A large-scale biological migration forecasting method based on dual-polarized weather radar

Abstract

The application relates to the technical field of weather radars, and discloses a large-scale biological migration flight prediction method based on a dual-polarized weather radar, which comprises the following steps: extracting polarization parameters of dual-polarized weather radar data; adopting a method of polar coordinate to rectangular coordinate conversion, converting the radar data into a unified 512*512 rectangular coordinate data format, and reconstructing the spatial resolution; constructing a biological echo classification model, classifying biological echoes and meteorological echoes according to the polarization parameter data; based on the classified biological echo data, constructing a spatio-temporal convolutional neural network prediction model, and performing short-time near-future prediction of biological migration flight; and outputting biological migration flight dynamic images at multiple time points in the future, and providing migration trend prediction results. Through the dual-polarized weather radar, multi-source data fusion and a generative adversarial network, high-precision monitoring and dynamic trend prediction of migration biological organisms are realized, and the data continuity and prediction reliability are ensured.

Description

Technical Field

[0001] This invention relates to the field of weather radar technology, specifically to a large-scale biological migration prediction method based on dual-polarization weather radar. Background Technology

[0002] Biological migration refers to the large-scale migration activities of organisms such as insects and birds driven by seasonal changes or environmental factors. Accurately predicting the dynamic distribution and trends of migratory organisms is of great significance for agricultural pest and disease control, ecological protection, and flight safety.

[0003] Current large-scale biological migration prediction methods largely rely on weather radar and single-band insect radar. These methods extract echo characteristics of target organisms (such as echo intensity and velocity) and combine them with machine learning algorithms or physical modeling to predict migration trends. For example, methods based on recurrent neural networks (RNNs) can capture the temporal dynamics of migratory organisms and achieve short-term predictions by combining historical data; while methods based on radar echo classification technology have high accuracy in distinguishing migratory targets from meteorological targets. These methods can effectively monitor biological migration behavior in specific application scenarios, providing technical support for agricultural and forestry pest and disease control and ecological monitoring.

[0004] However, existing methods have the following limitations: First, single-band radar or traditional weather radar is not good at distinguishing between migratory organisms and complex meteorological targets (such as heavy precipitation), which limits the prediction accuracy. Second, the lack of multi-source data fusion means it is difficult to fully describe the dynamic environmental characteristics of the migratory area, and the prediction effect decreases in heavy precipitation or data missing scenarios. Third, the blank areas in the input data are not fully filled, resulting in insufficient spatiotemporal continuity, which in turn affects the prediction ability and reliability of the model. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a large-scale biological migration prediction method based on dual-polarization weather radar. This method solves the problems of low accuracy in distinguishing migratory organisms from meteorological targets, incomplete dynamic description of migration areas due to lack of fusion of multi-source data, and insufficient spatiotemporal continuity caused by incomplete filling of blank areas in radar data.

[0006] To achieve the above objectives, the present invention provides a large-scale biological migration prediction method based on dual-polarization weather radar, comprising the following steps:

[0007] Extracting polarization parameters from dual-polarization weather radar data;

[0008] The radar data was converted into a unified 512×512 rectangular coordinate data format by using a polar coordinate to rectangular coordinate conversion method, and the spatial resolution was reconstructed.

[0009] A biological echo classification model was constructed to classify biological echoes and meteorological echoes from polarization parameter data;

[0010] Based on the classified biological echo data, a spatiotemporal convolutional neural network prediction model is constructed to perform short-term proximity prediction of biological migration.

[0011] It outputs dynamic images of biological migration at multiple future time points, providing migration trend prediction results.

[0012] Preferably, the extracted polarization parameters include:

[0013] reflectivity factor , used to describe the scattering intensity of the echo;

[0014] Differential reflectivity , used to describe the shape and directionality of the echo;

[0015] Differential phase Used to analyze the phase characteristics of the target scatterer;

[0016] Correlation coefficient It is used to assess the similarity between horizontal and vertical polarization echoes;

[0017] Depolarization ratio This is used to improve the distinction between biological echoes and meteorological echoes;

[0018] Spectral width It is used to characterize the distribution range of target motion velocity, serving as an auxiliary parameter for the dynamic characteristics of biological targets.

