Radar echo intelligent extrapolation and evolution identification system for short-term and imminent forecast

Abstract

The invention discloses a radar echo intelligent extrapolation and evolution identification system for short-term and imminent forecasting in the technical field of meteorological monitoring. The radar echo intelligent extrapolation and evolution identification system comprises a data acquisition module used for generating an echo image covering the whole domain; the feature extraction module is used for extracting space-time evolution features in the echo image; the terrain power correction module is used for generating a terrain weight factor; the urban heat island correction module is used for generating a heat island weight factor; the prediction module is used for identifying potential severe convection weather and predicting the influence range and duration of the weather; the dynamic data module is used for feeding back the updated data to the prediction module; the optimization module is used for adjusting algorithm parameters for different rainfall types; the early warning module is used for generating early warning information. According to the invention, environmental factors are dynamically coupled through the terrain and heat island correction module, the influence of complex terrain retardation and urban heat island disturbance on radar echoes is solved, the physical consistency and intensity evolution precision of an extrapolation path are improved, and adaptive correction of sudden weather is realized in combination with dynamic data assimilation.

Description

Technical Field

[0001] This invention relates to the field of meteorological monitoring technology, specifically to a radar echo intelligent extrapolation and evolution identification system for short-term forecasting. Background Technology

[0002] In short-term weather forecasting, the core challenge of radar echo extrapolation lies in accurately modeling the dynamic impact of complex environmental factors on echo evolution. Existing technologies, such as traditional extrapolation algorithms (e.g., optical flow and cross-correlation methods), typically rely on the spatiotemporal translation assumption of the echo sequence, neglecting the hindering effect of terrain undulations on the echo path (e.g., the lifting and diversion of airflow by mountains) and the nonlinear enhancement of echo intensity by the urban heat island effect. For example, a radar echo extrapolation short-term forecasting method based on deep learning, as announced in patent publication number CN110967695A, includes: firstly, acquiring measured radar echo data and rainfall; using the radar echo data to obtain an equivalent reflectivity factor, and using the reflectivity factor and rainfall to obtain a ZR relationship; constructing a radar echo extrapolation model, preprocessing the radar echo data, and using the processed data to train the radar echo extrapolation model to predict radar echo data within a future period; finally, inputting the predicted radar echo data into the ZR relationship to obtain the predicted rainfall. This invention can predict precipitation over a large area using gridded data, and the accuracy of the 1-hour precipitation forecast is higher than that of numerical model forecasts, with more accurate predictions of the precipitation area.

[0003] In mountainous scenarios, fixed terrain elevation data is used for path correction, but the rate of change of terrain slope (such as surface deformation caused by landslides and snowmelt) is not integrated in real time, leading to the accumulation of echo offset errors. In urban areas, existing methods mostly use static building density parameters to estimate the impact of the heat island, ignoring minute-level fluctuations in surface temperature gradients (such as a sharp increase in heat island intensity in the afternoon), resulting in an underestimation of strong echo intensity. In addition, the independent processing mode of terrain and heat island factors (such as superimposing only through linear interpolation) makes it difficult to quantify their synergistic effect (such as the local outbreak of convection triggered by the heat island in valley terrain), causing a physical mismatch between extrapolated paths and intensity evolution.

[0004] Therefore, this invention proposes a radar echo intelligent extrapolation and evolution identification system for short-term forecasting to solve the above problems. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a radar echo intelligent extrapolation and evolution identification system for short-term forecasting. By dynamically coupling environmental factors through terrain and heat island correction modules, it resolves the impact of complex terrain obstruction and urban heat island disturbance on radar echoes, improves the physical consistency of the extrapolation path and the accuracy of intensity evolution, and achieves adaptive correction for sudden weather events by combining dynamic data assimilation.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: a radar echo intelligent extrapolation and evolution identification system for short-term forecasting, comprising a data acquisition module for acquiring echo data through a network of weather radars and updating the echo data in real time, preprocessing the acquired echo data, and then fusing and stitching multiple radar data using a radar mosaic method to generate an echo image covering the entire area.

[0007] The feature extraction module is used to extract spatiotemporal evolution features in echo images using a deep learning model, and introduces recurrent dynamic convolutional layers and attention mechanisms into the deep learning model.

