A daily green tide comprehensive coverage information fusion method based on satellite remote sensing and ecological model data

Abstract

The application provides a daily Enteromorpha prolifera green tide comprehensive coverage information fusion method based on satellite remote sensing and ecological model data, and belongs to the technical field of Enteromorpha prolifera green tide monitoring. The application forms a complete fusion process including data preprocessing, reliability evaluation, dynamic weight configuration, deep fusion and post-processing by means of radiation calibration and spectral consistency processing of multi-source satellite remote sensing data, reliability evaluation and dynamic weight configuration of satellite remote sensing and ecological model data, multi-dimensional deep fusion of spectral consistency enhancement models, bidirectional buffer and weighted fusion processing and Douglas-Pork boundary optimization, and solves the technical problems of insufficient continuity and low accuracy of Enteromorpha prolifera green tide monitoring caused by the lack of effective fusion mechanism of multi-source satellite remote sensing data and ecological model data.

Description

Technical Field

[0001] This invention belongs to the field of Ulva prolifera green tide monitoring technology, specifically, it relates to a method for integrating daily Ulva prolifera green tide comprehensive coverage information based on satellite remote sensing and ecological model data. Background Technology

[0002] Monitoring Ulva prolifera green tides has significant application value in marine ecological early warning. Traditional Ulva prolifera green tide monitoring mainly relies on the interpretation of single satellite remote sensing data, obtaining distribution information through direct observation of the Ulva prolifera green tide coverage area via satellite sensors. However, in practical applications, due to the differences in spectral response characteristics of different sensors, atmospheric interference, cloud cover, and other factors, the reliability of single remote sensing data is limited, making it difficult to provide continuous Ulva prolifera coverage monitoring results. Meanwhile, Ulva prolifera ecological dynamics models can simulate and predict based on environmental parameters such as sea surface temperature and wind field, but the accuracy of the model output results is constrained by the precision of parameter input and model assumptions. Existing technologies mostly use either remote sensing interpretation or ecological models separately to obtain Ulva prolifera green tide information, lacking an organic fusion mechanism between the two data sources. Especially in cases of severe cloud cover, solar flare pollution, or deterioration of single remote sensing data quality, it is impossible to effectively compensate for the deficiencies of remote sensing data; and when model parameters fluctuate or initial conditions are uncertain, it is also difficult to fully utilize reliable remote sensing observations to constrain model results. The isolated existence of information from the two data sources makes it difficult to guarantee the continuity and accuracy of Ulva prolifera monitoring. In other words, existing technologies suffer from a lack of dynamic weighting and deep fusion mechanisms for multi-source satellite remote sensing data and ecological model data, making it difficult to generate high-quality, highly continuous daily comprehensive coverage information of Ulva prolifera green tides. Summary of the Invention

[0003] In view of this, the present invention provides a method for integrating daily green tide coverage information of Ulva prolifera based on satellite remote sensing and ecological model data, which can solve the technical problem in the prior art that it is difficult to effectively integrate multi-source satellite remote sensing data and ecological model data in Ulva prolifera monitoring.

[0004] This invention is implemented as follows: It provides a method for fusing daily comprehensive coverage information of *Ulva prolifera* green tides based on satellite remote sensing and ecological model data. This involves acquiring multi-source satellite remote sensing image data and preprocessing it to obtain satellite remote sensing interpretation vector results. Simultaneously, a *Ulva prolifera* ecological dynamics model driven by the satellite remote sensing interpretation results is constructed to output simulated *Ulva prolifera* coverage vector results. A cross-calibration model between sensors is established to eliminate differences in spectral response between different sensors. The reliability of the satellite remote sensing interpretation vector results and the simulated *Ulva prolifera* coverage vector results is assessed, a reliability index value is calculated, and a dynamic weight allocation rule is established. A fusion rule based on expert knowledge is established, selecting the data source or fusion method according to cloud identifier values ​​and cloud coverage values. The satellite remote sensing interpretation vector results are then processed. The translated vector results and the simulated vector results of Ulva prolifera cover are used to create a buffer and perform a bidirectional spatial connection operation. The similarity coefficient value is calculated to screen valid matching pairs. The valid matching pairs are input into the spectral consistency enhancement model for processing. The spectral consistency enhancement model extracts features simultaneously in the spatial, spectral, and temporal dimensions by unfolding the convolutional layer in multiple dimensions. The attention mechanism is dynamically adjusted according to environmental parameters through a multi-head attention layer to achieve deep fusion of observation data from different sensors at the feature level. The weighted fusion method is used to calculate the fused attribute values ​​and smooth the boundaries. The symmetric difference method is used to remove overlapping parts. The fused data is merged with the independent parts to form a unified vector dataset, and the daily Ulva prolifera green tide comprehensive coverage product is output.

[0005] The preprocessing includes radiometric calibration, atmospheric correction, geometric correction, and cloud extraction. Radiometric calibration converts the digital quantization values ​​of satellite remote sensing image data into radiance values. Atmospheric correction eliminates the influence of atmospheric scattering and absorption on radiance values. Geometric correction registers radiance values ​​to a unified geographic coordinate system. Cloud extraction identifies cloud-covered areas in satellite remote sensing image data and generates cloud identifier vector data.

[0006] The Ulva prolifera ecodynamic model is driven by satellite remote sensing interpretation results. It takes sea surface temperature data, wind field data, nutrient concentration data, precipitation data and shortwave radiation data as input, and outputs the Ulva prolifera cover simulation vector results at the time of satellite transit on the same day. The satellite remote sensing interpretation vector results and the Ulva prolifera cover simulation vector results are projected onto the China Geodetic Coordinate System 2000 and the spatial resolution is unified to 50m and the data format is vector format.

[0007] Among these measures, a cross-calibration model among sensors was established to obtain the radiometric conversion coefficients among the sensors through the pseudo-invariant target method and synchronous observation experiments. Spectral resampling technology was used to unify the band configuration. Histogram matching method and radiometric normalization method were used to eliminate systematic radiometric bias. The observation values ​​of each sensor were converted to the radiometric reference of the reference sensor, which is the radiometric reference of the HY-1C satellite coastal zone imager.

[0008] Among them, the pseudo-invariant target method selects a target with long-term stable and uniform surface reflectance in the study area as the radiation benchmark. The targets include desert areas, saline-alkali areas and deep sea areas. By performing regression analysis on the observation data of multiple sensors relative to the same pseudo-invariant target at different times, the radiation conversion relationship between the sensors is established.

[0009] The reliability of the satellite remote sensing interpretation vector results is assessed based on cloud coverage and solar flare index values. The cloud coverage value is obtained by calculating the ratio of the cloud coverage area to the total area of ​​the fused region using cloud identifier vector data. The solar flare index value is obtained by extracting solar flare region vector data and calculating the ratio of the solar flare area to the total area of ​​the fused region.

[0010] The reliability of the simulated vector results of Ulva prolifera cover was assessed by comparing them with historical observation data and by parameter sensitivity analysis. The parameter sensitivity analysis was conducted by changing the numerical range of the input parameters of the Ulva prolifera ecodynamic model, observing the magnitude of change in the simulated vector results of Ulva prolifera cover, and evaluating the sensitivity of the simulated vector results of Ulva prolifera cover to changes in input parameters.

[0011] The reliability index of satellite remote sensing interpretation results is calculated by comprehensively considering cloud coverage, solar flare index, and image signal-to-noise ratio. The reliability index of ecological model output results is calculated by comprehensively considering the consistency between the simulated vector result of seaweed cover and historical observation data and the parameter sensitivity analysis result. The values ​​of the reliability index of satellite remote sensing interpretation results and the reliability index of ecological model output results range from 0 to 1.

[0012] Specifically, when the satellite remote sensing interpretation vector results show that the seaweed is covered and the cloud identifier value in the cloud identifier vector data is 0, the satellite remote sensing interpretation vector results are used as the daily comprehensive coverage information of the seaweed green tide. When the cloud identifier value in the cloud identifier vector data is 1, the seaweed coverage simulation vector results are used as the daily comprehensive coverage information of the seaweed green tide. When the satellite remote sensing interpretation vector results can monitor some sea areas, the satellite remote sensing interpretation vector results are used as the benchmark and the seaweed coverage simulation vector results are fused using a two-way buffer overlay analysis method.

[0013] Among them, the bidirectional buffer overlay analysis method creates buffers for the Ulva prolifera cover vector data in the satellite remote sensing interpretation vector results and the Ulva prolifera cover vector data in the Ulva prolifera cover simulation vector results. The buffer distance value of the buffer is determined according to the square root of the area value of each polygon in the Ulva prolifera cover vector data. A bidirectional spatial connection operation is performed on the two buffers, and the common coverage area of ​​the two buffers is output as the overlapping part.

[0014] In this process, a similarity coefficient is calculated for each pair of matching elements, and a similarity coefficient threshold of 0.6 is set to filter valid matching pairs. The similarity coefficient is the ratio of the intersection area to the union area of ​​the matching element pairs. The filtered valid matching pairs are then input into the spectral consistency enhancement model for processing.

[0015] The input layer of the spectral consistency enhancement model receives multi-source remote sensing feature vectors and environmental parameter vectors of the matching region. The multi-source remote sensing feature vectors include normalized vegetation index, normalized water index, and seaweed cover density. The environmental parameter vectors include cloud cover, solar flare index, and sea surface temperature gradient.

[0016] The input layer is connected to a multi-dimensional unfolded convolutional layer, which expands the two-dimensional convolution into three-dimensional space, and performs feature extraction simultaneously in the spatial, spectral, and temporal dimensions. The output of the multi-dimensional unfolded convolutional layer is connected to a multi-head attention layer, and the attention head value of the multi-head attention layer is determined based on the cloud cover value, solar flare index value, and sea surface temperature gradient value.

[0017] Specifically, a weighted fusion method is used to calculate the fused attribute values ​​for the matched area. The weighted fusion method calculates the attribute values ​​of each pixel or vector element in the fused area based on the weight values ​​of the satellite remote sensing interpretation results and the weight values ​​of the ecological model output results. The weight values ​​of the satellite remote sensing interpretation results and the weight values ​​of the ecological model output results are dynamically adjusted according to the comprehensive pollution index value.

