Cloud parameter inversion methods, media, and electronic equipment based on satellite thermal infrared data

Abstract

This disclosure provides a method, medium, and electronic equipment for cloud parameter inversion based on satellite thermal infrared data. The method includes: acquiring thermal infrared data from a geostationary satellite, cloud parameters from the geostationary satellite, and auxiliary meteorological field data; training a deep learning model based on the thermal infrared data from the geostationary satellite, the cloud parameters from the geostationary satellite, and the auxiliary meteorological field data to generate a preliminary model; acquiring thermal infrared data from a geostationary satellite, cloud parameters from a polar-orbiting satellite, and auxiliary meteorological field data; training the preliminary model based on the thermal infrared data from the geostationary satellite, the cloud parameters from the polar-orbiting satellite, and the auxiliary meteorological field data to obtain a transfer model; and inverting all-weather cloud parameters based on the transfer model. This invention fully combines the observational advantages of geostationary and polar-orbiting satellites, and through a surface-to-surface approach, it can compensate for the inaccuracy of inverting thick clouds at night, achieving higher-precision all-weather geostationary satellite cloud product inversion.

Description

Technical Field

[0001] This disclosure relates to the field of meteorological satellite technology, specifically to cloud parameter inversion methods, media, and electronic equipment based on satellite thermal infrared data. Background Technology

[0002] Clouds are an important component of the atmosphere and play a vital role in the global climate system. Changes in cloud parameters directly affect longwave and shortwave radiative transmission, regulating global or regional energy balances and thus influencing global climate change. Cloud parameters such as cloud phase state (CLP), cloud top height (CTH), cloud effective radius (CER), and cloud optical thickness (COT) possess unique optical properties; therefore, accurately acquiring cloud parameters is crucial for improving our understanding of Earth's weather and climate system.

[0003] With the rapid development of meteorological satellite technology, passive remote sensing satellites such as H8, FY4A, and MODIS, with their advantages of wide coverage, high timeliness, and comprehensive and comparable data, have gradually become the main methods for retrieving cloud parameters. Geostationary orbit satellites such as H8 and FY4A can continuously observe one-third of the Earth's surface, enabling them to observe the diurnal variation characteristics of clouds, which plays a crucial role in cloud retrieval. Early studies often selected visible light and shortwave infrared channels, using the dual-spectral solar reflectance method to retrieve cloud parameters. However, due to the lack of visible light and shortwave infrared measurement results at night, nighttime cloud products could not be retrieved. Considering that the thermal infrared channel can provide effective measurements both day and night, it can be selected to retrieve all-weather cloud parameters using the infrared split-window method. However, due to the limitations of the blackbody principle, these physical methods using the thermal infrared channel have poor retrieval results for clouds with large optical thickness.

[0004] In recent years, artificial intelligence technology has developed rapidly, and many machine learning models have been widely applied to cloud parameter inversion from satellite remote sensing. Machine learning algorithms such as k-nearest neighbors, support vector machines, random forests, and gradient boosting decision trees outperform traditional physical algorithms in cloud-to-cloud (CTH) inversion. However, these algorithms employ a point-to-point processing approach, which is not only slow but also unable to capture the spatial structure features of clouds. In contrast, deep learning algorithms such as convolutional neural networks (CNNs) use an image-to-image processing approach, acquiring the spatial structure features of clouds through convolution operations. This approach offers higher accuracy and faster speed, making it more suitable for cloud parameter inversion from geostationary satellites. Importantly, this surface-to-surface learning method can compensate for the insufficient inversion capability of thick clouds at night, thereby improving the model's ability to invert thick clouds.

[0005] Furthermore, geostationary satellites, being farther from Earth, may introduce greater errors during observation, but they can conduct continuous observations over large areas, acquiring high-frequency, wide-ranging spatiotemporal characteristic information. Polar satellites, being closer to Earth, offer more accurate observations, but because their transit time for a given area is fixed, they cannot provide continuous, large-scale observations. Summary of the Invention

[0006] This application discloses a cloud parameter inversion method, medium, and electronic equipment based on satellite thermal infrared data, which is used to solve the technical problem that existing technologies cannot obtain high-precision cloud parameters around the clock.