[0008] The terrain dynamics correction module is used to fuse digital elevation model data and slope change rate in real time to generate terrain weight factors; the terrain weight factors characterize the obstruction effect of terrain undulation on the echo movement path.

[0009] The urban heat island correction module integrates building density data with surface temperature gradient to generate a heat island weighting factor; the heat island weighting factor quantifies the enhancement effect of the urban heat island effect on echo intensity.

[0010] The prediction module uses deep learning to analyze the spatiotemporal evolution characteristics, generate initial radar echo extrapolation results for the next 0-3 hours in 6-minute increments, and then dynamically corrects the initial radar echo extrapolation results based on the joint influence function of terrain weight factor and heat island weight factor, generating final radar echo extrapolation results for the next 0-3 hours in 6-minute increments. Finally, it analyzes the echo morphology and movement trajectory based on the final radar echo extrapolation results, identifies potential severe convective weather, and predicts its impact range and duration.

[0011] The dynamic data module updates the terrain slope change rate and heat island intensity gradient every 5 minutes and feeds the updated data back to the prediction module to adjust the extrapolation parameters in real time.

[0012] The optimization module is used to optimize precipitation estimation by utilizing the reflectance-rainfall intensity relationship, adjust algorithm parameters for different precipitation types, and correct extrapolation biases by using real-time data feedback.

[0013] The early warning module is used to identify danger signals in the echo extrapolation results and generate early warning information.

[0014] Furthermore, meteorological physical constraints are embedded in the prediction module, including mass conservation and energy conservation equations.

[0015] Furthermore, the precipitation types include severe convection and stratiform clouds.

[0016] Furthermore, danger signals include echo areas ≥45dBZ and thunderstorm cells with a hail probability >80%.

[0017] Furthermore, the spatiotemporal evolution characteristics include echo movement speed, morphological change trends, and intensity gradient.

[0018] Furthermore, the deep learning model is either SwinAt-UNet or a recurrent dynamic convolutional neural network.

[0019] Furthermore, in the dynamic data module, the rate of change of terrain slope is inverted in real time through InSAR surface deformation data, and the heat island intensity gradient is calculated by fusing meteorological satellite infrared channel data with urban building GIS layers.

[0020] Furthermore, the prediction module performs the following operations:

[0021] An initial displacement field is generated based on optical flow, and then a spatiotemporal convolutional network is used to predict the next 6 frames of echo images. Morphological analysis algorithms are used to identify vortex structures and bow-shaped echo features in the next 6 frames of echo images, and the probability of severe convective weather is calculated by combining the movement trajectory. The life cycle stage of the identified thunderstorm cells is labeled, and the boundary of the influence range is predicted by adjusting the environmental wind field.

[0022] Furthermore, in the dynamic data module, the update steps for the terrain slope change rate are as follows:

[0023] a. Receive InSAR surface deformation data every 5 minutes and extract the temporal changes in terrain elevation;

[0024] b. Calculate the slope change rate based on the time series change and the time interval; multiply the time interval by the horizontal resolution, and use the product as the numerator and the time series change as the denominator to obtain the slope change rate;

[0025] c. Input the slope change rate into the terrain dynamics correction module to adjust the calculation parameters of the terrain weight factor in real time.

[0026] Furthermore, in the dynamic data module, the update steps for the heat island intensity gradient are as follows:

[0027] a. Acquire infrared channel data from meteorological satellites every 5 minutes to retrieve the Earth's surface temperature field;

[0028] b. Calculate the heat island intensity gradient by combining the building density distribution in the urban building GIS layer;

[0029] c. Input the heat island intensity gradient into the urban heat island correction module to dynamically adjust the gain coefficient of the heat island weight factor.

[0030] The above-mentioned approach has the following beneficial effects: 1. By integrating digital elevation models and InSAR surface deformation data in real time through the terrain dynamic correction module, the slope change rate is dynamically calculated to generate terrain weight factors, quantifying the terrain blocking effect (such as the lifting and diversion of airflow by mountains), overcoming the path offset accumulation problem caused by the reliance on static elevation data in traditional methods. Simultaneously, the urban heat island correction module integrates meteorological satellite infrared data and building GIS layers, updating the heat island intensity gradient at a minute-level frequency, constructing a nonlinear mapping relationship between the heat island weight factors and echo intensity, accurately capturing the enhancing effect of urban heat islands on echoes (such as an echo intensity gain of 3-5 dBZ in densely built-up areas). The terrain and heat island factors work synergistically in the prediction module through a joint influence function. For example, in mountainous urban areas, the terrain blocking effect forces echoes to detour, while the heat island effect triggers enhanced local convection. The system achieves a physical balance between the two effects through dynamic weight allocation, reducing extrapolation path error and decreasing the root mean square error of intensity prediction, significantly improving the forecast reliability for complex underlying surface areas.

