Satellite and radar fused isolated mature convective cloud cluster dynamic identification system and method

Abstract

The application belongs to the technical field of artificial intelligence, and specifically provides a dynamic identification system and method for isolated mature convective clouds combining satellites and radars, which comprises: a data processing module, which is used for pre-processing wind-weather stationary meteorological satellite data and radar data obtained; a convective cloud sample extraction module, which is used for identifying convective cloud samples; an isolated convective cloud sample identification module, which is used for determining isolated convective cloud samples; an isolated mature convective cloud sample screening module, which is used for distinguishing mature convective clouds from non-mature convective clouds for the isolated convective cloud samples, and screening out isolated mature convective cloud samples; a sample dynamic inspection module, which is used for obtaining final isolated mature convective cloud samples; and an identification model construction module, which is used for combining and verifying satellite features and radar features of the final isolated mature convective cloud samples, and establishing an intelligent identification model; the technical scheme of the application can realize tracking of the whole life cycle of isolated mature convective systems.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a satellite and radar fused isolated mature convection cloud cluster dynamic identification system and method. BACKGROUND

[0002] Severe convective weather mainly includes short-time heavy rain, thunderstorm gale, tornado and hail, etc., and is one of the most serious and most frequently occurring disastrous weather systems in summer. The accurate prediction of severe convective weather has important significance for public safety, disaster reduction and military operations, etc. Convective initiation (CI) is a symbol of the beginning of the activity of severe convective weather and the state of the early stage of severe convective storm, and isolated mature convection (IMC) is a performance of the development of the convection system to the mature stage.

[0003] Severe convective weather has the characteristics of strong burst, small spatial scale and short life cycle, and the identification of isolated mature convection is particularly challenging. At present, the identification of local convective initiation (CI) mainly relies on satellite remote sensing and radar observation, but each has its own limitations: satellite observation has the advantages of large-scale and long-time continuous monitoring, and can identify early convection signals (such as cloud optical thickness, brightness temperature change, etc.) about 2 hours before the convection triggers, but it is difficult to accurately depict the microphysical and dynamic characteristics inside the cloud. Radar observation can provide high-precision cloud internal structure and precipitation information, such as the first appearance of 35 dBZ echo which can be used as a symbol of strong precipitation potential, but radar data is greatly affected by terrain obstruction, and for isolated convective cells, the spatial resolution and scanning period may cause monitoring lag. The existing problems mainly include two aspects: (1) Satellite and radar data are not fully fused, making it difficult to comprehensively utilize the macroscopic monitoring ability of satellite and the fine description ability of radar to accurately identify isolated mature convection. (2) There is uncertainty in the identification and tracking of convective cells, which cannot accurately determine whether the convection remains isolated or further organizes and develops, affecting the accuracy of short-time heavy rain, thunderstorm and other disastrous weather.

[0004] Existing researches set fixed thresholds to identify convective clouds based on radar or satellite data, but fixed thresholds are difficult to adapt to the complexity of convective development under different weather backgrounds. Due to the influence of different weather conditions, terrain, humidity and other factors on the convective characteristics under the same threshold, some convective clouds are not identified (false negatives), and weak convective clouds may be misjudged as strong convective clouds (false positives). In particular, there is a lack of dynamic tracking of the life cycle of convective systems, which makes it difficult to provide sustained short-time prediction, resulting in a short prediction time. The nonlinear evolution of convective systems is ignored, i.e. the convective intensity may change due to environmental conditions (such as wind shear, water vapor transport), rather than simple spatial displacement. SUMMARY

[0005] In order to solve the above problems, the application provides a satellite and radar fused isolated mature convective cloud cluster dynamic identification system and method, which combines new generation Fengyun geostationary meteorological satellite and dual-polarization weather radar data to monitor the whole life cycle from convective initial stage (CI) to mature stage, so as to solve the problem of low accuracy of isolated convective system.

[0006] The satellite and radar fused isolated mature convective cloud cluster dynamic identification system comprises:

[0007] A data processing module is configured to pre-process the acquired Fengyun geostationary meteorological satellite data and radar data.

[0008] A convective cloud cluster sample extraction module is configured to obtain a cloud mask of a convective cloud by using the pre-processed Fengyun geostationary meteorological satellite data, and identify a convective cloud cluster sample by using the cloud mask.

[0009] An isolated convective cloud cluster sample identification module is configured to determine an isolated convective cloud cluster sample based on the convective cloud cluster sample in combination with a cloud cluster growth region.

