Temperature and humidity profile inversion method and device combining infrared and microwave data
By combining infrared and microwave data to retrieve temperature and humidity profiles, and using a deep learning model to fuse feature extraction and regression, the problem of insufficient training data and computing resources was solved, and high-precision temperature and humidity profile retrieval was achieved under both cloudy and clear sky conditions.
Patent Information
- Application Number
- CN202411108134.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing methods for inverting temperature and humidity profiles struggle to achieve high accuracy when training data and computational resources are limited.
A method for inverting temperature and humidity profiles using combined infrared and microwave data is employed. Feature vectors of radiation and brightness temperature data are extracted through a feature extraction module, and after feature concatenation, they are input into a regression module for inversion. A deep learning model is used to fuse infrared and microwave data.
With limited training data and computing resources, the accuracy of temperature and humidity profile inversion has been improved, especially under cloudy and clear sky conditions, the inversion results are more accurate.
Smart Images

Figure CN118940226A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a temperature and humidity profile inversion method, and in particular to a temperature and humidity profile inversion method and device combining infrared and microwave data. Background Art
[0002] Microwave radiometer (MWR) is mainly used in the field of atmospheric remote sensing to measure the intensity of atmospheric microwave radiation. The instrument can obtain the intensity of microwave radiation in the atmosphere, and then invert parameters such as the temperature and humidity profile of the atmosphere. Ground-based microwave radiometers are based on microwave radiation transmission theory and thermodynamic equilibrium principles, and their measurement range usually covers the troposphere and stratosphere in the atmosphere. Due to the long wavelength of microwaves, they can penetrate clouds, haze, and dust in the atmosphere, and provide information on cloud characteristics and above the clouds. At present, MWR has been widely used in the field of meteorology. Among them, two-channel and three-channel MWR are mainly used to detect column totals such as Precipitable Water Vapor (PWV) and Liquid Water Path (LWP). Multi-channel MWR can be used for the inversion of atmospheric temperature and humidity profiles, mainly obtaining microwave radiation data of 22-30GHz water vapor sensitive channels and 50-60GHz oxygen sensitive channels.
[0003] On this basis, the commonly used inversion algorithms for temperature and humidity profiles using microwave radiometers include the least squares inversion algorithm, iterative inversion algorithm, Bayesian inversion algorithm, artificial neural network inversion algorithm and other methods. These algorithms each have their own advantages and disadvantages, and the appropriate algorithm needs to be selected according to the actual situation.
[0004] There are many methods for inverting temperature and humidity profiles using ground-based microwave radiometers, and the most representative ones are the least squares inversion algorithm, iterative inversion algorithm, Bayesian inversion algorithm, and artificial neural network inversion algorithm. This section briefly introduces the relevant algorithms and their characteristics. The least squares method is a commonly used inversion algorithm that can estimate the temperature and humidity profiles by minimizing the residual sum of squares between the observed value and the simulated value. The advantages of this method are that it is simple and easy to understand and has a fast calculation speed, but the disadvantage is that it is sensitive to measurement errors and data quality control is required to ensure the accuracy of the inversion results. The iterative inversion algorithm can optimize the inversion results through multiple iterative calculations, including iterative inversion algorithms based on vertical polarization and horizontal polarization channels and iterative inversion algorithms based on multiple channels. The advantage of this method is that it can reduce the impact of measurement errors, but the disadvantage is that the calculation time is long. The Bayesian inversion algorithm can use prior information and observational data to infer the temperature and humidity profiles, where the prior information can be obtained through meteorological models, etc. The advantage of this method is that it can effectively use prior information to improve the accuracy and credibility of the inversion results, but the disadvantage is that it requires modeling and calculation of prior information, and the calculation complexity is high. The artificial neural network inversion algorithm can realize the inversion of temperature and humidity profiles by training neural networks, where the input data is the microwave radiation intensity measured by the microwave radiometer, and the output data is the temperature and humidity profile. The advantage of this method is that the inversion results can be adjusted adaptively, but the disadvantage is that it requires a large amount of training data and computing resources.
[0005] It can be seen from this that the existing temperature and humidity profile inversion methods are difficult to obtain high-precision temperature and humidity profiles when there are less training data and computing resources. Summary of the invention
[0006] The technical problem to be solved by the present invention is that the above-mentioned methods commonly used in the prior art are difficult to obtain temperature and humidity profiles with high accuracy when there are less training data and computing resources. In order to solve the above-mentioned problem, the present invention provides a temperature and humidity profile inversion method and device combining infrared and microwave data.
[0007] The content of the present invention includes:
[0008] In a first aspect, an embodiment of the present invention provides a method for inverting a temperature and humidity profile by combining infrared and microwave data, comprising:
[0009] Inputting the radiation value data and the brightness temperature data into a feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data;
[0010] Inputting the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data into a feature splicing module for feature splicing to obtain a joint feature vector;
[0011] The joint feature vector is input into a regression module for inversion to obtain an inversion result.
[0012] Optionally, the feature extraction module includes an AERI feature extraction module and an MWR feature extraction module, and the inputting of the radiation value data and the brightness temperature data into the feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data includes:
[0013] Inputting the radiation value data into the AERI feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data;
[0014] The brightness temperature data is input into the MWR feature extraction module for feature extraction to obtain a feature vector corresponding to the brightness temperature data.
[0015] Optionally, the step of inputting the radiation value data into the AERI feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data includes:
[0016] The radiation value data is input into the AERI feature extraction module for nonlinear feature extraction to obtain a first output feature; wherein the AERI feature extraction module includes:
[0017] A first convolution block, the first convolution block comprising a first convolution layer, a first processing layer group and a first maximum pooling layer connected in sequence, the first processing layer group comprising a first batch normalization layer and a first output function layer;
[0018] A second convolution block, the second convolution block comprising a second convolution layer and the first processing layer group connected in sequence;
[0019] A third convolution block, the third convolution block comprising a third convolution layer and the first processing layer group connected in sequence;
[0020] A fourth convolution block, wherein the fourth convolution block includes a fourth convolution layer and the first processing layer group connected in sequence.
[0021] Optionally, the step of inputting the brightness temperature data into the MWR feature extraction module for feature extraction to obtain a feature vector corresponding to the brightness temperature data includes:
[0022] The brightness temperature data is input into the MWR feature extraction module for nonlinear feature extraction to obtain a first output feature; wherein the MWR feature extraction module includes:
[0023] a fifth convolutional block, the fifth convolutional block comprising a fifth convolutional layer, a second processing layer group and a first upsampling layer connected in sequence, the second processing layer group comprising a second batch normalization layer and a second output function layer;
[0024] a sixth convolution block, the sixth convolution block comprising a sixth convolution layer, the second processing layer group, and a second upsampling layer connected in sequence;
[0025] a seventh convolutional block, the seventh convolutional block comprising a seventh convolutional layer and the second processing layer group connected in sequence;
[0026] An eighth convolution block, wherein the eighth convolution block includes an eighth convolution layer and the second processing layer group connected in sequence.
