Thunderstorm early warning method based on combined observation of microwave radiometer and millimeter wave cloud radar

Abstract

The invention discloses a thunderstorm early warning method based on combined observation of a microwave radiometer and a millimeter wave cloud radar. The thunderstorm early warning method comprises the following steps: step 1, synchronously acquiring and preprocessing data of the microwave radiometer and the millimeter wave cloud radar; 2, calculating a thunderstorm index in real time; 3, training a thunderstorm early warning model; and step 4, outputting an early warning result. Compared with a traditional single detection means, the thunderstorm early warning method disclosed by the invention has the remarkable technical advantages that the early warning accuracy is remarkably improved through multi-parameter fusion, all-weather continuous automatic monitoring is realized, the operation and maintenance cost is remarkably reduced, and the complex terrain adaptability is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological monitoring and disaster warning, and in particular relates to a short-term approaching thunderstorm warning method based on combined observation data of a microwave radiometer and a millimeter-wave cloud radar in this field. Background Art

[0002] Existing thunderstorm warning methods mainly include weather radar monitoring, sounding observations, satellite remote sensing, and ground meteorological station data analysis, but these methods still have certain limitations in real-time, continuity, or detection accuracy.

[0003] Weather radar warning technology determines thunderstorm development by analyzing Doppler radar echo intensity, vertically integrated liquid content (VIL), and storm tracking information. However, weather radar can only detect the echo signals of precipitation particles and cannot directly obtain the atmospheric temperature and humidity structure. This results in poor recognition of early stages of thunderstorm development (such as cumulus clouds). There are also blind spots in the lower atmosphere, resulting in insufficient accuracy in measuring cloud base height. Furthermore, the limited deployment of equipment makes coverage difficult in some areas.

[0004] Sounding observations combined with numerical forecasting methods use sounding data to calculate convective parameters (such as the CAPE and LI index) for thunderstorm potential forecasting. While this method can reflect atmospheric instability, the low temporal resolution of sounding observations (typically twice daily) makes them inadequate for short-term, near-term warnings. Furthermore, their limited spatial coverage makes them applicable only to areas with sounding stations, making it difficult to monitor local convective development in real time.

[0005] Satellite remote sensing technology uses infrared and water vapor channels from geostationary meteorological satellites to identify thunderstorm clouds. However, satellite observations are susceptible to interference from high-level clouds, and their ability to detect the height of low-level cloud bases is insufficient. Furthermore, their temporal resolution (5 to 15 minutes) is insufficient to capture the rapid evolution of thunderstorms.

[0006] The microwave radiometer single-device warning method uses microwave radiometers to retrieve atmospheric temperature and humidity profiles and calculate the Modified Jefferson Index (MJI) for thunderstorm warning. While this method provides continuous atmospheric stratification data, it cannot directly observe cloud physical properties (such as cloud base height and cloud phase). Furthermore, the microwave radiometer's accuracy may decrease during low-altitude temperature inversions or haze conditions.

[0007] In summary, methods such as Doppler weather radar, sounding observations, and satellite remote sensing all have limitations to varying degrees. While weather radar can effectively monitor precipitation echoes, it has limited ability to identify early stages of thunderstorm development (such as the cumulus phase). Furthermore, due to the influence of the Earth's curvature, there are blind spots in the lower atmosphere, resulting in inaccurate cloud base height measurements. While sounding data can provide information on atmospheric temperature and humidity structure, their temporal resolution (typically only twice daily) is low, making them inadequate for short-term, near-miss warnings. Furthermore, their limited spatial coverage makes it difficult to monitor local convective development in real time. Satellite remote sensing, while capable of observing over large areas, is susceptible to interference from high-level clouds, resulting in inaccurate detection of low-level cloud base heights. Furthermore, its temporal resolution (5–15 minutes) still makes it difficult to capture the rapid evolution of thunderstorms. Furthermore, while microwave radiometers can continuously invert atmospheric temperature and humidity profiles and calculate thunderstorm indices, they cannot directly observe cloud physical properties (such as cloud base height and cloud phase) when used alone, resulting in a high false alarm rate. Furthermore, inversion accuracy may decrease during low-altitude temperature inversions or haze conditions. Therefore, existing technologies are difficult to simultaneously meet the needs of high timeliness, high accuracy and continuous unmanned thunderstorm warning. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a thunderstorm warning method based on the joint observation of microwave radiometer and millimeter wave cloud radar, which integrates multi-source data, has strong real-time performance, is suitable for unmanned operation, and can improve the accuracy and timeliness of warning.