[0007] In a first aspect, embodiments of this disclosure provide a cloud parameter inversion method based on satellite thermal infrared data, comprising: acquiring thermal infrared data of a geostationary satellite, cloud parameters of the geostationary satellite, and auxiliary meteorological field data; training a deep learning model based on the thermal infrared data of the geostationary satellite, the cloud parameters of the geostationary satellite, and the auxiliary meteorological field data to generate a preliminary model; acquiring thermal infrared data of a geostationary satellite, cloud parameters of a polar-orbiting satellite, and auxiliary meteorological field data; training the preliminary model based on the thermal infrared data of the geostationary satellite and the auxiliary data to obtain a transfer model; and inverting all-weather cloud parameters based on the transfer model.

[0008] In one implementation of the first aspect, the cloud parameters include: cloud effective particle radius, cloud optical thickness, cloud top height, and cloud phase.

[0009] In one implementation of the first aspect, before training the deep learning model, the method further includes: processing the spatial resolution of thermal infrared data and auxiliary meteorological field data at the same time to form a dataset with uniform spatial resolution.

[0010] In one implementation of the first aspect, the method further includes: segmenting the image of the thermal infrared data to form a training sample of several pixels; and filtering the training sample of several pixels to remove training samples of pixels with confidence levels lower than a confidence threshold.

[0011] In one implementation of the first aspect, the deep learning model includes: an encoder and a decoder; the encoder performs convolution and pooling on the input data to obtain an intermediate feature vector; the decoder performs sampling convolution on the intermediate feature vector to obtain an output vector.

[0012] In one implementation of the first aspect, during the training of the deep learning model, the model parameters of the deep learning model are evaluated using a cross-entropy loss function and / or a squared loss function.

[0013] In one implementation of the first aspect, the method further includes: evaluating the accuracy of the transfer model based on at least one evaluation parameter; the evaluation parameter includes one or more of the following: overall accuracy, root mean square error, mean absolute error, mean deviation, and Pearson coefficient.

[0014] In one implementation of the first aspect, the deep learning model is a ResUnet neural network model.

[0015] In a second aspect, embodiments of this disclosure provide a computer-readable storage medium having a computer program stored thereon, the computer program being executed to implement the cloud parameter inversion method based on satellite thermal infrared data according to any implementation of the first aspect of this disclosure.

[0016] Thirdly, embodiments of this disclosure provide an electronic device, the electronic device comprising: a memory configured to store a computer program; and a processor configured to invoke the computer program to execute the cloud parameter inversion method based on satellite thermal infrared data according to any implementation of the first aspect of this disclosure.

[0017] As described above, the cloud parameter inversion method based on satellite thermal infrared data provided in this disclosure has the following beneficial effects:

[0018] 1. This invention can fully combine the observation advantages of geostationary satellites and polar-orbiting satellites, and through a surface-to-surface approach, it can make up for the inaccuracy of nighttime thick cloud inversion as much as possible, and achieve higher precision all-weather satellite cloud parameter inversion.

[0019] 2. This invention also integrates meteorological field data into the model, which effectively improves the model's inversion accuracy.

[0020] 3. This invention establishes a surface-to-surface mapping relationship between input features and real targets based on deep learning methods, which can fully extract the spatial structure information of variables, thereby making up for the deficiency of the inversion ability of thick clouds.

[0021] 4. In this invention, the deep learning model is accelerated by GPU, which can complete the inversion more efficiently and significantly reduce the time of a single full disk inversion.

[0022] 5. This invention, through transfer learning, fully combines the advantages of geostationary satellite and polar-orbiting satellite observations. The cloud parameters obtained by inversion not only meet the requirements of high frequency and large area, but also significantly outperform the official geostationary satellite cloud products in terms of accuracy. Attached Figure Description

[0023] Figure 1 The flowchart shown is a cloud parameter inversion method based on satellite thermal infrared data in an embodiment of this disclosure.

[0024] Figure 2 The diagram shown is a schematic of a deep learning model for a cloud parameter inversion method based on satellite thermal infrared data in an embodiment of this disclosure.

[0025] Figure 3 The diagram shown illustrates the principle of the cloud parameter inversion method based on satellite thermal infrared data in this embodiment of the present disclosure.

