Three-dimensional cloud type classification method and device based on multi-source meteorological data fusion and medium

Abstract

The application provides a three-dimensional cloud type classification method and device based on multi-source meteorological data fusion, equipment and medium, the method comprises the following steps: acquiring satellite data, radar data, numerical model data and ground observation data and preprocessing; adopting a three-dimensional cloud amount analysis method, taking the three-dimensional cloud amount of the numerical model data as the background field, forming a three-dimensional cloud amount grid point field of multi-source data fusion; using the XGBOOST multi-classification method to establish and train multiple cloud type classification models, based on the solar elevation angle of each grid point of the three-dimensional cloud amount grid point field of multi-source data fusion at the monitoring time and the diagnostic result of the vertical direction cloud layer distribution, matching the corresponding cloud type classification model, obtaining the vertical direction cloud type classification of each cloud layer, generating the three-dimensional cloud type classification result, to solve the problem that the three-dimensional cloud type classification monitoring is limited due to the fact that the satellite data cannot penetrate the uppermost cloud layer to obtain the vertical structure and lacks effective training samples.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloud classification monitoring, and particularly relates to a three-dimensional cloud type classification method and device based on multi-source meteorological data fusion, equipment and medium. BACKGROUND

[0002] Clouds cover most of the earth's surface all year round, and play an important role in weather forecasting, meteorological disaster warning and aviation safety. Generally, clouds can be divided into three categories from height, i.e. low cloud, middle cloud and high cloud. Further, the International Satellite Cloud Climatology Program (ISCCP) divides clouds into nine cloud types, i.e. cirrus cloud Ci, cirrostratus cloud Cs, deep convective cloud (cumulonimbus cloud) Cb, high cumulus cloud Ac, high-level cloud As, nimbostratus cloud Ns, cumulus cloud Cu, strato-cumulus cloud Sc and stratus cloud St. Accurate cloud classification monitoring has important indicative significance for aviation flight. For example, cumulonimbus cloud often causes severe turbulence, ice accumulation, lightning, heavy rain, downburst and other severe weather, which seriously affects flight safety; stratus cloud has low cloud base height and poor visibility below the cloud, which seriously affects the take-off and landing of aircraft; nimbostratus cloud is often accompanied by continuous rainfall, and the visibility in the cloud is low, and the flight in the cloud may have moderate to severe ice accumulation, which also affects flight safety.

[0003] Current cloud classification monitoring is mostly based on data produced by meteorological satellites, and the commonly used technical methods can be basically divided into two categories:

[0004] One category is traditional threshold screening and statistical methods. For example, Purbantoro et al. selected different threshold values for brightness temperature and brightness temperature difference of different channels, and obtained cloud classification monitoring in summer and winter by using sliding window method (Purbantoro, B., Aminuddin, J., Manago, N., et al., 2018: Comparison of Cloud Type Classification with Split Window Algorithm Based on Different Infrared Band Combinations of Himawari-8 Satellite. Adv. Remote Sens., 7, 218-234.); Zhang et al. assumed that the input features conform to normal distribution, and obtained cloud classification monitoring based on satellite visible and infrared data by using maximum likelihood estimation method (Zhang, C., Zhuge, X., Yu, F., 2019: Development of a High Spatiotemporal Resolution Cloud-Type Classification Approach Using Himawari-8 and CloudSat. Int. J. Remote Sens, 40, 6464-6481.).

[0005] In recent years, with the rise of artificial intelligence technology, machine learning and deep learning methods have played an increasingly important role in cloud classification monitoring, and have achieved more accurate results than traditional methods. Here we list the implementation schemes based on domestic meteorological satellites. For example, Yu et al. selected the visible channel reflectance, infrared channel brightness temperature, and cloud top height, cloud particle effective radius, and other cloud products of FY-4A geostationary meteorological satellite as input features, and the cloud classification products of CloudSat as label values. First, the random forest model was used to obtain the classification results of eight cloud types and two-layer clouds. Then, a new random forest model was used to further classify the cloud types of two-layer clouds, and the classification results of 12 cloud type combinations of two-layer clouds were obtained (Yu, Z. F., Ma, S., Ding, H., et al., 2021: A cloud classification method based on random forest for FY-4A, International Journal of Remote Sensing, 42, 3357-3383.). He Xiaodong et al. selected the visible channel reflectance, infrared channel brightness temperature, and infrared channel brightness temperature difference of FY-4A satellite as input features, and the cloud classification products of Fengyun-8 satellite as label values. The UNet++ model was used to obtain the cloud classification monitoring results of nine cloud types (He Xiaodong, You Xiaogang, Chen Zhen, et al. ISCCP cloud classification method based on FY-4A satellite, system, medium and terminal: 202110263471[P]). Jiang et al. selected the visible channel reflectance and infrared channel brightness temperature of FY-4A geostationary meteorological satellite as input features, and the cloud classification products of Fengyun-8 satellite as label values. Based on the Unet network architecture, combined with attention mechanism and empty space convolution pyramid module, the CLP-CNN model for nine cloud type cloud classification was built (Jiang, Y., Cheng, W., Gao, F., et al., 2022: A Cloud Classification Method Based on a Convolutional Neural Network for FY-4A Satellites. Remote Sensing, 14, 10.3390 / rs14102314.). Lin et al. selected the visible channel reflectance, infrared channel brightness temperature, and infrared channel brightness temperature difference of FY-4A geostationary meteorological satellite as input features, and the cloud classification products of Fengyun-8 satellite as label values. The LightGBM model was used to obtain the cloud classification monitoring results of nine cloud types (Lin, J., Bao, Y., Petropoulos, G. P., et al.,2023: Cloud-Type Classification for Southeast China Based on Geostationary Orbit EODatasets and the LighGBM Model. Remote Sensing, 15, 10.3390 / rs15245660.

[0006] The prior art is mostly based on high-resolution cloud classification grid monitoring results obtained from geostationary meteorological satellite data. The visible and infrared sensors carried by geostationary meteorological satellites can only obtain information on the upper part of the cloud layer and cannot penetrate the uppermost cloud layer to obtain the vertical structure information of the cloud. Therefore, the obtained cloud classification monitoring results are only for the uppermost cloud layer, and if there are multiple layers of clouds in the vertical direction, the cloud classification monitoring results of the lower layers of clouds cannot be obtained, and a three-dimensional cloud classification grid monitoring field cannot be formed. Satellites equipped with active detection radars (such as CloudSat and CALIPSO) can penetrate the cloud layer to obtain the vertical structure of the cloud and obtain the cloud classification results of multiple layers of clouds, but they belong to polar-orbit meteorological satellites and have very narrow single scanning bandwidth, which cannot directly form a high-resolution grid monitoring field. Through machine learning models, the visible and infrared information of geostationary meteorological satellites is linked to the classification results of multiple layers of clouds of such satellites, thereby obtaining the classification results of multiple layers of clouds in the vertical direction based on geostationary meteorological satellites. The disadvantage of this scheme is that it cannot classify each layer of cloud in the vertical direction individually, but only obtains the combined classification of multiple layers of clouds through the visible and infrared information of the cloud top. The number of combined classification categories depends on the training data set and cannot cover all possible combinations. Moreover, this scheme can generally only support the combined classification of two layers of clouds. In addition, the two commonly used active detection satellites, CloudSat and CALIPSO, have currently stopped service, and there is no effective spatiotemporal matching sample data set for the latest generation of geostationary meteorological satellites (such as FY-4B) to perform model training.

