X-band dual-polarization weather radar nested intelligent extrapolation method and system

Abstract

The application provides an X-band dual-polarization rain measuring radar nested intelligent extrapolation method and system, and steps are as follows: S1, path integral attenuation correction, polarization parameter dynamic segmentation and rain area and non-rain area segmentation are carried out on radar echo data; S2, a multi-scale nested model of microscale, mesoscale and macroscale is constructed, and radar echo data of different resolutions are processed in each scale; S3, a multi-level LSTM network is used for intelligent extrapolation, and the confidence score of the polarization parameter is dynamically adjusted; S4, the extrapolation results of each scale are combined through a dynamic weight fusion method, and the weight coefficient is adjusted according to the signal-to-noise ratio and the life stage of the precipitation system; S5, short-term temporary precipitation prediction products are generated, and the products are verified and corrected through S-band radar real-time data inversion. The system can automatically correct the deviation, and ensure the high accuracy of the prediction results. The system has significant advantages in short-term and temporary precipitation prediction, strong convective monitoring and disaster warning.

Description

Technical Field

[0001] This invention belongs to the field of meteorological radar data processing technology, specifically relating to a nested intelligent extrapolation method and system for X-band dual-polarization rain measurement radar. Background Technology

[0002] Existing radar extrapolation methods generally face several technical challenges in short-term precipitation forecasting, particularly in terms of adaptability to complex precipitation systems and accuracy, where there is significant room for improvement. Traditional precipitation extrapolation methods mostly employ single-resolution radar data processing, lacking comprehensive consideration of multi-scale precipitation characteristics, resulting in limited forecast effectiveness.

[0003] 1. Limitations of single-scale extrapolation

[0004] Many current precipitation extrapolation methods, such as traditional optical flow methods or convolutional LSTM (Long Short-Term Memory) networks, are typically only applicable to single-resolution radar data processing, which limits their application across different spatial scales. Convective precipitation systems (such as thunderstorms and hail) often exhibit small spatial scales and rapid changes, while stratiform precipitation (such as continuous rainfall) belongs to larger-scale, more stable precipitation systems. In these precipitation systems of different scales, single-scale processing methods cannot effectively capture the fine structure of convective precipitation and the global characteristics of stratiform precipitation. Therefore, existing methods struggle to simultaneously capture the characteristics of small-scale and large-scale precipitation in short-term forecasts, resulting in low precipitation prediction accuracy, especially when complex weather systems occur, failing to provide sufficiently detailed forecasts.

[0005] 2. Insufficient utilization of polarization parameters

[0006] Current radar extrapolation methods, such as the algorithm in patent application CN113640832A, fail to fully utilize differential reflectivity (Z). dr ) and differential phase (K dp The advantages of polarization parameters such as differential reflectivity (Z0) in precipitation particle classification. Polarization radar data provides rich information about precipitation particle morphology, especially in the identification of mixed-phase precipitation, where it has unique advantages. dr It can effectively distinguish between different types of precipitation particles such as rain, snow, and hail, and is superior to differential phase (K-phase) precipitation. dp These polarization parameters can reflect the motion and size distribution of precipitation particles, thus aiding in accurate classification and estimation of precipitation. However, existing methods often fail to integrate these polarization parameters with other meteorological factors (such as radial velocity and echo reflectivity), leading to significant extrapolation errors in mixed-phase precipitation (such as hail and mixed rain and snow). Therefore, existing polarization parameters are underutilized, affecting the accuracy of precipitation forecasts.

[0007] 3. Poor dynamic adaptability

[0008] Most existing extrapolation methods employ fixed-weight fusion strategies (such as those mentioned in US20230194785A1 patent application), which lack flexibility in handling the evolution of precipitation systems. These methods typically use preset weighting coefficients to weight forecasts at different temporal and spatial scales, but cannot dynamically adjust based on the actual changes in the precipitation system. For example, in the spiral rainbands of a typhoon or the strong convective region of a supercell, the evolution of precipitation systems is highly complex and dynamic, and fixed-weight strategies cannot adapt to these changes. For severe convective weather (such as thunderstorms and hail) and large-scale weather systems (such as typhoons and frontal systems), fixed-weight strategies can cause significant deviations in precipitation amount and location, failing to meet the needs of real-time, detailed short-term forecasts. Therefore, the lack of dynamically adaptive extrapolation methods limits the accuracy and practicality of precipitation forecasts.