[0018] Among them, the boundary of the fused Ulva cover vector is optimized using the Douglas-Puk boundary smoothing algorithm. The Douglas-Puk boundary smoothing algorithm retains feature points on the boundary that deviate from the original boundary by a distance greater than the distance threshold by setting a distance threshold, and removes redundant points that deviate from the original boundary by a distance less than the distance threshold.

[0019] Specifically, the symmetric difference method is used to remove the overlapping part of the Ulva coverage area in the satellite remote sensing interpretation vector results and the Ulva coverage simulation vector results, while retaining the independent parts of the satellite remote sensing interpretation vector results and the Ulva coverage simulation vector results. The fused data is then combined with the independent parts of the satellite remote sensing interpretation vector results and the Ulva coverage simulation vector results to form a unified vector dataset, and the daily Ulva green tide comprehensive coverage product is output.

[0020] This invention establishes a dual-source data fusion framework combining satellite remote sensing interpretation results and the output of a *Ulva prolifera* ecodynamics model. It employs a dynamic weight allocation strategy based on reliability assessment and a spectral consistency enhancement model to achieve deep fusion of multi-source remote sensing data. In the data preprocessing stage, cross-sensor cross-calibration and radiometric normalization methods eliminate systematic spectral response differences between different sensors, ensuring the radiometric consistency of the multi-source data. In the fusion stage, the reliability of the two data sources is quantitatively assessed based on quality factors such as cloud cover and solar flare contamination, and the weight allocation rules are dynamically adjusted, giving higher-reliability data sources greater weight in the fusion process. This invention introduces a spectral consistency enhancement model, employing multi-dimensional unrolled convolution and multi-head attention mechanisms to simultaneously extract and fuse features in spatial, spectral, and temporal dimensions, enhancing the comprehensive understanding of *Ulva prolifera* cover information. A bidirectional buffer overlay analysis method accurately identifies overlapping and independent coverage areas between the two data sources, and a weighted fusion method flexibly adjusts weights according to the degree of regional contamination, overcoming the limitations of traditional single-weight fusion schemes that cannot adapt to changing environmental conditions. The Douglas-Puk boundary smoothing algorithm further optimizes the vector boundary of the fused *Ulva prolifera* coverage, removing fragmentation features. In summary, this invention addresses the technical problem of insufficient continuity and accuracy in *Ulva prolifera* monitoring caused by the lack of an effective fusion mechanism between multi-source satellite remote sensing data and ecological model data. This is achieved through unified radiometric calibration of multi-source remote sensing data, reliability-based dynamic weight allocation, deep fusion of a spectral consistency enhancement model, and intelligent boundary optimization. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a time series graph showing the changes in the coverage area of ​​*Ulva prolifera* in Example 2.

[0023] Figure 3 This is a comparison diagram of the space covered by Ulva prolifera before and after fusion in Example 2. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0025] like Figure 1 The diagram shows a flowchart of a method for integrating daily Ulva prolifera green tide coverage information based on satellite remote sensing and ecological model data, provided by this invention. This method includes the following steps:

[0026] S01. Collect multi-source satellite remote sensing image data and perform preprocessing. The preprocessing includes radiometric calibration, atmospheric correction, geometric correction, and cloud extraction. Radiometric calibration converts the digital quantization values ​​of the satellite remote sensing image data into radiance values. Atmospheric correction eliminates the influence of atmospheric scattering and absorption on radiance values. Geometric correction registers radiance values ​​to a unified geographic coordinate system. Cloud extraction identifies cloud-covered areas in the satellite remote sensing image data and generates cloud identifier vector data. The satellite remote sensing interpretation vector results of daily Ulva prolifera coverage are extracted, and data missing areas and low-quality areas are marked. At the same time, a Ulva prolifera ecological dynamics model driven by remote sensing data is constructed. The Ulva prolifera ecological dynamics model takes sea surface temperature data, wind field data, nutrient concentration data, precipitation data, and shortwave radiation data as inputs and outputs the simulated vector results of Ulva prolifera coverage at the time of satellite transit on the same day. The satellite remote sensing interpretation vector results and the simulated vector results of Ulva prolifera coverage are projected onto the China Geodetic Coordinate System 2000 and the spatial resolution is unified to 50m and the data format is vector format.

[0027] S02. Establish a cross-calibration model among sensors to address the differences in spectral responses among different sensors. Obtain the radiometric conversion coefficients among the sensors through the pseudo-invariant target method and synchronous observation experiments. Use spectral resampling technology to unify band configuration. Use histogram matching method and radiometric normalization method to eliminate systematic radiometric bias. Convert the observation values ​​of each sensor to the radiometric reference of the reference sensor. The radiometric reference of the reference sensor is the radiometric reference of the HY-1C satellite coastal zone imager to ensure the radiometric consistency of multi-source remote sensing data.

[0028] S03. Conduct a reliability assessment. The reliability of the satellite remote sensing interpretation vector results is assessed based on the cloud coverage value and the solar flare index value. The cloud coverage value is obtained by calculating the ratio of the cloud coverage area to the total area of ​​the fused region using cloud identifier vector data. The solar flare index value is obtained by extracting the solar flare region vector data and calculating the ratio of the solar flare area to the total area of ​​the fused region. The reliability of the simulated vector results of Ulva prolifera coverage is assessed by comparing with historical observation data and performing parameter sensitivity analysis. The reliability index values ​​of the satellite remote sensing interpretation results and the ecological model output results are calculated. A dynamic weight allocation rule is established based on the reliability index values ​​of the satellite remote sensing interpretation results and the ecological model output results.

[0029] S04. Establish fusion rules based on expert knowledge. When the satellite remote sensing interpretation vector result shows that it is covered by seaweed and the cloud identifier value in the cloud identifier vector data is 0, the satellite remote sensing interpretation vector result is used as the daily comprehensive coverage information of seaweed green tide. When the cloud identifier value in the cloud identifier vector data is 1, the seaweed coverage simulation vector result is used as the daily comprehensive coverage information of seaweed green tide. When the satellite remote sensing interpretation vector result can monitor part of the sea area, the satellite remote sensing interpretation vector result is used as the benchmark and the seaweed coverage simulation vector result is fused with the two results through a two-way buffer overlay analysis method. The comprehensive pollution index value is calculated based on the cloud coverage value and solar flare index value in the fused area. The weight value of the satellite remote sensing interpretation result and the weight value of the ecological model output result are dynamically adjusted based on the comprehensive pollution index value.

[0030] S05. Execute the fusion algorithm, creating buffers for the *Ulva prolifera* cover vector data in the satellite remote sensing interpretation vector results and the *Ulva prolifera* cover vector data in the simulated vector results. The buffer distance value of the buffer is determined based on the square root of the area value of each polygon in the *Ulva prolifera* cover vector data. Perform a bidirectional spatial connection operation on the two created buffers and output the common coverage area. Calculate the similarity coefficient value for each pair of matching elements and set the similarity coefficient threshold to 0.6 to filter valid matching pairs. Input the filtered valid matching pairs into the spectral consistency enhancement model for processing. The spectral consistency enhancement model determines the attention head value in the multi-head attention mechanism based on the cloud cover value, solar flare index value, and sea surface temperature gradient value. Calculate the fused attribute value for the matching area using a weighted fusion method. Optimize the boundary of the fused *Ulva prolifera* cover vector using the Douglas-Pock boundary smoothing algorithm to remove unreasonable boundaries and fragments.

[0031] S06. Create daily comprehensive coverage information for *Ulva prolifera* green tides. Use the symmetric difference method to remove the overlapping portion of *Ulva prolifera* coverage area in the satellite remote sensing interpretation vector results and the *Ulva prolifera* coverage simulation vector results. Retain the independent portions of the satellite remote sensing interpretation vector results and the independent portions of the *Ulva prolifera* coverage simulation vector results. Merge the fused data with the independent portions of the satellite remote sensing interpretation vector results and the independent portions of the *Ulva prolifera* coverage simulation vector results to form a unified vector dataset. Calculate the comprehensive coverage area value of *Ulva prolifera* and output daily comprehensive coverage information for *Ulva prolifera* green tides. The daily comprehensive coverage information for *Ulva prolifera* green tides includes coverage area data, comprehensive coverage area value, location coordinate data, and density level data. The format of the daily comprehensive coverage information for *Ulva prolifera* green tides is raster data format and vector boundary data format.

[0032] The pseudo-invariant target method selects a target with long-term stable and uniform surface reflectance within the study area as a radiation benchmark. The target includes desert areas, saline-alkali areas, and deep-sea areas. By performing regression analysis on the observation data of multiple sensors relative to the same pseudo-invariant target at different times, a radiation conversion relationship between the sensors is established. The radiation conversion relationship is used to correct the systematic radiation bias between different sensors.

[0033] The synchronous observation experiment involves conducting ground and sea surface measurements simultaneously during satellite transit to obtain the true spectral characteristic data of the target object. The true spectral characteristic data is then compared with the observation values ​​of each sensor to establish a quantitative relationship between the sensor observation values ​​and the true spectral characteristic data. This quantitative relationship is used to correct sensor observation biases.

[0034] The radiation conversion coefficient is a coefficient that describes the linear conversion relationship between observations from different sensors. The radiation conversion coefficient is obtained by least squares regression analysis on multi-temporal observation data of pseudo-invariant targets. The radiation conversion coefficient is used to convert the observations of the sensor to be calibrated into observations under the radiation reference of the reference sensor.

[0035] The spectral resampling technology converts the spectral bands of different sensors to a unified band configuration through mathematical interpolation methods. These mathematical interpolation methods include linear interpolation and cubic spline interpolation. By performing convolution operations on the spectral response functions of each sensor, the spectral response characteristics of a reference sensor are simulated, thereby achieving spectral consistency of observation data from different sensors.

[0036] The histogram matching method involves adjusting the histogram of the image to be corrected to match the histogram of the reference image. Figure 1 The process involves calculating the cumulative distribution function of the image to be corrected and the reference image, establishing a grayscale value mapping relationship, and reassigning the grayscale value of each pixel in the image to be corrected so that the grayscale distribution of the corrected image is consistent with that of the reference image.