[0007] Therefore, it is urgent to propose a three-dimensional cloud type classification method based on multi-source meteorological data fusion to solve the technical problem that the three-dimensional cloud type classification monitoring is limited due to the inability to penetrate the cloud layer to obtain the vertical structure and the lack of effective training samples when only satellite data is used. SUMMARY

[0008] To overcome the problems in the related art, the present disclosure provides a three-dimensional cloud type classification method, device, equipment and medium based on multi-source meteorological data fusion to solve the technical problem that the three-dimensional cloud type classification monitoring is limited due to the inability to penetrate the cloud layer to obtain the vertical structure and the lack of effective training samples when only satellite data is used in the related art.

[0009] The one or more embodiments of the specification provide a three-dimensional cloud type classification method based on multi-source meteorological data fusion, comprising the following steps:

[0010] Satellite data, radar data, numerical model data and ground observation data are acquired and preprocessed;

[0011] A three-dimensional cloud amount analysis method is adopted, taking the three-dimensional cloud amount of the numerical model data as a background field, and the ground observation data, the satellite data and the radar data are fused to form a three-dimensional cloud amount grid point field of multi-source data fusion;

[0012] Based on the satellite data, the numerical model data, the three-dimensional cloud amount grid point field of multi-source data fusion, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point fields, an XGBOOST multi-classification method is used to establish and train multiple cloud type classification models;

[0013] Based on the solar elevation angle of each grid point of the three-dimensional cloud amount grid point field of multi-source data fusion at the monitoring time and the diagnostic result of the vertical cloud layer distribution, the corresponding cloud type classification model is matched to obtain the vertical cloud type classification of each cloud layer, and a three-dimensional cloud type classification result is generated.

[0014] Preferably, the three-dimensional cloud amount analysis method is adopted, taking the three-dimensional cloud amount of the numerical model data as a background field, and the ground observation data, the satellite data and the radar data are fused to form a three-dimensional cloud amount grid point field of multi-source data fusion, specifically comprising the following steps:

[0015] Taking the three-dimensional cloud amount of the numerical model data as a background field, an objective analysis method is used to fuse the cloud amount vertical profile of ground observation layer by layer, correct the structural distribution of the three-dimensional cloud amount field, and update the three-dimensional cloud amount field;

[0016] Based on the updated three-dimensional cloud amount field, the satellite data is fused to correct the structural distribution of the three-dimensional cloud amount field, and the three-dimensional cloud amount field is updated;

[0017] Based on the updated three-dimensional cloud amount field, the radar data is fused to correct the structural distribution of the three-dimensional cloud amount field, and a three-dimensional cloud amount grid point field of multi-source data fusion is formed.

[0018] Preferably, the satellite data is fused based on the updated three-dimensional cloud amount field to correct the structural distribution of the three-dimensional cloud amount field, specifically comprising the following steps:

[0019] The satellite cloud detection data, cloud top pressure data and cloud top temperature data are fused using the cloud amount analysis method in GSI to correct the cloud top distribution of the three-dimensional cloud amount field;

[0020] The satellite L1 data is fused using the cloud amount analysis method in LAPS to correct the upper structure of the three-dimensional cloud amount field.

[0021] Preferably, based on the satellite data, the multi-source data fused three-dimensional cloud amount grid field, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid field, a plurality of cloud type classification models are established and trained using the XGBOOST multi-classification method, specifically including the following steps:

[0022] The first input feature and the second input feature are constructed using the satellite data as the sample data set, and the first cloud classification model and the second cloud classification model are constructed and trained using the XGBOOST multi-classification method;

[0023] The third input feature is constructed using the numerical model data, the multi-source data fused three-dimensional cloud amount grid field, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid field as the sample data set, and the third cloud classification model is constructed and trained using the XGBOOST multi-classification method.

[0024] Preferably, the third input feature constructed using the numerical model data, the multi-source data fused three-dimensional cloud amount grid field, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid field further includes the following steps:

[0025] Based on the three-dimensional cloud amount grid field, the position and average cloud amount of each cloud layer in the vertical direction of each grid point are diagnosed;

[0026] Based on the three-dimensional geopotential height and three-dimensional temperature of the numerical model data, the cloud layer top height, cloud layer top temperature, cloud layer thickness, cloud layer temperature, maximum geopotential temperature gradient in the cloud layer, and average geopotential temperature gradient in the cloud layer are calculated;

[0027] Based on the three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid field, the cloud liquid water path and cloud ice water path of each cloud layer are calculated;

[0028] The cloud layer average cloud amount, cloud layer top height, cloud layer top temperature, cloud layer thickness, cloud layer temperature, maximum geopotential temperature gradient in the cloud layer, average geopotential temperature gradient in the cloud layer, cloud layer cloud liquid water path, and cloud layer cloud ice water path are constructed as the third input feature;

[0029] The third cloud classification model is trained based on the third input feature of the uppermost cloud layer.

[0030] Preferably, the solar elevation angle of each grid point of the multi-source data fused three-dimensional cloud amount grid field at the monitoring time and the diagnostic results of the vertical cloud layer distribution are matched with the corresponding cloud type classification model to obtain the vertical cloud type classification of each cloud layer, and a three-dimensional cloud type classification result is generated, specifically including the following steps:

[0031] when a grid point is diagnosed to have cloud in the multi-source data fused three-dimensional cloud amount grid point field, calculating a solar elevation angle of the grid point at a monitoring time;

[0032] when the solar elevation angle is greater than 15°, inputting the first input feature into the first cloud classification model to obtain a cloud type classification result of the uppermost cloud layer, and when the solar elevation angle is not greater than 15°, inputting the second input feature into the second cloud classification model to obtain a cloud type classification result of the uppermost cloud layer;

[0033] inputting the third input feature of each cloud layer except the uppermost cloud layer into the third cloud classification model respectively to obtain a cloud type classification result of each cloud layer except the uppermost cloud layer.

[0034] One or more embodiments of the present specification provide a three-dimensional cloud type classification device based on multi-source meteorological data fusion, comprising a data acquisition module, a data fusion module, a model construction module and a real-time classification module;

[0035] The data acquisition module is configured to acquire satellite data, radar data, numerical model data and ground observation data and perform preprocessing;

[0036] The data fusion module is configured to adopt a three-dimensional cloud amount analysis method, take the three-dimensional cloud amount of the numerical model data as a background field, fuse the ground observation data, the satellite data and the radar data, and form a multi-source data fused three-dimensional cloud amount grid point field;

[0037] The model construction module is configured to establish and train a plurality of cloud type classification models based on the satellite data, the numerical model data, the multi-source data fused three-dimensional cloud amount grid point field and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point fields, and utilize an XGBOOST multi-classification method;

[0038] The real-time classification module is configured to match a corresponding cloud type classification model based on a solar elevation angle of each grid point of the multi-source data fused three-dimensional cloud amount grid point field at a monitoring time and a diagnostic result of vertical cloud layer distribution, obtain a cloud type classification of each cloud layer in the vertical direction, and generate a three-dimensional cloud type classification result.