[0027] Optionally, inputting the joint feature vector into a regression module for inversion to obtain an inversion result includes:
[0028] The joint feature vector is input into a regression module for regression calculation to obtain an inversion result, wherein the regression module includes in sequence: a first fully connected block, the first fully connected block includes a first fully connected layer, a third output function and a discard layer connected in sequence; a second fully connected block, the second fully connected block includes a second fully connected layer and a fourth output function connected in sequence.
[0029] Optionally, before inputting the radiation value data and the brightness temperature data into a feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data, the method further includes:
[0030] The radiation value data are obtained through infrared radiometer observations, and the brightness temperature data are obtained through microwave radiometer observations.
[0031] Optionally, before inputting the radiation value data and the brightness temperature data into a feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data, the method further includes:
[0032] The radiation value sample data is obtained through infrared radiometer observation, the brightness temperature sample data is obtained through microwave radiometer observation, and the corresponding supervision data is obtained through radiosonde;
[0033] The radiation value sample data and the brightness temperature sample data are used as training samples, and the supervision data are used as true labels to iteratively train the joint depth inversion model, which includes the feature extraction module, the feature splicing module and the regression module.
[0034] In a second aspect, an embodiment of the present invention provides a temperature and humidity profile inversion device combining infrared and microwave data, comprising:
[0035] A feature extraction module, used for inputting the radiation value data and the brightness temperature data into the feature extraction module for feature extraction, and obtaining a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data;
[0036] A feature splicing module, used for inputting the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data into the feature splicing module for feature splicing to obtain a joint feature vector;
[0037] The regression module is used to input the joint feature vector into the regression module for inversion to obtain an inversion result.
[0038] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory, a processor, and a program stored in the memory and executable on the processor; the processor is used to read the program in the memory to implement the steps in the method for inverting the temperature and humidity profile of the combined infrared and microwave data as described in the first aspect.
[0039] In a fourth aspect, an embodiment of the present invention provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the method for inverting the temperature and humidity profile of the combined infrared and microwave data as described in the first aspect.
[0040] In an embodiment of the present invention, the radiation value data and the brightness temperature data are input into a feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data; the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data are input into a feature splicing module for feature splicing to obtain a joint feature vector; the joint feature vector is input into a regression module for inversion to obtain an inversion result. The beneficial effects of the present invention are: by inverting the temperature and humidity profile by combining the radiation value data and the brightness temperature data, the accuracy of the obtained temperature and humidity profile can be improved. At the same time, since the model structure used is simple, a temperature and humidity profile with higher accuracy can be inverted with less training data and computing resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Attached Figure 1 A flow chart of a method for inverting temperature and humidity profiles of combined infrared and microwave data provided by an embodiment of the present invention;
[0042] Attached Figure 2 A schematic diagram of the architecture of a Joint Deep Retrieval Architecture (JDReA) model provided in an embodiment of the present invention;
[0043] Attached Figure 3a A schematic diagram of the structure of the JDReA model for temperature profile inversion provided by an embodiment of the present invention;
[0044] Attached Figure 3b A schematic diagram of the structure of the JDReA model for humidity profile inversion provided by an embodiment of the present invention;
[0045] Attached Figure 4 Scatter plot between the inversion results of temperature and the measurement results of radiosonde (clear sky and cloudy sky);
[0046] Attached Figure 5 It is the scatter plot between the inversion result of water vapor mixing ratio and the true value of radiosonde (clear sky and cloudy sky);
[0047] Attached Figure 6 is the temperature retrieval bias of DReA and JDReA at different altitudes (clear sky and cloudy);
[0048] Attached Figure 7 is the inversion deviation of water vapor mixing ratio at different altitudes (clear sky and cloudy sky);
[0049] Attached Figure 8a This is the time height map of the sounding reanalysis data when the cold front passes;
[0050] Attached Figure 8b It is the time height diagram of DReA method when the cold front passes;
[0051] Attached Figure 8c This is the time height diagram of the JDReA method when the cold front passes;
[0052] Attached Figure 9a It is the temperature profile at 05:30UTC when the cold front passes;
[0053] Attached Figure 9b It is the temperature profile at 23:30UTC when the cold front passes;
[0054] Attached Fig.9c It is the humidity profile at 05:30UTC when the cold front passes;
[0055] Attached Figure 9d It is the humidity profile at 23:30UTC when the cold front passes;
[0056] Attached Fig.10a It is the temperature profile at 11:30 UTC when the cold front passes;
[0057] Attached Fig.10b It is the temperature profile at 17:30UTC when the cold front passes;
[0058] Attached Fig.10c It is the humidity profile at 11:30 UTC when the cold front passes;
[0059] Attached Fig.10dIt is the humidity profile at 17:30UTC when the cold front passes;
[0060] Attached Fig.11a This is the time height map of the sounding reanalysis data when the warm front passes;
[0061] Attached Fig.11b This is the time height diagram of the DReA method when a warm front passes;
[0062] Attached Fig.11c This is the time height diagram of the JDReA method when the warm front passes;
[0063] Attached Fig.12 is a scatter plot between the inversion results of air temperature and the measurement results of radiosonde (with clouds);
[0064] Attached Fig.13 The scatter plot between the inversion result of water vapor mixing ratio and the true value of the radiosonde (with clouds);
[0065] Attached Fig.14 is the temperature inversion deviation at different altitudes (with clouds);
[0066] Attached Fig.15 is the water vapor mixing ratio inversion deviation at different altitudes (with clouds);
[0067] Attached Fig.16 is a scatter plot between the inversion results of temperature and the measurement results of radiosonde (clear sky);
[0068] Attached Fig.17 It is a scatter plot between the inversion result of water vapor mixing ratio and the true value of the radiosonde (clear sky);
[0069] Attached Fig.18 is the temperature inversion deviation at different altitudes (clear sky);
[0070] Attached Fig.19 is the water vapor mixing ratio inversion deviation at different altitudes (clear sky);
[0071] Attached Fig. 20 A schematic diagram of a temperature and humidity profile inversion device combining infrared and microwave data provided by an embodiment of the present invention;
[0072] Attached Fig.21 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0073] In the embodiments of the present application, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. In the embodiments of the present application, the term "multiple" refers to two or more, and other quantifiers are similar. The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first" and "second" are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more.
[0074] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0075] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0076] See also Figure 1 , Figure 1 A schematic flow chart of a method for inverting temperature and humidity profiles using combined infrared and microwave data provided by an embodiment of the present invention. The method specifically includes the following steps:
[0077] Step 101, inputting the radiation value data and the brightness temperature data into a feature extraction module for feature extraction, and obtaining a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data;
[0078] Step 102, inputting the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data into a feature splicing module for feature splicing to obtain a joint feature vector;
[0079] Step 103: input the joint feature vector into a regression module for inversion to obtain an inversion result.