[0026] Figure 4 The diagram shown is a schematic representation of the structure of an electronic device according to an embodiment of this disclosure. Detailed Implementation

[0027] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. This disclosure can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0028] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. Therefore, the drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0029] Geostationary satellites, orbiting far from Earth, may introduce greater errors during observation, but they can conduct continuous observations over large areas, acquiring high-frequency, wide-ranging spatiotemporal characteristic information. Polar satellites, orbiting closer to Earth, offer more accurate observations, but because their transit times for a given area are fixed, they cannot provide continuous, large-scale observations. Geostationary and polar satellite observations complement each other. Combining the two fully leverages the advantages of satellite remote sensing. Therefore, a preliminary deep learning model can be built using geostationary satellite observations of thermal infrared radiation and cloud products to obtain continuous, large-scale spatiotemporal characteristic information of clouds. Then, transfer learning can be used to integrate polar satellite cloud products into the model, resulting in more accurate all-weather cloud products.

[0030] This embodiment discloses a cloud parameter inversion method, medium, and electronic equipment based on satellite thermal infrared data, which is used to solve the technical problem that cloud parameters cannot be obtained around the clock in the prior art, and realize a large-scale, high-precision, all-weather deep learning inversion method based on satellite thermal infrared data.

[0031] The cloud parameter inversion method based on satellite thermal infrared data disclosed in this embodiment utilizes thermal infrared radiation data observed by geostationary satellites. It establishes a preliminary connection between the deep learning ResUnet model and geostationary satellites to obtain high-frequency, large-scale spatiotemporal characteristic information of clouds. Then, through transfer learning, it establishes a connection with polar-orbiting satellites to obtain high-precision cloud parameters for all weather conditions. This cloud parameter inversion method based on satellite thermal infrared data fully combines the observational advantages of geostationary and polar-orbiting satellites. By using a surface-to-surface approach, it can compensate for the inaccuracies in inversion of thick nighttime clouds, achieving higher-precision all-weather geostationary satellite cloud parameter inversion.

[0032] The technical solutions in the embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0033] Figure 1 This is a flowchart illustrating a cloud parameter inversion method based on satellite thermal infrared data provided according to an embodiment of this disclosure. Figure 1 As shown, the cloud parameter inversion method based on satellite thermal infrared data provided in this embodiment includes the following steps S100 to S300.

[0034] In step S100, thermal infrared data of the geostationary satellite, cloud parameters of the geostationary satellite, and auxiliary meteorological field data are acquired. Based on the thermal infrared data of the geostationary satellite, cloud parameters of the geostationary satellite, and auxiliary meteorological field data, the deep learning model is trained to generate a preliminary model.

[0035] In step S200, thermal infrared data of geostationary satellites, cloud parameters of polar-orbiting satellites, and auxiliary meteorological field data are acquired. Based on the thermal infrared data of geostationary satellites, cloud parameters of polar-orbiting satellites, and auxiliary meteorological field data, the preliminary model is trained to obtain a transfer model.

[0036] In step S300, all-weather cloud parameters are retrieved based on the migration model.

[0037] The cloud parameter inversion method based on satellite thermal infrared data in this embodiment uses satellite thermal infrared data as input. Based on deep learning models and transfer learning methods, it combines the observation advantages of geostationary satellites and polar-orbiting satellites to invert and obtain large-scale, high-precision all-weather cloud parameters.

[0038] The following provides a detailed description of steps S100 to S300 of the cloud parameter inversion method based on satellite thermal infrared data in this embodiment.

[0039] In step S100, thermal infrared data, cloud parameters, and auxiliary meteorological field data of the geostationary satellite are acquired. Based on the thermal infrared data, cloud parameters, and auxiliary meteorological field data of the geostationary satellite, the deep learning model is trained to generate a preliminary model.

[0040] Geostationary satellites can provide high-frequency, large-scale continuous observations of clouds. In this embodiment, thermal infrared radiation data observed by geostationary satellites are used to establish a preliminary connection between the deep learning model and the geostationary satellites, thereby obtaining high-frequency, large-scale spatiotemporal characteristic information of clouds.

[0041] When training the deep learning model, the main input data used for training is the thermal infrared data of the geostationary satellite collected during the day from the thermal infrared channel of the geostationary satellite. The training also uses auxiliary data such as satellite zenith angle and meteorological field data as input data.

