration) could be observed, and their movement trends could not be known. Attached Figure Description

[0035] Figure 1 This is a flowchart of the red tide dynamics evolution monitoring method based on geostationary satellites according to the present invention.

[0036] Figure 2 This is a roadmap for the red tide dynamics evolution monitoring technology based on geostationary satellites, as described in this invention. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0038] This embodiment uses the nearshore waters of the East China Sea (27°N-32°N, 120°E-126°E) as the demonstration area and utilizes GOCI-II geostationary orbit satellite data for red tide dynamics monitoring. The specific implementation steps are as follows:

[0039] like Figure 1 and Figure 2 As shown, a method for monitoring the dynamic evolution of red tides based on geostationary satellites includes the following steps:

[0040] S1: Acquire satellite multi-source data and perform spatiotemporal standardization processing on the multi-source data to generate a standardized time-series image matrix;

[0041] This step aims to address the issues of geometric distortion and lack of meshing in the raw data of geostationary satellites, and to construct a standard data field suitable for dynamic calculations.

[0042] S1 includes the following sub-steps:

[0043] S11: Acquire multi-source satellite data, and perform regional stitching and resampling on the multi-source satellite data to form a time-series image matrix;

[0044] Data access: Acquire GOCI-II satellite Level-2 chlorophyll a concentration (Chl-a) products with a time resolution of 1 hour. Simultaneously acquire GEBCO global high-resolution water depth data (15 arc-second resolution) as terrain constraints.

[0045] Regional stitching and resampling: Since GOCI-II data consists of multiple observation slots (such as slots S006-S011 in this embodiment), stitching is required. A standard latitude and longitude grid covering the target sea area is constructed, with a resolution set to [resolution missing]. (Approximately 250 meters). The unstructured scattered data of each slot is mapped to a standard grid using the nearest neighbor interpolation method to form a time series image matrix.

[0046] Furthermore, the original unstructured latitude and longitude coordinates of the satellite are mapped onto the target output grid. The formula is:

[0047]

[0048] in, and For the mapping result, The coordinates of the satellite observation point. The coordinates of the top left corner of the target grid. Spatial resolution (250m). This is the floor function operator. This formula enables seamless stitching of multi-slot data, resolving the spatial discontinuity issue caused by multi-slot scanning of geostationary satellites.

[0049] Image grayscale values ​​(DN) are converted to chlorophyll concentration for analysis. 8-bit grayscale values ​​(DN, 1~255, 0 indicates invalid pixels) and chlorophyll concentration are used. The conversion relationship is as follows:

[0050]

[0051] In the formula: These represent the minimum and maximum chlorophyll concentrations, respectively. This ensures that image enhancement processing is performed in the 0-255 space while retaining the ability to retrieve physical quantities, thus maintaining linear consistency between image processing (0-255 space) and physical analysis (concentration space).

[0052] To address the low contrast characteristic of water color images and enhance the visual features of sea surface eddies and fronts, thereby improving the matching success rate, a contrast-limited adaptive histogram equalization (CLAHE) is proposed. The core calculation is as follows:

[0053]

[0054]

[0055] In the formula: The cumulative distribution function within a local window indicates that the grayscale value within the local area is less than or equal to... The pixel accumulation probability is used to map the original grayscale value to a new grayscale value; gray levels within a local window The number of pixels; This represents the total number of pixels in the local window. The new grayscale value after enhancement; =0.02 is the contrast limiting factor; round() is the rounding operator.

[0056] S12: Based on the time series image matrix, perform quality control and hole filling to generate a standardized time series image matrix;

[0057] Quality control and hole filling: Data quality flags are read, and invalid cells affected by cloud cover, land cover, and solar flares are removed (set to NaN). To avoid small holes affecting subsequent differential calculations, a... The neighborhood sliding window performs interpolation to fill isolated NaN values, using the following formula:

[0058]

[0059] in, The normalized temporal image matrix after interpolation and filling. For effective pixels gather, This represents the number of effective pixels.

[0060] S2: A three-level adaptive processing strategy is used to perform flow field inversion on the standardized time-series image matrix and output a high-precision sea surface flow field vector.

[0061] To address the extremely uneven texture distribution in ocean color images, this invention abandons the traditional single FFT calculation mode and designs a three-level adaptive processing strategy that integrates Direct Cross-Correlation (DCC), Fast Fourier Transform (FFT), and Quick Interpolation Last-Line (QuickPIV). This strategy employs a step-by-step trial-and-error fault-tolerant mechanism to ensure reliable flow field vectors are obtained under various image quality conditions.