[0080] It should be understood that the inversion results include temperature profiles and / or humidity profiles. In specific implementations, the specific structures of the models used for inversion of temperature profiles and humidity profiles (i.e., the structure of the feature extraction module, the structure of the feature splicing module, and the structure of the regression module) may be the same or different. The corresponding inversion results can be obtained by using the corresponding model structure according to actual needs.
[0081] In this embodiment, the radiation value data and the brightness temperature data are combined to perform complex nonlinear projection to invert the atmospheric profile. Optionally, in some embodiments, before step 101, the method further includes:
[0082] The radiation value data are obtained through infrared radiometer observations, and the brightness temperature data are obtained through microwave radiometer observations.
[0083] It should be understood that the radiation value data is the real data observed by the infrared radiometer, which can also be called AERI data or AERI observations, and the brightness temperature data is the real data observed by the microwave radiometer, which can also be called MWR data or MWR observations. In the specific implementation, the radiation value data and the brightness temperature data are data that have undergone quality control and error processing, and the radiation value data and the brightness temperature data are matched in time and space.
[0084] In this embodiment, the radiation value data is obtained by observing with an infrared radiometer, and the brightness temperature data is obtained by observing with a microwave radiometer, so that real radiation value data and brightness temperature data can be obtained. Compared with the simulated data, the data obtained in this embodiment is more accurate and more realistic, thereby further improving the precision and accuracy of the temperature and humidity profile obtained by inversion.
[0085] See also Figure 2 In this embodiment, a joint deep retrieval architecture (JDReA) model is also provided. The JDReA model includes the feature extraction module, the feature concatenation module and the regression module. The above steps 101 to 103 can be regarded as the forward propagation process of the JDReA network.
[0086] Optionally, in some embodiments, before step 101, the method further includes: obtaining radiation value sample data through infrared radiometer observations, obtaining brightness temperature sample data through microwave radiometer observations, and obtaining corresponding supervision data through a radiosonde; using the radiation value sample data and the brightness temperature sample data as training samples, and using the supervision data as true labels to iteratively train a joint depth inversion model, wherein the joint depth inversion model includes the feature extraction module, the feature stitching module and the regression module.
[0087] In this embodiment, the supervision data obtained by the radiosonde is used as the supervision signal to perform backward feedback training on the JDReA network, and the parameters of the joint depth inversion model are optimized so that the joint depth inversion model is continuously approached to more correct parameter values. In this embodiment, the constructed JDReA uses the real data of the radiosonde as the supervision signal, and its potential nonlinear relationship with the joint observation data of AERI and MWR can be mined.
[0088] For the convenience of description, the parameters involved in this application are described below. Assume that the observed value (i.e., radiation value data) obtained by infrared radiometer AERI is The observation value (i.e. brightness temperature data) obtained by the microwave radiometer MWR is The corresponding true value (i.e., supervisory data) obtained from the radiosonde is in, It is used to characterize the observed value of infrared radiometer AERI at any time (time i), It is used to represent the observed value of the microwave radiometer MWR at any time (time i), x i A vector representing the radiosonde measurement at any time (i), N is the total number of data of a type obtained. In this embodiment, the symbol X is used to represent the temperature value X t Or humidity value X q , using the same x i To define an arbitrary temperature value vector or a humidity vector Among them, t j and q j They represent the air temperature and humidity (i.e., water vapor mixing ratio) at the jth vertical height, respectively. m represents the total number of vertical heights, and T represents the transposition operation.
[0089] The forward propagation process based on AERI observations and MWR observations will be introduced in detail below. The forward propagation process includes feature extraction process, feature splicing process and regression process.
[0090] The feature extraction module is used to extract the nonlinear features between the detected AERI observations and MWR observations and the true values of the radiosonde. Optionally, in some embodiments, the feature extraction module includes an AERI feature extraction module and an MWR feature extraction module, and step 101 includes:
[0091] The radiation value data is input into the AERI feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data; the brightness temperature data is input into the MWR feature extraction module for feature extraction to obtain a feature vector corresponding to the brightness temperature data.
[0092] The forward propagation process of JDReA is divided into feature extraction process, feature concatenation process and regression process, where the feature extraction process is completed by two modules: AERI feature extraction module and MWR feature extraction module. The AERI feature extraction module is represented by F θ , the trainable network parameter of AERI feature extraction module is θ. MWR feature extraction module is represented by G α , the trainable network parameter of the MWR feature extraction module is α.
[0093] The AERI feature extraction module can extract representative high-level abstract features from the radiation value data. The input data of the AERI feature extraction module is Y a , where y a Y a Input radiation value data and obtain the corresponding feature representation through the AERI feature extraction module, where the features generated by the AERI feature extraction module satisfy:
[0094]
[0095] Among them, F θ Used to characterize the AERI feature extraction module, It is used to characterize the observed value of the infrared radiometer AERI at any time (time i), and θ is used to characterize the trainable network parameters of the AERI feature extraction module. θ It consists of a one-dimensional convolutional layer, a batch normalization layer (Batch Normalization, BN), a pooling layer and a ReLU layer.
[0096] Optionally, as a specific embodiment, the AERI feature extraction module includes:
[0097] A first convolution block, the first convolution block comprising a first convolution layer, a first processing layer group and a first maximum pooling layer connected in sequence, the first processing layer group comprising a first batch normalization layer and a first output function layer;
[0098] A second convolution block, the second convolution block comprising a second convolution layer and the first processing layer group connected in sequence;
[0099] A third convolution block, the third convolution block comprising a third convolution layer and the first processing layer group connected in sequence;
[0100] A fourth convolution block, wherein the fourth convolution block includes a fourth convolution layer and the first processing layer group connected in sequence.
[0101] As a specific embodiment, the first convolution layer includes 200 channels, the convolution kernel size is 5, the second convolution layer includes 400 channels, the convolution kernel size is 5, the third convolution layer includes 500 channels, the convolution kernel size is 5, and the fourth convolution layer includes 550 channels, the convolution kernel size is 5. Each convolution layer is followed by BN and ReLU (i.e., the first processing layer group). The size of the first maximum pooling layer is 2, and its span is 2.
[0102] The MWR feature extraction module extracts representative high-level abstract features from brightness temperature data. The input data is Y m , where y m Y m As an example, input brightness temperature data and obtain the corresponding feature representation through the MWR feature extraction module. The features generated by the MWR feature extraction module meet the following requirements:
[0103]
[0104] Among them, G α Used to characterize the MWR feature extraction module, It is used to characterize the observed value of the microwave radiometer MWR at any time (time i), and α is used to characterize the trainable network parameters of the MWR feature extraction module.