[0042] In this embodiment, cloud parameters of geostationary satellites are used as the training target. These cloud parameters include, but are not limited to, cloud effective particle radius (CER), cloud optical thickness (COT), cloud top height (CTH), and cloud phase state (CLP).

[0043] Among them, the thermal infrared data of geostationary satellites, cloud parameters, such as the thermal infrared channel radiation data and secondary cloud parameters of the Himawari-8 (H8) or Fengyun-4A (FY4A) satellites.

[0044] In this embodiment, thermal infrared radiation from a geostationary satellite, satellite zenith angle, and other meteorological field data are selected as input features for the deep learning model.

[0045] In this embodiment, before training the deep learning model, the method further includes: processing the spatial resolution of thermal infrared data and auxiliary meteorological field data at the same time to form a dataset with a unified spatial resolution.

[0046] Since different types of data differ in spatial scale, it is necessary to unify all input features to the spatial resolution of geostationary satellite thermal infrared data at the same time (e.g., 0.05°×0.05°) to achieve spatiotemporal matching between different data.

[0047] In this embodiment, the method further includes: segmenting the image of the thermal infrared data to form a training sample of several pixels; and screening the training sample of several pixels to remove training samples of pixels with confidence levels lower than the confidence level threshold.

[0048] Because geostationary satellites have a large observation area, inputting the entire image into a deep learning model during training would overload the graphics card and reduce processing speed. Therefore, before applying the data to the model, the entire image needs to be cropped into several square samples containing 64×64 pixels. To ensure data accuracy, the data can be filtered based on the quality assurance labels required by data standards, removing data samples with an excessive proportion of low-confidence pixels. The remaining data samples are then used as the training set.

[0049] In this embodiment, the deep learning model is preferably the ResUnet neural network model, but other neural network models can also be used; this embodiment is not specifically limited. This embodiment uses the ResUnet neural network model as an example for illustration.

[0050] In this embodiment, the deep learning model includes an encoder and a decoder; the encoder performs convolution and pooling on the input data to obtain an intermediate feature vector; the decoder performs sampling convolution on the intermediate feature vector to obtain an output vector; wherein, the end of the decoder performs sampling convolution through a 1×1 convolution kernel to obtain the output.

[0051] Figure 2 This diagram illustrates the operational principle of a deep learning model for cloud parameter inversion based on satellite thermal infrared data, as shown in this embodiment of the disclosure. Specifically, as... Figure 2 As shown, the deep learning model is U-shaped and mainly consists of an encoder (left half) and a decoder (right half). In this embodiment, the encoder first processes the input image with x channels and a shape size of 64×64 through convolution and pooling to obtain an intermediate feature vector with 2048 channels and a shape size of 4×4. Then, in the decoder, this intermediate feature vector is upsampled and convolved to obtain an output vector with 64 channels and a shape size of 64×64. Finally, it is processed by a 1×1 convolution kernel to obtain an output image with n channels and a size of 64×64. It is worth noting that each stage in the decoder is connected to the corresponding stage in the encoder through a connection layer. This skip connection allows the feature vector to retain more original image features. It should also be noted that the inversion of CLP is essentially a multi-class classification problem, with its input and output images having 23 and 4 channels respectively (x = 23, n = 1); while the inversion of CTH, CER, and COT is essentially a regression problem. After the CLP product is obtained through inversion, it will be used as an input feature in the model, so its input and output images have 24 and 1 channels respectively (x = 24, n = 1).

[0052] In this embodiment, during the training of the deep learning model, the model parameters are evaluated using the cross-entropy loss function and / or the squared loss function (MSE). That is, CrossEntropyLoss and MSE can be selected as loss functions to adjust the parameters of the deep learning model, and the optimal model parameters are selected based on the deep learning model's performance on the training set.

[0053] Therefore, this embodiment uses the thermal infrared radiation and cloud products observed by geostationary satellites to train the deep learning model, establish a preliminary model, and based on the preliminary model, establish a preliminary connection between thermal infrared data and geostationary satellite cloud products to obtain continuous large-scale spatiotemporal characteristic information of clouds.