[0009] 4. Comprehensive Issues

[0010] The aforementioned problems demonstrate that existing radar extrapolation methods face significant challenges in handling short-term precipitation forecasts, particularly in capturing multi-scale precipitation characteristics, fully utilizing polarization parameters, and achieving dynamic adaptability, all of which require substantial optimization. As precipitation systems become more complex and weather events become more frequent and extreme, existing methods are increasingly revealing their inability to effectively address these issues. Therefore, a novel radar extrapolation method capable of simultaneously handling multi-scale, dynamic changes, and fine particle classification is urgently needed to improve the accuracy and reliability of precipitation forecasts. Summary of the Invention

[0011] This invention addresses the problems existing in the prior art by providing a nested intelligent extrapolation method for X-band dual-polarization rain-measuring radar, which can improve the accuracy and reliability of precipitation forecasting.

[0012] To solve the above technical problems, the present invention provides the following technical solution: a nested intelligent extrapolation method for X-band dual-polarization rain measurement radar, comprising the following steps:

[0013] S1. Perform path integral attenuation correction, dynamic polarization parameter segmentation, and segmentation of rain and non-rain areas on radar echo data.

[0014] S2. Construct multi-scale nested models of microscale, mesoscale, and macroscale to process radar echo data of different resolutions at microscale, mesoscale, and macroscale respectively.

[0015] S3. Employ a multi-level LSTM network for intelligent extrapolation, while dynamically adjusting the confidence score of polarization parameters.

[0016] S4. The extrapolation results at each scale are combined using a dynamic weighted fusion method, with the weighting coefficients adjusted according to the signal-to-noise ratio and the life stage of the precipitation system.

[0017] S5. Generate short-term temporary precipitation forecast products and verify and correct the errors using real-time S-band radar data.

[0018] Furthermore, the aforementioned step S1 includes the following sub-steps:

[0019] S1.1, Based on the phase difference Φ dp Path integral attenuation compensation is performed using the following formula:

[0020]

[0021] Among them, A Φdp Let L be the path integral attenuation, L be the distance, and K be the attenuation coefficient.

[0022] S1.2, through the correlation coefficient ρ hv Dynamic segmentation of rain and non-rain areas is performed, and the ice phase particle region is processed:

[0023]

[0024] Where E[] represents the expected value, Z h and Z v For horizontal and vertical radar reflectivity.

[0025] Furthermore, the aforementioned step S2 includes the following sub-steps:

[0026] S2.1 Construct a microscale model for the echo reflectivity Z h and differential reflectivity Z dr The fine structure of the convective monomer is extracted as follows:

[0027]

[0028] Among them, Z h It is the horizontal polarization reflectivity, Z v It is the vertical polarization reflectivity, Z dr Used to distinguish particle shapes.

[0029] S2.2 Construct a mesoscale model and analyze the differential phase K. dp Correcting particle trajectories and classifying different precipitation types, K dp The calculation formula is as follows:

[0030]

[0031] Where R is the distance. This represents the rate of phase change.

[0032] S2.3, Constructing a macro-scale model to fuse radial velocity V r By combining numerical model data with dynamic constraints, the evolution of large-scale precipitation systems can be captured. The extrapolation formula for the macroscale model is as follows:

[0033]

[0034] Among them, V r is the radial velocity, u is the wind speed component, and L is the horizontal distance calculated by the model.

[0035] Furthermore, the aforementioned step S3 includes the following sub-steps:

[0036] S3.1 In the scale model, a spatiotemporal convolutional LSTM network is used to capture the abrupt changes in echoes and to predict the rapid changes of convective cells.

[0037] S3.2 In the mesoscale model, environmental field parameters are introduced, and particle trajectories are optimized using a multi-layer LSTM network. The formula is as follows:

[0038]

[0039] Among them, T p Let T(z) be the saturation temperature, and T(z) be the air temperature. l and z u These are the heights of the bottom and top of the troposphere, respectively.

[0040] S3.3 In the macroscale model, large-scale dynamic constraints are provided by coupling the background field of the numerical model to assist the extrapolation process.

[0041] Furthermore, the aforementioned step S4 includes the following sub-steps:

[0042] S4.1. Adjust the weights of each scale based on the spatial distribution of radar signal-to-noise ratio. The calculation formula is as follows:

[0043]

[0044] Among them, SNR i Let be the signal-to-noise ratio at the i-th scale;

[0045] S4.2. Different fusion strategies are adopted for precipitation systems at different life stages. For strong convective regions, a micro- to mesoscale weighted geometric mean is used, as shown in the following formula:

[0046]

[0047] Among them, w micro w mesow macro These represent the weights for microscale, mesoscale, and large scale, respectively.