[0010] An isolated mature convective cloud cluster sample screening module is configured to use eight-connected region analysis to distinguish mature convective cloud from non-mature convective cloud for the isolated convective cloud cluster sample, and screen out an isolated mature convective cloud cluster sample.

[0011] A sample dynamic verification module is configured to verify a convective region of the isolated mature convective cloud cluster sample by using the pre-processed radar data, and obtain a final isolated mature convective cloud cluster sample.

[0012] An identification model construction module is configured to fuse satellite features and radar features of the verified final isolated mature convective cloud cluster sample, and construct an intelligent identification model.

[0013] An identification module is configured to complete dynamic identification of an isolated mature convective cloud cluster based on the intelligent identification model.

[0014] Preferably, the isolated convective cloud cluster sample identification module comprises:

[0015] A cold cloud nucleus identification unit is configured to identify a cold cloud nucleus by using an infrared channel in the pre-processed Fengyun geostationary meteorological satellite data.

[0016] A cloud cluster growth region acquisition unit is configured to check adjacent pixels of the cold cloud nucleus by using a growth search method, judge whether the adjacent pixels meet cloud cluster growth conditions, and obtain a cloud cluster growth region.

[0017] An isolated convective cloud cluster sample acquisition unit is configured to obtain an isolated convective cloud cluster sample based on the cloud cluster growth region.

[0018] Preferably, the sample dynamic verification module comprises:

[0019] A radar echo corresponding cloud mask pixel acquisition unit is configured to acquire the number of pixels of the isolated mature convective cloud sample cloud mask corresponding to the radar echo region by using the isolated mature convective cloud sample cloud mask extracted from the FY-4 satellite data, combining the binary function of the isolated mature convective cloud sample cloud mask and the radar echo region corresponding to the isolated mature convective cloud sample cloud mask.

[0020] A cloud mask total pixel acquisition unit is configured to acquire the total number of pixels of the isolated mature convective cloud sample cloud mask by using the binary function of the isolated mature convective cloud sample cloud mask.

[0021] A matching rate calculation unit is configured to calculate the matching rate of the radar echo region of different intensity and the isolated mature convective cloud sample cloud mask by using the number of pixels of the isolated mature convective cloud sample cloud mask corresponding to the radar echo region and the total number of pixels of the isolated mature convective cloud sample cloud mask.

[0022] A dynamic verification unit is configured to take the isolated mature convective cloud sample satisfying the preset threshold of the matching rate as the final isolated mature convective cloud sample.

[0023] Preferably, the identification model construction module comprises:

[0024] A sample construction unit is configured to perform convective initial identification verification on the final isolated mature convective cloud sample by using the radar data, mark the final isolated mature convective cloud sample passing the verification as a positive sample, and mark the isolated convective cloud sample failing to develop into a mature convective cloud as a negative sample.

[0025] A feature extraction unit is configured to extract feature parameters of the FY-4 satellite data and the radar data.

[0026] A model construction unit is configured to train a gradient boosting decision tree by using the feature parameters and construct an intelligent identification model.

[0027] Preferably, the process of performing convective initial identification verification on the final isolated mature convective cloud sample by using the radar data comprises:

[0028] The FY-4 satellite data and the radar data are time-matched according to a satellite-radar time matching span; wherein the satellite-radar time matching span includes 15 minutes, 30 minutes and 60 minutes.

[0029] When the satellite-radar time matching span is within 15 minutes, the convective cloud is in a development stage, and the observation quantity and the development stage threshold value are used to back-propagate the convective initial identification result of the previous 15 minutes; wherein the observation quantity includes reflectivity factor, differential reflectivity, correlation coefficient and specific differential phase rate.

[0030] When the satellite-radar time matching span is within 15-30 minutes, the convection is in the mature stage, and the observation quantity is used to back-propagate the convection initial identification result in the previous 30 minutes;

[0031] When the satellite-radar time matching span is within 60 minutes, the radar echo feature is used to verify the convection initial identification result.

[0032] The application further provides a dynamic identification method for isolated mature convection clouds by fusing satellites and radars, and application of the system, comprising:

[0033] Pretreating acquired Fengyun geostationary meteorological satellite data and radar data;

[0034] Using the pretreated Fengyun geostationary meteorological satellite data, obtaining a cloud mask of a convection cloud, and using the cloud mask to identify a convection cloud sample;

[0035] Based on the convection cloud sample, combining a cloud cluster growth region to determine an isolated convection cloud sample;

[0036] Using eight-connected region analysis, distinguishing the isolated convection cloud sample into mature convection and non-mature convection, and screening out an isolated mature convection cloud sample;

[0037] Using the pretreated radar data to verify the isolated mature convection cloud sample, and obtaining a final isolated mature convection cloud sample;

[0038] Fusing satellite features and radar features of the final isolated mature convection cloud sample to establish an intelligent identification model;

[0039] Based on the intelligent identification model, completing dynamic identification of the isolated mature convection cloud.

