A raindrop spectrum adaptive inversion method based on dual-polarization radar data

Abstract

The application provides a raindrop spectrum adaptive inversion method based on dual-polarization radar data, comprising the following steps: obtaining base data of the dual-polarization radar, pre-processing the base data to generate standard input data for inversion; constructing a discretized raindrop spectrum state vector, establishing a physical forward model based on T-matrix scattering theory, and associating the standard input data with the raindrop spectrum state vector through the physical forward model; constructing a joint cost function according to a maximum likelihood data fitting term and a Bayesian prior constraint term; optimizing and iteratively solving the joint cost function through an L-BFGS-B algorithm, and outputting an optimal raindrop spectrum state vector; fitting and generating a continuous raindrop spectrum curve according to the output optimal raindrop spectrum state vector; and taking the raindrop spectrum curve as the final raindrop spectrum obtained by inversion. The application can overcome the limitations of traditional raindrop spectrum inversion methods in information utilization and adaptability, and improve the adaptability and accuracy of raindrop spectrum inversion under different conditions.

Description

Technical Field

[0001] This invention relates to the field of raindrop spectrum prediction technology, and in particular to an adaptive inversion method for raindrop spectrum based on dual-polarization radar data. Background Technology

[0002] Raindrop spectra (DSD) are key parameters of precipitation microphysical processes, describing the distribution of the number of raindrops of different diameters per unit volume of air. Accurate measurement and inversion of DSD are crucial for a deeper understanding of precipitation formation mechanisms, improving the accuracy of quantitative precipitation estimation (QPE), and refining microphysical parameterization schemes in numerical weather prediction models. For a long time, DSD research has been a hot topic in atmospheric science and remote sensing. Traditional DSD measurements mainly rely on ground-based raindrop spectrometers, but this point-based measurement cannot provide large-scale and continuous precipitation information. With the development of remote sensing technology, especially the application of radar technology, large-scale, high spatiotemporal resolution precipitation detection has become possible. Radar acquires precipitation information by emitting microwave signals and receiving the scattered echoes from precipitation particles (such as raindrops). However, the physical quantities observed by radar (such as reflectivity factor Z, differential reflectivity Zdr, and differential propagation phase shift Kdp) are a comprehensive effect of DSD, not DSD itself. Therefore, how to accurately invert DSD from radar parameters is a core scientific problem in radar meteorology.

[0003] Existing DSD inversion methods typically assume that the DSD follows a specific mathematical function form, such as a gamma distribution or an exponential distribution. For example, a gamma-distributed DSD model is usually determined by intercept, shape, and slope parameters. This parameterization simplifies the inversion process, but in practical applications, the shape of the DSD can vary significantly due to various microphysical processes (such as raindrop collision, breakup, and evaporation), especially under different weather types (such as stratiform and convective cloud precipitation) and geographical regions (such as mountainous and marine areas). Pre-defined DSD models may fail to accurately capture its complexity and diversity. For instance, most studies use the gamma distribution, but research shows that the MP distribution performs better in certain situations.

[0004] In reality, raindrop spectral distributions are diverse, including not only unimodal distributions but also bimodal and trimodal distributions. The formation of these different modes is the result of multiple complex atmospheric microphysical processes, such as raindrop generation, growth, merging, fragmentation, and evaporation. The gamma function has significant limitations when fitting raindrop spectra (DSD), especially when dealing with bimodal and trimodal raindrop spectra. The gamma function is essentially an approximation of a unimodal distribution; when empirical raindrop spectrum data exhibits two or more significant peaks, a single gamma function cannot accurately fit these independent peaks and the valleys between them. Forcing a single gamma function for fitting results in a fitted curve that fails to accurately capture the complete shape of the raindrop spectrum, especially with significant deviations in the peak and valley regions. Therefore, more flexible and physically meaningful models are needed to better understand and predict precipitation microphysical processes. Summary of the Invention

[0005] This invention provides an adaptive inversion method for raindrop spectrum based on dual-polarization radar data, which can overcome the limitations of traditional raindrop spectrum inversion methods in terms of information utilization and adaptability, and improve the adaptability and accuracy of raindrop spectrum inversion under different weather systems and different geographical regions.