[0037] The radiation normalization method eliminates radiation differences caused by variations in solar elevation angle, atmospheric conditions, and sensor performance by establishing a radiation conversion model between images from different time phases or different sensors. The radiation conversion model is established by selecting invariant feature points in the images and performing linear or nonlinear regression to normalize the radiation values ​​of all images to a unified radiation reference.

[0038] The systematic radiation bias is the systematic difference between different sensor observations caused by differences in sensor design, calibration errors, and atmospheric correction residuals. The systematic radiation bias is eliminated by radiation normalization.

[0039] The cloud coverage value is the ratio of the cloud coverage area to the total area of ​​the fused region. The cloud coverage area is calculated using cloud identifier vector data. The cloud coverage value is used to evaluate the reliability of the satellite remote sensing interpretation vector results. When the cloud coverage value ∈ [0.5, 1], the reliability of the satellite remote sensing interpretation vector results is low, and when the cloud coverage value ∈ [0, 0.3), the reliability of the satellite remote sensing interpretation vector results is high.

[0040] The solar flare index is the ratio of the solar flare area calculated based on the solar flare region vector data to the total area of ​​the fused region. The solar flare index is used to evaluate the impact of image quality on the satellite remote sensing interpretation vector results. When the solar flare index value ∈ [0.3, 1], the satellite remote sensing interpretation vector results are significantly affected by solar flare contamination, and when the solar flare index value ∈ [0, 0.1), the impact of solar flare contamination is relatively small.

[0041] The parameter sensitivity analysis involves changing the numerical range of the input parameters of the Ulva prolifera ecodynamic model, observing the variation range of the Ulva prolifera cover simulation vector results, and evaluating the sensitivity of the Ulva prolifera cover simulation vector results to changes in input parameters. When the variation range of the input parameters is 20%, the variation range of the Ulva prolifera cover simulation vector results is less than 10%, and the stability of the Ulva prolifera cover simulation vector results is considered to be high.

[0042] The reliability index of the satellite remote sensing interpretation results is an evaluation index calculated by comprehensively considering cloud coverage, solar flare index, and image signal-to-noise ratio. The value of the reliability index ranges from 0 to 1. The larger the reliability index, the higher the reliability of the satellite remote sensing interpretation vector results. The reliability index is used to determine the weight allocation of the satellite remote sensing interpretation vector results during the fusion process.

[0043] The reliability index of the ecological model output is an evaluation index calculated by comprehensively considering the consistency between the simulated vector results of Ulva prolifera cover and historical observation data and the results of parameter sensitivity analysis. The reliability index of the ecological model output ranges from 0 to 1. The larger the reliability index of the ecological model output, the higher the reliability of the simulated vector results of Ulva prolifera cover. The reliability index of the ecological model output is used to determine the weight allocation of the simulated vector results of Ulva prolifera cover during the fusion process.

[0044] The comprehensive pollution index is the product of cloud cover value and first weighting coefficient, plus the product of solar flare index value and second weighting coefficient. The first and second weighting coefficients are dynamically adjusted according to regional and seasonal characteristics. When the region is in the rainy season, the value of the first weighting coefficient increases, and when the region is in the sunny season, the value of the second weighting coefficient increases. The comprehensive pollution index is used to determine high-pollution areas and low-pollution areas.

[0045] The weight value of the satellite remote sensing interpretation result is the ratio of the reliability index value of the satellite remote sensing interpretation result to the sum of the reliability index value of the satellite remote sensing interpretation result and the reliability index value of the ecological model output result. The weight value of the satellite remote sensing interpretation result is used to calculate the fused attribute value in the weighted fusion method.

[0046] The weight value of the ecological model output result is the ratio of the reliability index value of the ecological model output result to the sum of the reliability index values ​​of the satellite remote sensing interpretation result and the reliability index value of the ecological model output result. The weight value of the ecological model output result is used to calculate the fused attribute value in the weighted fusion method.

[0047] The bidirectional buffer overlay analysis method creates buffers for the *Ulva prolifera* cover vector data in the satellite remote sensing interpretation vector results and the *Ulva prolifera* cover simulation vector results, respectively. The buffer distance value of the buffer is determined based on the square root of the area value of each polygon in the *Ulva prolifera* cover vector data. A bidirectional spatial connection operation is performed on the two buffers. The bidirectional spatial connection operation includes a connection operation from the satellite remote sensing interpretation vector results to the *Ulva prolifera* cover simulation vector results and a connection operation from the *Ulva prolifera* cover simulation vector results to the satellite remote sensing interpretation vector results. The common coverage area of ​​the two buffers is output as the overlapping part.

[0048] The buffer distance value is a distance parameter used to create the buffer. The buffer distance value is obtained by calculating the square root of the area value of each polygon in the Ulva prolifera coverage vector data, and the unit is meters.

[0049] The bidirectional spatial connection operation involves performing a spatial connection from two data sources to each other, ensuring that the overlap between the two data sources is fully identified.

[0050] The similarity coefficient is the ratio of the intersection area to the union area of ​​the matched feature pairs. The similarity coefficient is used to quantify the spatial overlap of two vector features. When the similarity coefficient is ≥0.6, the two vector features are considered to be matched effectively. When the similarity coefficient is <0.6, the match is considered invalid and the matched feature pairs are removed.

[0051] The similarity coefficient threshold is a criterion used to filter valid matching pairs, and the similarity coefficient threshold is set to 0.6.

[0052] The structure of the spectral consistency enhancement model is as follows: the input layer receives multi-source remote sensing feature vectors and environmental parameter vectors of the matching region. The multi-source remote sensing feature vectors include normalized vegetation index (NVI), normalized water index (SVI), and *Ulva prolifera* cover density. The environmental parameter vectors include cloud cover, solar flare index, and sea surface temperature gradient. The input layer is connected to a multi-dimensional unfolded convolutional layer, which extends two-dimensional convolution to three-dimensional space, simultaneously extracting features in spatial, spectral, and temporal dimensions. The output of the multi-dimensional unfolded convolutional layer is connected to a multi-head attention layer. The attention head value of the multi-head attention layer is determined based on the cloud cover, solar flare index, and sea surface temperature gradient. The solar flare index and sea surface temperature gradient values ​​are determined. When the cloud cover value ∈ [0, 0.3) and the solar flare index value ∈ [0, 0.1), the attention head value is set to 8. When the cloud cover value ∈ [0.3, 0.5] or the solar flare index value ∈ [0.1, 0.3), the attention head value is set to 4. When the cloud cover value ∈ (0.5, 1) or the solar flare index value ∈ [0.3, 1], the attention head value is set to 2. The output of the multi-head attention layer is connected to the fully connected layer. The fully connected layer outputs the fused Ulva prolifera cover feature vector. The fused Ulva prolifera cover feature vector is processed by the activation function and then outputs the final Ulva prolifera cover information.

[0053] The steps for establishing the training dataset for the spectral consistency enhancement model include collecting historical multi-source satellite remote sensing data on *Ulva prolifera* and corresponding time-series simulation data of *Ulva prolifera* ecological dynamics models; screening the collected data for quality and removing samples with cloud cover values ​​>0.8 or severe data loss; dividing the screened data into training, validation, and test sets according to time series, with the training set accounting for 70% of the total data, the validation set accounting for 15%, and the test set accounting for 15%; extracting multi-source remote sensing feature vectors and environmental parameter vectors as inputs for each sample in the training set; and extracting manually labeled ground truth values ​​of *Ulva prolifera* cover as output labels, which are verified and determined through field observation data and high-resolution UAV imagery.

[0054] The training steps of the spectral consistency enhancement model include initializing the spectral consistency enhancement model parameters and setting the learning rate to 0.001; inputting the training set data into the spectral consistency enhancement model for forward propagation with a batch size of 32; calculating the loss function value between the output of the spectral consistency enhancement model and the manually labeled ground truth value of Ulva prolifera cover; the loss function value is a weighted combination of the mean squared error loss function value and the cross-entropy loss function value; calculating the gradient of the loss function value with respect to the spectral consistency enhancement model parameters using the backpropagation algorithm; updating the spectral consistency enhancement model parameters using the adaptive moment estimation optimization algorithm; evaluating the performance of the spectral consistency enhancement model on the validation set after each training cycle; stopping training when the loss function value on the validation set does not decrease for 10 consecutive training cycles; and saving the spectral consistency enhancement model parameters with the best performance on the validation set as the final spectral consistency enhancement model.

[0055] The multi-dimensional unfolded convolutional layer extends traditional two-dimensional convolution to three-dimensional space, simultaneously performing feature extraction and pattern recognition in spatial, spectral, and temporal dimensions, enabling multi-scale and multi-angle analysis of *Ulva prolifera* cover information. In the spatial dimension, the convolution kernels in the multi-dimensional unfolded convolutional layer capture the spatial continuity and boundary morphology features of *Ulva prolifera* cover, extracting the geometric, directional, and scale features of *Ulva prolifera* patches, enhancing the ability to identify spatial distribution patterns of *Ulva prolifera*. In the spectral dimension, the convolution kernels in the multi-dimensional unfolded convolutional layer learn the correlations between spectral bands of different sensors, uncovering the intrinsic relationship between *Ulva prolifera* spectral features and environmental parameters. Through spectral resampling techniques and radiometric normalization methods, the convolution operations in the multi-dimensional unfolded convolutional layer effectively fuse spectral information from different sensors, eliminating inconsistencies caused by differences in spectral responses between sensors. In the temporal dimension, the convolution kernels in the multi-dimensional unfolded convolutional layer capture the dynamic changes in *Ulva prolifera* cover over time, learning the temporal characteristics of *Ulva prolifera* growth, drift, and extinction processes. Combined with the time-series simulation results of the *Ulva prolifera* ecodynamics model, the temporal convolution in the multi-dimensional unfolded convolutional layer can identify the periodicity and trend of *Ulva prolifera* cover changes, improving the accuracy of predicting dynamic changes in *Ulva prolifera*. Through the synergistic effect of the three dimensions, the multi-dimensional unfolded convolutional layer constructs a high-dimensional feature space for *Ulva prolifera* cover information, enabling the spectral consistency enhancement model to simultaneously consider the comprehensive influence of spatial distribution, spectral characteristics, and temporal evolution, significantly improving the accuracy and robustness of multi-source remote sensing data fusion. When processing *Ulva prolifera* monitoring data under variable environmental conditions, the multi-dimensional unfolded convolutional layer exhibits excellent feature extraction capabilities and generalization performance, effectively overcoming the limitations of traditional fusion methods in handling differences in sensor spectral responses. It achieves deep fusion of observation data from different sensors at the feature level, providing key technical support for generating high-precision, highly continuous daily comprehensive *Ulva prolifera* green tide cover information.