[0105] The MWR data consists of two channels, and a dimensionality increase operation is performed on it. Specifically, G α It is composed of a one-dimensional convolutional layer, a BN layer, an upsampling layer and a ReLU layer. Optionally, as a specific embodiment, the MWR feature extraction module includes the following connected in sequence:
[0106] A fifth convolution block, the fifth convolution block includes a fifth convolution layer, a second processing layer group and a first upsampling layer connected in sequence, the second processing layer group includes a second batch normalization layer and a second output function layer; a sixth convolution block, the sixth convolution block includes a sixth convolution layer, the second processing layer group and a second upsampling layer connected in sequence; a seventh convolution block, the seventh convolution block includes a seventh convolution layer and the second processing layer group connected in sequence; an eighth convolution block, the eighth convolution block includes an eighth convolution layer and the second processing layer group connected in sequence.
[0107] As a specific embodiment, the fifth convolution layer includes 100 channels, the convolution kernel size is 1, the sixth convolution layer includes 150 channels, the convolution kernel size is 1, the seventh convolution layer includes 200 channels, the convolution kernel size is 1, and the eighth convolution layer includes 250 channels, the convolution kernel size is 1. Each convolution layer is followed by BN and ReLU (i.e., the second processing layer group). In view of the characteristics of MWR data, upsampling layers are connected to the ReLU of the fifth and sixth convolution layers respectively.
[0108] In step 102, the features generated by the AERI feature extraction module and the features generated by the MWR feature extraction module are input into the feature splicing module for splicing. θ What you get and G α What you get Input to the feature concatenation module, the obtained joint feature vector satisfies:
[0109]
[0110] Among them, Concat. represents the feature concatenation process, whose goal is to and Splice into vector z i , the splicing process is as follows:
[0111]
[0112] The result obtained by the feature concatenation module is passed as an input parameter to the regression module, which is represented by H φ , H φ It is composed of a fully connected layer (FC) and a ReLU layer. Optionally, as a specific embodiment, the regression module includes:
[0113] A first fully connected block, the first fully connected block comprising a first fully connected layer, a third output function, and a discard layer connected in sequence;
[0114] A second fully connected block, wherein the second fully connected block includes a second fully connected layer and a fourth output function connected in sequence.
[0115] The regression module is denoted as H φ , the trainable network parameter of the regression module is φ, and the joint feature vector z i Input the regression module for inversion, and the output results satisfy:
[0116] x i =H φ (z i );
[0117] As a specific embodiment, the first fully connected layer includes 1000 neurons, and the second fully connected layer includes 37 neurons. The first fully connected layer is followed by ReLU and Dropout in sequence, and the second fully connected layer is followed by ReLU and outputs the inverted temperature and humidity profile.
[0118] The specific process of JDReA feedback training is described in detail below. Radiosonde data is used as the supervisory signal, and JDReA parameters are optimized to make JDReA approach more correct parameter values. In the specific implementation, the model can be trained by minimizing the objective function to propagate the gradient.
[0119] The objective function of JDReA optimization satisfies:
[0120]
[0121] Among them, L(F,G,θ,α,φ) represents the learning function, which is used to optimize the network parameters θ,α,φ of each module. V represents the vector x i Length.
[0122] As a specific implementation method, the specific structure of the JDReA model can be trained separately to obtain two JDReA models with different model structures and / or parameters for inverting the temperature profile and humidity profile. For the inversion of the temperature profile, the objective function satisfies:
[0123]
[0124] Among them, θ t The network parameters, α, used to represent the AERI feature extraction module of the JDReA model for air temperature profile inversion t The network parameters, φ, used to represent the MWR feature extraction module of the JDReA model for temperature profile inversion t Network parameters used to represent the regression module of the JDReA model for air temperature profile inversion. is to use the corresponding observed value The inverted value of the temperature profile.
[0125] Similarly, for the inversion of humidity profile, the objective function satisfies:
[0126]
[0127] Among them, θ q The network parameters, α, used to represent the AERI feature extraction module of the JDReA model for air temperature profile inversion q The network parameters, φ, used to represent the MWR feature extraction module of the JDReA model for temperature profile inversion q Network parameters used to represent the regression module of the JDReA model for air temperature profile inversion. is to use the corresponding observed value The inverted value of the temperature profile.
[0128] For ease of understanding, a specific embodiment is used as an example for description below.
[0129] See also Figure 3a and Figure 3b The input data of the model are the 23.8 GHz and 31.4 GHz channels measured by the MWR instrument and the selected channel data of the ground-based infrared hyperspectral spectrum, and the output data are the corresponding atmospheric temperature and humidity profiles. As a specific embodiment, different JDReA models are trained for the inversion of temperature profile and humidity profile respectively. The two models are slightly different in structure and network parameters, as shown in the following example. Figure 3a and Figure 3b As shown, no further elaboration is given here.
[0130] The JDReA model consists of three parts: feature extraction module, feature splicing module and regression module. Figure 3a The schematic diagram of the structure of the JDReA model for temperature profile inversion. For example, the AERI feature extraction module has four one-dimensional convolutional layers: Conv1 (200 channels, convolution kernel size 5), Conv2 (400 channels, convolution kernel size 5), Conv3 (500 channels, convolution kernel size 5), Conv4 (550 channels, convolution kernel size 5). Each convolutional layer is followed by BN, ReLU and 1D maximum pooling layers. The size of the 1D maximum pooling layer is 2, and its stride is 2.
[0131] Figure 3b The schematic diagram of the structure of the JDReA model for humidity profile inversion is shown in Figure 1. For example, the MWR feature extraction module consists of four one-dimensional convolutional layers, namely Conv1 (100 channels, convolution kernel size is 1), Conv2 (150 channels, convolution kernel size is 1), Conv3 (200 channels, convolution kernel size is 1), and Conv4 (250 channels, convolution kernel size is 1). Each convolutional layer is followed by BN and ReLU in sequence. According to the characteristics of the MWR data, upsampling is performed after the ReLU of the first convolutional layer and the second convolutional layer. For the data of both, the feature splicing module is first used to splice them into a whole, and then they are processed by the linear regression module. The linear regression module used is composed of two fully connected layers, namely FC1 (1000 neurons) and FC2 (37 neurons). The FC1 fully connected layer is followed by ReLU and Dropout in sequence, and FC2 is followed by ReLU and outputs the inverted humidity profile.
[0132] In order to verify the effectiveness of inversion by introducing ground-based microwave radiometer data, in this embodiment, the inversion results are compared with those using only infrared radiometers (i.e., the DReA inversion method described below). By comparing and analyzing the results of the two inversion methods, the accuracy and stability of the inversion method by introducing ground-based microwave radiometer data can be evaluated more objectively, and the inversion algorithm can be further optimized.