[0048] Furthermore, the aforementioned step S5 includes the following sub-steps:

[0049] S5.1 After completing nested LSTM extrapolation and dynamic weight fusion, the system outputs forecast products covering short-term temporary precipitation changes. The output products are spatial distribution maps generated by grid points, representing the precipitation at each grid point within the short temporary period. The formula for calculating precipitation P at any time step t is:

[0050]

[0051] Among them, P i (t) represents the precipitation prediction result of the i-th scale model at time step t, w i , which is the dynamic weighting coefficient corresponding to the scale, and N is the number of scales participating in the fusion, including microscale, mesoscale, and macroscale;

[0052] S5.2 The confidence score for dynamically adjusting polarization parameters based on radar echo characteristics is calculated using the following formula:

[0053]

[0054] Where Var() is the variance, and the differential reflectance Z is the differential reflectance. dr , differential phase K dp Variance reflects the degree of fluctuation in the forecast;

[0055] S5.3 For strong convection regions, the confidence score of polarization parameters can be updated in real time during the prediction process of nested LSTM networks, and the polarization characteristics of strong signal regions can be prioritized through dynamic attention mechanism.

[0056] S5.4. Inversion verification and bias correction are performed using real-time S-band radar data. During the inversion process, the Z-band radar data will be used as the basis for the inversion. dr K dp and V r Parameters such as these are used to compare radar observation data with forecast results through an inversion model.

[0057] The inversion process formula is as follows:

[0058] P observed (t)=f(Z dr ,K dp V r (11)

[0059] Among them, P observed(t) represents the actual precipitation data from the S-band radar at time t, and f() is a function based on the inversion algorithm, utilizing polarization data including Z. dr and K dp and radial velocity V r To infer the amount of precipitation;

[0060] S5.5 During the verification process, if a systematic error is found between the forecast product and the S-band radar real-time data, it shall be corrected using a deviation correction algorithm. The deviation correction formula is as follows:

[0061] P corrected (t)=P forecast (t)+δP(t) (12)

[0062] Among them, P forecast δP(t) is the original forecast result, and δP(t) is the bias correction value. The bias correction value is obtained by comparing the measured data and the forecast result, and is optimized by the least squares method or other fitting methods.

[0063] S5.6. Bias correction strategies are divided into two categories: Conventional correction: For non-convective areas and areas with low precipitation, a linear correction strategy is used for bias correction; Strong convection area correction: In strong convection areas, a nonlinear correction model is used, combined with an LSTM network to fit local nonlinear characteristics, for refined correction, as shown in the following formula:

[0064] P corrected (t)=P forecast (t)·(1+γ·SNR local (13)

[0065] Among them, SNR local It is the local signal-to-noise ratio, and γ is the correction coefficient, which is dynamically adjusted according to the characteristics of radar observation.

[0066] Another aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described in the present invention.

[0067] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described in the present invention.

[0068] Compared with the prior art, the beneficial technical effects of the present invention using the above technical solution are as follows:

[0069] This invention demonstrates innovation in multi-scale nested modeling, differential utilization of polarization parameters, dynamic weight fusion mechanism, cascaded LSTM network, fusion of numerical model and radar data, vortex constraint, and polarization parameter inversion verification. First, by nesting microscale (100-500m), mesoscale (1-3km), and macroscale (5-10km) models, it solves the problem that traditional single-scale methods struggle to simultaneously capture convective and stratiform cloud precipitation characteristics. Second, this invention innovatively combines Z... dr and K dp Polarization parameters are used for particle classification, significantly improving the classification accuracy (92%) of mixed-phase precipitation. A dynamic weight fusion mechanism adaptively adjusts the weights at each scale based on the signal-to-noise ratio and precipitation lifecycle stage, avoiding the limitations of fixed-weight strategies. Through a cascaded LSTM network and polarization feature attention module, the system prioritizes strong convective regions, improving the accuracy of short-term precipitation forecasts. Furthermore, the fusion of the numerical model background field provides dynamic constraints for macroscale precipitation, while vorticity constraints and supercell kinematic models enhance the predictive ability for extreme weather phenomena. Finally, through inversion verification with S-band radar data, the system can automatically correct biases, ensuring high accuracy of forecast results. This invention has significant innovation and technical advantages in short-term precipitation forecasting, strong convection monitoring, and disaster early warning. Attached Figure Description

[0070] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0071] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0072] In this invention, various aspects of the invention are described with reference to the accompanying drawings, in which numerous illustrative embodiments are shown. Embodiments of the invention are not limited to those depicted in the drawings. It should be understood that the invention is implemented through any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0073] refer to Figure 1 This invention provides a nested intelligent extrapolation method for X-band dual-polarization rain measurement radar, comprising the following steps:

[0074] S1. Perform path integral attenuation correction, dynamic polarization parameter segmentation, and segmentation of rain and non-rain areas on radar echo data.

