A multi-modal radar echo extrapolation method based on Doppler dual-polarization weather radar

Abstract

This invention discloses a multimodal radar echo extrapolation method based on Doppler dual-polarization weather radar, belonging to the field of meteorological technology. Specifically, the method involves: first, acquiring Doppler dual-polarization weather baseline data and converting it into multimodal physical parameter data; then, constructing a multimodal radar echo extrapolation model including a multimodal encoder, a depth feature extraction module, and a decoder; the encoder, composed of a dual-branch structure, performs preliminary feature extraction on the original multimodal information and inputs it into the depth feature extraction module, which further extracts depth spatiotemporal features through a self-attention mechanism; finally, the decoder completes the layer-by-layer decoding of the depth spatiotemporal features and introduces preliminary features to enhance the transmission of shallow feature information; for newly observed radar information, it is input into the trained echo extrapolation model to predict radar echo images at future times. This invention provides strong technical support for real-time meteorological monitoring, precipitation forecasting, and extreme weather warnings.

Description

Technical Field

[0001] This invention belongs to the field of meteorological technology, specifically relating to a multi-mode radar echo extrapolation method based on a Doppler dual-polarization weather radar. Background Technology

[0002] Doppler weather radar is an advanced active remote sensing device widely used in atmospheric environment monitoring and weather forecasting. Its basic principle is based on the Doppler effect; it transmits electromagnetic waves and receives the reflected signals to measure the velocity of scattering objects in clouds, rain, and other meteorological phenomena relative to the radar in real time. Under certain conditions, by utilizing the phase and frequency changes of the echo signals, the radar can infer the distribution of atmospheric wind fields, changes in vertical airflow velocity, and local turbulence states, providing a scientific basis for monitoring and early warning of severe convective weather.

[0003] Traditional radar echo extrapolation methods mainly include cross-correlation, single-cell centroid, and optical flow methods. These methods rely on the correlation or motion characteristics between signals to predict the distribution of future echoes. However, due to the limitations of the algorithms themselves under complex meteorological conditions, their prediction accuracy and applicability are difficult to meet practical needs.

[0004] In recent years, with the significant improvement in the performance of computer hardware (such as GPUs and TPUs), deep learning technology has developed rapidly and has shown superior performance compared to traditional methods in many fields, including image processing and spatiotemporal sequence prediction. Against this backdrop, data-driven radar echo extrapolation models have become a current research hotspot and one of the optimal solutions.

[0005] Existing research typically simplifies the radar echo extrapolation problem into a deep learning-based spatiotemporal sequence prediction subproblem, focusing on modeling and representing the motion characteristics of the radar echo itself. While these models have achieved significant breakthroughs in short-term prediction accuracy, the evolution of radar echoes is influenced not only by internal dynamic processes but also by other physical parameters. Most current models fail to fully integrate this additional multimodal physical parameter information, resulting in insufficient predictive capabilities when dealing with complex weather systems.

[0006] Extrapolating predictions to unobserved areas enables global monitoring of large-scale weather conditions, leading to a more accurate understanding of weather system evolution, monitoring of precipitation distribution and wind field changes, and improved early warning capabilities for extreme weather. Constructing deep learning-based models, by fully mining and integrating various historical physical parameter information, not only promises to improve the accuracy of future radar echo predictions but also provides more precise data support for real-time monitoring and forecasting of surface precipitation information.

[0007] However, existing technologies lack effective methods for utilizing multimodal physical parameter information from weather radar to optimize radar echo extrapolation. Therefore, developing a novel radar echo extrapolation method and system, supported by deep learning algorithms and multi-source data fusion technology, has significant theoretical implications and broad application prospects. Summary of the Invention

[0008] To address the problem that existing technologies rely solely on single radar echo information for extrapolation prediction, resulting in insufficient prediction accuracy and a failure to fully integrate multiple physical parameter information, this invention discloses a multi-mode radar echo extrapolation method based on Doppler dual-polarization weather radar. This method fully integrates various physical parameter information acquired by the radar system, effectively captures the complex dynamic characteristics in the radar echo evolution process, significantly improves prediction accuracy and system robustness, and provides strong data support for large-scale weather system monitoring and extreme weather early warning.