[0062] The specific logic of the three-level adaptive processing strategy in S2 is as follows:

[0063] First-level strategy: Use the direct cross-correlation algorithm to calculate all image windows, count the proportion of effective vectors. If the proportion of effective vectors is ≥40%, accept the direct cross-correlation result and enter the sub-pixel fitting stage; otherwise, enter the second-level strategy.

[0064] This strategy directly calculates the cross-correlation function of the two windows in the spatial domain, avoiding the periodic assumption error caused by FFT, and can more reliably lock the true correlation peak under complex texture conditions.

[0065] DCC is used in regions with low signal-to-noise ratio and complex textures. It calculates directly pixel-by-pixel in the spatial domain without frequency domain transformation, thus maximally suppressing artifacts. The calculation formula is:

[0066]

[0067] The normalized formula is:

[0068]

[0069] in, This represents the direct cross-correlation coefficient, which measures the displacement of the search windows in the first and second frames. The similarity is calculated by the value of the two windows under the given displacement. The larger the value, the higher the matching degree between the two windows at that displacement. In short, it is used to find the best matching position and determine the flow field displacement vector. and These represent the first frame query window and the second frame search window at pixel points, respectively. The grayscale value at that location. This represents the horizontal and vertical displacement components of the search window relative to the query window. and This represents the average grayscale value of all pixels within the corresponding window, used to eliminate deviations caused by uneven illumination. This represents the set of pixels within the interrogation window, i.e., the local region involved in the calculation. The denominator is the image energy normalization factor, ensuring the correlation coefficient... Located in the [-1,1] range, it enhances noise resistance. The peak position corresponds to an integer pixel displacement vector. Although it requires a large amount of computation, it is more robust than the frequency domain method when the signal-to-noise ratio is extremely low (such as thin cloud cover).

[0070] The second-level strategy is to use Fast Fourier Transform to calculate the cross-correlation function in the frequency domain and perform quality control based on the anomaly capture mechanism. If no anomaly is thrown during the execution process, the result is accepted and the process enters the sub-pixel fitting stage. If an anomaly occurs (such as insufficient memory, numerical calculation anomaly, or parameter mismatch), the FFT is determined to be invalid and the process enters the third-level strategy.

[0071] This algorithm is switched when the proportion of effective vectors in the first level falls below a threshold (40%). It is used for high-efficiency, large-area background flow field inversion. According to the convolution theorem, correlation operations in the spatial domain are equivalent to conjugate products in the frequency domain:

[0072]

[0073] in, This represents the correlation function (cross-correlation result) calculated using the FFT method. This represents the Fast Fourier Transform. This represents the complex conjugate operation. This represents the Inverse Fourier Transform (FFT). It extracts the overall translation between two image frames through spectral multiplication, achieving a computation speed 1-2 orders of magnitude faster than the spatial domain DCC. The FFT significantly improves the processing speed of massive amounts of satellite data.

[0074] The third-level strategy is to use neighborhood reference interpolation as a fallback solution, use the spatial continuity assumption to estimate the flow velocity, and then enter the sub-pixel fitting stage.

[0075] When both DCC and FFT algorithms fail, a third-level strategy is triggered, employing the simplest neighborhood reference interpolation method as a fallback solution. This method utilizes the spatial continuity assumption for velocity estimation, ensuring the flow field is free of data gaps. The Quick PIV algorithm exhibits extremely high robustness and is virtually indestructible, providing a final guarantee of reliability for the entire system. Its core lies in the displacement prediction formula:

[0076]

[0077] The rough displacement obtained from the previous large window (e.g., 96 pixels). Based on this, the sub-windows are translated and corrected before the relevant calculations for the small window (24 pixels) are performed. This ensures that a physically continuous flow field can still be obtained in feature-sparse regions (such as pure water).

[0078] Its essence is an iterative approximation strategy "from global to local". (Prediction operator) represents the first round of large-scale query window (e.g. The displacement vector calculated within a pixel represents the background flow field or large-scale circulation structure within the ocean area. Due to its large window size, it can capture motions with large displacements, but its spatial resolution is low and it cannot reflect the details of the flow field (such as the sharp shear at the edge of a vortex). It provides a "basic reference value" or "initial momentum" for the next round of calculations. (Modification operator) indicates the second round of small-scale query window (e.g.) In pixels, the fine-tuned displacement is calculated after "translation correction" based on the results of the previous round. The algorithm is based on... The size of the second frame image is adjusted by shifting the corresponding window forward (Shift). The distance is calculated, and then a small-scale correlation is plotted at that location. This represents sub-mesoscale details, flow field shear, and corrections for deviations in the large background flow field caused by small-scale turbulence. The final composite vector represents the vector sum of the background flow field displacement and the local fine correction, which maintains the continuity of large-scale motion and achieves extremely high spatial resolution.