[0054] After establishing a preliminary connection between thermal infrared data and geostationary satellite cloud products based on an initial model, a connection is then established with polar-orbiting satellite cloud products through transfer learning, thereby achieving high-precision, all-weather cloud product inversion.

[0055] In step S200, thermal infrared data of geostationary satellites, cloud parameters of polar-orbiting satellites, and auxiliary meteorological field data are acquired. The preliminary model is trained based on the thermal infrared data of geostationary satellites, cloud parameters of polar-orbiting satellites, and auxiliary meteorological field data to obtain a transfer model.

[0056] Polar-orbiting satellites can acquire high-precision cloud parameter information within a small area. In this embodiment, the preliminary model is trained using thermal infrared data from geostationary satellites, cloud parameters from the polar-orbiting satellites, and auxiliary meteorological field data to establish the relationship between the spatiotemporal characteristics of clouds and the cloud parameters from polar-orbiting satellites, thereby achieving high-precision, all-weather cloud product inversion.

[0057] When training the preliminary model, the main input data used for training is the thermal infrared data of the geostationary satellite collected during the day from the thermal infrared channel of the geostationary satellite. The training also uses auxiliary data such as the zenith angle of the polar-orbiting satellite and meteorological field data as input data for the preliminary model.

[0058] In this embodiment, cloud parameters of polar-orbiting satellites are used as the training target. These cloud parameters include, but are not limited to, cloud effective particle radius (CER), cloud optical thickness (COT), cloud top height (CTH), and cloud phase state (CLP).

[0059] The cloud parameters for the polar-orbiting satellites are derived from secondary cloud products observed by the MODIS imager on the Aqua and Terra satellites.

[0060] In this embodiment, thermal infrared radiation from geostationary satellites, zenith angle from polar-orbiting satellites, and other meteorological field data are selected as input features for the preliminary model. The principle process of training the preliminary model into a transfer model is the same as that of training the deep learning model into a preliminary model, except that the cloud product data (cloud parameters) from geostationary satellites are replaced with cloud product data (cloud parameters) from polar-orbiting satellites. The principle process of training the deep learning model into a preliminary model has already been explained in detail above and will not be repeated here.

[0061] In this embodiment, before training the initial model, the method may further include: processing the spatial resolution of thermal infrared data and auxiliary meteorological field data at the same time to form a dataset with a unified spatial resolution.

[0062] Since different types of data differ in spatial scale, it is necessary to unify all input features to the spatial resolution of polar-orbiting satellite thermal infrared data at the same time (e.g., 0.05°×0.05°) to achieve spatiotemporal matching between different data.

[0063] In this embodiment, the method further includes: segmenting the image of the thermal infrared data to form a training sample of several pixels; and screening the training sample of several pixels to remove training samples of pixels with confidence levels lower than the confidence level threshold.

[0064] Before applying geostationary satellite thermal infrared data to the preliminary model, the entire image needs to be cropped into several square samples containing 64×64 pixels. To ensure data accuracy, the data can be filtered based on the quality assurance labels required by the data standards, removing data samples with an excessive proportion of low-confidence pixels. The remaining data samples are then divided into a test set.

[0065] In this embodiment, during the training of the preliminary model, the model parameters of the preliminary model are evaluated using the cross-entropy loss function and / or the squared loss function (MSE). That is, CrossEntropyLoss and MSE can be selected as loss functions to adjust the parameters of the preliminary model, and the optimal model parameters are selected based on the performance of the preliminary model on the test set.

[0066] like Figure 3 As shown, the generation process of the transfer model in this embodiment includes:

[0067] The first part is the pre-training of the preliminary model. This part uses the thermal infrared channel radiation, satellite zenith angle and other meteorological field data of geostationary satellites as input features, and CLP, CTH, CER and COT in geostationary satellite secondary cloud products as real targets. The ResUnet model is used for pre-training, and the preliminary model is determined after testing. This preliminary model aims to establish the relationship between thermal infrared data and geostationary satellite cloud parameters on a large spatial scale, and obtain continuous spatiotemporal characteristic information of clouds.