[0031] 2. The dynamic data module updates terrain slope and urban heat island gradient data every 5 minutes, achieving near real-time assimilation of environmental factors through InSAR deformation inversion and fusion with satellite temperature fields. For example, a sudden landslide causing a sharp increase in terrain slope (slope change rate > 15° / km) can trigger a dynamic upward adjustment of the terrain weight factor, forcibly correcting the echo path to avoid geological disaster areas; when the urban heat island intensity gradient rises sharply in the afternoon, the heat island gain coefficient increases nonlinearly, promptly reflecting the trend of enhanced echo intensity. This mechanism enables the extrapolation model to have minute-level environmental adaptability, combined with dynamic adaptation of the reflectivity-rainfall intensity relationship in the optimization module (e.g., using Z=300R for severe convective precipitation). 1.4 This can extend the lead time for localized heavy rain warnings to 40 minutes and reduce the false alarm rate for hail identification by 18%. In addition, the warning module uses danger signal thresholds (≥45dBZ strong echo, hail probability >80%) and geofencing technology to achieve targeted delivery of graded warning information. For example, it can automatically trigger emergency response instructions when a strong echo enters a sensitive area of ​​a reservoir.

[0032] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0033] Figure 1 This is a flowchart of an embodiment of the radar echo intelligent extrapolation and evolution identification system for short-term forecasting of the present invention. Detailed Implementation

[0034] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] The following detailed description illustrates the specific implementation method:

[0036] Example:

[0037] As attached Figure 1 As shown, a radar echo intelligent extrapolation and evolution identification system for short-term forecasting includes:

[0038] The data acquisition module is used to acquire high spatiotemporal resolution echo data through networked weather radars (such as C-band, S-band, and X-band radars). The echo data includes information such as reflectivity factor, radial velocity, and echo morphology. The module updates the echo data in real time and performs preprocessing on the acquired echo data, including noise reduction, calibration, and standardization, to eliminate the influence of non-meteorological echoes (such as ground clutter and electromagnetic interference). Then, the radar mosaic method is used to fuse and stitch together the data from multiple radars to generate an echo image covering the entire area.

[0039] The feature extraction module utilizes a deep learning model to extract spatiotemporal evolution features from echo images, incorporating recurrent dynamic convolutional layers and an attention mechanism into the model. These spatiotemporal evolution features include echo movement speed, morphological change trends, and intensity gradients. The introduction of recurrent dynamic convolutional layers and an attention mechanism (such as CBAM) enhances the sensitivity to changes in echo movement direction, speed, and intensity. For example, the recurrent dynamic subnetwork generates probability vectors as convolutional kernel weights, adaptively adjusting the extrapolation strategy. The deep learning model used is either SwinAt-UNet or a recurrent dynamic convolutional neural network.

[0040] The terrain dynamics correction module is used to fuse digital elevation model data with slope change rate (S) in real time to generate terrain weight factor (W). 地形 Here, the slope change rate refers to the change in slope over time. For example, a landslide can cause a significant change in local topographic elevation within a short period. The topographic weighting factor characterizes the hindering effect of topographic undulation on the echo path. The topographic weighting factor is dynamically calculated through the slope-hindering coefficient mapping relationship, specifically:

[0041] When S≤0.5m / (min·km), W 地形 =1.0 × S;

[0042] When 0.5 m / (min·km) < S ≤ 1.5 m / (min·km), W 地形 = 0.5 + 0.5×(S - 0.5);

[0043] When S > 1.5 m / (min·km), W 地形 = 1.0 + 0.1×(S - 1.5).

[0044] Moreover, the terrain weight factor is multiplied by the elevation gradient in the digital elevation model to generate a terrain blocking effect field for correcting the offset of the echo movement path.