[0040] Preferably, the method for identifying the isolated convection cloud sample comprises:

[0041] Using an infrared channel in the pretreated Fengyun geostationary meteorological satellite data to identify a cold cloud nucleus;

[0042] Using a growth search method to check adjacent pixels of the cold cloud nucleus, judging whether the adjacent pixels satisfy a cloud cluster growth condition, and obtaining a cloud cluster growth region;

[0043] Based on the cloud cluster growth region, obtaining an isolated convection cloud sample.

[0044] Preferably, the method for obtaining the final isolated mature convection cloud sample comprises:

[0045] The cloud mask of the isolated mature convective cloud sample is extracted by using the data of the Fengyun geostationary meteorological satellite, the binary function of the cloud mask of the isolated mature convective cloud sample is combined with the cloud mask of the isolated mature convective cloud sample, and the pixel number of the cloud mask of the isolated mature convective cloud sample corresponding to the radar echo area is obtained;

[0046] The total pixel number of the cloud mask of the isolated mature convective cloud sample is obtained by using the binary function of the cloud mask of the isolated mature convective cloud sample.

[0047] The matching rate of the radar echo area of different intensity and the cloud mask of the isolated mature convective cloud sample is calculated by using the pixel number of the cloud mask of the isolated mature convective cloud sample corresponding to the radar echo area and the total pixel number of the cloud mask of the isolated mature convective cloud sample.

[0048] The isolated mature convective cloud sample satisfying the preset threshold value of the matching rate is taken as the final isolated mature convective cloud sample.

[0049] Compared with the prior art, the beneficial effects of the present application are:

[0050] The present application improves the identification and prediction accuracy of the isolated mature convective system through data fusion, dynamic tracking and intelligent identification: (1) combined with the wide range of cloud top information provided by the stationary meteorological satellite (FY4A, etc.) and the high temporal and spatial resolution reflectivity of the ground-based Doppler radar, the joint identification of the isolated mature convective system is realized. (2) The adaptive threshold and multi-parameter fusion method is used to solve the misjudgment problem caused by single data source and improve the identification accuracy. (3) Through the machine learning model, the radar echo intensity evolution and satellite data characteristics are combined to realize the full life cycle tracking of the isolated mature convective system. Compared with the traditional extrapolation method, this method considers the nonlinear change of the convective system, can accurately predict the enhancement, weakening or merging of the convective system, and improves the short-term nowcasting ability. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0052] Figure 1 The structure diagram of the isolated mature convective cloud dynamic identification system of the present application is shown in the figure.

[0053] Figure 2 The detailed algorithm flowchart of the isolated mature convective cloud dynamic identification system of the present application is shown in the figure. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0055] In order to make the above objectives, characteristics and advantages of the present application more apparent, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0056] Embodiment one:

[0057] As shown in Figure 1 , Figure 2 , the satellite and radar fused isolated mature convective cloud cluster dynamic identification system comprises a data processing module, a convective cloud cluster sample extraction module, an isolated convective cloud cluster sample identification module, an isolated mature convective cloud cluster sample screening module, a sample dynamic verification module, an identification model construction module and an identification module.

[0058] The data processing module is configured to preprocess the acquired FY4A / AGRI L1-level satellite data and radar data. In the present embodiment, the satellite data mainly adopts FY4A / AGRI L1-level data, which is processed through longitude and latitude conversion, interpolation gridding and calibration. The radar data is denoised and radial velocity corrected. The satellite data is interpolated at 0.04° grid points to match the horizontal resolution of the radar data. The nearest neighbor interpolation method is adopted to align the radar and satellite data in space. The value of the nearest neighbor point is matched by calculating the Euclidean distance between the satellite and radar grid points, as shown in formula (1), and the radar data at the same time as the satellite is selected.

[0059] , (1)

[0060] In the above formula, x and y represent the longitude and latitude of the satellite grid point respectively, and x n and y n represent the longitude and latitude of the radar grid point respectively.

[0061] The convective cloud cluster sample extraction module is configured to obtain a cloud mask of a convective cloud by using the preprocessed FY4A / AGRI L1-level satellite data, and identify a convective cloud cluster sample by using the cloud mask.