[0006] The first aspect of this invention provides an adaptive inversion method for raindrop spectrum based on dual-polarization radar data, comprising the following steps: Acquire the basic data of the dual-polarization radar, including the reflectivity factor (Z). h ), differential reflectance (Z) dr ), differential phase shift (K) dp The base data undergoes preprocessing, including quality control, attenuation correction, and 3D meshing, to generate standard input data for inversion. Discretized raindrop spectral state vectors are constructed, and a physical forward model based on T-matrix scattering theory is established. The standard input data is then correlated with the raindrop spectral state vectors through the physical forward model. Set the maximum likelihood data fitting term and the Bayesian prior constraint term; construct the joint cost function based on the maximum likelihood data fitting term and the Bayesian prior constraint term; The joint cost function is optimized and iteratively solved using the L-BFGS-B algorithm, and the optimal raindrop spectral state vector is output. Based on the output of the optimal raindrop spectrum state vector, a continuous raindrop spectrum curve is generated by using cubic spline interpolation; the raindrop spectrum curve is then used as the final raindrop spectrum obtained through inversion.

[0007] Furthermore, the preprocessing of the base data, including quality control, attenuation correction, and three-dimensional meshing, includes noise filtering, ground clutter suppression, velocity ambiguity reduction, and phase ambiguity processing on the base data acquired by the dual-polarization radar. Signal attenuation correction is performed on the propagation path of the basic data acquired by the dual-polarization radar; The base data acquired by the dual-polarization radar is interpolated into a unified three-dimensional Cartesian grid coordinate system to generate standardized input data.

[0008] Furthermore, the construction of the discretized raindrop spectral state vector and the establishment of a physical forward model based on T-matrix scattering theory, and the correlation between the standard input data and the raindrop spectral state vector through the physical forward model, include the following steps: The diameter range of the raindrops to be inverted is divided into multiple continuous intervals, and the state vector of the raindrop spectrum is defined as the raindrop concentration value in each interval; Obtain the preprocessed base data, set the parameterized relation, and estimate the macroscopic parameters of the raindrop spectrum based on the parameterized relation; By substituting the macroscopic parameters into the normalized Gamma distribution formula, the theoretical concentration value for each diameter interval is calculated, and the theoretical concentration value is used as the initial state vector for optimization iteration. A physical forward model based on T-matrix scattering theory is established; the physical forward model calculates the theoretically corresponding radar observation parameter values ​​based on the given raindrop spectral state vector.

[0009] Furthermore, the maximum likelihood data fitting term is calculated using the following formula: in, For the maximum likelihood data fitting term; , and These are the error covariance matrices for each radar observation, used to assign different weights to observations of different accuracies. and These represent the original reflectivity factor and the calculated theoretical reflectivity factor, respectively. and These represent the original differential reflectance factor and the calculated theoretical differential reflectance factor, respectively. and These represent the original differential phase shift and the calculated theoretical differential phase shift, respectively.

[0010] Furthermore, the Bayesian prior constraint term is used to introduce physical rationality constraints; the physical rationality constraints include smoothness priors, nonnegativity priors, and spatiotemporal continuity priors; The smoothness prior is used to penalize the second derivative of the raindrop spectrum curve to avoid non-physical oscillations. The nonnegativity prior is used to enforce that the raindrop concentration value is greater than or equal to zero across all diameter ranges; The aforementioned spatiotemporal continuity prior, by using the inversion results of adjacent spatial grids or the previous observation time as a reference, ensures that the current inversion results maintain spatiotemporal continuity and consistency.

[0011] Furthermore, the step of optimizing and iteratively solving the joint cost function using the L-BFGS-B algorithm and outputting the optimal raindrop spectral state vector includes the following steps: Obtain the initial state vector and use it as the starting point for iteration; During the iterative loop, the theoretical radar observation value corresponding to the current iteration state vector is calculated using the physical forward model; the value and gradient of the joint cost function are calculated, and the initial state vector is updated. The iteration terminates when the decrease in the joint cost function value is less than a preset threshold or when the maximum number of iterations is reached, and the final optimal state vector is output.