[0045] The urban heat island correction module is used to integrate building density data and surface temperature gradient to generate a heat island weight factor (W 热岛 ); the heat island weight factor quantifies the enhancement of the urban heat island effect on the echo intensity. The calculation formula of the heat island weight factor (W 热岛 ) is as follows:

[0046]

[0047] where D is the building density (the value range is 0 - 1, for example, 80% building density corresponds to 0.8), is the heat island intensity gradient (quantifying the temperature difference between the city and the suburbs, unit: °C / km). The building density is obtained from the proportion of the building projection area in the GIS layer. The larger W 热岛 , the more significant the enhancement of the urban heat island effect on the echo intensity. [[ID=二十八]] [[ID=二十九]]

[0048] [[ID=三十]]The prediction module is used to analyze the spatio-temporal evolution characteristics by using the deep learning method to generate the initial radar echo extrapolation results every 6 minutes in the future 0 - 3 hours, and then dynamically correct the initial radar echo extrapolation results based on the joint influence function of the terrain weight factor and the heat island weight factor to generate the final radar echo extrapolation results every 6 minutes in the future 0 - 3 hours, and then analyze the echo morphology and movement trajectory according to the final radar echo extrapolation results to identify potential severe convective weather and predict its influence range and duration; meanwhile, meteorological physical constraints (mass conservation and energy conservation equations) are embedded in the prediction module to avoid generating unreasonable echo evolution paths. [[ID=三十一]] [[ID=三十二]]

[0049] [[ID=三十三]]The joint influence function is as follows:[[ID=三十四]] [[ID=三十五]]

[0050] [[ID=三十六]] [[ID=三十七]] [[ID=三十八]]

[0051] [[ID=三十九]]where ΔV is the correction amount of the echo movement speed,[[ID=四十]] [[ID=四十一]]is the terrain elevation gradient,[[ID=四十二]] [[ID=四十三]]is the heat island intensity gradient, and α, β are weight coefficients obtained by training historical data. [[ID=四十四]] [[ID=四十五]]

[0052] The basic execution process of the prediction module is as follows: an initial displacement field is generated based on the optical flow method, and then the future 6 frames of echo images are predicted through a spatiotemporal convolutional network. Morphological analysis algorithms are used to identify vortex structures and bow-shaped echo features in the future 6 frames of echo images, and the probability of severe convective weather is calculated by combining the movement trajectory. The life cycle stage of the identified thunderstorm cells is labeled, and the boundary of the influence range is adjusted according to the environmental wind field.

[0053] The spatiotemporal evolution features extracted in the feature extraction module (such as echo movement speed, morphological change trends, and intensity gradients) are encoded into high-dimensional feature vectors or probability distributions, which serve as input parameters for deep learning methods (such as optical flow methods and deep learning networks) in the prediction module. For example, multi-scale feature maps output by SwinAt-UNet (such as local echo enhancement and large-scale movement trends) are directly input into the prediction module, or adaptive convolutional kernel weights generated by a recurrent dynamic convolutional network are used to dynamically adjust the extrapolation direction (e.g., predicting the echo to move southward rather than mechanically translating).

[0054] Traditional optical flow methods rely on pixel-level correlations between echoes from adjacent frames (a physical assumption). However, deep learning extrapolation incorporates high-dimensional spatiotemporal evolution features extracted by the feature extraction module (such as the probability of storm cell splitting / merging and the probability of strong echo persistence), thus overcoming the dependence of traditional methods on linear assumptions. For example, when a deep learning model identifies "counterclockwise rotation" (a mesocyclone feature) in the echo from the feature extraction module, the extrapolation algorithm corrects the predicted trajectory to avoid misjudging it as linear movement.

[0055] Meanwhile, in order to further ensure that the feature extraction module can actively adapt to the needs of extrapolation tasks, the feature extraction model and the prediction module are usually trained end-to-end. For example, during the training process, the loss function of the prediction module (such as mean square error) will be backpropagated to the feature extraction module, forcing the feature extraction network to prioritize features that have a greater impact on the prediction results (such as the convergence line morphology of the echo front).

[0056] The dynamic data module updates the terrain slope change rate and urban heat island intensity gradient every 5 minutes, and feeds the updated data back to the terrain dynamic correction module and the urban heat island correction module to adjust the extrapolation parameters in real time. Specifically, the terrain slope change rate is retrieved in real time using InSAR surface deformation data, and the urban heat island intensity gradient is calculated by fusing meteorological satellite infrared channel data with urban building GIS layers.