[0019] Preferably, the depolarization ratio The calculation methods include:

[0020] Initial calculations are performed based on differential reflectance and correlation coefficient;

[0021] Introducing a weighting factor for the differential phase The optimization formula includes:

[0022] ;

[0023] in, Differential reflectivity, The correlation coefficient is defined as 0 ≤ ≤1.

[0024] Preferably, the polar coordinate data transformation includes:

[0025] Use the following formula to convert polar coordinate data to rectangular coordinate data:

[0026] ;

[0027] ;

[0028] in, For distance library, It is the azimuth angle. The angle of elevation;

[0029] The transformed data is reconstructed using bilinear interpolation to reconstruct the spatial resolution, ensuring spatial consistency and standardization of the data.

[0030] Preferably, the constructed biological echo classification model includes:

[0031] After extracting polarization parameters, the XGBoost algorithm was used to classify biological echoes and meteorological echoes.

[0032] Use convolutional neural networks to extract spatial texture features;

[0033] The weighted average method is used to fuse the outputs of XGBoost and convolutional neural networks to improve classification accuracy.

[0034] Preferably, the specific calculation formula for the weighted average method is as follows:

[0035] ;

[0036] in, These are the weighting coefficients, and the optimal values ​​are determined through cross-validation. : Represents the final classification result after weighted fusion, used to determine the probability of the input data being classified as biological echo or meteorological echo; : Represents the output of the XGBoost classification model, which is the probability value of the biological echo category; : Represents the output of the enhanced convolutional neural network, which is the probability value of the biological echo category; : Represents the weight coefficients of the CNN classification results, and Complementarity is used to balance the contributions of the two classification methods.

[0037] Preferably, the constructed spatiotemporal convolutional neural network model includes:

[0038] Input radar echo images from multiple past time points as spatiotemporal sequence data;

[0039] A 5-layer spatiotemporal convolutional neural network module is used for spatiotemporal dynamic modeling to predict radar echo images at multiple future time intervals.

[0040] The state update of a spatiotemporal convolutional neural network unit includes the following steps:

[0041] ;

[0042] in, , , These represent the states of the forget gate, input gate, and output gate, respectively. For the unit state, It is in a hidden state.

[0043] Preferably, the spatiotemporal convolutional neural network model further incorporates an attention mechanism, which includes:

[0044] ;

[0045] ;

[0046] in, and The weight matrix is ​​a learnable matrix. For bias, To score attention, This represents the normalized attention weights.

[0047] Preferably, the migration trend prediction result is optimized through multi-source data fusion, wherein the multi-source data includes:

[0048] Wind field data: Extract wind speed and direction information, combine it with dual-polarization weather radar data, and use it to correct the spatial path of migration trends;

[0049] Satellite remote sensing data: High-resolution satellite imagery is used to fill in the blank areas in radar echoes caused by precipitation removal, thereby enhancing the environmental characteristics description of the migration area.

[0050] Preferably, the method further includes:

[0051] To fill in the missing data in the blank areas of the input data, a generative adversarial network is used to generate highly reliable simulated data.

[0052] This invention provides a large-scale biological migration prediction method based on dual-polarization weather radar. It has the following beneficial effects:

[0053] 1. This invention combines the XGBoost classification model, spatiotemporal convolutional neural network and attention mechanism to achieve accurate monitoring and dynamic trend prediction of migratory organisms. The system can capture the temporal and spatial feature changes of migratory organisms, meet the requirements of high precision, and is suitable for real-time ecological monitoring in large-scale areas.

[0054] 2. This invention corrects migration paths using wind field data and fills in radar gaps caused by heavy precipitation using satellite remote sensing data, further optimizing migration trend prediction. The synergistic effect of multi-source data compensates for the limitations of a single data source, ensuring a complete description of the migration area and environmental adaptability of the path prediction, and exhibiting higher reliability and accuracy under complex weather conditions.

[0055] 3. By introducing generative adversarial networks, this invention generates highly reliable simulated data for blank areas in radar data, restoring the spatial continuity of the input data. The generated data has physical characteristics consistent with real radar echoes, enhancing the robustness and applicability of trend prediction, and providing high-quality monitoring results even in scenarios with heavy precipitation and missing data. Attached Figure Description

[0056] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see the appendix Figure 1 This invention provides a large-scale biological migration prediction method based on dual-polarization weather radar, comprising the following steps:

[0059] By extracting polarization parameters from dual-polarization weather radar data, we can comprehensively capture the characteristics of migratory organisms, effectively distinguish between biological echoes and meteorological echoes, and provide high-quality input data for subsequent classification and prediction.