[0009] The multi-mode radar echo extrapolation method based on Doppler dual-polarization weather radar includes the following steps:

[0010] Step 1: Based on the base data collected by the Doppler dual-polarization weather radar, the data is processed uniformly through inversion algorithm and transformed into multimodal physical parameter data;

[0011] Multimodal physical parameter data includes horizontally polarized radar reflectivity Z. H Differential radar reflectivity Z DR Specific differential phase K DP , partial correlation coefficient ρ HV and velocity spectrum width σ V wait.

[0012] Step 2: Construct a multi-mode radar echo extrapolation model for a Doppler dual-polarization weather radar and train it using multi-mode data;

[0013] The multimodal radar echo extrapolation model includes a multimodal encoder, a depth feature extraction module, and a decoder;

[0014] The encoder consists of a dual-branch structure. It performs preliminary feature extraction on the original multimodal information and inputs it into the depth feature extraction module. The module further extracts the depth spatiotemporal features of the multimodal information through a self-attention mechanism. Finally, the decoder completes the layer-by-layer decoding of the depth spatiotemporal features and introduces preliminary features in the last layer of the decoder to strengthen the transmission of shallow feature information. Finally, the extrapolation result of the radar echo image is obtained.

[0015] The multimodal encoder has two core branches: an inter-spatial feature extraction branch and an intra-spatial feature extraction branch.

[0016] The formula for the spatial feature extraction branch is as follows:

[0017]

[0018] in, Representing the The mode in the th ... Features of the layer Representing the One original modal information; SiLU represents a nonlinear activation function. Represents a two-dimensional convolution operation. This represents the stride number of the convolutional structure, if If it is an odd number, then Take 1, If it is an even number Take 2. The first layer that will be passed to the decoder The shallow, preliminary features of the information; RFF represents the inter-factor features extracted by the spatial feature extraction branch in the multimodal encoder. This represents the fourth layer feature of the first modality, and Concat represents feature concatenation along the channel dimension.

[0019] The formula for the feature extraction branch within the spatial area is as follows:

[0020]

[0021] This represents the information obtained by stacking all modal information along the channel dimension, ultimately when... When the maximum value is 4, The branch representing in-space feature extraction extracts in-factor features (AFF).

[0022] Finally, the inter-factor features RFF extracted by the inter-spatial feature extraction branch and the intra-factor features AFF extracted by the intra-spatial feature extraction branch are superimposed in the channel dimension to generate shallow aggregated features FF.

[0023] The formula for the deep feature extraction module is as follows:

[0024]

[0025]

[0026] when The initial value is 1. The module consists of L stacked layers to form the shallow aggregated features FF extracted from the Multimodal Encoder. .

[0027] Represents the attention formula, This represents a point-wise convolution with a kernel size of 1. Represents a gated self-attention unit. This represents a self-regularized non-monotonic neural activation function. The formula representing a multilayer perceptron, Represents depthwise separable convolution. Representing the The spatiotemporal characteristics of the layer's depth This represents the BatchNorm feature.

[0028] Gated spatiotemporal self-attention unit The formula is as follows:

[0029]

[0030] in, Represents dilated depthwise convolution. Features representing the input, Represents an attention map. The characteristics that represent the output.

[0031] Finally, the deep feature extraction module outputs deep spatiotemporal features as follows: .

[0032] The decoding module is responsible for integrating the shallow preliminary features transmitted from the multimodal encoder with the deep spatiotemporal features extracted by the deep feature extraction module; through layer-by-layer decoding, it generates a final result with high-precision spatiotemporal extrapolation capability.

[0033] The data flow formula is as follows:

[0034]

[0035] in, and This represents the features after the first and second upsampling. This represents the features after convolution following the first upsampling. Represents deconvolution. This is the final prediction result for the radar echo data.

[0036] Step 3: Perform the same data preprocessing operation on the newly observed radar physical information and input it into the trained multimodal radar echo extrapolation model to obtain the radar echo prediction image for future times.