[0057] The steps for updating the rate of change of terrain slope are as follows:

[0058] a. Receive InSAR surface deformation data every 5 minutes and extract the temporal change (ΔH) of topographic elevation;

[0059] b. Calculate the slope change rate (S) based on the temporal change of topographic elevation (ΔH), time interval (Δt), and horizontal resolution.

[0060] The calculation method is as follows:

[0061] Assume that the elevations of two adjacent points at time t are respectively and The elevations of these two points at time t+1 become respectively and Then the slopes at times t and t+1 are respectively Where ΔL is the horizontal resolution. From this, the slope change rate can be calculated as follows:

[0062]

[0063] c. Input the slope change rate into the terrain dynamics correction module to adjust the calculation parameters of the terrain weight factor in real time, so that the terrain weight factor is positively correlated with the terrain dynamic deformation rate.

[0064] For example, if a mudslide is triggered by heavy rainfall in a mountainous area, and InSAR detects a 0.5m drop in elevation within 5 minutes (horizontal resolution 0.1km), then the slope change rate S = 0.5 / (5 × 0.1) = 1.0m / (min·km), and the terrain weighting factor (W) is... 地形 The slope was increased from 0.6 (initial slope change rate was 0.7) to 0.75, and the system automatically corrected the echo path to shift eastward by ΔV = 800m to avoid geological disaster areas.

[0065] The update steps for the heat island intensity gradient are as follows:

[0066] a. Acquire infrared channel data from meteorological satellites every 5 minutes to retrieve the surface temperature field (T(x,y));

[0067] b. Calculate the heat island intensity gradient by combining the building density distribution (D(x,y)) in the urban building GIS layer;

[0068] The formula for the intensity gradient of the heat island is as follows:

[0069]

[0070] c. Heat island intensity gradient Input the urban heat island correction module to dynamically adjust the heat island weight factor (W). 热岛 The heat island gain coefficient (A) increases non-linearly with increasing heat island gain coefficient, as shown in the following formula:

[0071] A = k·(W) 热岛 -2.0), (W 热岛 >2.0)

[0072] Where k is a constant (in this embodiment, k = 1.5, which needs to be determined based on historical data training), and only when the heat island intensity is sufficiently large (W 热岛 When the gain is greater than 2.0, the system applies gain correction to avoid over-responding to weak heat island effects, and the maximum gain does not exceed 6 dBZ.

[0073] For example, at midday in summer, satellite inversion shows that the temperature gradient in the city center reaches 4℃ / km, and the building density is 0.8 (80%). The heat island gain coefficient was adjusted to 1.5×(3.2-2.0)=1.8dBZ, and the system predicted echo intensity increased from 40dBZ to 41.8dBZ, triggering the short-term heavy precipitation warning optimization module. This module is used to optimize precipitation estimation by utilizing the reflectivity-precipitation intensity relationship (ZR relationship), adjust algorithm parameters for different precipitation types (strong convection and stratiform clouds), and correct extrapolation bias through feedback from real-time data.

[0074] The early warning module is used to identify danger signals in the echo extrapolation results (danger signals include echo areas ≥45dBZ and thunderstorm cells with a hail probability >80%) and generate early warning information (such as short-term heavy precipitation warnings with an advance lead of up to 60 minutes).

[0075] Some experimental data are as follows:

[0076] Table 1 - Performance Comparison of Weather Forecasting Methods

[0077]

[0078]

[0079] Table 1 shows that in mountainous scenarios, the system dynamically integrates slope change rate and digital elevation data through the terrain dynamics correction module, reducing the echo path offset from 15.2 km to 9.8 km (a reduction of 35.5%), verifying the quantitative correction capability of the terrain weighting factor for the blocking effect. Simultaneously, the heat island correction module, combined with minute-level temperature gradient data, optimized the root mean square error (RMSE) of intensity prediction from 8.5 dBZ to 6.1 dBZ (an improvement of 28.2%), indicating that the heat island gain mechanism effectively captures the local convection enhancement effect. In urban agglomeration scenarios, the dynamic calculation of the heat island intensity gradient extends the hail warning lead time from 28 minutes to 48 minutes (an improvement of 71.4%), and reduces the false alarm rate by 38.7%, highlighting the refined response capability of the urban heat island correction module to sudden strong convection.