[0012] Furthermore, the step of using cubic spline interpolation to fit and generate a continuous raindrop spectrum curve based on the output optimal raindrop spectrum state vector includes the following steps: Obtain the optimal state vector and use it as an interpolation node; Apply cubic spline interpolation to the optimal state vector to perform smooth curve fitting; Generate a continuous function curve that is twice differentiable over the entire raindrop diameter range, and use the continuous function curve as the final raindrop spectrum obtained by inversion.

[0013] The second aspect of the present invention provides a raindrop spectrum adaptive inversion system based on dual-polarization radar data, including a first calculation module for acquiring the base data of the dual-polarization radar, performing preprocessing on the base data including quality control, attenuation correction and three-dimensional meshing, and generating standard input data for inversion; The second calculation module is used to construct a discretized raindrop spectral state vector, establish a physical forward model based on the T-matrix scattering theory, and associate the standard input data with the raindrop spectral state vector through the physical forward model; The third calculation module is used to set the maximum likelihood data fitting term and the Bayesian prior constraint term; and to construct the joint cost function based on the maximum likelihood data fitting term and the Bayesian prior constraint term. The fourth calculation module is used to optimize and iteratively solve the joint cost function using the L-BFGS-B algorithm and output the optimal raindrop spectral state vector. The fifth calculation module is used to fit and generate a continuous raindrop spectrum curve based on the output optimal raindrop spectrum state vector using cubic spline interpolation; the raindrop spectrum curve is used as the final raindrop spectrum obtained by inversion.

[0014] A third aspect of the present invention provides a computer device, comprising: Memory, transceiver, processor, and bus system; The memory is used to store programs; The processor is used to execute the program in the memory, including executing the raindrop spectrum adaptive inversion method based on dual-polarization radar data as described above. The bus system is used to connect the memory and the processor to enable communication between the memory and the processor.

[0015] A fourth aspect of the present invention provides a readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the raindrop spectrum adaptive inversion method based on dual-polarization radar data described above.

[0016] As can be seen from the above technical solutions, the present invention has the following advantages: This invention achieves adaptive inversion of raindrop spectrum morphology by discretizing the raindrop spectrum into a state vector and generating a continuous curve using spline fitting. This invention does not require pre-constraint of the spectrum shape and can flexibly invert various complex real-world morphologies, including single-peak and double-peak patterns. It overcomes the inherent bias of traditional parameterization methods when fitting atypical structures such as multi-peak distributions, improving the adaptability and accuracy of raindrop spectrum inversion under different weather systems and geographical regions.

[0017] Secondly, this invention constructs a rigorous physical forward model based on T-matrix scattering theory and designs a weighted maximum likelihood fitting term that integrates multiple observation parameters. This enables the coordinated and in-depth utilization of multi-parameter information from dual-polarization radar. By weighting multi-channel observation data such as horizontal reflectivity factor, differential reflectivity, and differential propagation phase shift, along with their different observation accuracies, through error covariance matrix weighting, it unifies them into an optimization framework. This fully explores the complex nonlinear cooperative constraint information between various observation parameters, thereby improving the accuracy of retrieving raindrop spectrum state from observation data and the reliability of information extraction.

[0018] Finally, by setting a maximum likelihood term, this invention achieves optimal matching between the function solution and the observation data. Furthermore, by introducing Bayesian prior constraints with smoothness, non-negativity, and spatiotemporal continuity, the inversion problem is regularized, suppressing the divergence or non-physical oscillations in the function solution that may be caused by observation noise and uncertainty. This makes it more flexible in dealing with complex and variable actual precipitation scenarios and can stably output physically reasonable and spatiotemporally continuous raindrop spectrum inversion results under a wider range of conditions.

