A microwave radiometer method for detecting supercooled clouds
By combining a multi-channel K-band radiometer and a four-polarized W-band radiometer with a BP neural network and image processing algorithms, the problem of phase identification between supercooled clouds and ice crystal clouds in ground detection was solved, achieving accurate supercooled cloud detection and improving aviation safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING GLARUN DEFENSE SYST CO LTD
- Filing Date
- 2026-04-16
- Publication Date
- 2026-07-17
AI Technical Summary
Existing ground-based detection technologies are unable to accurately identify the mixed phase of supercooled clouds and ice crystal clouds, leading to misjudgment of the phase and affecting aviation safety.
Brightness temperature data were acquired using a multi-channel K-band radiometer, and temperature and humidity were retrieved by combining the data with a BP neural network. DoLP and AoP images were generated using a quad-polarized W-band radiometer, and supercooled clouds and ice crystal clouds were distinguished by image processing algorithms. The results were then determined by combining the temperature and humidity data.
It enables accurate identification of supercooled clouds and ice crystal clouds, improves the accuracy of ground detection, and ensures aviation safety.
Smart Images
Figure CN122017850B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of microwave radiometer detection of supercooled clouds, specifically involving multi-channel radar detection in different bands, neural network inversion, application of temperature and liquid water profiles, and DoLP and AoP image processing technology. Background Technology
[0002] All matter radiates electromagnetic waves at a certain temperature. Microwave radiometers can capture the electromagnetic wave energy radiated, scattered, and reflected by objects within the antenna's field of view, converting it into blackbody temperature. A microwave radiometer mainly consists of an antenna, a receiver, and a recording and display device. It can probe deep below the surface of objects, acquiring information that is difficult to obtain with infrared and visible light.
[0003] Microwave radiometers have a certain ability to penetrate clouds, are less affected by meteorological factors such as fog, rain, and snow, and can work around the clock. They have a wide range of applications in various fields such as atmosphere, ocean, vegetation, and soil.
[0004] Clouds containing supercooled water are called supercooled clouds. When an aircraft passes through supercooled clouds, the supercooled water will freeze upon contact with the fuselage, forming an ice layer. This can damage aerodynamic performance, cause flight accidents, and threaten aviation safety.
[0005] Supercooled clouds lack condensation nuclei; their water content is in the form of small spherical liquid water condensates that radiate in an isotropic mode without polarization. In contrast, water in ice crystal clouds condenses into flat, dendritic crystals with distinct polarization.
[0006] Detecting the distribution of supercooled clouds can ensure flight safety. Existing technologies include three categories: meteorological satellites equipped with microwave radiometers retrieve cloud parameters, which can identify supercooled clouds over a wide area, but the resolution accuracy for low and medium cloud layers is not high; sampling aircraft equipped with microwave radiometers directly obtain microphysical parameters such as cloud particle phase and supercooled water content by penetrating clouds, but they are costly, have limited airspace, and are at risk of icing, making them difficult to operate routinely; ground-based detection uses cloud radar, microwave radiometers, and meteorological stations in a coordinated network, which has the advantage of low cost and all-weather, all-time operation, and has become the current mainstream monitoring method.
[0007] Ground-based detection relies on microwave inversion algorithms, which lack sufficient sensitivity to the mixed phases of supercooled clouds and ice crystal clouds, easily leading to misjudgments of phase. Accurately identifying water-ice mixed phases is a crucial problem that needs to be solved to improve ground-based detection results and build an aviation early warning system. Summary of the Invention
[0008] To address the technical challenge of ground-based supercooled cloud detection failing to effectively distinguish between supercooled water and ice crystals in mixed phases, a microwave radiometer-based supercooled cloud detection method was developed, offering advantages such as all-weather operation and low power consumption.
[0009] Step 1: Obtain brightness temperature using a multi-channel K-band radiometer, and use physical models and AI to retrieve temperature and humidity data.
[0010] First, the atmosphere of any range is divided into several altitude layers, atmospheric sounding data is obtained, and liquid water content profiles and temperature profiles for each altitude layer are plotted to train a neural network as a reference for judging supercooled clouds.