[0079] The subpixel fitting stage uses a two-dimensional Gaussian surface fitting method to locate the peak center based on the obtained integer pixel displacement peaks, and obtains the displacement amount with subpixel accuracy. A quality check is then performed to ensure that at least one valid vector exists, and finally, a high-precision sea surface current field vector is obtained.

[0080] Subpixel positioning: Regardless of whether the integer pixel displacement peaks are obtained through DCC, FFT or Quick PIV, the peak center is located using a two-dimensional Gaussian surface fitting method to obtain the displacement amount with subpixel accuracy.

[0081] By locally fitting the relevant plane, the displacement accuracy is improved from integers to the 0.1 pixel level, resulting in integer pixel displacements. The sub-pixel correction amount is obtained by parabolic fitting:

[0082]

[0083]

[0084] in, The maximum value of the correlation peak is the point with the highest value on the correlation plane during integer pixel-level search, representing the initially determined best matching position. and The values ​​represent the correlation strengths of adjacent peaks, respectively, indicating the maximum values ​​of the correlation peaks in the horizontal direction. Correlation coefficients at adjacent pixels on the left and right sides (direction). and The correlation strength is the vertical (y-direction) adjacent strength.

[0085] Natural logarithm operator: Since the peak of the cross-correlation function theoretically conforms to a two-dimensional Gaussian distribution, taking the logarithm can transform the nonlinear exponential distribution into a linear quadratic parabolic distribution.

[0086] The denominator is essentially the second-order central difference of the logarithm of the correlation function, representing the "curvature" or "sharpness" of the correlation peak.

[0087] Subpixel correction. It is the true physical peak value relative to the integer peak value. The offset, whose value range is usually within Between pixels. This breaks through the limitations of discrete grids, making the inverted velocity field smoother and directly improving the reliability of subsequent gradient calculations.

[0088] Pixel displacement is converted into physically meaningful vector fields and deformation fields. The physical flow velocity conversion formula is:

[0089]

[0090] In the formula: This refers to the physical velocity of the current in the horizontal direction (x-direction) (unit: m / s), which is the actual ocean current velocity in the east-west direction. This refers to the physical velocity of the current in the vertical direction (y-direction) (unit: m / s), that is, the actual ocean current velocity in the north-south direction. and These are the pixel displacements in the horizontal and vertical directions (including sub-pixel precision correction). =0.25 km, which is the pixel resolution of GOCI-II; =3600s=1h is the observation time interval; 1000 is the conversion factor from km to m.

[0091] Final quality check: Perform a unified verification on the final output of all algorithms to ensure that at least one valid vector exists. If a vector is completely invalid, skip the current image pair to avoid introducing invalid data.

[0092] Finally, high-precision sea surface current field vectors U and V are obtained through a three-level strategy and final quality check.

[0093] This three-level adaptive strategy has the following technical characteristics:

[0094] Algorithm complementarity: DCC is suitable for low signal-to-noise ratio regions, FFT is suitable for efficient batch processing, and Quick PIV provides a backup guarantee;

[0095] Quality tiered control: triple quality assurance at the algorithm level (40% threshold) + exception level (try-catch) + result level (>0 check);

[0096] A robust fault tolerance mechanism: a step-by-step degradation strategy ensures that reasonable results can still be output even under extreme conditions;

[0097] Computational efficiency optimization: Prioritize the use of high-precision DCC, switch to efficient FFT when necessary, and finally use QuickPIV.

[0098] S3: Based on high-precision sea surface current field vectors, the derivatives of the current field and its gradient are calculated using the central difference scheme, and the shear strain rate is extracted;

[0099] Shear strain rate This is not merely a mathematical result, but also a core technical feature of this invention for identifying marine dynamic characteristics. In a continuous fluid medium, the velocity gradient tensor describes the fluid's deformation. Shear strain rate Defined as the symmetric portion of the velocity gradient tensor, it characterizes the rate at which the fluid shape is distorted (rather than translated or rotated). This parameter is the core output metric of this invention, used for the precise location of sub-mesoscale fronts and shear zones in the ocean.