[0068] The second part is the generation of the transfer model. This part is based on some parameters in the preliminary model obtained from pre-training. It is trained with the same input features and the secondary cloud products of polar-orbiting satellites as real targets. After testing, the optimal transfer model is obtained. This transfer model aims to establish the relationship between the thermal infrared radiation observed by geostationary satellites and the cloud parameters of polar-orbiting satellites. While retaining the advantage of continuous large-scale observation by geostationary satellites, it inverts to obtain cloud parameters with higher accuracy.

[0069] In step S300, all-weather cloud parameters are retrieved based on the migration model.

[0070] During the training process of the transfer model, a relationship is established between daytime thermal infrared radiation data and cloud parameters, and then nighttime cloud parameters are obtained by inverting nighttime thermal infrared radiation, thus realizing all-weather cloud product inversion.

[0071] This embodiment combines the observational advantages of geostationary and polar-orbiting satellites, employing a surface-to-surface approach based on thermal infrared radiation data to retrieve high-precision all-weather cloud parameters. The surface-to-surface retrieval method establishes relationships between images based on a neural network model, acquiring more structural features of clouds at a spatial scale. This embodiment fully leverages the observational advantages of geostationary and polar-orbiting satellites, using a surface-to-surface approach to minimize the inaccuracies in retrieving thick nighttime clouds, achieving higher-precision all-weather geostationary satellite cloud product retrieval.

[0072] In this embodiment, the method further includes: evaluating the accuracy of the transfer model based on at least one evaluation parameter; the evaluation parameter includes one or more of the following: overall accuracy, root mean square error, mean absolute error, mean bias, and Pearson coefficient.

[0073] In order to quantitatively evaluate the effectiveness of the transfer model, statistical indicators such as overall accuracy (OA), root mean square error (RMSE), mean absolute error (MAE), mean bias (MBE), and Pearson coefficient (R) will be used to quantitatively evaluate the accuracy of the model.

[0074] The cloud parameter inversion method based on satellite thermal infrared data in this embodiment can obtain the all-weather distribution of clouds throughout the entire observation area. A single inversion of a complete image takes only about one minute, ensuring near real-time inversion of FY-4A or Himawari-8 geostationary satellite cloud products. This is of great significance for tracking the evolution of cloud systems under different dynamic conditions, understanding atmospheric physical processes, and improving weather and climate simulations.

[0075] The cloud parameter inversion method based on satellite thermal infrared data in this embodiment uses satellite thermal infrared channel radiation as input, enabling all-weather cloud product inversion. Furthermore, this embodiment integrates meteorological field data into the model, effectively improving the model's inversion accuracy. The method also establishes a surface-to-surface mapping relationship between input features and the real target using deep learning, fully extracting the spatial structure information of variables to compensate for the insufficient inversion capability of thick clouds. The deep learning model, accelerated by a GPU, can complete the inversion more efficiently, with a single full-disk inversion taking only about one minute. Finally, this method leverages transfer learning to fully combine the observation advantages of geostationary and polar-orbiting satellites, resulting in cloud products that not only meet the needs of high frequency and wide coverage but also significantly outperform official geostationary satellite cloud products in terms of accuracy.

[0076] The scope of protection for the cloud parameter inversion method based on satellite thermal infrared data described in this disclosure is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, subtracting, or replacing steps in the prior art based on the principles of this disclosure is included within the scope of protection of this disclosure.

[0077] This disclosure also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed to implement the cloud parameter inversion method based on satellite thermal infrared data provided in any embodiment of this disclosure.

[0078] In this disclosure, any combination of one or more storage media may be used. The storage media may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, RAM, ROM, an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0079] This disclosure also provides an electronic device, Figure 4 The diagram shown is a structural schematic of the electronic device 100 in an embodiment of this disclosure. Figure 4 As shown, in this embodiment, the electronic device 100 includes a memory 102 and a processor 101.

[0080] The memory 102 is used to store computer programs. In some embodiments, the memory 102 includes various media capable of storing program code, such as ROM, RAM, magnetic disk, USB flash drive, memory card, or optical disk.

[0081] Specifically, memory 102 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. Electronic device 100 may further include other removable / non-removable, volatile / non-volatile computer system storage media. Memory 102 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of various embodiments of the present invention.

[0082] The processor 101 is connected to the memory 102 and is used to execute the computer program stored in the memory 102 so that the electronic device 100 performs a cloud parameter inversion method based on satellite thermal infrared data.