[0019] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from an examination of the following, or may be learned from the practice of the invention. Attached Figure Description

[0020] Figure 1 The method flowchart provided by the present invention. Detailed Implementation

[0021] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] Example 1 The implementation method in this embodiment can be implemented in a system, on a server, or on a terminal, without explicit limitation. The method in this application will be described below from the perspective of system implementation. The first aspect of this invention provides a raindrop spectrum adaptive inversion method based on dual-polarization radar data, comprising the following steps: Acquire the basic data of the dual-polarization radar, including the reflectivity factor (Z). h ), differential reflectance (Z) dr ), differential phase shift (K) dp The base data undergoes preprocessing, including quality control, attenuation correction, and 3D meshing, to generate standard input data for inversion. Discretized raindrop spectral state vectors are constructed, and a physical forward model based on T-matrix scattering theory is established. The standard input data is then correlated with the raindrop spectral state vectors through the physical forward model. Set the maximum likelihood data fitting term and the Bayesian prior constraint term; construct the joint cost function based on the maximum likelihood data fitting term and the Bayesian prior constraint term; The joint cost function is optimized and iteratively solved using the L-BFGS-B algorithm, and the optimal raindrop spectral state vector is output. Based on the output of the optimal raindrop spectrum state vector, a continuous raindrop spectrum curve is generated by using cubic spline interpolation; the raindrop spectrum curve is then used as the final raindrop spectrum obtained through inversion.

[0023] The standard input data generated for inversion involves quality control and physical correction of the original radar base data, such as core parameters like reflectivity, differential reflectivity, and differential phase shift, and unification them into a regular three-dimensional spatial grid. This eliminates observation errors and system biases, generating standardized input data with high consistency and reliability.

[0024] Subsequently, the continuous raindrop spectrum to be solved is discretized into a state vector consisting of concentration values ​​for each diameter interval. A physical forward model based on T-matrix scattering theory is established as a physics engine to calculate the theoretical radar observation values, such as reflectivity and differential reflectivity, corresponding to any given discrete raindrop spectrum state vector; thus establishing a physical link between the raindrop spectrum and radar observations.

[0025] Next, a joint cost function is set as the optimization objective of the entire inversion process, consisting of two parts: a maximum likelihood data fitting term, which calculates the theoretical observation value corresponding to the inverted raindrop spectrum, which is statistically close to the actual radar observation value; and a Bayesian prior constraint term, which is used to introduce prior knowledge about the physical rationality of the raindrop spectrum, such as smoothness, non-negativity and spatiotemporal continuity.

[0026] The L-BFGS-B numerical optimization algorithm minimizes the joint cost function, starting with an estimated initial raindrop spectrum, such as assuming a Gamma distribution, and iteratively adjusts the state vector. In each iteration, the physical forward model calculates the theoretical value, the optimization algorithm evaluates the fit and updates the spectrum shape using gradient information, ensuring that the theoretical predictions continuously approach the actual observations while satisfying physical constraints. Ultimately, the algorithm converges to a discrete raindrop spectrum state vector that achieves an optimal balance between data fit and physical plausibility.

[0027] Finally, the optimal discrete state vector obtained through optimization is used as a set of data points, and a smooth and continuous raindrop spectrum curve is generated by fitting the data using cubic spline interpolation, which serves as the result of raindrop spectrum inversion.

[0028] Example 2 The difference between this embodiment and Embodiment 1 is that the preprocessing of the base data, including quality control, attenuation correction and three-dimensional meshing, includes noise filtering, ground clutter suppression, velocity ambiguity reduction and phase ambiguity processing on the base data acquired by the dual polarization radar. Signal attenuation correction is performed on the propagation path of the basic data acquired by the dual-polarization radar; The base data acquired by the dual-polarization radar is interpolated into a unified three-dimensional Cartesian grid coordinate system to generate standardized input data.

[0029] The base data includes the reflectivity factor, differential reflectivity, differential phase shift, and correlation coefficient of the dual-polarization radar. During data processing, noise filtering, ground clutter suppression, and deblurring are used to identify and remove non-meteorological echoes and observation artifacts, thus refining the base data. Next, physical correction is applied to the signal attenuation caused by electromagnetic waves propagating in rain areas to restore the true intensity of the radar parameters. Finally, the quality-controlled and corrected data is interpolated into a unified three-dimensional Cartesian grid.