[0056] The Normalized Difference Vegetation Index (NDVI) is an index reflecting vegetation growth status, calculated by dividing the difference between near-infrared and red light reflectance by the sum of their respective reflectance values. The Normalized Difference Water Index (NDI) reflects water body distribution, calculated by dividing the difference between green and near-infrared reflectance by the sum of their reflectance values. The *Ulva prolifera* cover density is the density of *Ulva prolifera* cover per unit area, calculated by the ratio of *Ulva prolifera* cover area to the area of ​​the statistical unit in the satellite remote sensing interpretation vector results. The sea surface temperature gradient is the rate of change of sea surface temperature in space, calculated by dividing the difference in sea surface temperature data between adjacent pixels by the distance between pixels, with units of °C / km. The attention head value is the number of parallel attention mechanisms in the multi-head attention layer, and the attention head value is determined based on the cloud cover value, solar flare index value, and sea surface temperature gradient value.

[0057] The weighted fusion method calculates the attribute value of each pixel or vector element within the fusion area based on the weight values ​​of the satellite remote sensing interpretation results and the ecological model output results. When the fusion area is a high-pollution area, the weight value of the satellite remote sensing interpretation results is set to 0.3 and the weight value of the ecological model output results is set to 0.7. When the fusion area is a low-pollution area, the weight value of the satellite remote sensing interpretation results is set to 0.7 and the weight value of the ecological model output results is set to 0.3. When the fusion area is a transition area, the weight value of the satellite remote sensing interpretation results is calculated by linear interpolation as 0.7 minus 0.8 times the cloud cover value, and the weight value of the ecological model output results is calculated as 0.3 plus 0.8 times the cloud cover value. The fused attribute value is the product of the attribute value of the simulated vector result of Ulva prolifera cover and the weight value of the ecological model output results, plus the product of the attribute value of the satellite remote sensing interpretation vector result and the weight value of the satellite remote sensing interpretation results.

[0058] The high-pollution area is defined as a region with a cloud cover value ≥ 0.5 or a solar flare index value ≥ 0.3. The low-pollution area is defined as a region with a cloud cover value < 0.3 and a solar flare index value < 0.1. The transition area is defined as a region with a cloud cover value ∈ [0.3, 0.5) and a solar flare index value ∈ [0.1, 0.3).

[0059] The Douglas-Puk boundary smoothing algorithm simplifies the fused Ulva prolifera coverage vector boundary by using the Douglas-Puk algorithm. The Douglas-Puk algorithm retains feature points on the boundary that deviate from the original boundary by a distance greater than the distance threshold by setting a distance threshold, and removes redundant points that deviate by a distance less than the distance threshold. The distance threshold is determined based on the average scale value and spatial resolution of the Ulva prolifera patches. The simplified boundary is smoother and retains the main morphological features.

[0060] The symmetric difference method calculates the symmetric difference between the *Ulva prolifera* coverage area in the satellite remote sensing interpretation vector results and the *Ulva prolifera* coverage area in the simulated vector results. The symmetric difference includes a portion belonging only to the satellite remote sensing interpretation vector results and a portion belonging only to the simulated vector results. The overlapping portion of the two is removed, and the independent coverage areas of the satellite remote sensing interpretation vector results and the simulated vector results are retained. The independent coverage areas of the satellite remote sensing interpretation vector results and the simulated vector results are merged with the fused overlapping portion to form a complete integrated *Ulva prolifera* coverage area.

[0061] The comprehensive coverage area of ​​*Ulva prolifera* is the sum of the overlapping area after fusion, the independent coverage area of ​​*Ulva prolifera* from the satellite remote sensing interpretation vector result, and the independent coverage area of ​​*Ulva prolifera* from the simulated vector result.

[0062] When the Euclidean distance between the fused *Ulva prolifera* cover feature vector and the baseline *Ulva prolifera* cover vector is <0.05, the weight values ​​of the current satellite remote sensing interpretation results and the ecological model output results are maintained. If the Euclidean distance ∈ [0.05, 0.15], the learning rate convergence factor value during the training process of the spectral consistency enhancement model is adjusted from 0.1 to 0.05. If the Euclidean distance is >0.15, the multi-head attention layer parameters of the spectral consistency enhancement model are reinitialized. The baseline *Ulva prolifera* cover vector is a standard feature vector of *Ulva prolifera* cover obtained through statistical analysis of historical high-quality observation data. This baseline *Ulva prolifera* cover vector is used to evaluate the rationality of the fused *Ulva prolifera* cover feature vector. The Euclidean distance value is a distance metric between the fused *Ulva prolifera* cover feature vector and the baseline *Ulva prolifera* cover vector in the high-dimensional feature space. The Euclidean distance value is obtained by calculating the square root of the sum of the squares of the differences between corresponding elements of the two vectors. The learning rate convergence factor is a parameter that controls the rate of learning rate decay during the training of the spectral consistency enhancement model. The learning rate convergence factor is used to adjust the step size of parameter updates in the spectral consistency enhancement model.

[0063] When the detected Ulva prolifera cover exhibits a scattered distribution pattern and the average diameter of the Ulva prolifera patches ∈ [100m, 500m], a spatial interpolation prediction algorithm is used to predict the Ulva prolifera cover in uncovered areas; when the detected Ulva prolifera cover exhibits an aggregated distribution pattern, a kernel density estimation prediction algorithm is used to predict high-density aggregated areas of Ulva prolifera; when the detected Ulva prolifera cover exhibits a striped distribution pattern, a directional diffusion prediction algorithm is used to predict along the main drift direction of Ulva prolifera.

[0064] The dispersed distribution pattern is characterized by the spatially dispersed and independent distribution of *Ulva prolifera* patches. This pattern is identified by calculating the nearest neighbor distance and spatial aggregation index between patches. The average diameter of each *Ulva prolifera* patch is the average of the equivalent circle diameters of all patches, measured in meters. The spatial interpolation prediction algorithm predicts *Ulva prolifera* coverage in unknown areas based on the known spatial location and attribute values ​​of *Ulva prolifera* coverage points, using distance-weighted or Kriging interpolation methods. This algorithm fills in data gaps related to the dispersed distribution pattern. The clustered distribution pattern is characterized by the spatially concentrated and densely distributed *Ulva prolifera* patches. This pattern is identified by calculating local *Ulva prolifera* coverage density and spatial autocorrelation index. The kernel density estimation prediction algorithm estimates the probability of *Ulva prolifera* coverage in high-density clustered areas by calculating the kernel density distribution of *Ulva prolifera* patch points. This algorithm predicts the *Ulva prolifera* expansion trend within the clustered distribution pattern. The banded distribution pattern refers to the morphological characteristic of *Ulva prolifera* patches extending in a banded pattern along a main spatial direction. This pattern is identified by calculating the directional index and aspect ratio of the *Ulva prolifera* patches. The main drift direction of *Ulva prolifera* is the primary direction of movement of the patches under the influence of ocean currents and wind. This main drift direction is calculated by analyzing the positional changes in *Ulva prolifera* cover vector data over continuous multi-temporal periods. The directional diffusion prediction algorithm, based on the main drift direction and ocean current velocity data, predicts the diffusion range and arrival time of *Ulva prolifera* along this main drift direction. This algorithm is used to predict the drift path of *Ulva prolifera* under the banded distribution pattern.

[0065] As one embodiment of the present invention, a computer-based method is also provided to form a daily comprehensive information fusion system for the green tide of Ulva prolifera based on satellite remote sensing and ecological model data. The computer is equipped with a readable storage medium, which stores program instructions. When the program instructions are run on the computer, they execute the above-described method.

[0066] The specific implementation methods of the above steps are described in detail below.

[0067] The specific implementation of step S01 involves first acquiring satellite remote sensing image data of different bands through multiple platforms such as ocean satellites and land satellites. This data undergoes radiometric calibration to convert the original digital quantization values ​​into physically meaningful radiance values. This conversion process is based on the sensor's calibration coefficients and gain parameters, aiming to eliminate systematic errors caused by sensor response nonlinearity. Subsequently, atmospheric correction is performed, using the dark pixel method or radiative transfer model method to eliminate the influence of atmospheric molecular scattering and aerosol absorption on the radiance values, ensuring the acquisition of true surface reflectance information. Then, geometric correction is performed, using ground control points and digital elevation model data to register the images to the China Geodetic Coordinate System 2000, eliminating geometric distortions caused by satellite attitude changes and terrain undulations. Simultaneously, threshold segmentation and spectral feature recognition methods are used to extract cloud-covered areas, generating cloud marker vector data for data quality labeling. Finally, the characteristics of *Ulva prolifera* cover are calculated and identified using spectral indices such as the normalized vegetation index and the normalized water index. By combining texture analysis and morphological filtering methods to remove interference factors such as sea surface foam and suspended sediment, the daily satellite remote sensing interpretation vector results of Ulva prolifera cover were finally extracted. Data gaps and low-quality areas caused by cloud cover or solar flare pollution were marked. Simultaneously, an Ulva prolifera ecodynamic model was constructed. This model, driven by the satellite remote sensing interpretation results, is based on the Ulva prolifera growth dynamics equation and ocean circulation model. Inputs include sea surface temperature data reflecting the Ulva prolifera growth rate, wind field data driving the Ulva prolifera drift trajectory, nutrient concentration data affecting Ulva prolifera biomass accumulation, precipitation data altering the sea surface salinity environment, and shortwave radiation data controlling the intensity of Ulva prolifera photosynthesis. The model simulates the growth and spatial diffusion processes of Ulva prolifera in the marine environment using numerical integration methods, outputting the simulated vector result of Ulva prolifera cover at the time of satellite transit on the day. This result, along with the satellite remote sensing interpretation vector result, is projected onto the China Geodetic Coordinate System 2000 with a unified spatial resolution of 50m and a vector format, providing standardized data input for subsequent fusion processing.