[0075] S2. Construct multi-scale nested models of microscale, mesoscale, and macroscale to process radar echo data of different resolutions at microscale, mesoscale, and macroscale respectively.

[0076] S3. Employ a multi-level LSTM network for intelligent extrapolation, while dynamically adjusting the confidence score of polarization parameters.

[0077] S4. The extrapolation results at each scale are combined using a dynamic weighted fusion method, with the weighting coefficients adjusted according to the signal-to-noise ratio and the life stage of the precipitation system.

[0078] S5. Generate short-term temporary precipitation forecast products and verify and correct the errors using real-time S-band radar data.

[0079] In a preferred embodiment of the present invention, step S1 includes the following sub-steps:

[0080] S1.1, Based on the phase difference Φ dp Path integral attenuation compensation is performed using the following formula:

[0081]

[0082] Among them, A Φdp Let L be the path integral attenuation, L be the distance, and K be the attenuation coefficient.

[0083] S1.2, through the correlation coefficient ρ hv Dynamic segmentation of rain and non-rain areas is performed, and the ice phase particle region (ρ) is segmented. hv Process (value <0.85) accordingly:

[0084]

[0085] Where E[] represents the expected value, Z h and Z v For horizontal and vertical radar reflectivity.

[0086] In a preferred embodiment of the present invention, step S2 includes the following sub-steps:

[0087] S2.1 Construct a microscale model (100-500m grid) targeting the echo reflectivity Z. h and differential reflectivity Z dr The fine structure of the convective monomer is extracted as follows:

[0088]

[0089] Among them, Z h It is the horizontal polarization reflectivity, Z v It is the vertical polarization reflectivity, Z drUsed to distinguish particle shapes. Especially in regions of strong convection, it has a high degree of distinguishability for particles such as hail.

[0090] S2.2 Construct a mesoscale model (1-3km grid) and analyze the differential phase K. dp Correcting particle trajectories and classifying different precipitation types, including distinguishing between different phases of rain, snow, and hail, K dp The calculation formula is as follows:

[0091]

[0092] Where R is the distance. This represents the rate of phase change.

[0093] S2.3, Construct a macro-scale model (5-10km grid) and fuse radial velocity V r Dynamic constraints are applied by combining numerical model data (such as ECMWF wind fields) to capture the evolution of large-scale precipitation systems. The extrapolation formula for the macroscale model is as follows:

[0094]

[0095] Among them, V r is the radial velocity, u is the wind speed component, and L is the horizontal distance calculated by the model.

[0096] In a preferred embodiment of the present invention, step S3 includes the following sub-steps:

[0097] S3.1 In the scale model, a spatiotemporal convolutional LSTM network is used to capture the abrupt changes in echoes and to predict the rapid changes of convective cells.

[0098] S3.2 In the mesoscale model, environmental field parameters (such as CAPE and vertical wind shear) are introduced, and the particle trajectory is optimized through a multi-layer LSTM network. The formula is as follows:

[0099]

[0100] Among them, T p Let T(z) be the saturation temperature, and T(z) be the air temperature. l and z u These are the heights of the bottom and top of the troposphere, respectively.

[0101] S3.3 In the macroscale model, large-scale dynamic constraints are provided by coupling the background field of the numerical model to assist the extrapolation process.

[0102] In a preferred embodiment of the present invention, step S4 includes the following sub-steps:

[0103] S4.1 Adjust the weights of each scale based on the spatial distribution of radar signal-to-noise ratio (SNR). The calculation formula is as follows:

[0104]

[0105] Among them, SNR i Let be the signal-to-noise ratio at the i-th scale;

[0106] S4.2. Different fusion strategies are adopted for precipitation systems at different life stages (initial, mature, and dissipating). For strong convective regions, a micro- to mesoscale weighted geometric mean is used, as shown in the following formula:

[0107]

[0108] Among them, w micro w meso w macro These represent the weights for microscale, mesoscale, and large scale, respectively.