[0009] The present invention adopts the following technical solutions:

[0010] A thunderstorm early warning method based on combined observation of microwave radiometer and millimeter wave cloud radar is improved in that it includes the following steps:

[0011] Step 1: Synchronous acquisition and preprocessing of microwave radiometer and millimeter wave cloud radar data:

[0012] The data obtained by the microwave radiometer and the millimeter-wave cloud radar are time-synchronized to obtain the parameters output by the microwave radiometer and the millimeter-wave cloud radar in the same period;

[0013] The parameters output by the microwave radiometer include temperature profile, humidity profile, surface temperature, humidity and pressure data, path liquid water content, integrated water vapor content, K index and Shabbat index;

[0014] The parameter output by the millimeter-wave cloud radar is the cloud base height;

[0015] Step 2: Real-time calculation of thunderstorm index;

[0016] Step 3: Train the thunderstorm warning model:

[0017] Extract the time when thunderstorms occur;

[0018] The thunderstorm index, K index, Sha's index, path liquid water content, integrated water vapor content, and cloud base height data within 2 hours before the thunderstorm occurred were selected to establish the inversion model data set;

[0019] The nonlinear relationship between thunderstorm index, K index, Sha's index, path liquid water content, integrated water vapor content, cloud base height and thunderstorm is simulated using BP neural network.

[0020] Step 4: Output warning results:

[0021] The output of the neural network model is obtained and combined with the thunderstorm index. When the model output indicates that there is a thunderstorm and the thunderstorm index is greater than 28, a thunderstorm warning information is output.

[0022] Furthermore, in step 2, the thunderstorm index T mj It is represented by the following formula:

[0023]

[0024] In the above formula, θ w900 is the wet bulb potential temperature at 900hPa, T 500 is the temperature at 500hPa, The humidity condition is 700hPa, T d is the dew point temperature.

[0025] Furthermore, in step 3,

[0026] The network adopts a three-layer fully connected BP neural network structure, where the input vector X is determined by the input parameter L, and the output vector Y represents the value to be inverted.

[0027] Furthermore, in step 3, the neurons in both the hidden layer and the output layer adopt the logarithmic tangent sigmoid function:

[0028]

[0029] The output of any single neuron is:

[0030]

[0031] In the above formula, f is the output value of the neuron, w i is the input weight of the neuron, p i is the input value of the neuron, b i is the deviation of the neuron.

[0032] Furthermore, in step 3, the weight matrix and bias matrix of the BP neural network are determined by the error back propagation algorithm during the training process, which is to find a set of weight matrices W and bias matrices b so that the network error E(W,b) is minimized:

[0033]

[0034] In the above formula, K is the number of training samples, is the predicted value of the k-th sample thunderstorm warning output, y k is the true value of the k-th sample thunderstorm warning output;

[0035] Using the gradient descent method, the weights and biases of each layer of neurons are repeatedly adjusted through training to obtain a new weight matrix after each correction. and the new deviation matrix

[0036]

[0037] In the above formula, η is the learning rate parameter, represents the weight matrix between the neurons in the j-1th layer and the jth layer, Represents the bias matrix between the neurons in the j-1th layer and the jth layer.

[0038] The beneficial effects of the present invention are:

[0039] The thunderstorm warning method disclosed in this invention has significant technical advantages over traditional single detection methods, which are specifically reflected in the following four aspects:

[0040] First, multi-parameter fusion significantly improves warning accuracy. A physically clear thunderstorm warning model is constructed using the thunderstorm index, K index (KI), Scharnauer index (SI), path liquid water content, integrated water vapor content (inverted from microwave radiometers), and cloud base height measured by millimeter-wave cloud radar. Compared to relying solely on a single device, this significantly improves warning accuracy.