[0060] By employing a polar coordinate to rectangular coordinate conversion method, radar data is converted into a unified 512×512 rectangular coordinate data format, and the spatial resolution is reconstructed. Through the unified rectangular coordinate data format and spatial resolution, this invention solves the resolution inconsistency problem caused by the original radar data format, ensuring the uniformity and effectiveness of subsequent model inputs, thereby improving the accuracy and stability of prediction.

[0061] A biological echo classification model is constructed to classify biological echoes and meteorological echoes based on polarization parameter data. By using multi-model fusion classification, this invention can effectively improve the accuracy of biological echo and meteorological echo classification, while reducing noise and misclassification risks, providing high-precision classification results for subsequent dynamic prediction of biological migration.

[0062] Based on the classified biological echo data, a spatiotemporal convolutional neural network (ConvLSTM) prediction model is constructed to perform short-term proximity prediction of biological migration. Through the spatiotemporal convolutional neural network, the dynamic characteristics of migrating organisms in time and space can be captured, and the migration trend in the next 1 to 2 hours can be accurately predicted with a root mean square error of less than 0.035, which meets the needs of practical applications.

[0063] This invention outputs dynamic images of biological migration over multiple future time periods, providing migration trend prediction results. By outputting dynamic migration results, this invention provides real-time monitoring and early warning capabilities for biological migration, and can provide technical support for agricultural pest control, ecological protection, and biological migration research.

[0064] Please see the appendix Figure 1 In a preferred embodiment of the present invention, the extracted polarization parameters include:

[0065] reflectivity factor This is used to describe the scattering intensity of the echo, and the specific extraction process includes:

[0066] Echo acquisition: Extracting horizontally polarized echo intensity from radar base data

[0067] The distribution of the data is calculated based on the distance and azimuth; data cleaning: removing...

[0068] The region is considered in this area because signals in this range are typically noise or invalid data, and the reflectivity factor... This provides fundamental characteristics for identifying biological targets; the range of values ​​can distinguish between low-intensity bioechoes and high-intensity precipitation echoes, providing an important basis for subsequent classification; differential reflectance Used to describe the shape and directionality of echoes, through differential reflectivity. It can accurately describe the shape characteristics of migratory biological targets, and when used in combination with reflectivity factors, it can effectively distinguish between migratory organisms and precipitation targets.

[0069] Differential phase Used to analyze the phase characteristics of target scatterers, differential phase It provides unique information for identifying the internal scattering characteristics of a target, and can further improve the accuracy of target differentiation when combined with other polarization parameters;

[0070] Correlation coefficient This is used to assess the similarity between horizontally and vertically polarized echoes, through the correlation coefficient. It can clearly distinguish between regularly shaped precipitation targets and irregularly shaped migratory organisms, improving the reliability of target identification;

[0071] Depolarization ratio Used to improve the distinction between biological echoes and meteorological echoes, depolarization ratio By combining information from multiple polarization parameters, the ability to distinguish between biological and meteorological targets is further enhanced, especially in complex weather conditions (such as areas where precipitation and migrating organisms intertwine).

[0072] Spectral width

[0073] The spectral width is used to characterize the distribution range of target motion velocity, serving as an auxiliary parameter for the dynamic characteristics of biological targets.

[0074] It provides a quantitative description of the dynamic characteristics of target motion, which helps to further distinguish between migratory organisms and static targets (such as ground clutter or precipitation), and enhances the performance of classification models.

[0075] Please see the appendix Figure 1 In a preferred embodiment of the present invention, the depolarization ratio The calculation methods include:

[0076] The initial calculation is based on differential reflectance and correlation coefficient, and the initial calculation formula is as follows:

[0077] ;

[0078] By combining differential reflectance and correlation coefficient, the depolarization ratio can effectively capture the irregular shape and uniformity characteristics of migratory biological targets, providing a basis for further optimization.

[0079] Introducing a weighting factor for the differential phase The optimization formula includes:

[0080] ;

[0081] in, Differential reflectivity, The correlation coefficient is defined as 0 ≤ ≤1;

[0082] Optimization process: Calculate the differential phase change. :

[0083] ;

[0084] in, Representing time points, the dynamic characteristics of the target are quantified by taking the absolute value of the differential phase between adjacent time points, through the introduction of differential phase and weighting factors. The formula for calculating the depolarization ratio is more flexible and can adapt to the distinction between biological echoes and meteorological echoes in complex scenarios, especially in areas with high precipitation intensity, it can effectively improve classification performance.