[0039] Preferably, the data fusion module comprises a ground observation data fusion unit, a satellite data fusion unit and a radar data fusion unit;

[0040] The ground observation data fusion unit is configured to take the three-dimensional cloud amount of the numerical model data as a background field, fuse a cloud amount vertical profile of ground observation layer by layer using an objective analysis method, correct a structural distribution of the three-dimensional cloud amount field, and update the three-dimensional cloud amount field;

[0041] The satellite data fusion unit is configured to fuse the satellite data based on the updated three-dimensional cloud amount field, correct the structural distribution of the three-dimensional cloud amount field, and update the three-dimensional cloud amount field.

[0042] The radar data fusion unit is configured to fuse the radar data based on the updated three-dimensional cloud amount field, correct the structural distribution of the three-dimensional cloud amount field, and form a three-dimensional cloud amount grid field fused by multiple sources.

[0043] One or more embodiments of the present specification provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the three-dimensional cloud type classification method based on multi-source meteorological data fusion as described above when executing the computer program.

[0044] One or more embodiments of the present specification provide a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the three-dimensional cloud type classification method based on multi-source meteorological data fusion as described above.

[0045] The method, device, equipment and medium for three-dimensional cloud type classification based on multi-source meteorological data fusion are provided, and the advantages are as follows: satellite data, radar data, numerical model data and ground observation data are acquired and preprocessed, various sources of meteorological data are comprehensively used, information from high altitude to ground and from large range to local is covered, data is more comprehensive, and the accuracy of meteorological data is improved; a three-dimensional cloud amount analysis method is used, the three-dimensional cloud amount of the numerical model data is used as a background field, the ground observation data, the satellite data and the radar data are fused, a three-dimensional cloud amount grid field of multi-source data fusion is formed, the respective advantages of different data in cloud amount are fully utilized, the cloud amount distribution in the three-dimensional space can be more accurately described, the spatial structure and range of the cloud layer are more clearly presented, and the macroscopic characteristics of the cloud are more deeply understood; based on the satellite data, the numerical model data, the three-dimensional cloud amount grid field of multi-source data fusion and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid fields, an XGBOOST multi-classification method is used to establish and train multiple cloud type classification models, the models can learn more rich feature information of the cloud, the accuracy and reliability of cloud type classification are improved, and different types of clouds can be more accurately distinguished; based on the solar elevation angle of each grid point of the three-dimensional cloud amount grid field of multi-source data fusion at the monitoring moment and the diagnostic result of the vertical direction cloud layer distribution, a corresponding cloud type classification model is matched, the cloud type classification of each cloud layer in the vertical direction is obtained, a three-dimensional cloud type classification result is generated, the cloud layer in the vertical direction can be more finely classified, the distribution of the cloud layer at different altitudes and the type of the cloud are understood, and the method has important significance for understanding the vertical structure and evolution of the cloud in the fields of meteorological research and weather forecasting, and helps to more accurately predict weather changes. The establishment of multiple cloud type classification models and the matching of the models according to different conditions make the method adapt to different meteorological conditions and data characteristics. For meteorological data of different regions and different times, cloud type classification can be performed through appropriate models, the universality and adaptability of the method are improved, and the method has a wider application scenario. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the one or more embodiments of the present specification or the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present specification, and other drawings can be obtained by those skilled in the art without any creative labor.

[0047] Figure 1 A flowchart of a method for three-dimensional cloud type classification based on multi-source meteorological data fusion is provided.

[0048] Figure 2 A technical flowchart provided for one or more embodiments of the present specification;

[0049] Figure 3 A model training process schematic provided for one or more embodiments of the present specification;

[0050] Figure 4 A model parameter initialization setting schematic provided for one or more embodiments of the present specification;

[0051] Figure 5 A structure schematic of a three-dimensional cloud type classification device based on multi-source meteorological data fusion provided for one or more embodiments of the present specification;

[0052] Figure 6 A structure schematic of a computer device provided for one or more embodiments of the present specification. DETAILED DESCRIPTION

[0053] In order to make the person in the art better understand the technical scheme in one or more embodiments of the present specification, the technical scheme in one or more embodiments of the present specification will be described clearly and completely in conjunction with the drawings in one or more embodiments of the present specification. Obviously, the described embodiments are only part of the embodiments of the present specification, not all the embodiments. Based on one or more embodiments of the present specification, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present specification.

[0054] The present application will be described in detail below in conjunction with the specific embodiments and the drawings of the specification.

[0055] Method embodiments

[0056] According to the embodiments of the present application, a three-dimensional cloud type classification method based on multi-source meteorological data fusion is provided, as shown in the following. Figure 1 The flowchart of the three-dimensional cloud type classification method based on multi-source meteorological data fusion provided by the present embodiment is shown in the following. Figure 2 The technical flowchart provided by the present embodiment is shown in the following. The three-dimensional cloud type classification method based on multi-source meteorological data fusion according to the embodiments of the present application includes the following steps:

[0057] S110, satellite data, radar data, numerical model data and ground observation data are obtained and preprocessed, specifically, L1 data and cloud detection, cloud top pressure, cloud top temperature data of the new generation of geostationary meteorological satellite FY-4B, cloud classification data of HuanHua-9 satellite, three-dimensional reflectivity of new generation weather radar CINRAD, cloud amount, cloud base height and weather phenomenon of ground airport observation METAR message, surface pressure, surface temperature, three-dimensional temperature, three-dimensional specific humidity, three-dimensional potential height and three-dimensional cloud amount of GFS numerical model data, and terrain height static data. The obtained data are preprocessed, specifically including the following steps:

[0058] 1. Using the latitude and longitude lookup table of FY-4B satellite, the L1 data and cloud detection, cloud top pressure, cloud top temperature data of FY-4B are interpolated to the equi-latitude and longitude grid with a resolution of 0.05° by nearest neighbor interpolation, and the reflectivity of the first three visible channels is corrected by solar elevation angle, i.e. ALB = ALB / sinβ, wherein β represents the solar elevation angle; the cloud classification data of HuanHua-9 satellite is interpolated to the same equi-latitude and longitude grid by nearest neighbor interpolation.

[0059] 2. The three-dimensional temperature, three-dimensional specific humidity, three-dimensional potential height and three-dimensional cloud amount of GFS model data are interpolated to the following pressure layers using logarithmic linear interpolation: 1000hPa-850hPa every 15hPa, 825hPa-200hPa every 25hPa, 150hPa-50hPa every 50hPa. The interpolated model three-dimensional data, surface pressure and surface temperature of GFS model data, and terrain height static data are interpolated to the equi-latitude and longitude grid with a resolution of 0.05° by bilinear interpolation, and if the potential height of a point on the three-dimensional grid point field is less than the terrain height, all data on the point are set to default value.

[0060] 3. The three-dimensional reflectivity of the radar is converted from the polar coordinate system to the Cartesian coordinate system with the radar as the center point, and the nearest neighbor method is used in the radial and azimuth directions, and the linear interpolation method is used in the vertical direction during conversion;

[0061] If data of multiple radars are obtained at the same time, the reflectivity data of the multiple radar coordinate conversion needs to be spatially spliced, and the inverse distance exponential weighting method is adopted for splicing, and the reflectivity at a certain three-dimensional grid point can be represented as:

[0062]

[0063] Wherein, a i represents the reflectivity of different radars at the point, the weight r i represents the horizontal distance of the point from different radars, and R represents the influence radius, and if a grid point is not covered by any radar observation, the reflectivity of the point is set to default value.