[0030] Example 3 The difference between this embodiment and Embodiment 2 lies in the following steps: Constructing a discretized raindrop spectral state vector and establishing a physical forward model based on T-matrix scattering theory, then correlating the standard input data with the raindrop spectral state vector through the physical forward model: The diameter range of the raindrops to be inverted is divided into multiple continuous intervals, and the state vector of the raindrop spectrum is defined as the raindrop concentration value in each interval; Obtain the preprocessed base data, set the parameterized relation, and estimate the macroscopic parameters of the raindrop spectrum based on the parameterized relation; By substituting the macroscopic parameters into the normalized Gamma distribution formula, the theoretical concentration value for each diameter interval is calculated, and the theoretical concentration value is used as the initial state vector for optimization iteration. A physical forward model based on T-matrix scattering theory is established; the physical forward model calculates the theoretically corresponding radar observation parameter values ​​based on the given raindrop spectral state vector.

[0031] When discretizing the raindrop spectrum, the raindrop diameter range is discretized into M continuous intervals, and the state vector is defined as the raindrop concentration value in each interval: The raindrop spectrum is represented as a vector consisting of concentration values ​​in each interval, defining the state space to be solved.

[0032] Next, an initial value x0 is set, and a parameterized relation is established based on the preprocessed base data. This parameterized relation includes the use of differential reflectivity factors. Mass-weighted average diameter of raindrop spectrum The relational expression of the relationship: Differential phase shift With normalized raindrop concentration Relationship: in, , , , These are all empirical coefficients. Furthermore, assuming a constant shape parameter μ, the obtained... , Substituting μ into the normalized Gamma distribution formula, we can calculate the diameter interval for each range. Theoretical concentration value Finally, the calculated vector [N1, N2, ..., N_M] is used as the initial state vector x0 for the optimization iteration.

[0033] Finally, a physical forward model based on T-matrix theory is established. Used to calculate a given raindrop spectral state vector Theoretically, the radar parameters that should be observed at that time are: It can simulate the theoretical dual-polarization radar observation values ​​that should be generated by any given discrete raindrop spectrum state vector.

[0034] Example 4 The difference between this embodiment and Embodiment 3 is that the maximum likelihood data fitting term is calculated using the following formula: in, For the maximum likelihood data fitting term; , and These are the error covariance matrices for each radar observation, used to assign different weights to observations of different accuracies. and These represent the original reflectivity factor and the calculated theoretical reflectivity factor, respectively. and These represent the original differential reflectance factor and the calculated theoretical differential reflectance factor, respectively. and These represent the original differential phase shift and the calculated theoretical differential phase shift, respectively.

[0035] The maximum likelihood data fitting term quantifies and minimizes the overall deviation between theoretical predictions and multi-source observation data by taking into account the uncertainty differences of different observation parameters.

[0036] Example 5 The difference between this embodiment and embodiment four is that the Bayesian prior constraint term is used to introduce physical rationality constraints; the physical rationality constraints include smoothness prior, nonnegativity prior, and spatiotemporal continuity prior; The smoothness prior is used to penalize the second derivative of the raindrop spectrum curve to avoid non-physical oscillations. The nonnegativity prior is used to enforce that the raindrop concentration value is greater than or equal to zero across all diameter ranges; The aforementioned spatiotemporal continuity prior, by using the inversion results of adjacent spatial grids or the previous observation time as a reference, ensures that the current inversion results maintain spatiotemporal continuity and consistency.

[0037] The smoothness prior, by mathematically penalizing the second-order difference of the state vector, suppresses non-physical high-frequency oscillations that may occur during the optimization process. The formula for calculating the smoothness prior is: Where L is the Laplacian operator matrix, These are the weighting coefficients for the smoothness constraint.

[0038] The nonnegativity prior is used to perform the basic requirement of data measurement, ensuring that the raindrop concentration cannot be negative.

[0039] in, These are the weighting coefficients for nonnegativity constraints. Let be the raindrop concentration value for the i-th interval.

[0040] The spatiotemporal continuity prior introduces reliable inversion results from neighboring grids or previous moments as reference information by leveraging the spatiotemporal correlation of the precipitation field, guiding the current inversion results to maintain reasonable consistency with these results. The smoothness prior, nonnegativity prior, and spatiotemporal continuity prior work together to regularize the inversion problem and improve the reliability of the estimation.