[0100] The shear strain rate in S3 The calculation formula is:

[0101]

[0102] in, The velocity in the horizontal direction (x-direction) is expressed in m / s. The velocity in the vertical direction (y-direction) is expressed in m / s. and These are the spatial coordinates in the east-west (horizontal) and north-south (vertical) directions, respectively. and Each is the current pixel The vertical velocity values ​​in the upper and lower neighborhoods. and Each is the current pixel The flow velocity values ​​in the left and right neighborhoods in the horizontal direction, and The actual geographic spatial grid spacing is represented by this value. The higher the value, the stronger the shearing force on the water body, which can easily lead to the aggregation or stretching and diffusion of red tide algae along the front. In this invention, it is determined by the pixel resolution. horizontal speed Rate of change along the latitudinal (vertical) direction; Vertical velocity The rate of change along the longitudinal (horizontal) direction. To solve this problem in a digitized satellite grid field, this invention employs a second-order accurate central difference operator.

[0103] In the post-processing algorithm, the core objective is to improve the completeness and reliability of ocean current field and strain rate data. First, the mean and standard deviation of each velocity vector in its neighborhood are calculated using a neighborhood standard deviation filtering method. Outlier vectors are then filtered out and marked using a threshold of twice the standard deviation, eliminating invalid data that deviates from a reasonable range. Second, for missing values ​​in the data, an inverse distance weighted interpolation method is used. Weights are constructed based on the distance between missing pixels and surrounding valid pixels (introducing a minimum value to avoid a zero denominator). The missing velocity vector values ​​are then filled in using a weighted average. Finally, through outlier filtering and missing value completion, the usability and accuracy of the data are significantly improved, providing a high-quality basic dataset for subsequent visualization enhancement and statistical quality assessment.

[0104] The formula for vector validation standard deviation filtering (Outlier Detection) is as follows:

[0105]

[0106] In the formula, The mean velocity of the neighborhood; The flow velocity value of the center pixel to be verified; The threshold coefficient is set to 2. Satellite chlorophyll images often exhibit false variations in brightness at cloud edges. The PIV algorithm may incorrectly capture the movement of these clouds instead of water. This algorithm is based on the fact that cloud movement typically moves much faster than ocean currents and in inconsistent directions. Through local statistical comparison, it can accurately identify these "pseudo-vectors" that are completely contrary to the direction of surrounding water flow, thereby automatically identifying and eliminating mismatched pseudo-vectors caused by cloud obstruction.

[0107] S4: Construct a binary logic mask matrix for the 75m isobath, and perform an AND operation based on the binary logic mask matrix, shear strain rate, and flow velocity amplitude to obtain the final red tide dynamic risk index, thereby realizing the monitoring of red tide dynamic evolution based on geostationary satellites.

[0108] This step aims to address the high false alarm rate problem of pure dynamic inversion in offshore areas by using seabed topographic data to impose strict spatial constraints on the monitoring range.

[0109] Limitations of purely kinetic indices: The shear strain rate calculated in step S3 above ( The shear rate and velocity amplitude only represent intense deformation and transport in the water body, providing the dynamic conditions (necessary conditions) for red tide accumulation. However, a sufficient material basis (adequate condition) is also required for a red tide outbreak, namely, a sufficient supply of nutrients. In the open ocean or outer continental shelf waters with depths greater than 75m, although strong shear current fields (such as branches of the Kuroshio Current or deep-sea eddies) also exist, the lack of terrestrial nutrient input and deep water mixing result in a scarcity of surface nutrients, making it extremely difficult for high-concentration red tides to occur. Relying solely on shear strain rate for identification would misreport dynamic processes in the open ocean as red tide risks.

[0110] Construction of the 75m Depth Masking: Based on the aforementioned "dynamic-trophic coupling" mechanism, this invention selects the 75m isobath as the key threshold for distinguishing between eutrophic nearshore waters and oligotrophic offshore waters. The 75m threshold is selected based on historical red tide observation records and the geographical distribution characteristics of shelf slope breaks in the East China Sea.

[0111] Data mapping: Read GEBCO water depth data and interpolate it to the same grid coordinate system as the satellite imagery, denoted as... .

[0112] Binary mask generation: Constructing a binary logical mask matrix Areas with a water depth of 75m or less are defined as "potential red tide outbreak zones" (set to 1), and areas with a water depth greater than 75m are defined as "non-red tide background zones" (set to 0). The mathematical expression is:

[0113]

[0114] Combining the previous formula, the red tide potential index is obtained as follows:

[0115]

[0116] This formula is based on the modern ocean turbulence-ecology coupling theory. Its core mechanism is that high shear strain rate promotes the vertical mixing and transport of deep nutrients to the surface, and high flow velocity accelerates the horizontal advection transport of nutrients. The product of the two represents the nutrient supply capacity of the ocean dynamic system. In nutrient-limited sea areas, the areas with greater dynamic intensity (such as upwelling areas, frontal convergence areas, and estuarine plume areas) have more abundant nutrient supply, thus providing a material basis for the rapid proliferation of algae. This allows the formula to accurately identify key sea areas with high red tide incidence, which is consistent with the red tide distribution patterns in nutrient-limited sea areas such as the East China Sea and modern marine ecological understanding.