[0133] The JDReA model is trained using the Adam optimizer with a learning rate of 0.0001. The learning rate annealing strategy used is: η p =η 0 (1+αp) -β , where p represents the change in training progress from 0 to the maximum number of iterations, where α = 0.001 and β = 0.75. η 0 is the initial learning rate, and its value is 0.0001.
[0134] Please refer to Table 1, which shows the average inversion error of temperature under different weather conditions. The root mean square error (RMSE) and mean absolute error (MAE) are used to measure the inversion results of DReA and JDReA. The data inverted by DReA only includes AERI data, while the data inverted by JDReA provided in the embodiment of the present invention is the joint data of MWR and AERI.
[0135] Table 1 Average temperature inversion error under different conditions (including overall, cloudy, and clear sky conditions)
[0136]
[0137] First, a horizontal analysis is conducted on Table 1. Under the overall (cloudy and clear) conditions, the RMSE of the temperature inverted by the DReA method is 1.2801K, and the MAE is 0.7084K. The RMSE of the temperature inverted by the JDReA method is 1.2425K, and the MAE is 0.7020K. Through simple calculation, under the overall (cloudy and clear) conditions, the RMSE of JDReA is 0.0424K less than that of DReA, and the MAE of JDReA is 0.0064K less than that of DReA, indicating that JDReA has a better inversion effect under the overall (cloudy and clear) conditions. In cloudy conditions, the RMSE of the temperature inverted by the DReA method is 1.9571K, and the MAE is 0.9793K. The RMSE of the temperature inverted by the JDReA method is 1.8941K, and the MAE is 0.9877K. Through simple calculation, in cloudy conditions, the RMSE of JDReA is 0.063K less than that of DReA, and the MAE of JDReA is 0.0084K higher than that of DReA. Under clear sky conditions, the RMSE of the temperature inverted by the DReA method is 0.8928K, and the MAE is 0.6043K. The RMSE of the temperature inverted by the JDReA method is 0.8711K, and the MAE is 0.5923K. Through simple calculations, under clear sky conditions, the RMSE of JDReA is 0.0218K less than that of DReA, and the MAE of JDReA is 0.012K lower than that of DReA, indicating that JDReA has a better inversion effect under clear sky conditions. The RMSE of the JDReA method is lower than that of the DReA method in the overall (cloudy and clear sky) and cloudy and clear sky conditions, and it can be concluded that the effect of joint inversion is better than that of infrared alone.
[0138] Then, a longitudinal analysis is conducted on Table 1. The RMSE of AERI and the joint RMSE is close to 2K and the MAE is close to 1K in the case of clouds. In the case of overall (cloudy and clear sky), the RMSE is around 1.2K and the MAE is around 0.7K. In the case of clear sky, the RMSE is less than 1K and the MAE is around 0.6K. It can be found that under the conditions of clear sky, overall (cloudy and clear sky), and clear sky, the RMSE and MAE gradually increase, and the inversion accuracy gradually decreases.
[0139] According to the data comparison and analysis, the joint inversion of MWR and AERI data provided by this method can improve the temperature accuracy, but the improvement is not obvious. The possible reason is that it is limited by the weight function in physics, and the temperature information contained in the MWR data is not rich enough, resulting in an insignificant improvement. According to different situations, the inversion effect is ranked as follows: clear sky > overall (clear sky and clouds) > clouds, that is, the effect is best in clear sky.
[0140] Table 2 Average humidity inversion error under different conditions (including overall, cloudy, and clear sky conditions)
[0141]
[0142] Table 2 shows the average inversion error of humidity under different weather conditions. RMSE and MAE are used to measure the results of DReA and JDReA inversion using the deep learning framework in the overall, clear sky, and cloudy conditions. The data inverted by DReA also includes AERI data, and the data inverted by JDReA is the joint data of MWR and AERI. Analysis of Table 2 shows that in the overall (cloudy and clear sky) case, the RMSE of humidity inverted by the DReA method is 1.0446K, and the MAE is 0.7220K. The RMSE of humidity inverted by the JDReA method is 1.0134K, and the MAE is 0.7115K. Through simple calculation, in the overall (cloudy and clear sky) case, the RMSE of JDReA is 0.0312K less than that of DReA, and the MAE of JDReA is 0.0105K lower than that of DReA, indicating that JDReA has a better inversion effect in the overall (cloudy and clear sky) case. In cloudy conditions, the RMSE of the temperature retrieved by the DReA method is 1.0918K, and the MAE is 0.7582K. The RMSE of the humidity retrieved by the JDReA method is 1.0093K, and the MAE is 0.7285K. In cloudy conditions, the RMSE of JDReA is 0.0825K lower than that of DReA, and the MAE of JDReA is 0.0297K lower than that of DReA. Under clear sky conditions, the RMSE of humidity inverted by the DReA method is 1.0259K, and the MAE is 0.7081K. The RMSE of humidity inverted by the JDReA method is 1.0150K, and the MAE is 0.7049K. Under clear sky conditions, the RMSE of JDReA is 0.0109K less than that of DReA, and the MAE of JDReA is 0.0032K lower than that of DReA, indicating that JDReA has a better inversion effect under clear sky conditions. For humidity inversion, the RMSE is lower than that of DReA in overall (cloudy and clear sky) and cloudy and clear sky conditions, and it can be concluded that the effect of joint inversion is better than that of infrared alone. According to data comparison and analysis, the joint inversion of MWR and AERI data provided by this method can effectively improve the humidity accuracy.
[0143] Figure 4 From left to right, the temperature scatter plots are drawn based on the inversion results of DReA and JDReA. The scatter plots are the scatter plots between the inversion results of temperature and the measurement results of radiosonde (clear sky and cloudy sky). This figure contains the data of the whole (clear sky and cloudy sky), a total of 1110 groups, each group contains 37 heights, a total of 41070 data. The red high kernel density value shows the location of most of the data. Figure 4It can be seen that there are some points that deviate from the target value, but these points account for a small proportion, and most of the points are evenly distributed on both sides of the black straight line "Y = X". Overall, the correlation coefficient R between the inverted value of the temperature inverted by JDReA and the true value of the radiosonde is 2 The correlation coefficient between the inverted value of DReA temperature and the true value of the radiosonde reaches 0.9719, and JDReA has a high accuracy, which proves that the JDReA method proposed in the embodiment of the present application is better than the DReA method in inverting temperature.
[0144] Figure 5 From left to right, the temperature scatter plots are drawn based on the inversion results of DReA and JDReA. The scatter plots are the scatter plots between the inversion results of water vapor mixing ratio and the true values of the radiosonde (clear sky and cloudy sky). The red high kernel density value shows the location of most of the data. Figure 4 Compared to the temperature scatter plot results shown in Figure 1, the humidity inversion results are more scattered. Figure 5 It can be seen that most of the points are evenly distributed on both sides of the black straight line "Y = X". Overall, the correlation coefficient R between the JDReA humidity inversion value and the sounding true value is 2 The value of JDReA reaches 0.9276, the correlation coefficient of DReA reaches 0.9199, and the accuracy of JDReA is relatively high, which proves that the overall humidity inversion effect of the JDReA method proposed in the embodiment of the present application is better than that of DReA.