[0011] Next, the MPM atmospheric radiation transfer model was used to calculate the simulated temperature values of the multi-channel K-band radiometer.
[0012] Then, a BP neural network was used as the inversion algorithm to establish a functional relationship between the brightness temperature value of the multi-channel K-band radiometer and the liquid water content and temperature.
[0013] Finally, based on the actual sky brightness temperature measured by the K-band microwave radiometer, the temperature profiles and liquid water content profiles of each altitude layer of the atmosphere in this range were obtained.
[0014] Step 2: Use a quad-polarized W-band radiometer to acquire different polarization brightness temperatures, generate DoLP and AoP images, and distinguish between supercooled clouds and ice crystal clouds.
[0015] First, a DoLP image is generated by simulation, which is a four-polarization image of a mixture of supercooled clouds and ice crystal clouds, including 0° polarization, 90° polarization, 45° polarization, and 135° polarization. Flat cylinders are identified as ice crystal clouds, and spheres are identified as supercooled clouds.
[0016] Next, the sky was scanned using a four-polarization W-band radiometer to obtain different polarization brightness temperatures, and the linear polarizability DoLP and polarization angle AoP were calculated.
[0017] Furthermore, using the formula Calculate the first component of the Stokes vector Second component Third component Fourth component ,in Indicates the brightness temperature value of horizontal polarization. This represents the brightness temperature value for vertical polarization. This represents the brightness temperature value for 45° polarization. This represents the brightness temperature value for 135° polarization. This represents the brightness temperature value of left-handed circular polarization. The brightness temperature value of right-hand circular polarization is expressed by the formula. and Calculate the linear polarizability DoLP and the polarization angle AoP.
[0018] Then, a DoLP local standard deviation image is generated, and the DoLP image is segmented using the Otsu thresholding algorithm. Cylinders and spheres are used to distinguish between supercooled clouds and ice crystal clouds.
[0019] Furthermore, let t represent the grayscale threshold in the DoLP image, where 0 ≤ t < L. This represents the probability of each gray level in the image. and This represents the two weights used to classify the data by the threshold t. and This represents the mean of the two parts divided by the threshold t. Let represent the variance between classes, and let , , , Iterate through the threshold t to determine the maximum inter-class variance. The DoLP image is segmented by setting pixels smaller than a threshold to 0 and pixels larger than a threshold to 1.
[0020] Finally, an AoP local standard deviation image was generated, and an automatic thresholding algorithm was used to segment the AoP image to verify supercooled clouds and ice crystal clouds.
[0021] Furthermore, a neighborhood window is set, with N×N representing the neighborhood size, and the value of each pixel in the AoP image is calculated. neighborhood mean With local standard deviation ,in Represents pixels The corresponding local standard deviation.
[0022] Step 3: Based on the combined temperature and humidity data, DoLP, and AoP, determine whether it is a supercooled cloud.
[0023] First, judge based on the temperature profile. If the brightness temperature of the region is within the preset range, then judge based on the liquid water content profile. If the liquid water content of the region is within the preset range, then judge based on the DoLP image and AoP image. If the segmented part of the region approaches zero, then it is determined to be a supercooled cloud. Attached Figure Description
[0024] Appendix Figure 1 It is a profile of liquid water content.
[0025] Appendix Figure 2 It is a temperature profile.
[0026] Appendix Figure 3 It is a DoLP image.
[0027] Appendix Figure 4 It is the segmented DoLP image. Detailed Implementation
[0028] The technical solution of the present invention will be described in detail with reference to the accompanying drawings.
[0029] The atmosphere from 0 to 10 km was divided into 33 altitude layers: 0, 15, 30, 60, 90, 120, 150, 200, 250, 300, 400, 500, 600, 700, 800, 900, 1000, 1200, 1400, 1600, 1800, 2000, 2500, 3000, 3500, 4000, 4500, 5000, 6000, 7000, 8000, 9000, and 10000 m. Atmospheric sounding data was acquired, and liquid water content profiles and temperature profiles from 0 to 10 km were plotted according to these 33 altitude layers. Figure 1 and Figure 2 The image shown is used to train the neural network as a reference for judging supercooled clouds and does not represent the actual measurement results.