[0085] Please see the appendix Figure 1In a preferred embodiment of the present invention, polar coordinate data conversion includes:

[0086] Use the following formula to convert polar coordinate data to rectangular coordinate data:

[0087] ;

[0088] ;

[0089] in, For distance library, It is the azimuth angle. As the elevation angle, through formula conversion, the original polar coordinate data can be accurately mapped to the rectangular coordinate system, providing a consistent spatial basis for subsequent data reconstruction and analysis, and avoiding the problem that the polar coordinate resolution decreases with increasing distance;

[0090] The transformed data is reconstructed using bilinear interpolation to reconstruct spatial resolution, ensuring data spatial consistency and standardization.

[0091] The transformed data is reconstructed using bilinear interpolation to reconstruct spatial resolution, ensuring data spatial consistency and standardization.

[0092] Interpolation formula: for any point Its value is obtained by taking the values ​​of four adjacent grid points. The calculation yielded:

[0093] ;

[0094] ;

[0095] By using bilinear interpolation to reconstruct and standardize the data, spatial consistency of the output data can be guaranteed, providing a unified input format for subsequent deep learning model training, thereby improving the accuracy of classification and prediction.

[0096] Please see the appendix Figure 1 In a preferred embodiment of the present invention, the constructed biological echo classification model includes:

[0097] After extracting the polarization parameters, the XGBoost algorithm is used to classify biological echoes and meteorological echoes. The specific implementation process is as follows:

[0098] Input features: Select the polarization parameters to be extracted, including reflectivity factor. Differential reflectivity Differential phase Correlation coefficient Depolarization ratio and spectral width , forming the feature vector: ;

[0099] Historical radar data was labeled to generate training sample sets for biological and meteorological echoes, with the following feature distribution:

[0100] Bio-echoes: Eigenvalues ​​typically have small values. lower and larger ;

[0101] Meteorological echoes: Characteristic values ​​typically have high higher and smaller ;

[0102] A gradient boosting decision tree is constructed using the XGBoost algorithm. The objective function is to minimize the classification error, and the loss function is defined as follows:

[0103] ;

[0104] in, and These are the actual value and the predicted value, respectively. It is a regularization term used to prevent overfitting. XGBoost's ability to process multi-dimensional features enables the classification model to efficiently capture the characteristic differences between biological echoes and meteorological echoes. The classification results have high accuracy and high computational efficiency, making it suitable for large-scale radar data processing scenarios.

[0105] Using convolutional neural networks (CNNs) to extract spatial texture features, CNNs automatically extract spatial texture features of radar echoes through deep learning, which can capture subtle differences in the spatial distribution of biological echoes and improve the overall performance of classification models.

[0106] To improve classification accuracy, the outputs of XGBoost and Convolutional Neural Network (CNN) are fused using a weighted averaging method. The final classification probability after fusion is calculated as follows:

[0107] ;

[0108] in: and These represent the classification probabilities of biological echoes by XGBoost and CNN, respectively. The initial weight for fusion is 0.5, which is then optimized and adjusted through cross-validation. By weighted fusion of the outputs of XGBoost and CNN, this invention significantly improves the accuracy of biological echo classification and enhances the robustness of the model in complex scenarios.

[0109] Please see the appendix Figure 1 In a preferred embodiment of the present invention, the specific calculation formula for the weighted average method is as follows:

[0110] ;

[0111] in, These are the weighting coefficients, and the optimal values ​​are determined through cross-validation. : Represents the final classification result after weighted fusion, used to determine the probability of the input data being classified as biological echo or meteorological echo; : Represents the output of the XGBoost classification model, which is the probability value of the biological echo category; : Represents the output of the enhanced convolutional neural network, which is the probability value of the biological echo category; : Represents the weight coefficients of the CNN classification results, and Complementarity is used to balance the contributions of the two classification methods;

[0112] Cross-validation allows for dynamic adjustment of the weight distribution between XGBoost and CNN, enabling the fusion model to achieve optimal classification performance across different feature scenarios. This is achieved by adjusting the weight coefficients in different scenarios.