[0145] Figure 6 From left to right, the deviations of different vertical height layers (37 layers in total) drawn based on the inversion results of DReA and JDReA are respectively, which are the temperature inversion deviations of DReA and JDReA at different altitudes (clear sky and cloudy sky). The black horizontal dotted line is the third quartile line, the black horizontal solid line is the first quartile line, the vertical black dotted line is the zero deviation scale line, the red dot represents the average deviation, and the blue short line represents the median. Figure 6 (a) Contrast Figure 6 The middle (b) is obviously the most deviated from the zero deviation scale line. Figure 6 (b) ratio Figure 6 The result in (a) is better. Based on the above analysis, under cloudy and clear sky conditions, the deviation profile of JDReA is closer to the zero deviation scale line, and the inversion result is good.
[0146] Figure 7 From left to right, the deviations of different vertical height layers (37 layers in total) drawn based on the inversion results of DReA and JDReA are the inversion deviations of water vapor mixing ratio at different altitudes (clear sky and cloudy sky). The black horizontal dotted line is the third quartile line, the black horizontal solid line is the first quartile line, the vertical black dotted line is the zero deviation scale line, the red dot represents the average deviation, and the blue short line represents the median. Figure 7The middle (a) is obviously the most deviated from the zero deviation scale line compared with other figures. Figure 7 (b) ratio Figure 7 The result in (a) is better. In summary, under cloudy and clear sky conditions, the deviation profile of the water vapor mixing ratio of JDReA is closer to the zero deviation scale line, and the inversion result is the best.
[0147] Figure 8a-8c Time-height diagrams using different methods when a cold front passed (December 30, 2017), with the temperature on top and humidity on the bottom. Figure 8a represents the time-altitude diagram of the sounding reanalysis data, Figure 8b A temporal height diagram representing the DReA method, Figure 8c The time-height diagram of the JDReA method. This figure can intuitively reflect that below 1KM, JDReA can better depict the changes in temperature and humidity over time compared to DReA. Above 1KM, due to the addition of microwave radiation data, the interference of clouds is reduced, and the temperature and humidity inverted by JDReA are more accurate, especially the inversion of high-level humidity, which is significantly better than DReA. Overall, in cloudy conditions, although JDReA has improved, the improvement is relatively limited, mainly because microwaves only provide two channel data. When more microwave channel information is introduced, the improvement effect of the deep fusion inversion algorithm proposed in this application will be more significant.
[0148] Figure 9a-9d It represents the temperature and humidity profiles at 05:30UTC and 23:30UTC during the cold front passing event, where 05:30UTC is a clear sky moment and 23:30UTC is a cloudy moment. The JDReA inversion result is not much different from the DReA in the clear sky moment, while the DReA profile is closer to the true value in the cloudy moment. Figure 9d This shows that when a cold front passes, the humidity inversion effect is improved compared to the single inversion.
[0149] Figure 10a-Figure 10d It represents the temperature and humidity profiles at 05:30UTC and 17:30UTC during the warm front passing event, where 05:30UTC is a clear sky moment and 17:30UTC is a cloudy moment. The JDReA inversion result is not much different from the DReA in the clear sky moment, while the DReA profile is closer to the true value in the cloudy moment. Fig.10b This shows that when a warm front passes, the humidity inversion effect is improved compared to the single inversion.
[0150] Figure 11a-Figure 11c Time height diagrams using different methods when a warm front passed (March 28, 2017). These diagrams also show the temperature on the top and the humidity on the bottom. Fig.11a represents the time-altitude graph of the sounding reanalysis data, Fig.11bA temporal height diagram representing the DReA method, Fig.11c The time-height diagram of the JDReA method can intuitively reflect that JDReA has been effectively improved in the humidity inversion part.
[0151] Fig.12 From left to right, the temperature scatter plots are drawn based on the inversion results of DReA and JDReA, respectively. The temperature scatter plot is a scatter plot between the inversion results of the temperature and the measurement results of the radiosonde (with clouds). Among these two methods, JDReA has a higher accuracy, reaching 0.9309. It proves that the JDReA method proposed in the embodiment of the present invention has a better effect in inverting the temperature under clouds.
[0152] Fig.13 From left to right, the water vapor mixing ratio scatter plots based on the inversion results of DReA and JDReA are shown. The scatter plots are the scatter plots between the inversion results of the water vapor mixing ratio and the true value of the radiosonde (with clouds). In the case of clouds, the JDReA method R 2 It is 0.0203 higher than DReA, which fully proves the superiority of JDReA under cloudy conditions and indicates that the addition of a microwave water vapor-sensitive channel can greatly improve the inversion effect under cloudy conditions.
[0153] In cloudy conditions, Fig.14 From left to right, the deviations of different vertical height layers (37 layers in total) drawn based on the inversion results of DReA and JDReA are respectively, which are the deviations of temperature inversion at different altitudes (with clouds). The black horizontal dotted line is the third quartile line, the black horizontal solid line is the first quartile line, the vertical black dotted line is the zero deviation scale line, the red dot represents the average deviation, and the blue short line represents the median. Fig.14 (a) and Fig.14 In (b), the inversion result of the DReA method deviates significantly from the zero line, while the inversion result of the JDReA method is closer to the zero line. It can be considered that the inversion effect of JDReA is better in cloudy conditions. Fig.14 The middle (a) is obviously the most deviated from the zero deviation scale line compared with other figures. Fig.14 (b) ratio Fig.14 The result in (a) is better. In summary, under cloud conditions, the JDReA deviation profile is closer to the zero deviation scale line, and the inversion result is good.
[0154] In cloudy conditions, Fig.15From left to right, the deviations of different vertical height layers (37 layers in total) drawn based on the inversion results of DReA and JDReA are respectively, which are the inversion deviations of water vapor mixing ratio at different altitudes (with clouds). The black horizontal dotted line is the third quartile line, the black horizontal solid line is the first quartile line, the vertical black dotted line is the zero deviation scale line, the red dot represents the average deviation, and the blue short line represents the median. Fig.15 The middle (a) is obviously the most deviated from the zero deviation scale line compared with other figures. Fig.15 (b) ratio Fig.15 The result in (a) is better. In summary, under clear sky, the JDReA deviation profile is closer to the zero deviation scale line, and the inversion result is good.
[0155] Fig.16 From left to right, the temperature scatter plots are drawn based on the inversion results of DReA and JDReA, which are scatter plots between the inversion results of temperature and the measurement results of radiosonde (clear sky). In the clear sky case, JDReA method R 2 It is higher than that of DReA, which fully proves the superiority of JDReA under clear sky conditions and indicates that the addition of a microwave water vapor-sensitive channel can improve the inversion effect under clear sky conditions.