[0030] The atmospheric data uses the fifth-generation global atmospheric reanalysis dataset developed by the European Centre for Medium-Range Weather Forecasts, which provides global meteorological data from 1940 to the present, with a horizontal resolution of approximately 31 kilometers and a temporal resolution of 1 hour, including hundreds of meteorological variables of the surface, near-surface, and upper atmosphere.
[0031] Using the MPM atmospheric radiative transfer model, based on molecular spectroscopy principles and experimental test data, the calculation equation for the microwave absorption coefficient was fitted, and the simulated temperature value of the 10-channel K-band radiometer was calculated.
[0032] A backpropagation neural network was used as the inversion algorithm to establish a functional relationship between the brightness temperature of a 10-channel K-band radiometer and the liquid water content and temperature. Input vector Output vector ,in to This represents the brightness temperature values of 10 channels. to This indicates the liquid water content at 33 altitude levels. to This indicates the temperature at 33 altitude levels.
[0033] Based on the actual sky brightness temperature measured by the K-band microwave radiometer, temperature profiles and liquid water content profiles for 33 altitude layers from 0 to 10 km were obtained.
[0034] Simulations generated DoLP four-polarization images of supercooled clouds and ice crystal clouds, including 0°, 90°, 45°, and 135° polarizations, representing mixtures of supercooled and ice crystal clouds, such as... Figure 3As shown, ice crystal clouds are formed by the freezing of supercooled water or the sublimation of water vapor on ice nuclei. They are not limited by spherical size and tend to be flat, represented by a total of 6 flat cylinders on the left and in the middle. Supercooled clouds are clouds composed of supercooled water droplets that remain liquid at temperatures below 0°C. They radiate in an isotropic manner and are represented by 3 spheres on the right.
[0035] Sky imaging was performed using a four-polarization W-band radiometer to obtain different polarization brightness temperatures, and the linear polarizability DoLP and polarization angle AoP were calculated.
[0036] Use the formula Calculate the first component of the Stokes vector Second component Third component Fourth component ,in Indicates the brightness temperature value of horizontal polarization. This represents the brightness temperature value for vertical polarization. This represents the brightness temperature value for 45° polarization. This represents the brightness temperature value for 135° polarization. This represents the brightness temperature value of left-handed circular polarization. The brightness temperature value of right-hand circular polarization is expressed by the formula. and Calculate the linear polarizability DoLP and the polarization angle AoP.
[0037] A DoLP local standard deviation image is generated, and the DoLP image is segmented using the Otsu thresholding algorithm. Let t represent the gray-level threshold in the DoLP image, where 0 ≤ t < L, and L is the upper limit. This represents the probability of each gray level in the image. and This represents the two weights used to classify the data by the threshold t. and This represents the mean of the two parts divided by the threshold t. Let represent the variance between classes, and let , , , Iterate through the threshold t to determine the maximum inter-class variance. Pixels smaller than the threshold are set to 0, and pixels larger than the threshold are set to 1. Figure 4 As shown, the left cylinder represents an ice crystal cloud, and the right sphere represents a supercooled cloud.
[0038] Generate a local standard deviation image of the Area of Effect (AoP). Use an automatic thresholding algorithm to segment the AoP image. Set a neighborhood window, with N×N representing the neighborhood size. If N=3, the neighborhood window is 3×3. Calculate the standard deviation of each pixel in the AoP image. neighborhood mean With local standard deviation ,in Represents pixels The corresponding local standard deviation.
[0039] First, determine the region's brightness temperature based on the temperature profile. If the region's brightness temperature is between -40℃ and 0℃, then determine the region's liquid water content based on the liquid water content profile. If the region's liquid water content is between 0.1 g / m³ and 0.42 g / m³, then determine the region's segmentation based on the DoLP and AoP images. If the segmented portion of the region approaches zero, then it is determined to be a supercooled cloud.