[0080] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A radar echo intelligent extrapolation and evolution identification system for short-term forecasting, characterized in that, include: The data acquisition module is used to acquire echo data through networked weather radar and update the echo data in real time. It preprocesses the acquired echo data and then fuses and stitches the data from multiple radars using the radar mosaic method to generate an echo image covering the entire area. The feature extraction module is used to extract spatiotemporal evolution features in echo images using a deep learning model, and introduces recurrent dynamic convolutional layers and attention mechanisms into the deep learning model. The terrain dynamics correction module is used to fuse digital elevation model data and slope change rate in real time to generate terrain weight factors; The terrain weighting factor characterizes the obstruction effect of terrain undulation on the echo movement path. The urban heat island correction module integrates building density data with surface temperature gradient to generate a heat island weighting factor; the heat island weighting factor quantifies the enhancement effect of the urban heat island effect on echo intensity. The prediction module uses deep learning to analyze the spatiotemporal evolution characteristics, generate initial radar echo extrapolation results for the next 0-3 hours in 6-minute increments, and then dynamically corrects the initial radar echo extrapolation results based on the joint influence function of terrain weight factor and heat island weight factor, generating final radar echo extrapolation results for the next 0-3 hours in 6-minute increments. Finally, it analyzes the echo morphology and movement trajectory based on the final radar echo extrapolation results, identifies potential severe convective weather, and predicts its impact range and duration. The dynamic data module updates the terrain slope change rate and urban heat island intensity gradient every 5 minutes, and feeds the updated data back to the terrain dynamic correction module and the urban heat island correction module respectively, adjusting the extrapolation parameters in real time. The optimization module is used to optimize precipitation estimation by utilizing the reflectance-rainfall intensity relationship, adjust algorithm parameters for different precipitation types, and correct extrapolation biases by using real-time data feedback. The early warning module is used to identify danger signals in the echo extrapolation results and generate early warning information.

2. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 1, characterized in that, Meteorological physical constraints are embedded in the prediction module, including mass conservation and energy conservation equations.

3. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 2, characterized in that, Precipitation types include severe convection and stratiform clouds.

4. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 3, characterized in that, Danger signals include echo areas ≥45dBZ and thunderstorm cells with a hail probability >80%.

5. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 4, characterized in that, The spatiotemporal evolution characteristics include echo movement speed, morphological change trend, and intensity gradient.

6. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 5, characterized in that, The deep learning model is either SwinAt-UNet or a recurrent dynamic convolutional neural network.

7. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 6, characterized in that, In the dynamic data module, the rate of change of terrain slope is inverted in real time through InSAR surface deformation data, and the heat island intensity gradient is calculated by fusing meteorological satellite infrared channel data with urban building GIS layers.

8. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 7, characterized in that, The prediction module performs the following operations: An initial displacement field is generated based on optical flow, and then a spatiotemporal convolutional network is used to predict the next 6 frames of echo images. Morphological analysis algorithms are used to identify vortex structures and bow-shaped echo features in the next 6 frames of echo images, and the probability of severe convective weather is calculated by combining the movement trajectory. The life cycle stage of the identified thunderstorm cells is labeled, and the boundary of the influence range is predicted by adjusting the environmental wind field.

9. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 8, characterized in that, In the dynamic data module, the update steps for the terrain slope change rate are as follows: a. Receive InSAR surface deformation data every 5 minutes and extract the temporal changes in terrain elevation; b. Calculate the slope change rate based on the time series change and the time interval; multiply the time interval by the horizontal resolution, and use the product as the numerator and the time series change as the denominator to obtain the slope change rate; c. Input the slope change rate into the terrain dynamics correction module to adjust the calculation parameters of the terrain weight factor in real time.

10. The intelligent extrapolation and evolution identification system for radar echoes for short-term forecasting according to claim 9, characterized in that, In the dynamic data module, the update steps for the heat island intensity gradient are as follows: a. Acquire infrared channel data from meteorological satellites every 5 minutes to retrieve the Earth's surface temperature field; b. Calculate the heat island intensity gradient by combining the building density distribution in the urban building GIS layer; c. Input the heat island intensity gradient into the urban heat island correction module to dynamically adjust the gain coefficient of the heat island weight factor.

Images