[0064] 4. Obtain the longitude and latitude information of the observation point from the ground airport observation METAR message, obtain the potential height of each pressure layer of the point by using the three-dimensional potential height of the model, and convert the cloud amount code in the ground airport observation METAR message into a cloud amount value between 0 and 1: for example, SKC / 0.0, CLR / 0.0 (below 3600 meters), FEW / 0.25, SCT / 0.5, BKN / 0.75, OVC / 1.0, if the cloud amount is not 0, convert the unit of cloud bottom height into meters, and assume that the cloud layer thickness is 300 meters (when weather phenomena such as rain, snow, hail, etc. occur, assume that the cloud layer thickness is 1000 meters, and when thunderstorm weather phenomenon occurs, assume that the cloud layer thickness is 10000 meters), and generate the vertical profile of the cloud amount of the point according to the cloud amount, cloud bottom height, cloud layer thickness of the observation point, and the potential height of each pressure layer.

[0065] S120, using a three-dimensional cloud amount analysis method, taking the three-dimensional cloud amount of the numerical mode data as a background field, fusing the ground observation data, the satellite data and the radar data, forming a three-dimensional cloud amount grid point field of multi-source data fusion.

[0066] S130, based on the satellite data, the numerical mode data, the three-dimensional cloud amount grid point field of multi-source data fusion, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point field, using XGBOOST multi-classification method to establish and train multiple cloud type classification models, wherein the Smith-Feddes model is used to calculate the three-dimensional cloud liquid water content and the three-dimensional cloud ice water content, and the specific steps include the following steps:

[0067] In the cloud coverage area, each cloud layer is divided into multiple height layers with a vertical resolution of 100m from the cloud bottom to the cloud top, and it is assumed that the cloud interior is a wet adiabatic process, so as to calculate the temperature, pressure and saturated water vapor mixing ratio of each height layer, and the difference between the saturated water vapor mixing ratios of adjacent two layers is taken as the sum of the initial cloud liquid water and cloud ice water content, denoted as CWI0.

[0068] Considering the dilution effect of the entrainment process between the cloud and the dry air on the cloud water content, the Warner curve is used to establish the relationship between the entrainment rate and the height, and the formula is as follows:

[0069]

[0070] Wherein, ht represents the height from the cloud bottom, and the unit is kilometer. The entrainment rate y is used to correct CWI0, that is, CWI1=y*CWI0.

[0071] The cloud liquid water content and the cloud ice water content are separated according to the temperature of each height layer, if the temperature is greater than -10℃, all is cloud liquid water, if the temperature is less than -30℃, all is cloud ice water, if between the two, the separation is carried out according to linear proportion, and the obtained cloud liquid water content and cloud ice water content of each height layer are interpolated to the pressure layer of the three-dimensional cloud amount field.

[0072] When the temperature on the pressure layer of the three-dimensional cloud amount field is less than 0℃, part of the cloud liquid water is converted into cloud ice water by using the radar reflectivity, and the proportion of the converted part in the cloud liquid water is:

[0073]

[0074] Wherein, REF represents the radar reflectivity, and the unit is dBZ.

[0075] S140, based on the solar elevation angle of each grid point of the multi-source data fusion three-dimensional cloud amount grid field at the monitoring moment and the diagnostic result of the vertical direction cloud layer distribution, the corresponding cloud type classification model is matched, the vertical direction cloud type classification of each cloud layer is obtained, and a three-dimensional cloud type classification result is generated.

[0076] The method provided by the embodiment comprehensively integrates meteorological data from multiple sources by acquiring satellite data, radar data, numerical model data and ground observation data and preprocessing, covers information from high altitude to ground and from large range to local, makes the data more comprehensive and improves the accuracy of meteorological data; a three-dimensional cloud amount analysis method is adopted, the three-dimensional cloud amount of the numerical model data is taken as a background field, the ground observation data, the satellite data and the radar data are fused to form a three-dimensional cloud amount grid field of multi-source data fusion, the respective advantages of different data in cloud amount are fully utilized, the cloud amount distribution in the three-dimensional space can be more accurately described, and therefore the spatial structure and range of the cloud layer are more clearly presented, which is helpful for more deeply understanding the macro features of the cloud; based on the satellite data, the numerical model data, the three-dimensional cloud amount grid field of multi-source data fusion and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid fields, an XGBOOST multi-classification method is used to establish and train multiple cloud type classification models, so that the models can learn more rich feature information of the cloud, thereby improving the accuracy and reliability of cloud type classification and more accurately distinguishing different types of clouds; based on the solar elevation angle of each grid point of the three-dimensional cloud amount grid field of multi-source data fusion at the monitoring moment and the diagnostic result of the vertical direction cloud layer distribution, a corresponding cloud type classification model is matched to obtain the vertical direction cloud type classification of each cloud layer and generate a three-dimensional cloud type classification result, which can more finely classify the cloud layer in the vertical direction, understand the distribution of the cloud layer at different altitudes and the type of the cloud, and has important significance for understanding the vertical structure and evolution of the cloud in the field of meteorological research, weather forecasting and the like, and is helpful for more accurately predicting weather changes. The establishment of multiple cloud type classification models and the matching of the models according to different conditions make the method adaptable to different meteorological conditions and data characteristics. For meteorological data of different regions and different times, cloud type classification can be performed through appropriate models, the generality and adaptability of the method are improved, and the method has a wider application scenario.

[0077] In one embodiment, S120, a three-dimensional cloud amount analysis method is adopted, the three-dimensional cloud amount of the numerical model data is taken as a background field, the ground observation data, the satellite data and the radar data are fused to form a three-dimensional cloud amount grid field of multi-source data fusion, and the following steps are specifically included:

[0078] Taking the kth pressure layer as an example, the weight ω of the cloud amount observation i on the grid point xy located on the pressure layer is calculated. i = (r i / r0) -5 wherein, ri r0 represents the horizontal distance from the observation location to the grid point, and r0 is the horizontal radius of influence.

[0079] Using the weights calculated in the previous step, the observations are fused to the background field grid points x and y, correcting the structure distribution of the 3D cloud field and updating the 3D cloud field. The fusion formula is as follows:

[0080]

[0081] Among them, C i Represents cloud cover observations. and Representing the background cloud cover field and the fused cloud cover analysis field at grid point xy, respectively, ω b Represents the background field weight.

[0082] Based on the updated 3D cloud field fused with FY-4B satellite data, the structural distribution of the 3D cloud field is corrected, and the 3D cloud field is updated. The specific steps include:

[0083] The cloud top distribution of the three-dimensional cloud field is corrected by fusing the satellite cloud detection data, cloud top pressure data, and cloud top temperature data using cloud cover analysis methods in GSI.

[0084] The upper and middle structures of the three-dimensional cloud field were corrected by fusing the satellite L1 data using cloud cover analysis methods in LAPS.