[0037] The advantages of this invention are:

[0038] This invention overcomes the limitations of traditional single-data-source extrapolation methods by making full use of multimodal physical parameter data and deep learning technology, providing strong technical support for real-time meteorological monitoring, precipitation forecasting and extreme weather warning. Attached Figure Description

[0039] Figure 1 This is a flowchart of a multi-mode radar echo extrapolation method based on a Doppler dual-polarization weather radar according to the present invention;

[0040] Figure 2 This is a schematic diagram illustrating the principle of constructing a multimode radar echo extrapolation model for a Doppler dual-polarization weather radar according to the present invention. Detailed Implementation

[0041] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are merely some, not all, embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0042] This invention discloses a multi-mode radar echo extrapolation method based on a Doppler dual-polarization weather radar. First, base data is acquired based on the Doppler dual-polarization weather radar, and the radar receiving system data undergoes quality control. Various physical parameter data are uniformly processed into multi-mode data at a specified altitude layer using a radar inversion algorithm. Second, a deep learning-based multi-mode radar echo extrapolation model is constructed and trained using multi-mode grayscale images. Finally, real-time acquired physical parameter data of different parameters are input into the model using the same inversion algorithm to extrapolate and obtain radar echo images for future times. This invention fully utilizes the unique advantages of the dual-polarization radar system and integrates multiple physical parameter information, enabling the extraction of key information from historical data and obtaining better radar echo extrapolation results, thereby significantly improving the accuracy and effectiveness of echo extrapolation.

[0043] like Figure 1 As shown, it includes the following steps:

[0044] Step 1: Based on the base data collected by the Doppler dual-polarization weather radar, the data is processed uniformly through inversion algorithm and transformed into multimodal physical parameter data;

[0045] First, a Doppler dual-polarization weather radar is used to scan precipitation, clouds, and other meteorological targets in the atmosphere in real time to collect raw baseline data.

[0046] Subsequently, the collected raw base data is processed uniformly using a preset inversion algorithm, transforming the raw data into structured multimodal physical parameter data.

[0047] Weather radar data receiving system receives data at fixed intervals. indivual Radar physical parameter information of size, including Z H (Horizontal polarization radar reflectivity), Z DR (Differential radar reflectivity), K DP (Specific differential phase), ρ HV (Partial co-correlation coefficient) and σ V (Velocity spectral width), etc. Radar physical observation data at time t are expressed as... Data processing is unified through inversion algorithms, transforming it into... Multimodal data.

[0048] Step 2: Construct a multi-mode radar echo extrapolation model for a Doppler dual-polarization weather radar and train it using multi-mode data;

[0049] The multimodal radar echo extrapolation model includes a multimodal encoder, a depth feature extraction module, and a decoder;

[0050] like Figure 2 As shown, multimodal data is fed into a dual-branch encoder for independent feature extraction. Each encoder consists of convolution and downsampling modules to extract shallow spatiotemporal features of different modalities. The features extracted by the first and second branches are then superimposed along the channel dimension to generate aggregated features. Subsequently, a layered spatiotemporal self-attention mechanism is used to learn the continuous spatiotemporal dynamic relationships between multimodal features, resulting in deep spatiotemporal features. Finally, continuous convolution and upsampling of multiple deep spatiotemporal features are decoded to obtain the radar echo prediction image for future timeframes.

[0051] The multimodal encoder has two core branches: an inter-spatial feature extraction branch and an intra-spatial feature extraction branch.

[0052] The inter-spatial feature extraction branch focuses on global feature interaction across modalities. It generates unified inter-factor features through feature stacking to deeply explore the collaborative relationships between different modalities and enhance the fusion effect of cross-modal information.

[0053] The spatial feature extraction branch focuses on the deep coupling relationships within multimodal data, generating intra-factor features to achieve a refined representation of information within multimodal data. The features extracted in the first step of the spatial feature extraction branch are then directly superimposed with those from the decoder module. This module integrates multiple physical parameters and, based on a dual-branch information fusion strategy, achieves preliminary extraction and fusion of multimodal features, aiming to efficiently capture and integrate key information from multimodal data.