[0068] The specific implementation of step S02 involves establishing a cross-calibration model between sensors to address the inconsistency in radiometric measurements caused by differences in the spectral response functions of different sensors. First, a desert or deep-sea region is selected as a radiometrically stable reference target using the pseudo-invariant target method. Linear regression analysis is then performed on observation data from multiple sensors relative to the same target at different times to establish the radiometric conversion relationship between sensors. The coefficients of this relationship are obtained through least-squares fitting and are used to convert the observation values ​​of the sensor to be calibrated into equivalent observation values ​​under the reference sensor's radiometric reference. Simultaneously, synchronous observation experiments are conducted, acquiring the true spectral characteristics of the target object through ground and sea surface measurements at the satellite's transit time. The measured data are compared and analyzed with the observation values ​​from each sensor to establish a quantitative sensor bias correction model. Finally, spectral resampling technology is used to convert the data from different sensors... The spectral bands of the sensors are converted to a unified band configuration using linear interpolation or cubic spline interpolation. This technique simulates the spectral response characteristics of the reference sensor by performing convolution operations on the spectral response functions of each sensor, achieving data consistency in the spectral dimension. Finally, the grayscale distribution of the image to be corrected is adjusted using histogram matching to make it consistent with the reference image. A grayscale value mapping relationship is established by calculating the cumulative distribution function, and a linear or nonlinear radiometric conversion model is established using the radiometric normalization method. Invariant feature points in the image are selected for regression analysis to eliminate radiometric bias caused by changes in solar altitude angle and atmospheric conditions. All sensor observations are converted to the radiometric reference of the HY-1C satellite coastal zone imager to ensure the radiometric consistency and comparability of multi-source remote sensing data. This step lays the data quality foundation for subsequent multi-source data fusion.

[0069] The specific implementation of step S03 involves conducting reliability assessments on both the satellite remote sensing interpretation vector results and the simulated Ulva prolifera cover vector results to determine the weight allocation strategy during data fusion. For the satellite remote sensing interpretation vector results, the cloud coverage value is obtained by calculating the ratio of the cloud coverage area to the total area of ​​the fusion region using cloud identifier vector data. A cloud coverage value between 0 and 0.3 indicates a high-reliability region, while a value between 0.5 and 1 indicates a low-reliability region. Simultaneously, the solar flare index value is obtained by extracting the solar flare region vector data and calculating the ratio of the solar flare area to the total area of ​​the fusion region. A solar flare index value between 0 and 0.1 indicates a small impact from flare pollution, while a value between 0.3 and 1 indicates a significant impact. The reliability index value of the satellite remote sensing interpretation results is calculated by comprehensively considering the cloud coverage value, the solar flare index value, and the image signal-to-noise ratio. This index value ranges from 0 to 1, with a higher value indicating higher reliability. For the simulated vector results of *Ulva prolifera* cover, the consistency value is calculated by spatially overlaying the simulated results with historical observation data from the same period. Simultaneously, parameter sensitivity analysis is performed, and the change in the simulated vector results of *Ulva prolifera* cover is observed by varying the input parameters of the *Ulva prolifera* ecological dynamics model by 20%. When the change in the output results is less than 10%, the model stability is considered high. The reliability index of the ecological model output results is calculated by combining the consistency value and the stability analysis results. This index value ranges from 0 to 1, with a larger value indicating higher reliability. Based on these two reliability index values, a dynamic weight allocation rule is established. The weight value of the satellite remote sensing interpretation result is equal to the ratio of the satellite remote sensing interpretation result reliability index value to the sum of the two reliability index values. The weight value of the ecological model output result is also equal to the ratio of the ecological model output result reliability index value to the sum of the two reliability index values. This dynamic weight allocation strategy can adaptively adjust the fusion weights according to data quality, improving the reliability and accuracy of the fusion results.

[0070] The specific implementation of step S04 involves establishing fusion rules based on expert knowledge to achieve an adaptive data fusion strategy under different data quality conditions. When the satellite remote sensing interpretation vector result shows Ulva lactuca coverage and the cloud identifier value in the cloud identifier vector data is 0, it indicates good observation conditions. In this case, the satellite remote sensing interpretation vector result is directly used as the daily comprehensive coverage information of Ulva lactuca green tide. When the cloud identifier value is 1, it indicates that the observation area is completely obscured by clouds. In this case, the Ulva lactuca coverage simulation vector result is used as the daily comprehensive coverage information of Ulva lactuca green tide to fill the data gap. When the satellite remote sensing interpretation vector result can detect part of the sea area, it indicates that the observation data is partially usable. In this case, the satellite remote sensing interpretation vector result is used as the benchmark, and the Ulva lactuca coverage simulation vector result is fused using a two-way buffer overlay analysis method. At the same time, a comprehensive pollution index value is calculated based on the cloud coverage value and the solar flare index value within the fused area. This index value is the product of the cloud coverage value and the first weighting coefficient plus the product of the solar flare index value and the second weighting coefficient. The first weighting coefficient and the second weighting coefficient are dynamically adjusted according to regional and seasonal characteristics. When the area is in During the rainy season, the first weighting coefficient increases to 0.7, and during the sunny season, the second weighting coefficient increases to 0.6. Based on the comprehensive pollution index, the merged area is divided into high-pollution and low-pollution areas. A high-pollution area is defined as one with a cloud cover value greater than or equal to 0.5 or a solar flare index value greater than or equal to 0.3; a low-pollution area is defined as one with a cloud cover value less than 0.3 and a solar flare index value less than 0.1; and a low-pollution area is defined as one with a cloud cover value between 0.3 and 0.5 and a solar flare index value between 0.1 and 0.3. The transition zone is designated as a region, and the weight values ​​of satellite remote sensing interpretation results and ecological model output results are dynamically adjusted for regions with different pollution levels. In high-pollution areas, the weight value of satellite remote sensing interpretation results is set to 0.3 and the weight value of ecological model output results is set to 0.7. In low-pollution areas, the weight value of satellite remote sensing interpretation results is set to 0.7 and the weight value of ecological model output results is set to 0.3. In the transition zone, the weight values ​​are dynamically calculated using a linear interpolation method. This fusion rule makes full use of expert knowledge and data quality assessment results to achieve an adaptive fusion strategy.

[0071] The specific implementation of step S05 involves executing a fusion algorithm to achieve spatial and attribute fusion of satellite remote sensing interpretation vector results and simulated Ulva prolifera cover vector results. First, buffers are created for the Ulva prolifera cover vector data in both the satellite remote sensing interpretation vector results and the simulated Ulva prolifera cover vector results. The buffer distance is determined by calculating the square root of the area of ​​each polygon in the Ulva prolifera cover vector data. This method can adaptively adjust the buffer range according to the scale of the Ulva prolifera patches. Then, a bidirectional spatial connection operation is performed on the two created buffers. This operation includes a connection operation from the satellite remote sensing interpretation vector results to the simulated Ulva prolifera cover vector results and a reverse connection operation, ensuring that the overlap between the two data sources is fully identified. The common coverage area of ​​the two buffers is output as a potential matching area. A similarity coefficient is calculated for each pair of matching elements. This coefficient is the ratio of the intersection area to the union area of ​​the matching element pair, used to quantify the spatial overlap of the two vector elements. A similarity coefficient threshold of 0.6 is set to filter valid matching pairs. When the similarity coefficient value is greater than or equal to 0.6, the two vector elements are considered to be validly matched. When the similarity coefficient value is less than or equal to 0.6, the matching is considered to be invalid. When the value is less than 0.6, the matching element pair is removed. The filtered valid matching pairs are then input into a spectral consistency enhancement model for processing. This model employs a deep learning architecture. The input layer receives multi-source remote sensing feature vectors of the matching region, including normalized vegetation index, normalized water index, and *Ulva prolifera* cover density, as well as environmental parameter vectors including cloud cover, solar flare index, and sea surface temperature gradient. The input layer is connected to a multi-dimensional unfolded convolutional layer, which simultaneously extracts features in spatial, spectral, and temporal dimensions, capturing the spatial continuity of *Ulva prolifera* cover. The continuous characteristics, spectral response characteristics, and temporal evolution characteristics are used to expand the output of the convolutional layer in multiple dimensions and connect it to a multi-head attention layer. The attention head value is determined based on the cloud cover value, solar flare index value, and sea surface temperature gradient value. When the cloud cover value is between 0 and 0.3 and the solar flare index value is between 0 and 0.1, the attention head value is set to 8. When the cloud cover value is between 0.3 and 0.5 or the solar flare index value is between 0.1 and 0.3, the attention head value is set to 4. When the cloud cover value is greater than 0.5 or the solar flare index value is greater than or equal to 0, the attention head value is set to 4.Attention heads were set to 2 at time 3. The multi-head attention layer enhances the model's adaptability to complex environmental conditions by learning the correlations between different features. The fully connected layer outputs the fused *Ulva prolifera* cover feature vector. This feature vector is processed by an activation function to output the final *Ulva prolifera* cover information. A weighted fusion method is used to calculate the fused attribute values ​​for the matching regions. The fused attribute value is the product of the attribute value of the simulated *Ulva prolifera* cover vector result and the weight value of the ecological model output, plus the product of the attribute value of the satellite remote sensing interpretation vector result and the weight value of the satellite remote sensing interpretation result. Finally, the boundary of the fused *Ulva prolifera* cover vector is optimized using the Douglas-Puk boundary smoothing algorithm. By setting a distance threshold, feature points on the boundary that deviate from the original boundary by a distance greater than the threshold are retained, while redundant points that deviate by a distance less than the threshold are removed. This distance threshold is determined based on the average scale and spatial resolution of the *Ulva prolifera* patches. The simplified boundary is smoother and retains the main morphological features, while removing unreasonable boundaries and fragments to improve data quality.