[0085] Specifically, the cloud cover analysis method in GSI (Gridpoint Statistical Interpolation) is first used to fuse satellite cloud detection data, cloud top pressure data, and cloud top temperature data to correct the cloud top distribution of three-dimensional clouds:

[0086] If the satellite cloud top pressure at a certain point is greater than the model surface pressure, the satellite observation is considered unreasonable and will not be used. If the satellite cloud top pressure is greater than 600 hPa, based on the model's atmospheric temperature vertical profile at that point, a pressure layer with atmospheric stability greater than a certain threshold is searched downwards from the satellite cloud top. Among these stable pressure layers, the pressure layer with the closest temperature to the satellite cloud top is found. If the absolute value of the deviation between the temperature of this pressure layer and the temperature of the satellite cloud top is less than a certain threshold, the cloud top pressure is corrected to the pressure corresponding to this pressure layer; otherwise, the corresponding satellite observation will not be used.

[0087] If the satellite cloud detection is no cloud, the cloud amount of all pressure layers in the vertical direction of the corresponding grid point is set to 0; if the satellite cloud detection is cloud, the cloud amount profile of the corresponding grid point is corrected using the satellite cloud top pressure: the pressure layer closest to the satellite cloud top is taken as the cloud top, the cloud thickness is assumed to be 50 hPa, the cloud amount of each pressure layer above the point in the cloud top is set to 0, the cloud amount of the cloud top and each pressure layer within the cloud thickness is set to 0.5, and the cloud amount of the remaining pressure layers remains unchanged.

[0088] Secondly, the satellite L1 data is fused using the cloud amount analysis method in LAPS (Local Analysis Prediction System) to further correct the middle and upper structure of the three-dimensional cloud. For FY-4B satellite, three channels C02, C08 and C13 are mainly used:

[0089] The brightness temperature of satellite C08 channel is recorded as t4, the brightness temperature of C13 channel is recorded as t11, ts represents the model surface temperature, and β represents the solar elevation angle. For any satellite grid point xy, if one of the following two conditions is met:

[0090] ① t11 xy -ts xy <-8℃;

[0091] ②t11 xy >-10℃,t4 xy -t11 xy <-2.5℃,β xy <0;

[0092] then the estimated brightness temperature of each pressure layer in the vertical direction of the corresponding grid point is calculated, and the formula of the estimated brightness temperature is:

[0093] tm xyz =IBRAD((1-cld xyz )*BRAD(ts xy )+cld xyz *BRAD(t xyz ));

[0094] Wherein, subscript z represents the pressure layer in the vertical direction, tm represents the estimated brightness temperature, t represents the model three-dimensional atmospheric temperature, cld represents the three-dimensional cloud amount, BRAD represents a function of converting temperature to radiance, and IBRAD represents the inverse function of BRAD.

[0095] If the satellite observed brightness temperature t11 xy is greater than the estimated brightness temperature tm xyz , then it is considered that the cloud amount of the point is too much, and the cloud amount cld xyzThe estimated brightness temperature is made consistent with the satellite observed brightness temperature (this method is not used for low-level warm clouds to avoid false reduction of low-level warm clouds).

[0096] All cloud layers in the vertical direction for each grid point are obtained by using the diagnostic algorithm: for each grid point, the first pressure layer with cloud cover greater than 0.2 is found from high to low in the vertical direction, and the middle height of the pressure layer and the pressure layer above it is taken as the cloud top height of the uppermost cloud layer, and then it is continued to be found downward until the first pressure layer with cloud cover less than or equal to 0.2 is found, and the middle height of the pressure layer and the pressure layer above it is taken as the cloud bottom height of the uppermost cloud layer. All cloud layers in the vertical direction for the grid point can be obtained by continuing to search downward in this way. The average cloud cover in each cloud layer (searching from the top of the cloud layer to the bottom of the cloud layer, only when the cloud cover of a certain pressure layer is greater than the maximum cloud cover of all pressure layers above it, the average cloud cover is calculated) and the cloud layer temperature (the temperature of the pressure layer with the maximum cloud cover in the cloud layer) are calculated. The average cloud cover of the i-th cloud layer from top to bottom is denoted as a i , and the cloud layer temperature is denoted as t i , and n is the number of cloud layers in the vertical direction. The effective temperature te is obtained by using the following formula:

[0097]

[0098] The objective function of the satellite observed brightness temperature t11 is established:

[0099]

[0100] Where m and n are the number of grids in the east-west direction and the north-south direction respectively. Using the step-by-step correction scheme, the same cloud cover value is iteratively increased or decreased in the cloud area, and the value is continuously adjusted until the objective function reaches a minimum value.

[0101] The reflectivity data of the C02 visible light channel after the solar elevation angle correction (only data with a solar elevation angle greater than 15 degrees is used) is converted into total cloud cover. For each grid point, if the cloud cover of a certain pressure layer in the vertical direction is greater than the total cloud cover, the cloud cover of the layer is reduced.

[0102] Based on the updated three-dimensional cloud cover field, the radar data is fused to correct the structural distribution of the three-dimensional cloud cover field, and a three-dimensional cloud cover grid point field of multi-source data fusion is formed. Specifically, the three-dimensional reflectivity grid point data of the radar is fused on the basis: in order to avoid the influence of ground object echo on the radar reflectivity, only the cloud cover above a certain height is considered to be corrected using the radar reflectivity. If the height of a certain three-dimensional grid point from the ground is greater than 2km, and it simultaneously satisfies the two conditions that it is above the cloud bottom and the corresponding radar reflectivity exceeds the set threshold value of 10dBZ, then the cloud cover of the point is set to 1.

[0103] The method provided by the embodiment realizes complementary advantages of multi-source data, objectively analyzes the ground observation data to correct the vertical structure of the bottom and the lower part of the cloud amount field layer by layer, the satellite data supplements large-scale spatial information and corrects the vertical structure of the top and the upper part of the cloud amount field, and the radar data further refines the details of the lower part of the cloud amount field in a larger spatial range, so that the three gradually fused data overcome the limitations of single data, greatly improve the accuracy of the three-dimensional cloud amount field in the vertical and horizontal directions, accurately present the characteristics such as the position, thickness and hierarchical structure of the cloud, and provide high-quality data basis for subsequent meteorological analysis such as cloud microphysical quantity calculation and cloud type classification, and effectively improve the quality and level of meteorological services.

[0104] In one embodiment, S130, based on the satellite data, the numerical mode data, the multi-source data fused three-dimensional cloud amount grid point field, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point field, a plurality of cloud type classification models are established and trained using an XGBOOST multi-classification method, as shown in Figure 3 The model training process provided by the embodiment is specifically as follows:

[0105] The first input feature and the second input feature are constructed using the satellite data as a sample data set, and the first cloud classification model and the second cloud classification model are constructed and trained using an XGBOOST multi-classification method, and the specific model training process is as follows:

[0106] The historical data of the FY-4B satellite and the cloud classification historical data of the Fengyun-9 satellite are spatio-temporally matched to obtain a sample data set. Since meteorological data has strong spatial correlation, to ensure the independence between the training set and the validation set, the training set and the validation set are not divided in a completely random manner, but the first 80% of the data in the sample data set is used as the training set, and the last 20% of the data is used as the validation set.

[0107] The input features of the cloud type classification model one are constructed: the reflectivity or brightness temperature data of C01-C15 channels, BTD10 (the brightness temperature difference of C10 and C13 channels), BTD11 (the brightness temperature difference of C11 and C13 channels), BTD12 (the brightness temperature difference of C12 and C13 channels), and BTD13 (the brightness temperature difference of C13 and C14 channels), a total of 19 features.