[0062] Specifically, the cloud mask of the convective cloud is designed based on the water vapor and infrared channels (6.25 μm, 7.1 μm, 10.7 μm and 12.0 μm) in the stationary meteorological satellite, wherein the threshold of the convective cloud mask adopted is as shown in Table 1:

[0063] Table 1

[0064]

[0065] Specifically, such as Figure 2 As shown, the convective cloud tracking and identification (CS) system is based on the Fengyun-4 geostationary meteorological satellite. Convective cloud identification is performed using ground-based automatic weather station / precipitation data, with identification criteria including: BT10.8 < 0℃, BT7.1-BT10.7 > -15℃, BT12.0-BT10.7 > -5℃, and BT6.25-BT7.1 > -10℃.

[0066] An isolated convective cloud sample identification module is used to identify isolated convective cloud samples based on the convective cloud samples and the cloud growth region.

[0067] Specifically, such as Figure 2 As shown, isolated convective cloud clusters are identified using the brightness temperature (BT10.8) of the 10.8 μm channel of the FY4A / AGRI satellite. The criteria are whether the BT10.8 temperature is less than or equal to 0℃ and whether the cloud area is greater than 50 pixels (grid points). If so, the corresponding grid point pixels are removed; otherwise, the cloud top brightness temperature BTcloud(T) is obtained. This allows the isolated convective cloud regions to be extracted, and these regions are then labeled with connected components to provide a basis for subsequent tracking and classification.

[0068] The isolated convective cloud sample identification module includes:

[0069] The cold cloud core identification unit is used to identify cold cloud cores using the infrared channel in preprocessed geostationary meteorological satellite data.

[0070] The cloud growth region acquisition unit is used to check the neighboring pixels of the pixel corresponding to the cold cloud core using a growth search method to determine whether the neighboring pixels meet the cloud growth conditions and obtain the cloud growth region. Specifically, the lowest brightness temperature (BT) is first selected as the center of the cold cloud core, with corresponding coordinates (x0, y0). The neighboring pixels (x, y) are then checked to determine whether they meet the cloud growth conditions.

[0071] (2)

[0072] Where BT(x, y) is the brightness temperature of the current pixel, and BT(x0, y0) is the brightness temperature of the cold cloud core.

[0073] (3)

[0074] In the above formula, k=2, This is the standard deviation of local brightness temperature calculated through a 5×5 pixel window. If the brightness temperature variation does not exceed this threshold, the pixels are considered to belong to the same cloud cluster and are assigned to the cloud cluster growth region.

[0075] An isolated convective cloud sample acquisition unit is used to obtain isolated convective cloud samples based on the cloud growth region.

[0076] The isolated mature convective cloud sample screening module is used to use eight-connected region analysis to distinguish between mature and immature convection in the isolated convective cloud samples, and to screen out isolated mature convective cloud samples.

[0077] Specifically, once the connectivity region is determined, short-wave near-infrared and long-wave infrared spectrometers are used to distinguish between mature and immature convection. The near-infrared channel (3.7 μm) reflects the size of cloud top particles; a lower albedo in this channel indicates larger cloud droplet particles, making them more likely to develop into mature convection. Combined with the brightness temperature of the infrared channel (10.8 μm), which reflects cloud top height, a higher brightness temperature also makes mature convection more likely. Figure 2 As shown, the maturity of isolated convective clouds identified by satellite is judged. The judgment conditions include: the cloud top brightness temperature (BTcloud(T)) is continuously below 0℃, and the temperature change before and after it is greater than 8℃ (i.e., BTcloud(t)-BTcloud(t-1) ≤ -8℃ and BTcloud(t+1)-BTcloud(t) ≤ -8℃). At the same time, the reflectivity of the 3.7 μm channel (Ref3.7) must be greater than 0.40. Only clouds that meet the above three indicators are considered mature convective clouds. Otherwise, grid pixels are removed.

[0078] The sample dynamic verification module is used to verify the convective region of the isolated mature convective cloud cluster sample using preprocessed radar data, and obtain the final isolated mature convective cloud cluster sample.

[0079] The sample dynamic testing module includes:

[0080] The radar echo corresponding cloud mask pixel acquisition unit is used to extract isolated mature convective cloud sample cloud masks from Fengyun geostationary meteorological satellite data, and combine the binary function of the isolated mature convective cloud sample cloud mask with the radar echo region corresponding to the isolated mature convective cloud sample cloud mask to obtain the number of pixels of the isolated mature convective cloud sample cloud mask corresponding to the radar echo region.

[0081] The cloud mask total pixel acquisition unit is used to obtain the total number of pixels of the isolated mature convective cloud sample cloud mask by using a binary function of the isolated mature convective cloud sample cloud mask.