[0113] The fusion model has scene adaptability, which greatly improves the flexibility and robustness of classification.

[0114] Please see the appendix Figure 1 In a preferred embodiment of the present invention, the constructed spatiotemporal convolutional neural network (ConvLSTM) model includes:

[0115] By inputting radar echo images from multiple past time periods as spatiotemporal sequence data, the model can capture the temporal dynamics and spatial variation patterns during biological migration, providing high-quality input data for subsequent spatiotemporal modeling.

[0116] Spatiotemporal dynamic modeling is performed using a 5-layer ConvLSTM module to predict radar echo images at multiple future times. By stacking the 5-layer ConvLSTM module, the model can fully capture the dynamic characteristics of migratory organisms in time and space, thereby improving the prediction accuracy of future echo images.

[0117] The state update of a ConvLSTM cell includes the following steps:

[0118] ;

[0119] in, , , These represent the states of the forget gate, input gate, and output gate, respectively. For the unit state, By hiding the state and updating the state through the forget gate, input gate, and output gate, the ConvLSTM unit can flexibly capture the temporal and spatial feature changes of migratory organisms, ensuring the efficiency and accuracy of the model when processing spatiotemporal sequence data. By predicting radar echo images at multiple future times, the model achieves short-term proximity prediction of migratory organisms, providing real-time early warning capabilities for agricultural control and ecological protection.

[0120] Please see the appendix Figure 1 In a preferred embodiment of the present invention, the ConvLSTM model further incorporates an attention mechanism, which includes:

[0121] ;

[0122] ;

[0123] in, and The weight matrix is ​​a learnable matrix. For bias, To score attention, As normalized attention weights, by introducing an attention mechanism, the model can automatically allocate computing resources, prioritizing spatiotemporal regions that contribute significantly to biological migration prediction, thus significantly improving the model's feature extraction capabilities and prediction accuracy. The attention mechanism enhances the model's robustness and adaptability, ensuring the reliability and accuracy of migration trend prediction results. Through the extension of the attention mechanism, the model can maintain high performance in more diverse and complex scenarios, providing broader technical support for the monitoring and prediction of migratory organisms.

[0124] Please see the appendix Figure 1 In a preferred embodiment of the present invention, the migration trend prediction results are optimized through multi-source data fusion, wherein the multi-source data includes:

[0125] Wind field data: Extract wind speed and direction information, combine it with dual-polarization weather radar data, and use it to correct the spatial path of migration trends;

[0126] Satellite remote sensing data: High-resolution satellite imagery is used to fill in the blank areas in radar echoes caused by precipitation removal, thereby enhancing the description of environmental features in the migrating areas;

[0127] The fusion formula includes:

[0128] ;

[0129] in, , and The contributions of radar data, wind field data, and satellite remote sensing data to migration trends are respectively:

[0130] , and It is a fusion weight, satisfying + + =1, determined through experimental optimization. By fusing multi-source data, this invention can effectively compensate for the limitations of a single data source and improve the accuracy and stability of migration trend prediction results.

[0131] The introduction of wind field data effectively captures the impact of external meteorological factors on biological migration behavior, significantly improving the accuracy and reliability of migration path prediction; while the introduction of satellite remote sensing data fills the gap in radar data in areas of heavy precipitation and enhances the description of environmental characteristics in migration areas, providing key data support for more accurate trend prediction.

[0132] Please see the appendix Figure 1 In a preferred embodiment of the present invention, the method further includes:

[0133] To address the blank areas in the input data, a generative adversarial network (GAN) generates highly reliable simulated data to fill in the missing data. These blank areas originate from various sources, including: areas of heavy precipitation where the polarization characteristics of echoes are difficult to distinguish from biological echoes, often resulting in signal rejection and large areas of blank data; areas within the short-range detection range of radar where strong echo clutter renders the data unusable; and instances of data loss due to radar hardware or network transmission failures. By filling in the blank areas, the spatial continuity of the input data can be restored, providing a complete and consistent input for subsequent models and improving the accuracy of migration trend predictions.

[0134] GAN models, through adversarial training between the generator and discriminator, enable the generated simulated data to have characteristics highly consistent with real radar echo data, ensuring the credibility and physical rationality of the completed data. Through adversarial training between the generator and discriminator, the generated simulated data can highly restore the radar echo characteristics of the blank area, providing a highly reliable solution for input data completion.