[0108] The cloud type classification model one is constructed using an XGBOOST multi-classification method, as shown in Figure 4Fig. 1 shows a schematic diagram of the initialization setting of the model parameters provided by the embodiment. The historical cloud classification results of the HJ-9 satellite in the training set are taken as label values, combined with the input features corresponding to the FY-4B satellite, and the cloud type classification model one is trained. In the training process, the Bayesian hyperparameter optimization method is used to select different combinations of the six hyperparameters learnig_rate, reg_alpha, reg_lambda, gamma, max_depth and min_child_weight to train the model, and the optimal hyperparameter combination and cloud type classification model one are selected according to the evaluation results of the validation set.

[0109] The input features for constructing the cloud type classification model two are the brightness temperature data of C07-C15 channels, BTD10 (the brightness temperature difference of C10 and C13 channels), BTD11 (the brightness temperature difference of C11 and C13 channels), BTD12 (the brightness temperature difference of C12 and C13 channels), and BTD13 (the brightness temperature difference of C13 and C14 channels), a total of 13 features.

[0110] The XGBOOST multi-classification method is used to construct the cloud type classification model two, and the initialization setting of the model parameters is as shown in Fig. 2. Figure 4 The historical cloud classification results of the HJ-9 satellite in the training set are taken as label values, combined with the input features corresponding to the FY-4B satellite, and the cloud type classification model two is trained. In the training process, the Bayesian hyperparameter optimization method is used to select different combinations of the six hyperparameters learnig_rate, reg_alpha, reg_lambda, gamma, max_depth and min_child_weight to train the model, and the optimal hyperparameter combination and cloud type classification model two are selected according to the evaluation results of the validation set.

[0111] The third input feature is constructed by taking the numerical mode data, the multi-source data fused three-dimensional cloud amount grid field, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid field as sample data set. The XGBOOST multi-classification method is used to construct and train the third cloud classification model, and the model training process includes the following steps:

[0112] The historical data of the numerical mode data, the multi-source data fused three-dimensional cloud amount grid field, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid field are spatiotemporally matched with the cloud classification historical data of the HJ-9 satellite to obtain a sample data set. Due to the strong spatial correlation of meteorological data, in order to ensure the independence between the training set and the validation set, instead of using a completely random way to divide the training set and the validation set, the first 80% of the data in the sample data set is taken as the training set, and the last 20% of the data is taken as the validation set.

[0113] The input features of the cloud type classification model three are constructed: cloud layer average cloud amount, cloud layer top height, cloud layer top temperature, cloud layer thickness, cloud layer temperature, cloud layer maximum temperature gradient, cloud layer average temperature gradient, cloud layer cloud liquid water path, and cloud layer cloud ice water path, totaling 9 features. The specific steps are as follows:

[0114] Based on the three-dimensional cloud amount grid point field, the positions and average cloud amounts of each cloud layer in the vertical direction of each grid point are diagnosed: for each grid point, the first pressure layer with a cloud amount greater than 0.2 is found from high to low in the vertical direction, the middle height of the pressure layer and the pressure layer above it is taken as the cloud top height of the uppermost cloud layer, and the process continues downward until the first pressure layer with a cloud amount less than or equal to 0.2 is found, the middle height of the pressure layer and the pressure layer above it is taken as the cloud bottom height of the uppermost cloud layer. According to this method, all cloud layers in the vertical direction of the grid point can be found. For each cloud layer, search from the cloud layer top to the cloud layer bottom, only when the cloud amount of a certain pressure layer is greater than the maximum value of the cloud amount of all pressure layers above it, it is involved in the calculation of the average cloud amount, and the cloud layer average cloud amount of each cloud layer is obtained;

[0115] Based on the three-dimensional potential height and three-dimensional temperature of the numerical model data, the cloud layer top height, cloud layer top temperature and cloud layer thickness (cloud layer top height-cloud layer bottom height) of each cloud layer are calculated, the temperature of the pressure layer with the maximum cloud amount in each cloud layer is calculated as the cloud layer temperature, the temperature of all pressure layers in each cloud layer is converted to potential temperature, and the maximum potential temperature gradient and average potential temperature gradient in each cloud layer are calculated;

[0116] Based on the three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point field, the vertical integration of cloud liquid water content and cloud ice water content in each cloud layer is calculated to obtain the cloud layer cloud liquid water path and cloud layer cloud ice water path.

[0117] XGBOOST multi-classification method is used to construct cloud type classification model three, and the initialization setting of model parameters is as shown in Figure 4 The historical cloud classification results of HJ-9 satellite in the training set are taken as the label value, combined with the input features of the corresponding uppermost cloud layer, and the cloud type classification model three is trained. In the training process, the Bayesian hyperparameter optimization method is used to select different combinations of the six hyperparameters learnig_rate, reg_alpha, reg_lambda, gamma, max_depth, and min_child_weight to train the model, and the optimal hyperparameter combination and cloud type classification model three are selected according to the evaluation results of the validation set.

[0118] The method provided by the embodiment shows significant technical effects based on multi-source meteorological data, and uses an XGBOOST multi-classification method to establish and train multiple cloud type classification models. First, the first and second input features are constructed using satellite data, and the first and second cloud classification models are constructed and trained by the XGBOOST multi-classification method, which can fully mine the information in the satellite data and effectively classify the cloud type of the uppermost cloud layer, providing a basis for subsequent in-depth analysis. Second, the third input feature is constructed by using the numerical model data, the three-dimensional cloud amount grid point field obtained by multi-source data fusion, and the three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point fields calculated, and the third cloud classification model is trained. The model integrates the advantages of multi-source data and can more comprehensively grasp the characteristics of the cloud from a three-dimensional perspective. The combination of the three models can greatly improve the accuracy and comprehensiveness of cloud type classification, provide strong support for cloud type identification and related analysis in the fields of meteorological research and weather forecasting, and improve the cognitive ability of cloud-related phenomena.

[0119] In one embodiment, based on the diagnostic results of the solar elevation angle of each grid point of the multi-source data fusion three-dimensional cloud amount grid point field at the monitoring time and the vertical direction cloud layer distribution, the corresponding cloud type classification model is matched to obtain the vertical direction cloud type classification of each cloud layer, and a three-dimensional cloud type classification result is generated, which specifically includes the following steps:

[0120] In real-time operation, it is determined from the multi-source data fusion three-dimensional cloud amount grid point field whether there is cloud at the grid point, and the judgment standard for whether there is cloud is whether the maximum value of the cloud amount of all pressure layers in the vertical direction of the grid point is greater than 0.2. If there is cloud at the grid point, the solar elevation angle of the grid point at the monitoring time is calculated.

[0121] When the solar elevation angle is greater than 15°, 19 input features are constructed using FY-4B satellite data as the first input feature, the first input feature is input into the first cloud classification model, and the cloud type classification result of the uppermost cloud layer is obtained. When the solar elevation angle is not greater than 15°, 13 input features are constructed using FY-4B satellite data as the second input feature, the second input feature is input into the second cloud classification model, and the cloud type classification result of the uppermost cloud layer is obtained.

[0122] 9 input features are constructed using numerical model data, multi-source data fusion three-dimensional cloud amount grid point field, and three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point fields calculated as the third input feature, the third input feature of each cloud layer except the uppermost cloud layer is input into the third cloud classification model, and the cloud type classification result of each cloud layer except the uppermost cloud layer is obtained. Finally, the grid monitoring result of three-dimensional cloud type classification is obtained.