[0041] Example 6 The difference between this embodiment and Embodiment 5 is that the step of optimizing and iteratively solving the joint cost function using the L-BFGS-B algorithm and outputting the optimal raindrop spectral state vector includes the following steps: Obtain the initial state vector and use it as the starting point for iteration; During the iterative loop, the theoretical radar observation value corresponding to the current iteration state vector is calculated using the physical forward model; the value and gradient of the joint cost function are calculated, and the initial state vector is updated. The iteration terminates when the decrease in the joint cost function value is less than a preset threshold or when the maximum number of iterations is reached, and the final optimal state vector is output.

[0042] The joint cost function consists of two parts: in, For the joint cost function, For the maximum likelihood data fitting term, These are Bayesian prior constraints.

[0043] During iteration, the state vector is initialized first. The L-BFGS-B algorithm is employed to minimize the joint cost function. In each iteration, the following steps are performed: Call the physical forward model Calculate the theoretical radar value; Calculate the joint cost function and the corresponding gradient ; Update the state vector according to the L-BFGS-B algorithm: ; The iteration terminates when the cost function decreases by less than a threshold or when the maximum number of iterations is reached, and the optimal discrete raindrop spectrum state vector is output. .

[0044] In each iteration, based on the currently guessed raindrop spectrum, the corresponding radar observation value is predicted using a physical forward model. Then, the system evaluates the merits of the current guess by calculating the joint cost function and its gradient at the current point, indicating the direction for improvement. The L-BFGS-B algorithm constructs an approximate quadratic model of the objective function based on the function values ​​and gradient information from the current and historical iterations. On this model, it calculates the search direction and step size to decrease the function value, thereby updating the state vector. Finally, the iteration terminates when two conditions are met: one is that the improvement in the joint cost function is below a threshold; the other is setting a maximum number of iterations to prevent infinite loops.

[0045] Example 7 The difference between this embodiment and Embodiment Six is ​​that the step of using cubic spline interpolation to fit and generate a continuous raindrop spectrum curve based on the output optimal raindrop spectrum state vector includes the following steps: Obtain the optimal state vector and use it as an interpolation node; Apply cubic spline interpolation to the optimal state vector to perform smooth curve fitting; Generate a continuous function curve that is twice differentiable over the entire raindrop diameter range, and use the continuous function curve as the final raindrop spectrum obtained by inversion.

[0046] The optimal state vector from the discrete inversion result As data points, i.e., point sets These can be considered as control points or nodes describing the morphology of the raindrop spectrum. Next, cubic spline interpolation is used to fit the nodes, generating a smooth, globally second-order continuous, differentiable curve over the entire raindrop spectrum diameter. This avoids the oscillations or uneven curves that may occur with linear interpolation.

[0047] Example 8 A raindrop spectrum adaptive inversion system based on dual-polarization radar data includes a first calculation module for acquiring the base data of the dual-polarization radar, performing preprocessing on the base data including quality control, attenuation correction and three-dimensional meshing, and generating standard input data for inversion. The second calculation module is used to construct a discretized raindrop spectral state vector, establish a physical forward model based on the T-matrix scattering theory, and associate the standard input data with the raindrop spectral state vector through the physical forward model; The third calculation module is used to set the maximum likelihood data fitting term and the Bayesian prior constraint term; and to construct the joint cost function based on the maximum likelihood data fitting term and the Bayesian prior constraint term. The fourth calculation module is used to optimize and iteratively solve the joint cost function using the L-BFGS-B algorithm and output the optimal raindrop spectral state vector. The fifth calculation module is used to fit and generate a continuous raindrop spectrum curve based on the output optimal raindrop spectrum state vector using cubic spline interpolation; the raindrop spectrum curve is used as the final raindrop spectrum obtained by inversion.

[0048] Example 9 A computer device, comprising: Memory, transceiver, processor, and bus system; The memory is used to store programs; The processor is used to execute the program in the memory, including executing the raindrop spectrum adaptive inversion method based on dual-polarization radar data as described above. The bus system is used to connect the memory and the processor to enable communication between the memory and the processor.