[0156] Fig.17 From left to right, they are the water vapor mixing ratio scatter plots based on the inversion results of DReA and JDReA, the scatter plots between the inversion results of the water vapor mixing ratio of the temperature scatter plot and the true value of the radiosonde (clear sky). In the case of clear sky, the JDReA method R 2 It is higher than that of DReA, which fully proves the superiority of JDReA under clear sky conditions and indicates that the addition of a microwave water vapor-sensitive channel can improve the inversion effect under clear sky conditions.
[0157] Under clear sky conditions, Fig.18 From left to right, the deviations of different vertical height layers (37 layers in total) drawn based on the inversion results of DReA and JDReA are respectively, which are the deviations of temperature inversion at different altitudes (clear sky). The black horizontal dotted line is the third quartile line, the black horizontal solid line is the first quartile line, the vertical black dotted line is the zero deviation scale line, the red dot represents the average deviation, and the blue short line represents the median. Fig.18 The middle (a) is obviously the most deviated from the zero deviation scale line compared with other figures. Fig.18 (b) ratio Fig.18 The result in (a) is better. In summary, under clear sky, the JDReA deviation profile is closer to the zero deviation scale line, and the inversion result is good.
[0158] Under clear sky conditions, Fig.19From left to right, the deviations of different vertical height layers (37 layers in total) drawn based on the inversion results of DReA and JDReA are respectively, which are the inversion deviations of water vapor mixing ratio at different altitudes (clear sky). The black horizontal dotted line is the third quartile line, the black horizontal solid line is the first quartile line, the vertical black dotted line is the zero deviation scale line, the red dot represents the average deviation, and the blue short line represents the median. Fig.19 The middle (a) is obviously the most deviated from the zero deviation scale line compared with other figures. Fig.19 (b) ratio Fig.19 The result in (a) is better. In summary, under clear sky, the JDReA deviation profile is closer to the zero deviation scale line, and the inversion result is good.
[0159] In order to evaluate the effect of the inversion method provided in this application, an overall experimental analysis, experimental analysis under cloudy and clear sky conditions were designed, and the temperature and humidity profile inversion method of combined infrared and microwave data provided in this application was compared with the DReA method in related technologies. The main conclusions are as follows:
[0160] (1) By comparing the experimental results with the DReA method under overall, cloudy and clear sky conditions, the experimental results show that the observation data of the microwave channel is used as a supplement to the infrared channel, and the inversion method of the temperature and humidity profile of the combined infrared and microwave data provided in this application is used for inversion, which can integrate the advantages of the inversion data and improve the accuracy of temperature and humidity inversion. (2) By drawing a time-height diagram with the sounding reanalysis data as the true value for analysis, it is found that the humidity inversion effect at 2000 to 3000 meters is significantly improved using the JDReA method provided in this application. The results prove that the accuracy of the JDReA inversion model established using a neural network in this application is improved compared to DReA, and the inversion humidity has a great advantage.
[0161] For the inversion of ground-based temperature and humidity profiles, the existing technology usually only uses infrared radiation data, because infrared radiation data contains a lot of information. In addition, there are relatively few works using deep learning methods. Most of the previous work is based on physical inversion methods. Physical inversion methods require the use of atmospheric radiation transmission models. It takes nearly an hour to invert a profile, and the inversion efficiency is very slow. However, infrared radiometers cannot penetrate clouds, so they cannot effectively invert temperature and humidity profiles in cloudy conditions. Microwave radiometers can penetrate clouds, but they contain less information. Experimental verification shows that only two channels are sensitive to temperature and humidity, so they are not widely used. The method provided by the embodiment of the present invention not only fuses microwave data for inversion, but also fuses infrared data and microwave data from different channels using a deep network, thereby improving the inversion accuracy of temperature and humidity profiles.
[0162] See also Fig. 20The embodiment of the present invention provides a temperature and humidity profile inversion device 2000 combining infrared and microwave data, comprising:
[0163] A feature extraction module 2001 is used to input the radiation value data and the brightness temperature data into a feature extraction module for feature extraction, and obtain a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data;
[0164] A feature splicing module 2002 is used to input the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data into a feature splicing module for feature splicing to obtain a joint feature vector;
[0165] The regression module 2003 is used to input the joint feature vector into the regression module for inversion to obtain an inversion result.
[0166] Optionally, the feature extraction module includes an AERI feature extraction module and an MWR feature extraction module, and the feature extraction module 2001 includes:
[0167] The first input unit is used to input the radiation value data into the AERI feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data; the second input unit is used to input the brightness temperature data into the MWR feature extraction module for feature extraction to obtain a feature vector corresponding to the brightness temperature data.
[0168] Optionally, the first input unit is used to:
[0169] The radiation value data is input into the AERI feature extraction module for nonlinear feature extraction to obtain a first output feature; wherein the AERI feature extraction module includes:
[0170] A first convolution block, wherein the first convolution block includes a first convolution layer, a first processing layer group and a first maximum pooling layer connected in sequence, wherein the first processing layer group includes a first batch normalization layer and a first output function layer; a second convolution block, wherein the second convolution block includes a second convolution layer and the first processing layer group connected in sequence; a third convolution block, wherein the third convolution block includes a third convolution layer and the first processing layer group connected in sequence; and a fourth convolution block, wherein the fourth convolution block includes a fourth convolution layer and the first processing layer group connected in sequence.
[0171] Optionally, the second input unit is specifically used for:
[0172] The brightness temperature data is input into the MWR feature extraction module for nonlinear feature extraction to obtain a first output feature; wherein the MWR feature extraction module includes:
[0173] A fifth convolution block, the fifth convolution block includes a fifth convolution layer, a second processing layer group and a first upsampling layer connected in sequence, the second processing layer group includes a second batch normalization layer and a second output function layer; a sixth convolution block, the sixth convolution block includes a sixth convolution layer, the second processing layer group and a second upsampling layer connected in sequence; a seventh convolution block, the seventh convolution block includes a seventh convolution layer and the second processing layer group connected in sequence; an eighth convolution block, the eighth convolution block includes an eighth convolution layer and the second processing layer group connected in sequence.
[0174] Optionally, the regression module 2003 is specifically used for:
[0175] The joint feature vector is input into a regression module for regression calculation to obtain an inversion result, wherein the regression module includes in sequence: a first fully connected block, the first fully connected block includes a first fully connected layer, a third output function and a discard layer connected in sequence; a second fully connected block, the second fully connected block includes a second fully connected layer and a fourth output function connected in sequence.
[0176] Optionally, the temperature and humidity profile inversion device 2000 combining infrared and microwave data further includes: an observation module for obtaining radiation value data through infrared radiometer observation, and obtaining brightness temperature data through microwave radiometer observation.