[0123] The method provided by the embodiment calculates the solar elevation angle of a grid point in a three-dimensional cloud amount grid point field of multi-source data fusion at a monitoring moment, and for different solar elevation angles, a first or second input feature is selected to input a corresponding first or second cloud classification model, so as to obtain the cloud type classification result of the uppermost cloud layer. The classification strategy varies with the solar elevation angle, which can fully utilize the satellite data features under different illumination conditions and improve the accuracy of cloud type recognition of the uppermost cloud layer. The third input feature of each cloud layer except the uppermost cloud layer is input into a third cloud classification model, so as to realize the classification of the cloud type of other cloud layers. All cloud layers in the vertical direction can be comprehensively considered, and the generated three-dimensional cloud type classification result covers the cloud type information of each cloud layer, which provides detailed and comprehensive cloud type distribution data for the field of meteorological research, numerical simulation and the like, and helps to improve the cognition and understanding of complex cloud systems.

[0124] Apparatus embodiment

[0125] According to the embodiment of the application, a three-dimensional cloud type classification device based on multi-source meteorological data fusion is provided. Figure 5 As shown in the figure, it is a structural schematic diagram of the three-dimensional cloud type classification device based on multi-source meteorological data fusion provided by the embodiment. The three-dimensional cloud type classification device based on multi-source meteorological data fusion according to the embodiment of the application comprises a data acquisition module 51, a data fusion module 52, a model construction module 53 and a real-time classification module 54.

[0126] The data acquisition module 51 is configured to acquire satellite data, radar data, numerical model data and ground observation data and perform preprocessing.

[0127] The data fusion module 52 adopts a three-dimensional cloud amount analysis method, takes the three-dimensional cloud amount of the numerical model data as a background field, fuses the ground observation data, the satellite data and the radar data, and forms a three-dimensional cloud amount grid point field of multi-source data fusion.

[0128] The model construction module 53 is configured to establish and train a plurality of cloud type classification models by using an XGBOOST multi-classification method based on the satellite data, the numerical model data, the three-dimensional cloud amount grid point field of multi-source data fusion and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point fields.

[0129] The real-time classification module 54 is configured to match a corresponding cloud type classification model based on the solar elevation angle of each grid point of the three-dimensional cloud amount grid point field of multi-source data fusion at a monitoring moment and the diagnostic result of the vertical cloud layer distribution, so as to obtain the cloud type classification of each cloud layer in the vertical direction and generate a three-dimensional cloud type classification result.

[0130] The device provided by the embodiment comprehensively integrates meteorological data from multiple sources, covers information from high altitude to the ground and from a wide range to a local area, makes the data more comprehensive, and improves the accuracy of meteorological data; the data fusion module 52 adopts a three-dimensional cloud amount analysis method, takes the three-dimensional cloud amount of the numerical mode data as a background field, fuses the ground observation data, the satellite data, and the radar data, forms a three-dimensional cloud amount grid field of multi-source data fusion, fully utilizes the respective advantages of different data in cloud amount, can more accurately describe the cloud amount distribution in the three-dimensional space, and thus has a clearer presentation of the spatial structure and range of the cloud layer, which is helpful for a deeper understanding of the macro features of the cloud; the model construction module 53 establishes and trains multiple cloud type classification models by using an XGBOOST multi-classification method based on the satellite data, the numerical mode data, the three-dimensional cloud amount grid field of multi-source data fusion, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid fields, so that the model can learn more rich feature information of the cloud, thereby improving the accuracy and reliability of cloud type classification and being able to more accurately distinguish different types of clouds; the real-time classification module 54 matches the corresponding cloud type classification model based on the solar elevation angle of each grid point of the three-dimensional cloud amount grid field of multi-source data fusion at the monitoring moment and the diagnostic result of the vertical cloud layer distribution, obtains the cloud type classification of each cloud layer in the vertical direction, generates a three-dimensional cloud type classification result, can more meticulously classify the cloud layer in the vertical direction, understands the distribution of the cloud layer at different altitudes and the type of the cloud, and has important significance in understanding the vertical structure and evolution of the cloud in the field of meteorological research, weather forecasting, and the like, and is helpful for more accurately predicting weather changes. The establishment of multiple cloud type classification models and the matching of the model according to different conditions make the method be able to adapt to different meteorological conditions and data characteristics. For meteorological data of different regions and different times, cloud type classification can be performed through a suitable model, the generality and adaptability of the method are improved, and the method has a wider application scenario.

[0131] In one embodiment, the data fusion module 52 includes a ground observation data fusion unit, a satellite data fusion unit, and a radar data fusion unit.

[0132] The ground observation data fusion unit is configured to use an objective analysis method to fuse a cloud amount vertical profile of ground observation by pressure layer, correct the structural distribution of the three-dimensional cloud amount field, and update the three-dimensional cloud amount field, with the three-dimensional cloud amount of the numerical mode data as a background field.

[0133] The satellite data fusion unit is configured to fuse the satellite data based on the updated three-dimensional cloud amount field, correct the structural distribution of the three-dimensional cloud amount field, and update the three-dimensional cloud amount field.

[0134] The radar data fusion unit is configured to fuse the radar data based on the updated three-dimensional cloud amount field, correct the structural distribution of the three-dimensional cloud amount field, and form a three-dimensional cloud amount grid field fused by multiple sources of data.

[0135] The device provided in the embodiment is used to fuse ground observation data, satellite data and radar data in sequence to form a three-dimensional cloud amount grid field fused by multiple sources of data, taking the three-dimensional cloud amount of the numerical mode data as a background field, so that the advantages of multiple sources of data are complementary, the ground observation data is used to correct the vertical structure of the bottom and the lower part of the cloud amount field by layer by layer using an objective analysis method, the satellite data is used to supplement large-scale spatial information and correct the vertical structure of the top and the upper part of the cloud amount field, and the radar data is used to further refine the details of the lower part of the cloud amount field in a larger spatial range, so that the three are gradually fused to overcome the limitations of single data, greatly improve the accuracy of the three-dimensional cloud amount field in the vertical and horizontal directions, accurately present the characteristics such as the position, thickness and hierarchical structure of the cloud, provide a high-quality data basis for subsequent meteorological analysis such as cloud microphysical quantity calculation and cloud type classification, and effectively improve the quality and level of meteorological services.

[0136] The embodiment of the application is a device corresponding to the above-mentioned method embodiment, and the specific operations of each module processing step can be understood with reference to the description of the method embodiment, which will not be repeated here.

[0137] As shown in Figure 6 The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the three-dimensional cloud type classification method based on the fusion of multiple sources of meteorological data in the above-mentioned embodiment, or the computer program is executed by the processor to implement the three-dimensional cloud type classification method based on the fusion of multiple sources of meteorological data in the above-mentioned embodiment.

[0138] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0139] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, it is described more simply, and the relevant part can be referred to the part of the method embodiment. The above-described device and system embodiments are only illustrative, and the units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0140] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can still be modified, or some or all of the technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and the contents not described in detail in the specification of the present application belong to the common knowledge of those skilled in the art.