[0109] In a preferred embodiment of the present invention, step S5 includes the following sub-steps:

[0110] S5.1 After completing nested LSTM extrapolation and dynamic weight fusion, the system outputs forecast products covering short-term temporary precipitation changes within 0-2 hours. The output products are spatial distribution maps generated by grid points (e.g., 5km or 1km), representing the precipitation at each grid point within the short temporary period. The formula for calculating precipitation P at any time step t is:

[0111]

[0112] Among them, P i (t) represents the precipitation prediction result of the i-th scale model at time step t, w i , which is the dynamic weighting coefficient corresponding to the scale, and N is the number of scales participating in the fusion, including microscale, mesoscale, and macroscale;

[0113] S5.2 The confidence score for dynamically adjusting polarization parameters based on radar echo characteristics reflects the reliability and effectiveness of polarization parameters in forecasting. The confidence score C for polarization parameters is calculated using the following formula:

[0114]

[0115] Where Var() is the variance, and the differential reflectance Z is the differential reflectance. dr , differential phase K dp Variance reflects the degree of fluctuation in the forecast; a smaller variance indicates that the polarization parameter has better predictive stability in the region, and therefore a higher confidence level.

[0116] S5.3 For strong convection regions (such as hail and thunderstorms), the confidence score of polarization parameters can be updated in real time during the prediction process of the nested LSTM network. A dynamic attention mechanism is used to prioritize the polarization characteristics of strong signal regions (e.g., Z-axis polarization). dr >1dB,K dp >0.5° / km).

[0117] S5.4 To ensure the accuracy of the short-term precipitation forecast output by this invention, an inversion verification method with S-band radar real-time data is designed in the embodiment. S-band radar is usually used to provide high-precision precipitation data, so it can be used as standard reference data.

[0118] Inversion verification and bias correction were performed using real-time S-band radar data. During the inversion process, data from Z-band radar was used as the basis for the inversion. dr K d and V r Parameters such as these are used to compare radar observation data with forecast results through an inversion model.

[0119] The inversion process formula is as follows:

[0120] P observed (t)=f(Z dr ,K dp V r (11)

[0121] Among them, P observed (t) represents the actual precipitation data from the S-band radar at time t, and f() is a function based on the inversion algorithm, utilizing polarization data including Z. dr and K dp and radial velocity V r To infer the amount of precipitation;

[0122] S5.5 During the verification process, if a systematic error is found between the forecast product and the actual S-band radar data (e.g., the predicted value is too high or too low in some precipitation areas), it will be corrected using a bias correction algorithm.

[0123] The deviation correction formula is:

[0124] P corrected (t)=P forecast (t)+δP(t) (12)

[0125] Among them, P forecast δP(t) is the original forecast result, and δP(t) is the bias correction value. The bias correction value is obtained by comparing the measured data and the forecast result, and is optimized by the least squares method or other fitting methods.

[0126] S5.6. Bias correction strategies are divided into two categories: Conventional correction: For non-convective areas and areas with low precipitation, a linear correction strategy is used for bias correction; Strong convective area correction: In strong convective areas, such as hail and thunderstorms, a nonlinear correction model is used, combined with an LSTM network to fit local nonlinear characteristics, for refined correction, as shown in the following formula:

[0127] P corrected (t)=P forecast (t)·(1+γ·SNR local (13)

[0128] Among them, SNR local It is the local signal-to-noise ratio, and γ is the correction coefficient, which is dynamically adjusted according to the characteristics of radar observation.

[0129] Another aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described in this embodiment.

[0130] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described in this embodiment.