[0041] Second, it enables continuous, automatic monitoring around the clock. The microwave radiometer uses passive remote sensing to output a set of atmospheric profile data every two minutes, while the millimeter-wave cloud radar uses active detection to update cloud parameters every two minutes. Both are capable of 24-hour unattended operation. Compared to satellite cloud image analysis that relies on manual interpretation and sounding observations with insufficient temporal resolution, the thunderstorm warning method of this invention significantly improves temporal resolution and is particularly suitable for critical locations such as airports and key areas where real-time warnings are required.

[0042] 3. Significantly Reduced Operation and Maintenance Costs. While the construction cost of a single traditional Doppler weather radar station exceeds 10 million yuan, microwave radiometers offer long-term operation. The cloud radar system has lower annual operation and maintenance costs, making it suitable for deployment in areas with sparse meteorological observation sites.

[0043] Fourth, enhanced adaptability to complex terrain. Millimeter-wave cloud radar has a narrower beamwidth than weather radar, effectively reducing the impact of ground obstruction in complex terrain such as mountainous areas. Microwave radiometers, through multi-channel observations, infer atmospheric parameters that are not restricted by terrain. The combination of the two offers excellent applicability in experiments in specialized areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of the thunderstorm early warning method disclosed in the present invention;

[0045] Figure 2 This is a comparison chart of the thunderstorm index calculated by the microwave radiometer and the sounding data during the test. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] In view of the deficiencies of thunderstorm warning methods in the prior art in terms of spatiotemporal resolution, early recognition capability, equipment cost and data fusion, the present invention aims to provide a thunderstorm warning method based on the joint observation of microwave radiometers and millimeter-wave cloud radars, so as to solve the following core problem: how to improve the accuracy and timeliness of thunderstorm warnings while ensuring high spatiotemporal resolution through multi-source data fusion. Specifically, the present invention needs to overcome the problem of insufficient early thunderstorm recognition capability of weather radar, make up for the defect of low temporal resolution of sounding data, and solve the problem of false alarms caused by the lack of cloud physics information when microwave radiometers are used alone. At the same time, it is also necessary to adapt to the needs of unmanned automated observation, reduce equipment operation and maintenance costs, and make it suitable for areas with sparse meteorological stations or special terrain environments (such as mountainous areas, airports, etc.).

[0048] The present invention provides a thunderstorm warning method based on the combined observation of microwave radiometer and millimeter wave cloud radar. By fusing the thunderstorm index (modified Jefferson index) inverted by microwave radiometer and the cloud base height and cloud shape information output in real time by millimeter wave cloud radar, a multi-parameter superposition criterion is constructed to achieve high-precision and high-efficiency thunderstorm warning. The specific tasks include: (1) designing a calculation model for the thunderstorm index, using the atmospheric temperature and humidity profile data continuously observed by the microwave radiometer to quantitatively characterize atmospheric instability; (2) combining the high-resolution cloud base height detection data (400-1000 meters threshold) and cloud shape recognition technology of the millimeter wave cloud radar to eliminate the interference of non-thunderstorm clouds; (3) establishing a multi-source data fusion algorithm, and improving the warning accuracy and reducing the false alarm rate through the coordinated judgment of thunderstorm index, cloud base height and cloud shape information; (4) building an all-weather, automated, low-cost thunderstorm monitoring system suitable for real-time warning needs in various complex environments.

[0049] Example 1. This example discloses a thunderstorm warning method based on the joint observation of microwave radiometer and millimeter wave cloud radar. Parameters closely related to thunderstorm activity are selected as prediction factors. Combined with the neural network algorithm, a neural network model based on microwave radiometer and millimeter wave cloud radar data is established. At the same time, the thunderstorm index (improved Jefferson index) observed in real time by the microwave radiometer is introduced as a dynamic criterion to achieve dual judgment on the forecast results, improve the timeliness and accuracy of thunderstorm warnings, and provide technical support for disaster prevention and mitigation. Figure 1 As shown, the specific steps include:

[0050] Step 1: Synchronous acquisition and preprocessing of microwave radiometer and millimeter wave cloud radar data:

[0051] The data obtained by the microwave radiometer and the millimeter-wave cloud radar are time-synchronized to obtain the parameters output by the microwave radiometer and the millimeter-wave cloud radar in the same period. Currently, the time resolution of the data from the two devices is 2 minutes, and the clock synchronization error is less than 1 second.