[0049] Example 10 A readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the raindrop spectrum adaptive inversion method based on dual-polarization radar data described above.

[0050] In summary, this invention achieves adaptive inversion of raindrop spectral morphology by discretizing the raindrop spectrum into a state vector and generating a continuous curve using spline fitting. This invention does not require pre-constraint of the spectral shape and can flexibly invert various complex real-world morphologies, including single-peak and double-peak patterns. It overcomes the inherent bias of traditional parameterization methods when fitting atypical structures such as multi-peak distributions, thus improving the adaptability and accuracy of raindrop spectral inversion under different weather systems and geographical regions.

[0051] Secondly, this invention constructs a rigorous physical forward model based on T-matrix scattering theory and designs a weighted maximum likelihood fitting term that integrates multiple observation parameters. This enables the coordinated and in-depth utilization of multi-parameter information from dual-polarization radar. By weighting multi-channel observation data such as horizontal reflectivity factor, differential reflectivity, and differential propagation phase shift, along with their different observation accuracies, through error covariance matrix weighting, it unifies them into an optimization framework. This fully explores the complex nonlinear cooperative constraint information between various observation parameters, thereby improving the accuracy of retrieving raindrop spectrum state from observation data and the reliability of information extraction.

[0052] Finally, by setting a maximum likelihood term, this invention achieves optimal matching between the function solution and the observation data. Furthermore, by introducing Bayesian prior constraints with smoothness, non-negativity, and spatiotemporal continuity, the inversion problem is regularized, suppressing the divergence or non-physical oscillations in the function solution that may be caused by observation noise and uncertainty. This makes it more flexible in dealing with complex and variable actual precipitation scenarios and can stably output physically reasonable and spatiotemporally continuous raindrop spectrum inversion results under a wider range of conditions.

[0053] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0054] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0055] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0056] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive inversion method for raindrop spectrum based on dual-polarization radar data, characterized in that, Includes the following steps: Acquire the basic data of the dual-polarization radar, perform preprocessing on the basic data including quality control, attenuation correction and three-dimensional meshing, and generate standard input data for inversion; Discretized raindrop spectral state vectors are constructed, and a physical forward model based on T-matrix scattering theory is established. The standard input data is then correlated with the raindrop spectral state vectors through the physical forward model. Set the maximum likelihood data fitting term and the Bayesian prior constraint term; Construct a joint cost function based on the maximum likelihood data fitting term and the Bayesian prior constraint term; The joint cost function is optimized and iteratively solved using the L-BFGS-B algorithm, and the optimal raindrop spectral state vector is output. Based on the output of the optimal raindrop spectrum state vector, a continuous raindrop spectrum curve is generated by using cubic spline interpolation; the raindrop spectrum curve is then used as the final raindrop spectrum obtained through inversion.

2. The raindrop spectrum adaptive inversion method based on dual-polarization radar data according to claim 1, characterized in that, The preprocessing of the base data includes quality control, attenuation correction and three-dimensional meshing, including noise filtering, ground clutter suppression, velocity ambiguity reduction and phase ambiguity processing on the base data acquired by the dual polarization radar. Signal attenuation correction is performed on the propagation path of the basic data acquired by the dual-polarization radar; The base data acquired by the dual-polarization radar is interpolated into a unified three-dimensional Cartesian grid coordinate system to generate standardized input data.

3. The raindrop spectrum adaptive inversion method based on dual-polarization radar data according to claim 1, characterized in that, The process of constructing a discretized raindrop spectral state vector and establishing a physical forward model based on T-matrix scattering theory, and then correlating the standard input data with the raindrop spectral state vector through the physical forward model, includes the following steps: The diameter range of the raindrops to be inverted is divided into multiple continuous intervals, and the state vector of the raindrop spectrum is defined as the raindrop concentration value in each interval; Obtain the preprocessed base data, set the parameterized relation, and estimate the macroscopic parameters of the raindrop spectrum based on the parameterized relation; By substituting the macroscopic parameters into the normalized Gamma distribution formula, the theoretical concentration value for each diameter interval is calculated, and the theoretical concentration value is used as the initial state vector for optimization iteration. A physical forward model based on T-matrix scattering theory is established; the physical forward model calculates the theoretically corresponding radar observation parameter values ​​based on the given raindrop spectral state vector.