[0131] While the present invention has been described above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A nested intelligent extrapolation method for X-band dual-polarization rain-measuring radar, characterized in that, Includes the following steps: S1. Perform path integral attenuation correction, dynamic polarization parameter segmentation, and segmentation of rain and non-rain areas on radar echo data. S2. Construct multi-scale nested models of microscale, mesoscale, and macroscale to process radar echo data of different resolutions at microscale, mesoscale, and macroscale respectively. Specifically, it includes the following sub-steps: S2.1 Construct a microscale model targeting echo reflectivity. Differential reflectivity The fine structure of the convective monomer is extracted as follows: ， in, It is the horizontal polarization reflectivity. It is the vertical polarization reflectivity. Used to distinguish particle shapes; S2.2 Construct a mesoscale model and analyze the differential phase ratio. The particle trajectory is corrected, and different precipitation types are classified. The calculation formula is as follows: , in, For distance, The rate of change of phase; S2.3 Constructing a macro-scale model to fuse radial velocity By combining numerical model data with dynamic constraints, the evolution of large-scale precipitation systems can be captured. The extrapolation formula for the macroscale model is as follows: ; in, It is radial velocity. It is the wind speed component. It is the horizontal distance calculated by the model; S3. Employ a multi-level LSTM network for intelligent extrapolation, while dynamically adjusting the confidence score of polarization parameters. Specifically, it includes the following sub-steps: S3.1 In the scale model, a spatiotemporal convolutional LSTM network is used to capture the abrupt changes in echoes and to predict the rapid changes of convective cells. S3.2 In the mesoscale model, environmental field parameters are introduced, and particle trajectories are optimized using a multi-layer LSTM network. The formula is as follows: , in, The saturation temperature For temperature, and These are the heights of the bottom and top of the troposphere, respectively. S3.3 In the macroscale model, large-scale dynamic constraints are provided by coupling the background field of the numerical model to assist the extrapolation process; S4. The extrapolation results at each scale are combined using a dynamic weighted fusion method, with the weighting coefficients adjusted according to the signal-to-noise ratio and the life stage of the precipitation system. S5. Generate short-term temporary precipitation forecast products and verify and correct the errors using real-time S-band radar data.

2. The nested intelligent extrapolation method for X-band dual-polarization rain measurement radar according to claim 1, characterized in that, Step S1 includes the following sub-steps: S1.1, Based on phase difference Path integral attenuation compensation is performed using the following formula: , in, This is the path integral attenuation. For distance, The attenuation coefficient; S1.2, through correlation coefficient Dynamic segmentation of rain and non-rain areas is performed, and the ice phase particle region is processed: , in, Indicates the expected value. and For horizontal and vertical radar reflectivity.

3. The nested intelligent extrapolation method for X-band dual-polarization rain measurement radar according to claim 1, characterized in that, Step S4 includes the following sub-steps: S4.

1. Adjust the weights of each scale based on the spatial distribution of radar signal-to-noise ratio. The calculation formula is as follows: , in, Let be the signal-to-noise ratio at the i-th scale; S4.

2. Different fusion strategies are adopted for precipitation systems at different life stages. For strong convective regions, a micro- to mesoscale weighted geometric mean is used, as shown in the following formula: , in, , , These represent the weights for microscale, mesoscale, and large scale, respectively.

4. The nested intelligent extrapolation method for X-band dual-polarization rain measurement radar according to claim 1, characterized in that, Step S5 includes the following sub-steps: S5.1 After completing nested LSTM extrapolation and dynamic weight fusion, the system outputs forecast products covering short-term temporary precipitation changes. The output products are spatial distribution maps generated by grid points, representing the precipitation at each grid point within the short temporary period. The formula for calculating precipitation P at any time step t is: ， in, This is the precipitation prediction result of the i-th scale model at time step t. , which is the dynamic weighting coefficient corresponding to the scale, and N is the number of scales participating in the fusion, including microscale, mesoscale, and macroscale; S5.2 The confidence score for dynamically adjusting polarization parameters based on radar echo characteristics is calculated using the following formula: , in, It is variance, differential reflectance. , differential phase Variance reflects the degree of fluctuation in the forecast; S5.3 For strong convection regions, the confidence score of polarization parameters can be updated in real time during the prediction process of nested LSTM networks, and the polarization characteristics of strong signal regions can be prioritized through dynamic attention mechanism. S5.

4. Inversion verification and bias correction are performed using real-time S-band radar data. During the inversion process, based on... , and The parameters are used to compare radar observation data with forecast results through an inversion model. The inversion process formula is as follows: , in, This is the real-time precipitation data from the S-band radar at time t. It is a function based on an inversion algorithm, utilizing polarization data including and and radial velocity To infer the amount of precipitation; S5.5 During the verification process, if a systematic error is found between the forecast product and the actual S-band radar data, it shall be corrected using a deviation correction algorithm. The deviation correction formula is: , in, This is the original forecast result. It is the deviation correction value, which is obtained by comparing the measured data and the prediction results, and is optimized by the least squares method or other fitting methods. S5.

6. Bias correction strategies are divided into two categories: Conventional correction: For non-convective areas and areas with low precipitation, a linear correction strategy is used for bias correction; Strong convection area correction: In strong convection areas, a nonlinear correction model is used, combined with an LSTM network to fit local nonlinear characteristics, for refined correction, as shown in the following formula: , in, It is the local signal-to-noise ratio. It is a correction factor that is dynamically adjusted based on the characteristics of radar observation.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images