[0052] The parameters output by the microwave radiometer include temperature profile, humidity profile, ground temperature, humidity and pressure data, path liquid water content, integrated water vapor content, K index and Sha'a index; these meteorological parameters are highly correlated with the occurrence of thunderstorms.

[0053] The parameter output by the millimeter-wave cloud radar is the cloud base height;

[0054] Step 2: Real-time calculation of the thunderstorm index using the microwave radiometer;

[0055] The thunderstorm index is an indicator that assesses whether atmospheric conditions are favorable for thunderstorms. It uses the temperature profile, relative humidity profile, and ground temperature, humidity, and pressure data output by microwave radiometers to assess whether atmospheric conditions are favorable for thunderstorms. The modified Jefferson index (T) is usually used to represent the thunderstorm index. mj :

[0056]

[0057] In the above formula, θ w900 is the wet bulb potential temperature at 900hPa, T 500 is the temperature at 500hPa, The humidity condition is 700hPa, T d is the dew point temperature.

[0058] Step 3: Train the thunderstorm warning model:

[0059] Using historical microwave radiometer and millimeter wave cloud radar measurement data and lightning locator data, a neural network model is trained to determine the connection weights and thresholds of the network model, and the test set sample data is used to test the forecast effect.

[0060] Select the ground-to-ground lightning data from the lightning locator, including the time, longitude, latitude and other parameters of the lightning occurrence, and extract the time of the thunderstorm occurrence;

[0061] The thunderstorm index, K index (KI), Schaffner index (SI), path liquid water content, integrated water vapor content, and cloud base height data within 2 hours before the thunderstorm occurred were selected to establish the inversion model data set;

[0062] The nonlinear relationship between thunderstorm index, K index (KI), Scharnauer index (SI), path liquid water content, integrated water vapor content, cloud base height and thunderstorm is simulated using BP neural network.

[0063] The network adopts a three-layer fully connected BP neural network structure, wherein the input vector X is determined by the input parameter L. In the model of this embodiment, L=6, and the output vector Y represents the value to be inverted.

[0064] The neurons in the hidden layer and the output layer use the logarithmic tangent S function:

[0065]

[0066] The output of any single neuron is:

[0067]

[0068] In the above formula, f is the output value of the neuron, w i is the input weight of the neuron, p i are the input values ​​of the neuron (i.e., thunderstorm index (MJI), K index (KI), Scharnauer index (SI), path liquid water content (LWP), integrated water vapor content (IWV), and cloud base height (HCB)), b i is the deviation of the neuron.

[0069] The weight matrix and bias matrix of the BP neural network are determined by the error back propagation algorithm during the training process. By continuously adjusting the weights and biases of the network to reduce the deviation between the output vector calculated by the network input vector and the actual training target output vector, a set of weight matrices W and bias matrices b are obtained to minimize the network error E(W,b):

[0070]

[0071] In the above formula, K is the number of training samples, is the predicted value of the k-th sample thunderstorm warning output, y k is the true value of the k-th sample thunderstorm warning output;

[0072] Using the gradient descent method, the weights and biases of each layer of neurons are repeatedly adjusted through training to obtain a new weight matrix after each correction. and the new deviation matrix

[0073]

[0074] In the above formula, η is the learning rate parameter, represents the weight matrix between the neurons in the j-1th layer and the jth layer, Represents the bias matrix between the neurons in the j-1th layer and the jth layer.

[0075] Step 4: Output warning results:

[0076] The thunderstorm index is calculated in real time based on the temperature and humidity profiles output by the microwave radiometer.

[0077] The output results of the neural network model are obtained and combined with the thunderstorm index to achieve a double judgment on the forecast results. When the model output indicates that there is a thunderstorm and the thunderstorm index is greater than 28, a thunderstorm warning information is output.

[0078] The following simulation verification is carried out using observation data from the microwave radiometer and millimeter-wave cloud radar in Qingdao, as follows.