[0054] First, the spatial feature extraction branch formula of the multimodal encoder is as follows.

[0055]

[0056] in, Representing the The mode in the th ... Features of the layer Representing the One original modal information; SiLU represents a nonlinear activation function. Represents a two-dimensional convolution operation. This represents the stride number of the convolutional structure, if If it is an odd number, then Take 1, If it is an even number Take 2. The first layer that will be passed to the decoder The shallow, preliminary features of the information; RFF represents the inter-factor features extracted by the spatial feature extraction branch in the multimodal encoder. This represents the fourth layer feature of the first modality, and Concat represents feature concatenation along the channel dimension.

[0057] Secondly, the spatial feature extraction branch formula of the multimodal encoder is as follows:

[0058]

[0059] This represents the information obtained by stacking all modal information along the channel dimension, ultimately when... When the maximum value is 4, The branch representing in-space feature extraction extracts in-factor features (AFF).

[0060] Finally, the inter-factor features RFF extracted by the inter-spatial feature extraction branch and the intra-factor features AFF extracted by the intra-spatial feature extraction branch are superimposed in the channel dimension to generate shallow aggregated features FF.

[0061] The deep feature extraction module, through a layered spatiotemporal self-attention mechanism, deeply learns the continuous spatiotemporal dynamic relationships between multimodal features, enhancing the model's ability to model spatiotemporal dependencies. The formula is as follows:

[0062]

[0063]

[0064] when The initial value is 1. The module consists of L stacked layers to form the shallow aggregated features FF extracted from the Multimodal Encoder. . Represents the attention formula, This represents a point-wise convolution with a kernel size of 1. Represents a gated self-attention unit. This represents a self-regularized non-monotonic neural activation function. The formula representing a multilayer perceptron, Represents depthwise separable convolution. Representing the The spatiotemporal characteristics of the layer's depth This represents the BatchNorm feature.

[0065] A gated spatiotemporal self-attention (GSSA) unit was designed to address the complex multimodal spatiotemporal features in this task, aiming to further extract deep features effective for extrapolation tasks. The formula is as follows.

[0066]

[0067] in, Represents dilated depthwise convolution. Features representing the input, Represents an attention map. The characteristics that represent the output.

[0068] Finally, the output features of the deep feature extraction module are: .

[0069] The decoding module integrates the shallow radar features from the multimodal encoder with the multimodal features extracted by the deep feature extraction module. Through layer-by-layer decoding, this module generates a final result with high-precision spatiotemporal extrapolation capabilities. The overall architecture effectively combines shallow and deep features of multimodal information, aiming to improve the model's predictive performance in complex dynamic environments. The data flow formula is as follows:

[0070]

[0071] in, and This represents the features after the first and second upsampling. This represents the features after convolution following the first upsampling. Represents deconvolution. This is the final prediction result for the radar echo data.

[0072] Step 3: Perform the same data preprocessing operation on the real-time observed radar physical information, and input it into the trained multi-mode radar echo extrapolation model to obtain the prediction result of the future radar echo state at the current moment.

[0073] For the current observation time t, the input to the multimode radar echo extrapolation model is: Observational data representing the past T historical moments, To predict the future Radar echo data at each moment.

[0074] This invention utilizes various physical parameter data, such as reflectivity, differential reflectivity, correlation coefficient, and differential phase, acquired by a Doppler dual-polarization weather radar. Through preprocessing and feature extraction, the multimodal data is transformed into input features suitable for deep learning modeling. Subsequently, a deep learning-based multimodal radar echo extrapolation model is constructed to achieve high-precision extrapolation of radar echo images for future times. By fully integrating various physical parameter information acquired by the radar system, this method effectively captures the complex dynamic characteristics of radar echo evolution, significantly improving prediction accuracy and system robustness, and providing strong data support for large-scale weather system monitoring and extreme weather early warning.