[0072] The specific implementation of step S06 involves creating daily integrated coverage information for *Ulva prolifera* green tides to form a complete *Ulva prolifera* monitoring product. First, the symmetric difference method is used to calculate the symmetric difference between the *Ulva prolifera* coverage area in the satellite remote sensing interpretation vector results and the *Ulva prolifera* coverage simulation vector results. This symmetric difference includes a portion belonging only to the satellite remote sensing interpretation vector results and a portion belonging only to the *Ulva prolifera* coverage simulation vector results. The overlapping portions are removed, retaining the independent coverage areas of the satellite remote sensing interpretation vector results and the independent coverage areas of the *Ulva prolifera* coverage simulation vector results. Then, the fused data is merged with the independent portions of the satellite remote sensing interpretation vector results and the independent portions of the *Ulva prolifera* coverage simulation vector results to form a unified vector dataset. This merging operation is performed through spatial topological relationship processing. To ensure data integrity and consistency, the comprehensive coverage area of ​​*Ulva prolifera* is calculated. This area is calculated by adding the area of ​​the overlapping portion after fusion to the independent portion of the *Ulva prolifera* coverage area from the satellite remote sensing interpretation vector results, and adding the independent portion of the *Ulva prolifera* coverage simulation vector results. A daily comprehensive *Ulva prolifera* green tide coverage product is output. This product includes coverage range data describing the spatial distribution boundary of *Ulva prolifera*, a comprehensive coverage area value quantifying the total coverage scale, location coordinate data labeling the center location of *Ulva prolifera* patches, and density level data reflecting the density of *Ulva prolifera* coverage. The product format includes raster data format for spatial analysis and vector boundary data format for accurate mapping. This product can provide high-precision and timely *Ulva prolifera* green tide monitoring information for marine environmental monitoring and disaster early warning.

[0073] It should be noted that this invention also addresses the following technical problem: the difficulty in coordinating multi-source remote sensing data due to systematic spectral response differences between different satellite sensors. This invention employs a pseudo-invariant target method to establish the radiometric conversion coefficients between sensors, utilizes synchronous observation experiments to obtain the true spectral characteristics of the target object, unifies band configuration through spectral resampling technology, and eliminates systematic radiometric biases using histogram matching and radiometric normalization methods. This series of methods ensures that the observations from each sensor are converted to a unified reference standard, making the *Ulva prolifera* monitoring data from different sensors such as HY-1C and Landsat comparable, laying a radiometric consistency foundation for subsequent fusion.

[0074] This invention also addresses the problem of unreasonable weight allocation in the case of missing and unstable data on *Ulva prolifera* cover. By establishing multi-dimensional quality assessment indicators such as cloud cover, solar flare index, and image signal-to-noise ratio, and combining the consistency of historical observation data with parameter sensitivity analysis, the reliability of remote sensing interpretation results and ecological model output results is quantified. Based on this, a weight allocation strategy is dynamically adjusted according to the pollution level of the fusion area (high pollution, low pollution, transitional zone), ensuring that data under different quality conditions receive appropriate weight allocation, thus improving the adaptability and robustness of the fusion results under changing environmental conditions.

[0075] Specifically, the principle of this invention is as follows: The reason why the fusion scheme of this invention can solve this technical problem is that it systematically eliminates various obstacles that restrict the fusion of multi-source data. First, through cross-sensor calibration methods such as the pseudo-invariant target method, the establishment of radiometric conversion coefficients, and histogram matching, the observations of different remote sensing sensors are converted to a unified radiometric benchmark, fundamentally eliminating the data incomparability caused by differences in spectral response. Second, the invention introduces a two-dimensional evaluation index system, namely the reliability index of satellite remote sensing interpretation results and the reliability index of ecological model output results. These comprehensively consider multiple quality factors such as cloud cover, solar flare pollution, image signal-to-noise ratio, and the consistency between the model and historical data, quantitatively characterizing the credibility of the two types of data sources under specific spatiotemporal conditions. The dynamic weight allocation rules calculated based on these reliability indices enable the fusion process to adapt to changing data quality conditions. Third, the spectral consistency enhancement model, through the synergistic effect of the three-dimensional feature space, not only fuses the spectral information of data from different sources, but also deeply explores their intrinsic connections in spatial distribution, temporal evolution, and other dimensions. This multi-dimensional joint processing method has a stronger information integration capability than traditional pixel-level or feature-level fusion. Finally, post-processing steps such as bidirectional buffer overlay analysis, weighted fusion calculation, and boundary smoothing, while preserving the independent coverage information of the two data sources, scientifically handle the attribute fusion of overlapping areas, ensuring the integrity and logical consistency of the final product. This systematic, multi-layered fusion process enables the organic unification of multi-source heterogeneous data, generating truly comprehensive coverage information.

[0076] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0077] In this embodiment, the specific implementation of step S01 is the same as described above, and will not be repeated in detail here.

[0078] The formula for calculating the radiation conversion coefficient of the pseudo-invariant target method in step S02 is as follows:

[0079] ;

[0080] In the formula, For reference sensor radiance values, the unit is... ; The radiance value of the sensor to be calibrated, in units of ; The slope parameter in the radiation conversion coefficient; The intercept parameter in the radiation conversion coefficient, in units of .

[0081] The parameter acquisition method is as follows: and The results were obtained through least squares regression analysis of multi-temporal observation data of pseudo-invariant targets. The specific calculation steps were as follows: First, at least 30 desert areas, saline-alkali areas, and deep-sea areas within the study area were selected as pseudo-invariant target points. Observation data of these target points from the reference sensor and the sensor to be calibrated were acquired at different temporal phases, and an observation data sequence was established. ,in The sample number. To determine the number of observation samples, the least squares method is then used to solve for the sum of squared residuals. Minimum parameter and The value of .

[0082] The formula for calculating cloud cover value in step S03 is as follows:

[0083] ;

[0084] In the formula, This represents the cloud cover value. Cloud coverage area, unit: ; This is the maximum reference value for cloud coverage area, with a default value of 1000. ; To represent the total area of ​​the merged region, the unit is... ; The maximum reference value for the total area of ​​the merged region is 10000 by default. .

[0085] The formula for calculating the solar flare index in step S03 is as follows:

[0086] ;

[0087] In the formula, This refers to the solar flare index value; The area of ​​a solar flare is expressed in units of... ; This is the maximum reference value for the area of ​​a solar flare, with a default value of 500. .

[0088] The formula for calculating the reliability index of the satellite remote sensing interpretation results in step S03 is as follows:

[0089] ;

[0090] In the formula, This refers to the reliability index value of satellite remote sensing interpretation results; This is the cloud cover weighting coefficient, with an empirical value of 0.4; This is the weighting factor for solar flares, with an empirical value of 0.3; This is the signal-to-noise ratio weighting coefficient, with an empirical value of 0.3; This represents the maximum cloud cover, typically set to 1. This is the maximum value of the solar flare index, which is usually taken as 1. Image signal-to-noise ratio, unit: ; This is the maximum reference value for image signal-to-noise ratio, with a default value of 50. .

[0091] The formula for calculating the degree of agreement between the simulated vector result of Ulva prolifera cover and historical observation data in step S03 is as follows:

[0092] ;

[0093] In the formula, This represents the degree of agreement. This represents the number of historical comparison samples; The sample number; For the first The area of ​​overlap between simulation results and observed data in each sample, in units of ; This is a reference value for the overlapping area, with a default value of 100. ; For the first The area of ​​the simulated results in each sample, in units of ; For the first The area of ​​observed data in each sample, in units of ; This is a reference value for the area, with a default value of 100. .

[0094] The formula for calculating the reliability index value of the ecological model output in step S03 is as follows:

[0095] ;

[0096] In the formula, Output a reliability index value for the results of the ecological model; This is the matching degree weighting coefficient, with an empirical value of 0.6; This is the parameter sensitivity weighting coefficient, with an empirical value of 0.4; This is the maximum reference value for the degree of fit, and is usually set to 1. Output the magnitude of change in the parameter sensitivity analysis results, in percentage form; The maximum reference value for output variation is 20 by default.

[0097] The formula for calculating the comprehensive pollution index value in step S04 is as follows:

[0098] ;

[0099] In the formula, This is the comprehensive pollution index value; The first weighting coefficient; This is the second weighting coefficient.

[0100] The parameter acquisition method is as follows: and Dynamic adjustments are made based on regional and seasonal characteristics; when the region is in the rainy season... Set to 0.7 and Set to 0.3 when the region is in a sunny season. Set to 0.3 and Set to 0.7.

[0101] The formula for calculating the weight value of the satellite remote sensing interpretation results in step S04 is as follows:

[0102] ;

[0103] In the formula, Weight values ​​for satellite remote sensing interpretation results; This is the maximum reliability index for satellite remote sensing interpretation results, typically set to 1. This is the maximum value of the reliability index of the ecological model output results, which is usually set to 1.

[0104] The formula for calculating the weight values ​​of the ecological model output results in step S04 is as follows:

[0105] ;

[0106] In the formula, Output weight values ​​for the ecological model results.

[0107] The formula for calculating the weight value of the transition region in step S04 is as follows:

[0108] ;

[0109] ;

[0110] In the formula, Weight values ​​for satellite remote sensing interpretation results in the transition area; The formula outputs weight values ​​for the ecological model of the transition area; in the formula, 0.7 is the satellite weight benchmark value for low-pollution areas, 0.3 is the model weight benchmark value for low-pollution areas, and 0.8 is the cloud coverage adjustment coefficient.

[0111] The formula for calculating the buffer distance value in step S05 is as follows:

[0112] ;

[0113] In the formula, This is the buffer distance value, in units of ; This represents the area values ​​of each polygon in the vector data of *Ulva prolifera* cover, in units of... ; This is a reference value for the polygon area, and the default value is [value missing]. ; This is a reference value for the buffer distance; the default value is 250. .