[0040] The above are embodiments of the present invention and do not limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention are included within the protection scope of the present invention.
Claims
1. A method for detecting supercooled clouds using a microwave radiometer, characterized in that, include: Step 1: Obtain brightness temperature using a multi-channel K-band radiometer, and use physical models and AI to retrieve temperature and humidity data, including: The atmosphere of any range is divided into several altitude layers, atmospheric sounding data is obtained, and liquid water content profiles and temperature profiles of each altitude layer are plotted for training a neural network as a reference for judging supercooled clouds. The MPM atmospheric radiative transfer model was used to calculate the simulated temperature values of a multi-channel K-band radiometer. A BP neural network was used as the inversion algorithm to establish a functional relationship between the brightness temperature of a multi-channel K-band radiometer and the liquid water content and temperature. Based on the actual sky brightness temperature measured by the K-band microwave radiometer, the temperature profiles and liquid water content profiles of each altitude layer of the atmosphere in this range are obtained. Step 2: Use a quad-polarized W-band radiometer to acquire different polarization brightness temperatures, generate DoLP and AoP images, and distinguish between supercooled clouds and ice crystal clouds, including: The simulation generates DoLP images, which are four-polarization images of a mixture of supercooled clouds and ice crystal clouds, including 0° polarization, 90° polarization, 45° polarization, and 135° polarization. Flat cylinders are identified as ice crystal clouds, and spheres are identified as supercooled clouds. Sky imaging was scanned using a four-polarization W-band radiometer to obtain different polarization brightness temperatures, and the linear polarizability DoLP and polarization angle AoP were calculated. A DoLP local standard deviation image is generated, and the DoLP image is segmented using the Otsu thresholding algorithm. Cylinders and spheres are used to distinguish between supercooled clouds and ice crystal clouds. A local standard deviation image of AoP is generated, and the AoP image is segmented using an automatic thresholding algorithm to verify supercooled clouds and ice crystal clouds. Step 3: Based on the relationship between temperature and humidity data, DoLP, and AoP, determine whether it is a supercooled cloud, including: First, judge based on the temperature profile. If the brightness temperature of the region is within the preset range, then judge based on the liquid water content profile. If the liquid water content of the region is within the preset range, then judge based on the DoLP image and AoP image. If the segmented part of the region approaches zero, then it is determined to be a supercooled cloud.
2. The microwave radiometer method for detecting supercooled clouds according to claim 1, characterized in that, The calculation of linear polarizability DoLP and polarization angle AoP includes: using the formula Calculate the first component of the Stokes vector Second component Third component Fourth component ,in Indicates the brightness temperature value of horizontal polarization. This represents the brightness temperature value for vertical polarization. This represents the brightness temperature value for 45° polarization. This represents the brightness temperature value for 135° polarization. This represents the brightness temperature value of left-handed circular polarization. The brightness temperature value of right-hand circular polarization is expressed by the formula. and Calculate the linear polarizability DoLP and the polarization angle AoP.
3. The microwave radiometer method for detecting supercooled clouds according to claim 2, characterized in that, The Otsu thresholding algorithm is used to segment the DoLP image, which includes: Let t represent the grayscale threshold in the DoLP image, 0 ≤ t < L, where L is the upper limit. This represents the probability of each gray level in the image. and This represents the two weights used to classify the data by the threshold t. and This represents the mean of the two parts divided by the threshold t. Let represent the variance between classes, and let , , , Iterate through the threshold t to determine the maximum inter-class variance. The DoLP image is segmented by setting pixels smaller than a threshold to 0 and pixels larger than a threshold to 1.
4. The microwave radiometer method for detecting supercooled clouds according to claim 2, characterized in that, The automatic thresholding algorithm for AoP image segmentation includes: setting a neighborhood window, using N×N to represent the neighborhood size, and calculating the value of each pixel in the AoP image. neighborhood mean With local standard deviation ,in Represents pixels The corresponding local standard deviation.