Claims

1. A three-dimensional cloud type classification method based on multi-source meteorological data fusion, characterized in that, The method comprises the following steps: acquiring satellite data, radar data, numerical model data and ground observation data and preprocessing the data; adopting a three-dimensional cloud amount analysis method, taking three-dimensional cloud amount of the numerical model data as a background field, fusing the ground observation data, the satellite data and the radar data to form a three-dimensional cloud amount grid field of multi-source data fusion; based on the satellite data, the numerical model data, the three-dimensional cloud amount grid field of multi-source data fusion and calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid fields, establishing and training multiple cloud type classification models by using an XGBOOST multi-classification method; based on the solar elevation angle of each grid point of the three-dimensional cloud amount grid field of multi-source data fusion at a monitoring time and the diagnostic result of vertical cloud layer distribution, matching a corresponding cloud type classification model to obtain vertical cloud type classification of each cloud layer and generate a three-dimensional cloud type classification result. 2.The method of claim 1, wherein, The method of adopting a three-dimensional cloud amount analysis method, taking three-dimensional cloud amount of the numerical model data as a background field, fusing the ground observation data, the satellite data and the radar data to form a three-dimensional cloud amount grid field of multi-source data fusion comprises the following steps: taking the three-dimensional cloud amount of the numerical model data as a background field, using an objective analysis method to fuse a cloud amount vertical profile of ground observation at each pressure layer, correcting the structural distribution of the three-dimensional cloud amount field to obtain an updated three-dimensional cloud amount field of ground observation; fusing the satellite data based on the updated three-dimensional cloud amount field of ground observation, correcting the structural distribution of the three-dimensional cloud amount field to obtain an updated three-dimensional cloud amount field of satellite data; fusing the radar data based on the updated three-dimensional cloud amount field of satellite data, correcting the structural distribution of the three-dimensional cloud amount field to form a three-dimensional cloud amount grid field of multi-source data fusion. 3.The method of claim 2, wherein, The method of fusing the satellite data based on the updated three-dimensional cloud amount field of ground observation, correcting the structural distribution of the three-dimensional cloud amount field comprises the following steps: using a cloud amount analysis method in GSI to fuse satellite cloud detection data, cloud top pressure data and cloud top temperature data to correct the cloud top distribution of the three-dimensional cloud amount field; using a cloud amount analysis method in LAPS to fuse satellite L1 data to correct the upper structure of the three-dimensional cloud amount field. 4.The method of claim 1, wherein, The method of establishing and training multiple cloud type classification models based on the satellite data, the numerical model data, the three-dimensional cloud amount grid field of multi-source data fusion and calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid fields by using an XGBOOST multi-classification method comprises the following steps: constructing first input features and second input features by taking the satellite data as a sample data set, constructing and training first cloud classification models and second cloud classification models by using an XGBOOST multi-classification method; constructing third input features by taking the numerical model data, the three-dimensional cloud amount grid field of multi-source data fusion and calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid fields as a sample data set, constructing and training third cloud classification models by using an XGBOOST multi-classification method.

5. The method of claim 4, wherein the three-dimensional cloud type classification based on multi-source meteorological data fusion is characterized by, The third input feature is constructed based on the sample data set of the numerical mode data, the multi-source data fused three-dimensional cloud amount grid point field, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point field, and further includes the following steps: Diagnose the position and average cloud amount of each cloud layer in the vertical direction based on the three-dimensional cloud amount grid point field; Calculate the cloud layer top height, cloud layer top temperature, cloud layer thickness, cloud layer temperature, maximum temperature gradient in the cloud layer, and average temperature gradient in the cloud layer based on the three-dimensional potential height and three-dimensional temperature of the numerical mode data; Calculate the cloud liquid water path and cloud ice water path of each cloud layer based on the three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point field; Construct the cloud layer average cloud amount, cloud layer top height, cloud layer top temperature, cloud layer thickness, cloud layer temperature, maximum temperature gradient in the cloud layer, average temperature gradient in the cloud layer, cloud layer cloud liquid water path, and cloud layer cloud ice water path as the third input feature; Train the third cloud classification model based on the third input feature of the uppermost cloud layer. 6.The method of claim 4, wherein, The solar elevation angle of each grid point of the multi-source data fused three-dimensional cloud amount grid point field at the monitoring time and the diagnostic result of the vertical cloud layer distribution are matched with the corresponding cloud type classification model to obtain the cloud type classification of each cloud layer in the vertical direction and generate a three-dimensional cloud type classification result, which specifically includes the following steps: When it is diagnosed in the multi-source data fused three-dimensional cloud amount grid point field that there is cloud at the grid point, calculate the solar elevation angle of the grid point at the monitoring time; When the solar elevation angle is greater than 15°, input the first input feature into the first cloud classification model to obtain the cloud type classification result of the uppermost cloud layer, and when the solar elevation angle is not greater than 15°, input the second input feature into the second cloud classification model to obtain the cloud type classification result of the uppermost cloud layer; Input the third input feature of each cloud layer except the uppermost cloud layer into the third cloud classification model to obtain the cloud type classification result of each cloud layer except the uppermost cloud layer.

7. A three-dimensional cloud type classification device based on multi-source meteorological data fusion, characterized by, It includes a data acquisition module, a data fusion module, a model construction module, and a real-time classification module; The data acquisition module is used to acquire satellite data, radar data, numerical mode data, and ground observation data and perform preprocessing; The data fusion module is used to adopt a three-dimensional cloud amount analysis method, take the three-dimensional cloud amount of the numerical mode data as a background field, fuse the ground observation data, the satellite data, and the radar data, and form a multi-source data fused three-dimensional cloud amount grid point field; The model construction module is used to establish and train multiple cloud type classification models based on the satellite data, the numerical mode data, the multi-source data fused three-dimensional cloud amount grid point field, and the calculated three-dimensional cloud liquid water content and three-dimensional cloud ice water content grid point field, and utilize an XGBOOST multi-classification method; The real-time classification module is used to match the corresponding cloud type classification model based on the diagnostic result of the solar elevation angle of each grid point of the multi-source data fused three-dimensional cloud amount grid point field at the monitoring time and the vertical cloud layer distribution, obtain the cloud type classification of each cloud layer in the vertical direction, and generate a three-dimensional cloud type classification result. 8.The three-dimensional cloud type classification device based on multi-source meteorological data fusion of claim 7, wherein, The data fusion module comprises a ground observation data fusion unit, a satellite data fusion unit and a radar data fusion unit; The ground observation data fusion unit is configured to use a three-dimensional cloud amount of the numerical mode data as a background field, fuse a cloud amount vertical profile of ground observation by using an objective analysis method, correct a structure distribution of the three-dimensional cloud amount field, and obtain an updated three-dimensional cloud amount field of the ground observation; The satellite data fusion unit is configured to fuse the satellite data based on the updated three-dimensional cloud amount field of the ground observation, correct the structure distribution of the three-dimensional cloud amount field, and obtain an updated three-dimensional cloud amount field of the satellite data; The radar data fusion unit is configured to fuse the radar data based on the updated three-dimensional cloud amount field of the satellite data, correct the structure distribution of the three-dimensional cloud amount field, and form a three-dimensional cloud amount grid field of multi-source data fusion.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the three-dimensional cloud type classification method based on multi-source meteorological data fusion according to any one of claims 1 to 6.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the steps of the three-dimensional cloud type classification method based on multi-source meteorological data fusion according to any one of claims 1 to 6.

Images