[0114] The formula for calculating the similarity coefficient in step S05 is as follows:

[0115] ;

[0116] In the formula, The similarity coefficient value; To match the intersection area of ​​feature pairs, in units of ; The maximum reference value for the intersection area is , which is the default value. ; To match the union area of ​​feature pairs, in units of ; The maximum reference value for the union area is , which is the default value. .

[0117] The formula for calculating the normalized vegetation index in step S05 is as follows:

[0118] ;

[0119] In the formula, The value is the normalized vegetation index. Reflectivity in the near-infrared band; Reflectivity in the red light band; This is a reference value for reflectivity, typically set to 1.

[0120] The formula for calculating the normalized water index value in step S05 is as follows:

[0121] ;

[0122] In the formula, This is the normalized water index value; This refers to the reflectivity in the green light band.

[0123] The formula for calculating the coverage density of *Ulva prolifera* in step S05 is as follows:

[0124] ;

[0125] In the formula, This represents the density value of the *Ulva prolifera* cover. The area of ​​*Ulva prolifera* coverage in the vector results interpreted by satellite remote sensing, in units of... ; This is the maximum reference value for the coverage area of ​​*Ulva prolifera*, the default value is... ; For the area of ​​the statistical unit, the unit is... ; This is the maximum reference value for the area of ​​the statistical unit, and the default value is [value missing]. .

[0126] The formula for calculating the sea surface temperature gradient in step S05 is as follows:

[0127] ;

[0128] In the formula, This represents the sea surface temperature gradient, in °C. ; The data represents the sea surface temperature of adjacent pixels, in °C. This is the sea surface temperature data for the current pixel, in °C. This is a temperature reference value, with a default value of 30℃. The distance between pixels, in units of ; in the formula Convert distance units to ; This is a distance reference value, with a value of 1. .

[0129] The formula for calculating the fused attribute values ​​using the weighted fusion method in step S05 is expressed as follows:

[0130] ;

[0131] In the formula, The attribute values ​​after fusion; Attribute values ​​of the simulated vector results for Ulva prolifera cover; This is the maximum reference value for the vector result attribute of the simulated Ulva prolifera cover; the default value is 100. Interpret the attribute values ​​of vector results from satellite remote sensing; This is the maximum reference value for the vector result attributes interpreted by satellite remote sensing; the default value is 100.

[0132] The formula for calculating the comprehensive coverage area of ​​*Ulva prolifera* in step S06 is as follows:

[0133] ;

[0134] In the formula, This represents the total coverage area of ​​*Ulva prolifera*, in units of... ; This is the area benchmark value, which is 1. ; The area of ​​the overlapping portion after merging is expressed in units of... ; This is the maximum reference value for the overlapping area, with a default value of 1. ; The value of the *Ulva prolifera* coverage area for the independent portion of the vector results interpreted by satellite remote sensing, in units of... ; This is the maximum reference value for the area of ​​an independent part of the satellite remote sensing data; the default value is 1. ; This represents the area of ​​*Ulva prolifera* coverage, an independent portion of the simulated vector result, in units of... ; This is the maximum reference value for the area of ​​the independent parts of the model, with a default value of 1. .

[0135] The formula for calculating the Euclidean distance value in the document is as follows:

[0136] ;

[0137] In the formula, This is the Euclidean distance value; The dimension of the feature vector; For feature serial number; The first feature vector of the fused *Ulva prolifera* coverage One element; The first of the baseline Ulva prolifera coverage vectors One element; For the first The reference value for each feature is the maximum observed value of each feature by default.

[0138] The formula for calculating the average diameter of *Ulva prolifera* patches in the document is as follows:

[0139] ;

[0140] In the formula, This represents the average diameter of the *Ulva prolifera* patches, in units of... ; This represents the total number of *Ulva prolifera* patches; For plaque serial numbers; For the first Area of ​​each *Ulva prolifera* patch, in units of ; This is a reference value for the area of ​​the equivalent circle, and the default value is [value missing]. ; This is a reference value for the equivalent circle diameter, with a default value of 100. ; Pi, with a value of 3.14159.

[0141] It should be explained that the reliability index calculation formula for satellite remote sensing interpretation results proposed in this scheme establishes a quantitative reliability evaluation system by comprehensively considering three key quality assessment factors: cloud cover value, solar flare index value, and image signal-to-noise ratio value. This formula uses a weighted normalization method to unify each quality factor into a dimensionless space, and achieves objective evaluation of remote sensing data quality under different environmental conditions through a dynamic weight allocation mechanism. This ensures the rationality and accuracy of weight allocation during the fusion process, effectively improving the accuracy and continuity of the daily comprehensive coverage information of *Ulva prolifera* green tides. The reliability index calculation formula for the ecological model output results introduces two dimensions of evaluation indicators: consistency value and parameter sensitivity analysis results. This achieves a quantitative evaluation of the stability and accuracy of the ecological model simulation results. The formula eliminates dimensional differences between different evaluation indicators through normalization processing, ensuring the consistency and comparability of reliability assessments. This provides reliable data support for dynamic weight adjustment in the fusion algorithm, significantly improving the adaptability and predictive accuracy of the fusion results to complex marine environmental changes.

[0142] To better understand and implement this invention, a specific application scenario is provided below as Example 2: A technical team undertook the daily monitoring task of Ulva prolifera green tides in the Yellow Sea. Large-scale Ulva prolifera green tide disasters frequently occur in this area from May to August, severely impacting the marine ecological environment and coastal aquaculture. The technical team used the method of this invention, integrating multi-source satellite remote sensing data with Ulva prolifera ecological dynamics model data, to generate a daily comprehensive coverage information product of Ulva prolifera green tides.

[0143] The technical team first acquired multi-source satellite remote sensing imagery data from June 15, 2024, including data from the HY-1C satellite coastal zone imager, Sentinel-3 satellite OLCI data, and Landsat-8 satellite OLI data. Radiometric calibration was performed on the acquired imagery data, converting digital quantization values ​​into radiance values. The radiometric calibration coefficient for the HY-1C data was 0.0876. The radiometric calibration factor for Sentinel-3 data is 0.0954. After atmospheric correction, the 6S radiative transfer model was used to eliminate the effects of atmospheric scattering and absorption, achieving an atmospheric correction accuracy of less than 5%. Geometric correction used the China Geodetic Coordinate System 2000, registering all images to a unified geographic coordinate system, with registration errors controlled within 0.5 pixels. Cloud extraction employed a threshold-based segmentation and morphological processing method to identify the cloud coverage area within the monitoring region. The cloud cover value is 0.15. The extracted satellite remote sensing interpretation vector results show that the area covered by *Ulva prolifera* is... .

[0144] The technical team constructed an ecodynamic model of *Ulva prolifera* based on the Navier-Stokes equation and the *Ulva prolifera* growth kinetic equation. Input data included sea surface temperature, wind field, nutrient concentration, precipitation, and shortwave radiation. Sea surface temperature data for June 15, 2024, showed a temperature range of 18.5℃ to 22.3℃, with an average temperature of 20.4℃. Wind field data showed the prevailing wind direction was southeast, with wind speeds ranging from 3.2 m / s to 8.7 m / s. Nutrient concentration data showed nitrate concentration of 12.6 μmol / L and phosphate concentration of 0.87 μmol / L. Precipitation data showed a daily precipitation of 2.3 mm. Shortwave radiation data showed a radiation intensity of 456. The simulation vector results of the *Ulva prolifera* cover output by the model show that the *Ulva prolifera* coverage area is... .

[0145] The technical team established a cross-calibration model between sensors, selecting the Taklamakan Desert and the deep-sea area of ​​the Yellow Sea as pseudo-invariant targets. Through regression analysis of multi-temporal observation data throughout 2023, the radiative conversion coefficients between Sentinel-3 and HY-1C were established as 1.089, and between Landsat-8 and HY-1C as 1.052. A synchronous observation experiment was conducted on May 28, 2024, obtaining spectral reflectance values ​​for *Ulva prolifera* at 0.085 in the 555nm band, 0.032 in the 665nm band, and 0.178 in the 865nm band. Spectral resampling technology employed cubic spline interpolation to unify the spectral bands of different sensors to the band configuration of HY-1C. After histogram matching, the correlation coefficient of grayscale distribution between Sentinel-3 and HY-1C images reached 0.94. The radiative normalization method selected 156 invariant feature points for linear regression, and the coefficient of determination of the regression equation was... The value is 0.91, and the root mean square error is 0.023.

[0146] The technical team conducted a reliability assessment of the satellite remote sensing interpretation vector results, calculating a cloud cover value of 0.15, a solar flare index of 0.08, and an image signal-to-noise ratio of 42.6 dB. Based on the assessment criteria, the reliability index of the satellite remote sensing interpretation results was calculated to be 0.82. The reliability assessment of the simulated Ulva prolifera cover vector results, compared with historical observation data from the same period in 2023, showed a consistency value of 0.76. Parameter sensitivity analysis showed that when the sea surface temperature parameter changed by 20%, the simulated Ulva prolifera cover vector results changed by 8.3%; when the wind speed parameter changed by 20%, the simulated Ulva prolifera cover vector results changed by 6.7%, indicating high stability. The reliability index of the ecological model output results was calculated to be 0.73. Based on the two reliability index values, a dynamic weight allocation rule was established, with a weight of 0.53 for the satellite remote sensing interpretation results and a weight of 0.47 for the ecological model output results.

[0147] The technical team established fusion rules based on expert knowledge. For areas with a cloud identifier value of 0, the vector results interpreted directly from satellite remote sensing were used. For areas with a cloud identifier value of 1, the area was... The simulated vector results of *Ulva prolifera* cover were used. For areas with a cloud cover value of 0.15 and a solar flare index value of 0.08, the calculated comprehensive pollution index value was 0.19, classifying them as transitional areas. A two-way buffer overlay analysis method was used for fusion analysis within these transitional areas.