[0075] Finally, it should be noted that although this specification contains many specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily used to describe the features of specific embodiments of a particular invention. Certain features described in the various embodiments of this specification may also be implemented in combination in a single embodiment. On the other hand, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation of a sub-combination.

[0076] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0077] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0078] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the technical scope of the conclusions of the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-mode radar echo extrapolation method based on a Doppler dual-polarization weather radar, characterized in that, Includes the following steps: First, based on the baseline data collected by the Doppler dual-polarization weather radar, the data is processed uniformly through an inversion algorithm and transformed into multimodal physical parameter data; Multimodal physical parameter data includes horizontally polarized radar reflectivity Differential radar reflectivity Specific differential phase Co-partial correlation coefficient and velocity spectrum ; Then, a multi-mode radar echo extrapolation model of a Doppler dual-polarization weather radar was constructed and trained using multi-mode data; The real-time observed radar physical information is converted into multi-mode physical parameter data and then input into a trained multi-mode radar echo extrapolation model to obtain the radar echo prediction image for future moments. The multimodal radar echo extrapolation model includes a multimodal encoder, a depth feature extraction module, and a decoder; The multimodal encoder consists of a dual-branch structure: an inter-spatial feature extraction branch and an intra-spatial feature extraction branch. It performs preliminary feature extraction on the original multimodal physical parameter data, and superimposes the inter-factor features RFF extracted by the inter-spatial feature extraction branch and the intra-factor features AFF extracted by the intra-spatial feature extraction branch in the channel dimension to generate shallow aggregated features FF, which are then input into the deep feature extraction module. The deep feature extraction module further extracts multimodal deep spatiotemporal features through a self-attention mechanism. Finally, the depth spatiotemporal features are decoded layer by layer using a decoder, and preliminary features are introduced in the last layer of the decoder. This is done to enhance the transmission of shallow feature information, ultimately obtaining the extrapolation results of the radar echo image.

2. The multi-mode radar echo extrapolation method based on Doppler dual-polarization weather radar as described in claim 1, characterized in that, The formula for the spatial feature extraction branch is as follows: Representing the The mode in the th ... Features of the layer ; Representing the One original modal information, ; SiLU represents a nonlinear activation function. Represents a two-dimensional convolution operation. This represents the stride number of the convolutional structure, if If it is an odd number, then Take 1, If it is an even number Take 2; The first layer that will be passed to the decoder The superficial, preliminary characteristics of information; RFF represents the inter-factor features extracted by the spatial feature extraction branch in a multimodal encoder; This represents the fourth layer feature of the first modality; Concat represents feature concatenation along the channel dimension. The formula for the feature extraction branch within the spatial area is as follows: This represents the information obtained by stacking all modal information along the channel dimension, ultimately when... When the maximum value is 4, The branch representing in-space feature extraction extracts in-factor features (AFF).

3. The multi-mode radar echo extrapolation method based on Doppler dual-polarization weather radar as described in claim 1, characterized in that, The formula for the deep feature extraction module is as follows: when The initial value is 1. The shallow aggregated features FF extracted from the multimodal encoder are stacked in L layers. ; Represents the attention formula, This represents a point-wise convolution with a kernel size of 1. Represents a gated self-attention unit. This represents a self-regularized non-monotonic neural activation function. The formula representing a multilayer perceptron, Represents depthwise separable convolution. Representing the The spatiotemporal characteristics of the layer's depth Represents the BatchNorm function; Finally, the deep feature extraction module outputs deep spatiotemporal features as follows: .

4. The multi-mode radar echo extrapolation method based on Doppler dual-polarization weather radar as described in claim 3, characterized in that, The gated self-attention unit The formula is as follows: in, Represents dilated depthwise convolution. Features representing the input, Represents an attention map. The characteristics that represent the output.

5. The multi-mode radar echo extrapolation method based on Doppler dual-polarization weather radar as described in claim 2, characterized in that, The decoder data stream formula is as follows: in, and This represents the features after the first and second upsampling. This represents the features after the first upsampling and convolution. Represents deconvolution. This is the final prediction result for the radar echo data.

Images