[0135] To better understand the present invention, the above content will be described in detail below with reference to specific embodiments.

[0136] Example 1: Prediction of Migrating Organisms Based on Dual-Polarized Radar Data and Traditional Classification Methods

[0137] In this embodiment, only dual-polarization weather radar data is used, and biological echoes and meteorological echoes are classified using traditional machine learning classification methods (such as random forests or support vector machines).

[0138] The specific steps include:

[0139] Extract the basic polarization parameters (reflectivity factor, differential reflectivity, etc.) of the dual-polarization radar and classify them;

[0140] The random forest algorithm is used to classify radar data;

[0141] Without incorporating multi-source data and completion methods, biological migration trends are predicted solely based on single radar data.

[0142] Results: This method can classify biological echoes in simple scenarios, but its classification effect is poor under complex weather conditions. It cannot fill in the blank areas in areas of heavy precipitation, and its accuracy in predicting migration trends is insufficient, resulting in limited applicability.

[0143] Example 2: Predicting Migratory Organisms by Introducing Multi-Source Data Fusion

[0144] In this embodiment, in addition to dual-polarization weather radar data, wind field data and satellite remote sensing data are also introduced to perform multi-source data fusion;

[0145] The specific steps include:

[0146] Extract the polarization parameters of the dual-polarization radar and combine them with wind field data to correct the migration path;

[0147] Use satellite remote sensing imagery to fill in the gaps in radar data for areas of heavy precipitation;

[0148] The XGBoost algorithm combined with CNN is used to classify radar data, and multi-source data weighted fusion is performed to optimize migration trend prediction.

[0149] Results: This method enhances the dynamic environmental description of migratory areas through multi-source data fusion and improves the classification and prediction accuracy under complex weather conditions, but it still does not solve the problem of insufficient spatiotemporal continuity of input caused by missing data.

[0150] Example 3: Prediction of Migrating Organisms Based on Generative Adversarial Networks (GANs)

[0151] In this embodiment, a generative adversarial network (GAN) is further introduced based on embodiment 2 to fill in the blank areas of radar data with high reliability;

[0152] The specific steps include:

[0153] Using a GAN generator to generate simulated data consistent with real radar echoes, fill in the blank areas caused by heavy precipitation or missing data.

[0154] By integrating and supplementing radar data with multi-source data (wind field and satellite remote sensing), the classification and trend prediction models are further optimized.

[0155] Using a ConvLSTM model with an attention mechanism, key spatiotemporal features are dynamically captured to accurately predict migration trends.

[0156] Results: This embodiment improves the continuity of data input and the robustness of the model by using GAN to fill in the blank areas of radar data, ensuring high accuracy in migration trend prediction in scenarios with heavy precipitation or missing data, and providing comprehensive technical support for large-scale migratory biological monitoring.

[0157] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A large-scale biological migration prediction method based on dual-polarization weather radar, characterized in that, Includes the following steps: Extracting polarization parameters from dual-polarization weather radar data; The radar data was converted into a unified 512×512 rectangular coordinate data format by using a polar coordinate to rectangular coordinate conversion method, and the spatial resolution was reconstructed. A biological echo classification model was constructed to classify biological echoes and meteorological echoes from polarization parameter data; Based on the classified biological echo data, a spatiotemporal convolutional neural network prediction model is constructed to perform short-term proximity prediction of biological migration. Output dynamic images of biological migration at multiple future time points, providing migration trend prediction results; The polarization parameters include: Depolarization ratio (DR) is used to improve the distinction between biological echoes and meteorological echoes. The method for calculating the depolarization ratio DR includes: Initial calculations are performed based on differential reflectance and correlation coefficient; An optimization is performed by introducing a weighting factor k for the differential phase, and the formula includes: Among them, Z dr For differential reflectivity, ρ hv The correlation coefficient is defined as 0 ≤ ρ. hv ≤1; ρ hv The correlation coefficient is Δφ. DP The calculated differential phase change: Where t represents a time point, and φ is obtained by taking the absolute value of the differential phase between adjacent time points. DP Differential phase; The construction of the biological echo classification model includes: After extracting polarization parameters, the XGBoost algorithm was used to classify biological echoes and meteorological echoes. Use convolutional neural networks to extract spatial texture features; The weighted average method is used to fuse the outputs of the XGBoost algorithm and the convolutional neural network to improve classification accuracy. After extracting polarization parameters, the XGBoost algorithm was used to classify biological echoes and meteorological echoes, including: The polarization parameters include reflectivity factor Z and differential reflectivity Z0. dr Differential phase φ DP Correlation coefficient ρ hv Depolarization ratio (DR) and spectral width (W) s , forming the feature vector: X=[Z,Z dr ,φ DP ,ρ hv ,DR,W s ] Historical data collected by radar is labeled to generate training sample sets for biological and meteorological echoes; A gradient boosting decision tree is constructed using the XGBoost algorithm. The objective function is to minimize the classification error, and the loss function is defined as follows: Among them, y i and Let λ be the actual value and ω be the predicted value, respectively. 2 It is a regularization term; The specific calculation formula for the weighted average method is as follows: F final =αF XGBoost +(1-α)F CNN Where α is the weighting coefficient, and its optimal value is determined through cross-validation; F final : Represents the final classification result after weighted fusion, used to determine the probability of the input data being classified as biological echo or meteorological echo; F XGBoost : Represents the output of the XGBoost classification model, which is the probability value of the biological echo category; F CNN : Represents the output of the enhanced convolutional neural network, which is the probability value of the biological echo category; 1-α: Represents the weight coefficient of the CNN classification result, which is complementary to α and is used to balance the contributions of the two classification methods.