[0148] The technical team implemented a fusion algorithm to create a buffer for the *Ulva prolifera* cover vector data in the satellite remote sensing interpretation results. The buffer distance value was determined based on the square root of the area of ​​each polygon. The area range of the *Ulva prolifera* patches was... to The corresponding buffer distance ranges from 152m to 755m. A buffer was created for the *Ulva prolifera* cover vector data in the simulated vector results, with a buffer distance range of 168m to 812m. A bidirectional spatial join operation was performed on the two buffers, identifying 326 pairs of matching features. The similarity coefficient for each pair of matching features was calculated, ranging from 0.42 to 0.89. A similarity coefficient threshold of 0.6 was set, and 214 valid matching pairs were selected, as shown in Table 1.

[0149] Table 1. Similarity coefficient values ​​of some valid matching pairs

[0150]

[0151] The technical team input the 214 selected valid matching pairs into a spectral consistency enhancement model for processing. The input layer of the spectral consistency enhancement model receives multi-source remote sensing feature vectors of the matching region, with normalized vegetation index values ​​ranging from 0.32 to 0.68, normalized water index values ​​ranging from -0.45 to -0.12, and *Ulva prolifera* cover density values ​​ranging from 0.18 to 0.87. Environmental parameter vectors include cloud cover (0.15), solar flare index (0.08), and sea surface temperature gradient (0.34℃ / km). Based on the cloud cover and solar flare index values, the attention head value of the multi-head attention layer was determined to be 8. The multi-dimensional unfolded convolutional layer extracts the geometric shape features of *Ulva prolifera* patches in the spatial dimension, identifying a striped distribution pattern accounting for 43%, a clustered distribution pattern for 31%, and a dispersed distribution pattern for 26%. In the spectral dimension, the convolutional kernel learns the correlation between different sensor spectral bands, with eigenvalues ​​of 2.87, 1.64, and 0.93 for the fusion coefficient matrix. In terms of time dimension, combining the time series data of Ulva prolifera cover from the previous 5 days, the rate of change in Ulva prolifera cover was identified as increasing daily. .

[0152] like Figure 2 As shown, the technical team used a weighted fusion method to calculate the fused attribute values. For the transition area, the weight value of the satellite remote sensing interpretation result was calculated to be 0.58, and the weight value of the ecological model output result was calculated to be 0.42. The fused *Ulva prolifera* cover vector boundary was optimized using the Douglas-Puk boundary smoothing algorithm, with a distance threshold set at 125m. After optimization, areas smaller than [a certain size] were removed. The number of fragment patches decreased from 15,642 to 8,973. For example... Figure 3 As shown, the technical team used the symmetric difference method to remove overlapping parts, retaining the area of ​​the independent part of the satellite remote sensing interpretation vector result. The area of ​​the independent part of the simulated vector result of Ulva prolifera cover is The area of ​​the overlapping portion after fusion is The three parts were merged to form a unified vector dataset, and the comprehensive coverage area of ​​*Ulva prolifera* was statistically obtained. .

[0153] The daily green tide coverage product for seaweed produced by the technical team includes coverage range data and comprehensive seaweed coverage area values. Location coordinate data and density level data. Density levels are divided into high-density areas and their coverage areas. medium-density area coverage Low-density area coverage The product formats include raster data and vector boundary data. The raster data has a spatial resolution of 50m, while the vector boundary data is stored in Shapefile format. The technical team evaluated the fused *Ulva prolifera* cover feature vector and calculated its Euclidean distance from the baseline *Ulva prolifera* cover vector to be 0.087. Based on the evaluation rules, the learning rate convergence factor during the training process of the spectral consistency enhancement model was adjusted from 0.1 to 0.05.

[0154] The technical team performed spatial prediction based on the distribution pattern of *Ulva prolifera*. The average diameter of *Ulva prolifera* patches in the dispersed distribution pattern area was detected to be 286 μm. Using the Kriging spatial interpolation prediction algorithm, the potential distribution area of ​​*Ulva prolifera* was predicted to be [missing information]. The kernel density estimation algorithm was used to predict the expansion trend of high-density aggregation areas of *Ulva prolifera*, based on the detected aggregation pattern. The area of ​​*Ulva prolifera* aggregation areas is expected to increase within the next 24 hours. A striped distribution pattern was detected, with the main drift direction of the seaweed being northeast at an azimuth of 78°. A directional diffusion prediction algorithm was used to predict the drift path of the seaweed, and it was predicted that the seaweed front would reach an area 56 km away from its current location in 48 hours.

[0155] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for integrating daily Ulva prolifera green tide coverage information based on satellite remote sensing and ecological model data, characterized in that, Multi-source satellite remote sensing image data was collected and preprocessed to obtain satellite remote sensing interpretation vector results. Simultaneously, a *Ulva prolifera* ecological dynamics model was constructed to output simulated *Ulva prolifera* cover vector results. A cross-calibration model between sensors was established to eliminate differences in spectral response between different sensors. The reliability of the satellite remote sensing interpretation vector results and the simulated *Ulva prolifera* cover vector results was assessed, reliability index values ​​were calculated, and dynamic weight allocation rules were established. A fusion rule based on expert knowledge was established, selecting the data source or fusion method according to cloud identifier values ​​and cloud cover values. A buffer is created for the satellite remote sensing interpretation vector results and the simulated vector results of Ulva prolifera cover, and a bidirectional spatial connection operation is performed. Similarity coefficient values ​​are calculated to screen valid matching pairs. The valid matching pairs are then input into the spectral consistency enhancement model for processing. The spectral consistency enhancement model extracts features simultaneously in the spatial, spectral, and temporal dimensions by unfolding convolutional layers in multiple dimensions. A multi-head attention layer dynamically adjusts the attention mechanism according to environmental parameters to achieve deep fusion of observation data from different sensors at the feature level. A weighted fusion method is used to calculate the fused attribute values ​​and smooth the boundaries. The symmetric difference method is used to remove overlapping parts. The fused data is merged with the independent parts to form a unified vector dataset, outputting a daily comprehensive Ulva prolifera green tide coverage product.

2. The method according to claim 1, characterized in that, Preprocessing includes radiometric calibration, atmospheric correction, geometric correction, and cloud extraction. Radiometric calibration converts the digital quantization values ​​of satellite remote sensing image data into radiance values. Atmospheric correction eliminates the influence of atmospheric scattering and absorption on radiance values. Geometric correction registers radiance values ​​to a unified geographic coordinate system. Cloud extraction identifies cloud-covered areas in satellite remote sensing image data and generates cloud-identifying vector data.

3. The method according to claim 2, characterized in that, The Ulva prolifera ecosystem dynamics model is driven by satellite remote sensing interpretation results under clear weather and high data quality conditions. The inputs are sea surface temperature data, wind field data, nutrient concentration data, precipitation data, and shortwave radiation data. The output is the simulated vector result of Ulva prolifera cover at the time of satellite transit on the same day. The satellite remote sensing interpretation vector result and the simulated vector result of Ulva prolifera cover are projected onto the China Geodetic Coordinate System 2000 and the spatial resolution is unified to 50m and the data format is vector format.

4. The method according to claim 3, characterized in that, A cross-calibration model among sensors was established. The radiometric conversion coefficients among the sensors were obtained through the pseudo-invariant target method and synchronous observation experiments. The band configuration was unified by spectral resampling technology. The systematic radiometric bias was eliminated by histogram matching method and radiometric normalization method. The observation values ​​of each sensor were converted to the radiometric reference of the reference sensor, which is the radiometric reference of the HY-1C satellite coastal zone imager.

5. The method according to claim 4, characterized in that, The pseudo-invariant target method selects targets with long-term stable and uniform surface reflectance within the study area as radiation benchmarks. These targets include desert areas, saline-alkali areas, and deep-sea areas. By performing regression analysis on observation data from multiple sensors relative to the same pseudo-invariant target at different times, the radiation conversion relationship between the sensors is established.

6. The method according to claim 5, characterized in that, The reliability of the satellite remote sensing interpretation vector results is assessed based on cloud coverage and solar flare index values. The cloud coverage value is obtained by calculating the ratio of the cloud coverage area to the total area of ​​the fused region using cloud identifier vector data. The solar flare index value is obtained by extracting solar flare region vector data and calculating the ratio of the solar flare area to the total area of ​​the fused region.

7. The method according to claim 6, characterized in that, The reliability of the simulated vector results of Ulva prolifera cover was assessed by comparing them with historical observation data and by parameter sensitivity analysis. The parameter sensitivity analysis was conducted by changing the numerical range of the input parameters of the Ulva prolifera ecodynamic model and observing the magnitude of the changes in the simulated vector results of Ulva prolifera cover, thus evaluating the sensitivity of the simulated vector results of Ulva prolifera cover to changes in the input parameters.

8. The method according to claim 7, characterized in that, The reliability index of satellite remote sensing interpretation results is calculated by comprehensively considering cloud cover value, solar flare index value, and image signal-to-noise ratio value. The reliability index of ecological model output results is calculated by comprehensively considering the consistency between the simulated vector result of seaweed cover and historical observation data and the parameter sensitivity analysis result. The values ​​of the reliability index of satellite remote sensing interpretation results and the reliability index of ecological model output results range from 0 to 1.

9. The method according to claim 8, characterized in that, When the satellite remote sensing interpretation vector results show that the seaweed is covered and the cloud identifier value in the cloud identifier vector data is 0, the satellite remote sensing interpretation vector results are used as the daily comprehensive coverage information of the seaweed green tide. When the cloud identifier value in the cloud identifier vector data is 1, the seaweed coverage simulation vector results are used as the daily comprehensive coverage information of the seaweed green tide. When the satellite remote sensing interpretation vector results can monitor part of the sea area, the satellite remote sensing interpretation vector results are used as the benchmark and the seaweed coverage simulation vector results are fused using the two-way buffer overlay analysis method.

10. The method according to claim 9, characterized in that, The bidirectional buffer overlay analysis method creates buffers for the Ulva prolifera cover vector data in the satellite remote sensing interpretation vector results and the Ulva prolifera cover vector data in the Ulva prolifera cover simulation vector results. The buffer distance value of the buffer is determined according to the square root of the area value of each polygon in the Ulva prolifera cover vector data. A bidirectional spatial connection operation is performed on the two buffers, and the common coverage area of ​​the two buffers is output as the overlapping part.

Images