[0082] The matching rate calculation unit is configured to calculate the matching rate of the radar echo region and the isolated mature convective cloud sample cloud mask by using the number of pixels of the isolated mature convective cloud sample cloud mask corresponding to the radar echo region and the total number of pixels of the isolated mature convective cloud sample cloud mask.

[0083] Specifically, the extracted sample is dynamically verified in combination with the radar. Considering that the satellite can observe the convective information earlier than the radar, regions with different radar echo intensities (15 dBZ, 20 dBZ, 25 dBZ, and 30 dBZ) are matched with the cloud mask. The matching rate of the satellite cloud mask and the radar echo is defined as follows:

[0084] , (4)

[0085] wherein M d represents the matching rate of the cloud mask and the radar echo region at the satellite observation time t s and the radar observation time t r . C(x, y, t s ) is a binary function of the cloud mask observed by the satellite at t s , and is 1 if there is cloud at the pixel (x, y), otherwise 0; R d (x, y, t r ) is the echo mask observed by the radar at t r . The numerator represents the number of pixels corresponding to the radar echo that meets the condition in the cloud mask region. The denominator is the total number of pixels in the cloud mask region.

[0086] The dynamic verification unit is configured to take the isolated mature convective cloud sample with the matching rate satisfying the preset threshold as the final isolated mature convective cloud sample. In this embodiment, the preset threshold is M d > 0.5.

[0087] Since the factors of convection initiation and triggering can affect the development of convection, they are cross-over with the physical factors representing the evolution of convection. The development and disappearance of convection are caused by the changes of atmospheric dynamic and thermal conditions due to the evolution of weather situation, and the interaction and feedback relationship between convection systems, which leads to the rapid growth or disappearance of convection groups. Because of the complex mechanism, it is difficult to simply express the convection with individual factors, but after the convection is generated, some physical parameters can still predict the development of convection. For example, the water vapor channel with a center wavelength of 6.7 μm is very sensitive to the water vapor radiation in the upper troposphere, and by comparing the brightness temperature difference between the water vapor channel and the IR window channel, the position and development speed of the cumulus cloud top relative to the upper troposphere can be estimated. In addition to this, visible and near-infrared (0.6 μm-3.9 μm) data are also used to quantitatively describe or invert the evolution characteristics of cloud physical features (such as cloud optical thickness, cloud top phase, cloud particle effective radius and cloud top shape) in the development process of convective related cumulus clouds in the daytime.

[0088] The recognition model construction module is configured to fuse satellite features and radar features of the verified final isolated mature convection cloud cluster sample to establish an intelligent recognition model.

[0089] The recognition model construction module comprises a sample construction unit, a feature extraction unit and a model construction unit.

[0090] The sample construction unit is configured to perform convection initiation recognition verification on the final isolated mature convection cloud cluster sample by using radar data, mark the final isolated mature convection cloud cluster sample that passes the verification as a positive sample, and mark the isolated convection cloud cluster sample that fails to develop into a mature convection as a negative sample; track and analyze the life cycle of the cloud cluster to extract information such as the duration and evolution track of the cloud cluster, and pair the radar echo intensity with the cloud cluster, thereby constructing a positive and negative sample set for training an intelligent recognition model of the isolated mature convection cloud cluster. In this embodiment, the positive and negative samples are also balanced by random sampling and other strategies according to the number of positive and negative samples, and the ratio of positive and negative samples is not less than 1:2, so as to avoid model bias towards the mainstream category in the training process.

[0091] A further embodiment is to perform convection initiation recognition verification on the final isolated mature convection cloud cluster sample by using radar data, which comprises the following steps:

[0092] The Fengyun geostationary meteorological satellite data and the radar data are time-matched according to a satellite-radar time matching span; wherein the satellite-radar time matching span comprises 15 minutes, 30 minutes and 60 minutes; for example, Figure 2As shown, in the radar data, if the reflectivity factor (dBZ) satisfies any of the following conditions: dBZ≥25 for 15 minutes, dBZ≥35 for 30 minutes, or dBZ≥45 for 60 minutes, it is considered that there is a potential convective system; at the same time, combined with ground precipitation observation, the spatiotemporal continuity of the identification is enhanced.

[0093] Specifically, the satellite has an advantage in early identification of convection, but how to prove whether the convection successfully develops into a convective system, the present technology further introduces ground weather radar observation data as verification. The principle is that the radar can observe larger reflectivity factor Z, stronger differential reflectivity factor ZDR, and lower correlation coefficient (CC) in the development and mature stages of the convection. Among them, the correlation coefficient is an inversion parameter in the dual-polarization radar. These observation quantities are selected to verify the occurrence of convection. For different satellite-radar time matching spans (15 minutes, 30 minutes, and 60 minutes), according to the general physical process of convection development, threshold combination standards under different time windows are set.