[0117] Final Risk Index Calculation: The original dynamic index is obtained by performing an AND operation (spatial filtering) on ​​the generated depth mask to obtain the final red tide dynamic risk index. :

[0118]

[0119] in, Shear strain rate For the velocity amplitude, It is a binary logic mask matrix for the 75m isobath.

[0120] This step forces the risk values ​​in deep water areas east of the 75m isobath to be reset to zero, retaining only high-risk signals with "strong shear + strong current velocity" characteristics in nearshore shallow water areas. This not only aligns with the biological mechanisms of red tide occurrence but also significantly reduces the false alarm rate of the monitoring system.

[0121] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of the invention.

Claims

1. A method for monitoring the dynamic evolution of red tides based on geostationary satellites, characterized in that, Includes the following steps: S1: Acquire satellite multi-source data and perform spatiotemporal standardization processing on the multi-source data to generate a standardized time-series image matrix; S2: A three-level adaptive processing strategy is used to perform flow field inversion on the standardized time-series image matrix and output a high-precision sea surface flow field vector. S3: Based on high-precision sea surface current field vectors, the derivatives of the current field and its gradient are calculated using the central difference scheme, and the shear strain rate is extracted; S4: Construct a binary logic mask matrix for the 75m isobath, and perform an AND operation based on the binary logic mask matrix, shear strain rate, and flow velocity amplitude to obtain the final red tide dynamic risk index, thereby realizing the monitoring of red tide dynamic evolution based on geostationary satellites.

2. The method for monitoring the dynamic evolution of red tides based on geostationary satellites according to claim 1, characterized in that, S1 includes the following sub-steps: S11: Acquire multi-source satellite data, and perform regional stitching and resampling on the multi-source satellite data to form a time-series image matrix; S12: Based on the time series image matrix, perform quality control and hole filling to generate a standardized time series image matrix.

3. The method for monitoring the dynamic evolution of red tides based on geostationary satellites according to claim 1, characterized in that, The specific logic of the three-level adaptive processing strategy in S2 is as follows: First-level strategy: Use the direct cross-correlation algorithm to calculate all image windows, count the proportion of effective vectors. If the proportion of effective vectors is ≥40%, accept the direct cross-correlation result and enter the sub-pixel fitting stage; otherwise, enter the second-level strategy. The second-level strategy is to use fast Fourier transform to calculate the cross-correlation function in the frequency domain and perform quality control based on the anomaly capture mechanism. If no anomaly is thrown during the execution process, the result is accepted and the process enters the sub-pixel fitting stage. If an anomaly occurs, proceed to the third-level strategy; The third-level strategy is to use neighborhood reference interpolation as a fallback solution, use the spatial continuity assumption to estimate the flow velocity, and then enter the sub-pixel fitting stage. The subpixel fitting stage uses a two-dimensional Gaussian surface fitting method to locate the peak center based on the obtained integer pixel displacement peaks, and obtains the displacement amount with subpixel accuracy. A quality check is then performed to ensure that at least one valid vector exists, and finally, a high-precision sea surface current field vector is obtained.

4. The method for monitoring the dynamic evolution of red tides based on geostationary satellites according to claim 1, characterized in that, The shear strain rate in S3 The calculation formula is: in, The flow velocity is in the horizontal direction. The vertical flow velocity. and These are the spatial coordinates in the horizontal and vertical directions, respectively. and Each is the current pixel The vertical velocity values ​​in the upper and lower neighborhoods. and Each is the current pixel The flow velocity values ​​in the left and right neighborhoods in the horizontal direction, and This represents the actual geographic spatial grid spacing.

5. The method for monitoring the dynamic evolution of red tides based on geostationary satellites according to claim 1, characterized in that, The method for constructing the binary logic mask matrix in S4 is as follows: If a region with a water depth of less than or equal to 75m is defined as a potential red tide outbreak zone, then the binary logic mask matrix is ​​1. If the area with a water depth greater than 75m is defined as the non-red tide background area, then the binary logic mask matrix is ​​0.

6. The method for monitoring the dynamic evolution of red tides based on geostationary satellites according to claim 1, characterized in that, The red tide dynamics risk index in S4 The calculation formula is: in, Shear strain rate For the velocity amplitude, It is a binary logic mask matrix for the 75m isobath.

Images