[0177] Optionally, the temperature and humidity profile inversion device 2000 for combining infrared and microwave data further includes:
[0178] An acquisition module is used to obtain radiation value sample data through infrared radiometer observation, brightness temperature sample data through microwave radiometer observation, and corresponding supervision data through radiosonde;
[0179] A training module is used to use the radiation value sample data and the brightness temperature sample data as training samples and the supervision data as true labels to iteratively train the joint depth inversion model, wherein the joint depth inversion model includes the feature extraction module, the feature splicing module and the regression module.
[0180] The temperature and humidity profile inversion device 2000 for combined infrared and microwave data provided in the embodiment of the present application can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated in this embodiment.
[0181] It should be noted that the division of units in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional unit in each embodiment of the present application may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0182] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.
[0183] like Fig.21 As shown, an embodiment of the present application provides an electronic device 2100, including: a memory 2102, a processor 2101, and a program stored in the memory 2102 and executable on the processor 2101; the processor 2101 is used to read the program in the memory 2102 to implement the steps in the temperature and humidity profile inversion method of the combined infrared and microwave data as described above.
[0184] An embodiment of the present application also provides a readable storage medium, on which a program is stored. When the program is executed by a processor, the various processes of the above-mentioned embodiment of the temperature and humidity profile inversion method for combined infrared and microwave data are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here. Among them, the readable storage medium can be any available medium or data storage device that can be accessed by the processor, including but not limited to magnetic storage (such as floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical storage (such as compact disk (CD), digital video disc (DVD), Blu-ray Disc (BD), high-definition versatile disc (HVD), etc.), and semiconductor memory (such as read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read only memory (EEPROM), non-volatile memory (NAND FLASH), solid-state drive (Solid State Disk or Solid State Drive, SSD)), etc.
[0185] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0186] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.
[0187] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A method for inverting temperature and humidity profiles by combining infrared and microwave data, characterized in that: include: Inputting the radiation value data and the brightness temperature data into a feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data; Inputting the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data into a feature splicing module for feature splicing to obtain a joint feature vector; The joint feature vector is input into a regression module for inversion to obtain an inversion result.
2. The method according to claim 1, characterized in that: The feature extraction module includes an AERI feature extraction module and an MWR feature extraction module. The radiation value data and the brightness temperature data are input into the feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data, including: Inputting the radiation value data into the AERI feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data; The brightness temperature data is input into the MWR feature extraction module for feature extraction to obtain a feature vector corresponding to the brightness temperature data.
3. The method according to claim 2, characterized in that: The step of inputting the radiation value data into the AERI feature extraction module for feature extraction to obtain a feature vector corresponding to the radiation value data includes: The radiation value data is input into the AERI feature extraction module for nonlinear feature extraction to obtain a first output feature; wherein the AERI feature extraction module includes: A first convolution block, the first convolution block comprising a first convolution layer, a first processing layer group and a first maximum pooling layer connected in sequence, the first processing layer group comprising a first batch normalization layer and a first output function layer; A second convolution block, the second convolution block comprising a second convolution layer and the first processing layer group connected in sequence; A third convolution block, the third convolution block comprising a third convolution layer and the first processing layer group connected in sequence; A fourth convolution block, wherein the fourth convolution block includes a fourth convolution layer and the first processing layer group connected in sequence.
4. The method according to claim 2, characterized in that: The step of inputting the brightness temperature data into the MWR feature extraction module for feature extraction to obtain a feature vector corresponding to the brightness temperature data includes: The brightness temperature data is input into the MWR feature extraction module for nonlinear feature extraction to obtain a first output feature; wherein the MWR feature extraction module includes: a fifth convolutional block, the fifth convolutional block comprising a fifth convolutional layer, a second processing layer group and a first upsampling layer connected in sequence, the second processing layer group comprising a second batch normalization layer and a second output function layer; a sixth convolution block, the sixth convolution block comprising a sixth convolution layer, the second processing layer group, and a second upsampling layer connected in sequence; a seventh convolutional block, the seventh convolutional block comprising a seventh convolutional layer and the second processing layer group connected in sequence; An eighth convolution block, wherein the eighth convolution block includes an eighth convolution layer and the second processing layer group connected in sequence.
5. The method according to claim 2, characterized in that: The step of inputting the joint feature vector into a regression module for inversion to obtain an inversion result includes: The joint feature vector is input into a regression module for regression calculation to obtain an inversion result, wherein the regression module sequentially includes: A first fully connected block, the first fully connected block comprising a first fully connected layer, a third output function, and a discard layer connected in sequence; A second fully connected block, wherein the second fully connected block includes a second fully connected layer and a fourth output function connected in sequence.
6. The method according to claim 5, characterized in that: Before inputting the radiation value data and the brightness temperature data into the feature extraction module for feature extraction to obtain the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data, the method further includes: The radiation value data are obtained through infrared radiometer observations, and the brightness temperature data are obtained through microwave radiometer observations.
7. The method according to any one of claims 1 to 6, characterized in that: Before inputting the radiation value data and the brightness temperature data into the feature extraction module for feature extraction to obtain the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data, the method further includes: The radiation value sample data is obtained through infrared radiometer observation, the brightness temperature sample data is obtained through microwave radiometer observation, and the corresponding supervision data is obtained through radiosonde; The radiation value sample data and the brightness temperature sample data are used as training samples, and the supervision data are used as true labels to iteratively train the joint depth inversion model, which includes the feature extraction module, the feature splicing module and the regression module.
8. A temperature and humidity profile inversion device combining infrared and microwave data, characterized in that: include: A feature extraction module, used for inputting the radiation value data and the brightness temperature data into the feature extraction module for feature extraction, and obtaining a feature vector corresponding to the radiation value data and a feature vector corresponding to the brightness temperature data; A feature splicing module, used for inputting the feature vector corresponding to the radiation value data and the feature vector corresponding to the brightness temperature data into the feature splicing module for feature splicing to obtain a joint feature vector; The regression module is used to input the joint feature vector into the regression module for inversion to obtain an inversion result.
9. An electronic device, comprising: A memory, a processor, and a program stored in the memory and executable on the processor; wherein the processor is used to read the program in the memory to implement the steps in the temperature and humidity profile inversion method of combined infrared and microwave data as described in any one of claims 1 to 7.
10. A readable storage medium for storing a program, characterized in that: When the program is executed by a processor, the steps in the method for inverting the temperature and humidity profile of combined infrared and microwave data as described in any one of claims 1 to 7 are implemented.
Citation Information
Patent Citations
Microwave and infrared combined cable joint internal temperature inversion method
CN114462301A
Cited By
Sea ice concentration inversion method and device based on brightness temperature data and electronic equipment
CN122334055A