[0094] When the satellite-radar time matching span is within 15 minutes, the convection is in the development stage, and the observation quantity is used to reverse the convection initial identification result in the previous 15 minutes using the development stage threshold value; wherein the observation quantity includes reflectivity factor, differential reflectivity, correlation coefficient, and specific differential phase shift rate.

[0095] When the satellite-radar time matching span is within 15-30 minutes, the convection is in the mature stage, and the observation quantity is used to reverse the convection initial identification result in the previous 30 minutes using the mature stage threshold value. The specific preliminary threshold value setting is shown in Table 2, which can be adjusted according to the region subsequently:

[0096] Table 2

[0097]

[0098] When the satellite-radar time matching span is within 60 minutes, the radar echo characteristics are used to verify the convection initial identification result. That is, based on the radar echo characteristics in the subsequent time window after the initial identification, false identification caused by satellite inversion error, convection development suppression, etc. is excluded, and the false positive rate and false negative rate are significantly reduced.

[0099] Finally, based on the star-ground verification, a relatively objective final isolated mature convective cloud sample is established, which is used as a standard for subsequent intelligent identification model training and evaluation.

[0100] The identification model construction module is used to fuse the satellite features and radar features of the verified isolated mature convective cloud sample to establish an intelligent identification model.

[0101] Further embodiments are that the identification model construction module includes:

[0102] The feature extraction unit is configured to extract feature parameters of the weather satellite data and the radar data.

[0103] Specifically, the input features include not only traditional information such as satellite channel brightness temperature, brightness temperature difference, cooling rate, cloud microphysical features (such as liquid water path LWP, ice water content IWC), but also radar feature parameters, environmental parameters (such as CAPE, CIN), and other multi-source data. The radar feature parameters include the maximum reflectivity factor derived from the reflectivity factor, and the extraction process is to find the maximum reflectivity factor at each height of the convective location.

[0104] The model construction unit is configured to train a gradient boosting decision tree using the feature parameters to construct an intelligent recognition model. Specifically, a binary classification model XGBClassifier is selected, the loss function is the error rate 'error', and the hyperparameters are set as follows: weak learner 10-100, maximum depth 6-10, learning rate 0.01-0.1, and sample ratio for each learning 0.8-1.0. The accuracy is set as the evaluation criterion to evaluate the model performance, and the model is saved after training. The model combines cloud top brightness temperature, brightness temperature change rate, reflectivity, and cloud cluster geometric shape and other feature information to achieve efficient recognition of isolated mature convective clouds and improve the early recognition ability of potential severe convective events.

[0105] In this embodiment, a threshold method is provided to iteratively improve the model.

[0106] The output results of the intelligent recognition model are compared and analyzed with the original empirical threshold-based method.

[0107] If the model performs better than the traditional method, the optimization and adjustment of the empirical threshold can be guided in reverse based on the feature importance analysis of the model, and a more intelligent and dynamic recognition system can be gradually formed.

[0108] The training set is continuously updated by periodically introducing new samples and online learning during user use, so as to realize continuous iteration and upgrading of the model.

[0109] The recognition module is configured to complete dynamic recognition of isolated mature convective clouds based on the intelligent recognition model. Finally, an isolated mature convective cell detection system combining physical law discrimination and data-driven intelligent recognition is formed.

[0110] This embodiment realizes automatic tracking and recognition of convective systems by combining FY4A / AGRI and ground minute precipitation data. Through comparison with the combined reflectivity detected by the Doppler weather radar, the algorithm can give a convective occurrence signal 1.5 hours in advance. Based on the near-infrared, water vapor channels, and other channels of FY4A / AGRI, automatic extraction of isolated mature convective clouds is realized, which is consistent with the spatial position of the strong echo detected by the Doppler radar.

[0111] Embodiment two:

[0112] The application also provides a satellite and radar fused isolated mature convective cloud cluster dynamic identification method and application system, comprising:

[0113] The acquired Fengyun geostationary meteorological satellite data and radar data are preprocessed.

[0114] The preprocessed Fengyun geostationary meteorological satellite data are used to obtain a cloud mask of a convective cloud, and the cloud mask is used to identify a convective cloud cluster sample.

[0115] Based on the convective cloud cluster sample, an isolated convective cloud cluster sample is determined in combination with a cloud cluster growth region.