4. The raindrop spectrum adaptive inversion method based on dual-polarization radar data according to claim 1, characterized in that, The maximum likelihood data fitting term is calculated using the following formula: in, This is the term used for fitting the maximum likelihood data; , and These are the error covariance matrices for each radar observation, used to assign different weights to observations of different accuracies.

5. The raindrop spectrum adaptive inversion method based on dual-polarization radar data according to claim 1, characterized in that, The Bayesian prior constraint term is used to introduce physical rationality constraints; the physical rationality constraints include smoothness prior, nonnegativity prior, and spatiotemporal continuity prior; The smoothness prior is used to penalize the second derivative of the raindrop spectrum curve to avoid non-physical oscillations. The nonnegativity prior is used to enforce that the raindrop concentration value is greater than or equal to zero across all diameter ranges; The aforementioned spatiotemporal continuity prior, by using the inversion results of adjacent spatial grids or the previous observation time as a reference, ensures that the current inversion results maintain spatiotemporal continuity and consistency.

6. The raindrop spectrum adaptive inversion method based on dual-polarization radar data according to claim 1, characterized in that, The process of optimizing and iteratively solving the joint cost function using the L-BFGS-B algorithm and outputting the optimal raindrop spectral state vector includes the following steps: Obtain the initial state vector and use it as the starting point for iteration; During the iterative loop, the theoretical radar observation value corresponding to the current iteration state vector is calculated using the physical forward model; the value and gradient of the joint cost function are calculated, and the initial state vector is updated. The iteration terminates when the decrease in the joint cost function value is less than a preset threshold or when the maximum number of iterations is reached, and the final optimal state vector is output.

7. The raindrop spectrum adaptive inversion method based on dual-polarization radar data according to claim 1, characterized in that, The step of generating a continuous raindrop spectrum curve by fitting the optimal raindrop spectrum state vector using cubic spline interpolation includes the following steps: Obtain the optimal state vector and use it as an interpolation node; Apply cubic spline interpolation to the optimal state vector to perform smooth curve fitting; Generate a continuous function curve that is twice differentiable over the entire raindrop diameter range, and use the continuous function curve as the final raindrop spectrum obtained by inversion.

8. A raindrop spectrum adaptive inversion system based on dual-polarization radar data, characterized in that, It includes a first calculation module, which is used to acquire the basic data of the dual-polarization radar, perform preprocessing on the basic data including quality control, attenuation correction and three-dimensional meshing, and generate standard input data for inversion; The second calculation module is used to construct a discretized raindrop spectral state vector, establish a physical forward model based on the T-matrix scattering theory, and associate the standard input data with the raindrop spectral state vector through the physical forward model; The third calculation module is used to set the maximum likelihood data fitting term and the Bayesian prior constraint term. Construct a joint cost function based on the maximum likelihood data fitting term and the Bayesian prior constraint term; The fourth calculation module is used to optimize and iteratively solve the joint cost function using the L-BFGS-B algorithm and output the optimal raindrop spectral state vector. The fifth calculation module is used to fit and generate a continuous raindrop spectrum curve based on the output optimal raindrop spectrum state vector using cubic spline interpolation; the raindrop spectrum curve is used as the final raindrop spectrum obtained by inversion.

9. A computer device, characterized in that, include: Memory, transceiver, processor, and bus system; The memory is used to store programs; The processor is used to execute a program in the memory, including executing the raindrop spectrum adaptive inversion method based on dual-polarization radar data as described in any one of claims 1 to 7; The bus system is used to connect the memory and the processor to enable communication between the memory and the processor.

10. A readable storage medium storing computer-readable instructions, characterized in that, When the computer-readable instructions are executed by the processor, they implement the steps of the raindrop spectrum adaptive inversion method based on dual-polarization radar data as described in any one of claims 1 to 7.

Images