[0079] The corresponding period of the selected data is from January 1, 2023 to December 31, 2024, of which the data from June to July 2024 are used for effect testing. A total of 14 thunderstorm cases occurred in June and July 2024. The measured data of microwave radiometer and millimeter wave cloud radar were selected and substituted into the forecast model for forecast, and judged in combination with the thunderstorm index. The 14 thunderstorm weather cases were used as independent samples for forecast testing. If the warning result indicates that there is a thunderstorm, and a thunderstorm occurs within 2 hours of the warning, it is judged that the thunderstorm is correctly predicted; if no thunderstorm occurs within 2 hours, it is judged to be an empty report; if a thunderstorm is not predicted but a thunderstorm occurs, it is judged to be a missed report. Since the accuracy of the thunderstorm index directly affects the test accuracy in thunderstorm warning, the comparison of the thunderstorm index calculated by the microwave radiometer and the sounding data during the test was analyzed. Figure 2 It can be seen that the thunderstorm index calculated from the microwave radiometer output product is consistent with the thunderstorm index calculated from the sounding data and can be used as a real-time criterion.

[0080] From the test results in the table below, we can see that when using neural network forecasts and thunderstorm index for dual judgment, the hit rate is 78.6%, the missed rate is 21.4%, and the false alarm rate is 15.4%, which significantly improves the accuracy of thunderstorm warnings and provides strong data support for thunderstorm warnings.

[0081]

Claims

1. A thunderstorm early warning method based on combined observation of microwave radiometer and millimeter wave cloud radar, characterized in that: The steps include: Step 1: Synchronous acquisition and preprocessing of microwave radiometer and millimeter wave cloud radar data: The data obtained by the microwave radiometer and the millimeter-wave cloud radar are time-synchronized to obtain the parameters output by the microwave radiometer and the millimeter-wave cloud radar in the same period; The parameters output by the microwave radiometer include temperature profile, humidity profile, surface temperature, humidity and pressure data, path liquid water content, integrated water vapor content, K index and Shabbat index; The parameter output by the millimeter-wave cloud radar is the cloud base height; Step 2: Real-time calculation of thunderstorm index; Step 3: Train the thunderstorm warning model: Extract the time when thunderstorms occur; The thunderstorm index, K index, Sha's index, path liquid water content, integrated water vapor content, and cloud base height data within 2 hours before the thunderstorm occurred were selected to establish the inversion model data set; The nonlinear relationship between thunderstorm index, K index, Sha's index, path liquid water content, integrated water vapor content, cloud base height and thunderstorm is simulated using BP neural network. Step 4: Output warning results: The output of the neural network model is obtained and combined with the thunderstorm index. When the model output indicates that there is a thunderstorm and the thunderstorm index is greater than 28, a thunderstorm warning information is output.

2. The thunderstorm early warning method based on combined observation of microwave radiometer and millimeter wave cloud radar according to claim 1, characterized in that: In step 2, the thunderstorm index T mj It is represented by the following formula: In the above formula, θ w900 is the wet bulb potential temperature at 900hPa, T 500 is the temperature at 500hPa, The humidity condition is 700hPa, T d is the dew point temperature.

3. The thunderstorm early warning method based on combined observation of microwave radiometer and millimeter wave cloud radar according to claim 1, characterized in that: In step 3, The network adopts a three-layer fully connected BP neural network structure, where the input vector X is determined by the input parameter L, and the output vector Y represents the value to be inverted.

4. The thunderstorm early warning method based on combined observation of microwave radiometer and millimeter wave cloud radar according to claim 3 is characterized by: In step 3, the neurons in both the hidden layer and the output layer use the logarithmic tangent sigmoid function: The output of any single neuron is: In the above formula, f is the output value of the neuron, w i is the input weight of the neuron, p i is the input value of the neuron, b i is the deviation of the neuron.

5. The thunderstorm early warning method based on combined observation of microwave radiometer and millimeter wave cloud radar according to claim 4, characterized in that: In step 3, the weight matrix and bias matrix of the BP neural network are determined by the error back propagation algorithm during the training process, which is to find a set of weight matrix W and bias matrix b so that the network error E(W,b) is minimized: In the above formula, K is the number of training samples, is the predicted value of the k-th sample thunderstorm warning output, y k is the true value of the k-th sample thunderstorm warning output; Using the gradient descent method, the weights and biases of each layer of neurons are repeatedly adjusted through training to obtain a new weight matrix after each correction. and the new deviation matrix In the above formula, η is the learning rate parameter, represents the weight matrix between the neurons in the j-1th layer and the jth layer, Represents the bias matrix between the neurons in the j-1th layer and the jth layer.

Images