[0116] Eight-connected region analysis is adopted to distinguish mature convective clouds from non-mature convective clouds for the isolated convective cloud cluster sample, and an isolated mature convective cloud cluster sample is screened out.

[0117] The preprocessed radar data are used to verify the isolated mature convective cloud cluster sample, and a final isolated mature convective cloud cluster sample is obtained.

[0118] Satellite features and radar features of the final isolated mature convective cloud cluster sample are fused to establish an intelligent identification model.

[0119] Based on the intelligent identification model, dynamic identification of the isolated mature convective cloud cluster is completed.

[0120] A further embodiment is that the method for identifying an isolated convective cloud cluster sample comprises:

[0121] The infrared channel in the preprocessed Fengyun geostationary meteorological satellite data is used to identify a cold cloud nucleus.

[0122] A growth search method is adopted to check adjacent pixels of the cold cloud nucleus, to determine whether the adjacent pixels meet cloud cluster growth conditions, and to obtain a cloud cluster growth region.

[0123] Based on the cloud cluster growth region, an isolated convective cloud cluster sample is obtained.

[0124] A further embodiment is that the method for obtaining a final isolated mature convective cloud cluster sample comprises:

[0125] The isolated mature convective cloud cluster sample cloud mask extracted from the Fengyun geostationary meteorological satellite data is used in combination with a binary function of the isolated mature convective cloud cluster sample cloud mask and a radar echo region corresponding to the isolated mature convective cloud cluster sample cloud mask to obtain a pixel number of the isolated mature convective cloud cluster sample cloud mask corresponding to the radar echo region.

[0126] The binary function of the isolated mature convective cloud cluster sample cloud mask is used to obtain a total pixel number of the isolated mature convective cloud cluster sample cloud mask.

[0127] The matching rate of the radar echo region of different intensity and the isolated mature convective cloud sample cloud mask is calculated by using the pixel number of the isolated mature convective cloud sample cloud mask corresponding to the radar echo region and the total pixel number of the isolated mature convective cloud sample cloud mask.

[0128] The isolated mature convective cloud sample satisfying the preset threshold of the matching rate is taken as the final isolated mature convective cloud sample.

[0129] The above-described embodiments are only descriptions of the preferred modes of the present application and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements of the technical solutions of the present application made by those skilled in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. A system for dynamically identifying isolated mature convective cloud clusters using a fusion of satellite and radar, characterized in that, The application relates to a method for dynamically identifying isolated mature convective cloud clusters. The method comprises the following steps: a data processing module is used for pre-processing acquired Fengyun geostationary meteorological satellite data and radar data; a convective cloud sample extraction module is used for obtaining a cloud mask of a convective cloud by using the pre-processed Fengyun geostationary meteorological satellite data, and identifying a convective cloud sample by using the cloud mask; an isolated convective cloud sample identification module is used for identifying an isolated convective cloud sample based on the convective cloud sample and in combination with a cloud cluster growth region; the isolated convective cloud sample identification module comprises a cloud cluster growth region acquisition unit, which is used for checking adjacent pixels of a cold cloud nucleus corresponding pixel by using a growth search method, judging whether the adjacent pixels satisfy a cloud cluster growth condition, and obtaining a cloud cluster growth region; an isolated mature convective cloud sample screening module is used for screening an isolated mature convective cloud sample by using eight-connected region analysis to distinguish mature convective cloud from non-mature convective cloud for the isolated convective cloud sample; a sample dynamic inspection module is used for verifying a convective region of the isolated mature convective cloud sample by using the pre-processed radar data, and obtaining a final isolated mature convective cloud sample; an identification model construction module is used for fusing satellite features and radar features of the final isolated mature convective cloud sample, and establishing an intelligent identification model; 2. The system of claim 1, wherein, an identification module is used for completing dynamic identification of the isolated mature convective cloud cluster based on the intelligent identification model. The isolated convective cloud sample identification module comprises: a cold cloud nucleus identification unit is used for identifying a cold cloud nucleus by using an infrared channel in the pre-processed Fengyun geostationary meteorological satellite data; 3. The system of claim 2, wherein, an isolated convective cloud sample acquisition unit is used for obtaining an isolated convective cloud sample based on a cloud cluster growth region. The sample dynamic inspection module comprises: a radar echo corresponding cloud mask pixel acquisition unit is used for obtaining a pixel number of an isolated mature convective cloud sample cloud mask corresponding to a radar echo region by using an isolated mature convective cloud sample cloud mask extracted from the Fengyun geostationary meteorological satellite data, in combination with a binary function of the isolated mature convective cloud sample cloud mask and the radar echo region corresponding to the isolated mature convective cloud sample cloud mask; a cloud mask total pixel acquisition unit is used for obtaining a total pixel number of the isolated mature convective cloud sample cloud mask by using a binary function of the isolated mature convective cloud sample cloud mask; a matching rate calculation unit is used for calculating matching rates of different intensity radar echo regions and the isolated mature convective cloud sample cloud mask by using the pixel number of the isolated mature convective cloud sample cloud mask corresponding to the radar echo region and the total pixel number of the isolated mature convective cloud sample cloud mask; 4. The system of claim 1, wherein, a dynamic inspection unit is used for taking the isolated mature convective cloud sample with the matching rate satisfying a preset threshold value as the final isolated mature convective cloud sample. The identification model construction module comprises: a sample construction unit is used for identifying and verifying a convective initial stage of the final isolated mature convective cloud sample by using radar data, marking the final isolated mature convective cloud sample passing the verification as a positive sample, and marking an isolated convective cloud sample failing to develop into a mature convective cloud as a negative sample; a feature extraction unit is used for extracting feature parameters of the Fengyun geostationary meteorological satellite data and the radar data; A model construction unit is configured to train a gradient boosting decision tree using the characteristic parameters and construct an intelligent identification model.