2. The method for large-scale biological migration prediction based on dual-polarization weather radar according to claim 1, characterized in that, The extracted polarization parameters include: The reflectivity factor Z is used to describe the scattering intensity of the echo. Differential reflectivity Z dr , used to describe the shape and directionality of the echo; Differential phase φ DP Used to analyze the phase characteristics of the target scatterer; Correlation coefficient ρ hv It is used to assess the similarity between horizontal and vertical polarization echoes; Depolarization ratio (DR) is used to improve the distinction between biological echoes and meteorological echoes. Spectral width W s It is used to characterize the distribution range of target motion velocity, serving as an auxiliary parameter for the dynamic characteristics of biological targets.

3. The method for large-scale biological migration prediction based on dual-polarization weather radar according to claim 1, characterized in that, Polar coordinate data transformation includes: Use the following formula to convert polar coordinate data to rectangular coordinate data: x=rcosθsinφ y=rsinθsinφ Where r is the distance, θ is the azimuth angle, and φ is the elevation angle; The transformed data is reconstructed using bilinear interpolation to reconstruct the spatial resolution, ensuring spatial consistency and standardization of the data.

4. The method for large-scale biological migration prediction based on dual-polarization weather radar according to claim 1, characterized in that, The construction of the spatiotemporal convolutional neural network model includes: Input radar echo images from multiple past time points as spatiotemporal sequence data; A 5-layer spatiotemporal convolutional neural network module is used for spatiotemporal dynamic modeling to predict radar echo images at multiple future time intervals. The state update of a spatiotemporal convolutional neural network unit includes the following steps: C t =f t ⊙C t-1 +i t ⊙tanh(W xc *x t +W hc *h t-1 +b c ) h t =o t ⊙tanh(C t ) Among them, f t i t o t These represent the states of the forget gate, input gate, and output gate, respectively, C. t For the unit state, h t It is in a hidden state.

5. The method for large-scale biological migration prediction based on dual-polarization weather radar according to claim 4, characterized in that, The spatiotemporal convolutional neural network model further introduces an attention mechanism, which includes: e ij = tanh(W q h ij +W k x ij +b e ) Among them, W q and W k b is a learnable weight matrix. e For bias, e ij For attention score, α ij This represents the normalized attention weights.

6. The method for large-scale biological migration prediction based on dual-polarization weather radar according to claim 1, characterized in that, The method further includes: To fill in the missing data in the blank areas of the input data, a generative adversarial network is used to generate highly reliable simulated data.

7. The method for large-scale biological migration prediction based on dual-polarization weather radar according to claim 1, characterized in that, The migration trend prediction results are optimized through multi-source data fusion, and the multi-source data includes: Wind field data: Extract wind speed and direction information, combine it with dual-polarization weather radar data, and use it to correct the spatial path of migration trends; Satellite remote sensing data: High-resolution satellite imagery is used to fill in the blank areas in radar echoes caused by precipitation removal, thereby enhancing the environmental characteristics description of the migration area.

Images