5. The system of claim 4, wherein, The process of using radar data to verify the convection initial identification of the final isolated mature convection cloud sample includes: The Fengyun geostationary meteorological satellite data and the radar data are time-matched according to a satellite-radar time matching span, wherein the satellite-radar time matching span includes 15 minutes, 30 minutes and 60 minutes; When the satellite-radar time matching span is within 15 minutes, the convection is in the development stage, and the observation quantity and the development stage threshold value are used to back-propagate the convection initial identification result of the previous 15 minutes, wherein the observation quantity includes the reflectivity factor, the differential reflectivity, the correlation coefficient and the specific differential phase rate; When the satellite-radar time matching span is within 15-30 minutes, the convection is in the mature stage, and the observation quantity and the mature stage threshold value are used to back-propagate the convection initial identification result of the previous 30 minutes; When the satellite-radar time matching span is within 60 minutes, the radar echo characteristics are used to verify the convection initial identification result.

6. A method for dynamically identifying isolated mature convective cloud clusters using a fusion of satellite and radar, applying the system of any of claims 1-5, characterized in that, The method includes: preprocessing the acquired Fengyun geostationary meteorological satellite data and radar data; using the preprocessed Fengyun geostationary meteorological satellite data to obtain a cloud mask of a convection cloud, and using the cloud mask to identify a convection cloud sample; based on the convection cloud sample, isolated convection cloud samples are determined in combination with cloud cluster growth regions; using eight-connected region analysis, the isolated convection cloud samples are discriminated into mature convection and non-mature convection, and isolated mature convection cloud samples are screened out; the isolated mature convection cloud samples are verified by using the preprocessed radar data, and final isolated mature convection cloud samples are obtained; satellite features and radar features of the final isolated mature convection cloud samples are fused to establish an intelligent identification model; based on the intelligent identification model, dynamic identification of isolated mature convection cloud samples is completed.

7. The method of claim 6, wherein, The method of identifying the isolated convection cloud sample includes: using the infrared channel in the preprocessed Fengyun geostationary meteorological satellite data to identify a cold cloud nucleus; using a growth search method to check adjacent pixels of the cold cloud nucleus, and judging whether the adjacent pixels satisfy cloud cluster growth conditions to obtain a cloud cluster growth region; based on the cloud cluster growth region, isolated convection cloud samples are obtained.

8. The method of claim 7, wherein, The method of obtaining the final isolated mature convection cloud sample includes: using the isolated mature convection cloud sample cloud mask extracted from the Fengyun geostationary meteorological satellite data, in combination with a binary function of the isolated mature convection cloud sample cloud mask and a radar echo region corresponding to the isolated mature convection cloud sample cloud mask, the number of pixels of the isolated mature convection cloud sample cloud mask corresponding to the radar echo region is obtained; using the binary function of the isolated mature convection cloud sample cloud mask, the total number of pixels of the isolated mature convection cloud sample cloud mask is obtained; using the number of pixels of the isolated mature convection cloud sample cloud mask corresponding to the radar echo region and the total number of pixels of the isolated mature convection cloud sample cloud mask, the matching rate of the radar echo region of different intensity and the isolated mature convection cloud sample cloud mask is calculated; The isolated mature convective cloud sample satisfying the preset threshold value in the matching rate is taken as the final isolated mature convective cloud sample.

Images