[0083] In some embodiments, the processor 101 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0084] In some embodiments, the electronic device 100 may further include a display. The display is communicatively connected to the memory 102 and the processor 101, and is used to display a GUI interface related to the cloud parameter inversion method based on satellite thermal infrared data.

[0085] In summary, the cloud parameter inversion method based on satellite thermal infrared data provided in this disclosure uses satellite thermal infrared channel radiation as the model input, which can fully combine the observation advantages of geostationary satellites and polar-orbiting satellites. Through a surface-to-surface approach, it can compensate for the inaccuracy of inversion from thick nighttime clouds as much as possible, achieving higher accuracy in all-weather satellite cloud parameter inversion. This invention also integrates meteorological field data into the model, effectively improving the model's inversion accuracy. This invention establishes a surface-to-surface mapping relationship between input features and real targets based on deep learning methods, which can fully extract the spatial structure information of variables, thereby compensating for the insufficient inversion capability of thick clouds. In this invention, the deep learning model is accelerated by GPU, which can complete the inversion more efficiently and significantly reduce the time for a single full-disk inversion. This invention, through transfer learning, fully combines the observation advantages of geostationary satellites and polar-orbiting satellites. The cloud parameters obtained by this invention not only meet the requirements of high frequency and large area, but also significantly outperform the official cloud products of geostationary satellites in terms of accuracy. Therefore, this disclosure effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0086] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0087] The above embodiments are merely illustrative of the principles and effects of this disclosure and are not intended to limit this disclosure. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this disclosure. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this disclosure should still be covered by the claims of this disclosure.

Claims

1. A cloud parameter inversion method based on satellite thermal infrared data, characterized in that, include: Acquire thermal infrared data, cloud parameters, and auxiliary meteorological field data from geostationary satellites. Train a deep learning model based on the thermal infrared data, cloud parameters, and auxiliary meteorological field data to generate a preliminary model and obtain continuous, large-scale spatiotemporal feature information of clouds. Acquire thermal infrared data from geostationary satellites, cloud parameters from polar-orbiting satellites, and auxiliary meteorological field data. Train the preliminary model based on the thermal infrared data from geostationary satellites, cloud parameters from polar-orbiting satellites, and auxiliary meteorological field data to obtain a transfer model and establish the relationship between the spatiotemporal characteristics of clouds and the cloud parameters of polar-orbiting satellites. The process of generating the migration model includes: The first part is the pre-training part of the preliminary model. This part uses at least the thermal infrared channel radiation and satellite zenith angle of the geostationary satellite as input features, and CLP, CTH, CER and COT in the geostationary satellite secondary cloud products as real targets. The ResUnet model is used for pre-training, and the preliminary model is determined after testing. The preliminary model aims to establish the relationship between thermal infrared data and geostationary satellite cloud parameters on a large spatial scale, and obtain continuous spatiotemporal characteristic information of clouds. The second part is the generation of the transfer model. This part is based on some parameters in the preliminary model obtained from the pre-training. It is trained with the same input features and the secondary cloud products of polar-orbiting satellites as real targets. After testing, the optimal transfer model is obtained. This transfer model aims to establish the relationship between the thermal infrared radiation observed by geostationary satellites and the cloud parameters of polar-orbiting satellites. While retaining the advantage of continuous large-scale observation by geostationary satellites, it retrieves cloud parameters with higher accuracy. All-weather cloud parameters are retrieved based on the migration model.

2. The cloud parameter inversion method based on satellite thermal infrared data according to claim 1, characterized in that, The cloud parameters include: cloud effective particle radius, cloud optical thickness, cloud top height, and cloud phase.

3. The cloud parameter inversion method based on satellite thermal infrared data according to claim 1 or 2, characterized in that, Before training a deep learning model, the following steps are also required: The spatial resolution of thermal infrared data and auxiliary meteorological field data at the same time is processed to form a dataset with a unified spatial resolution.

4. The cloud parameter inversion method based on satellite thermal infrared data according to claim 3, characterized in that, Also includes: The thermal infrared data image is segmented to form a training sample of several pixels; The training samples of the aforementioned pixels are screened, and training samples of pixels with confidence scores lower than the confidence score threshold are removed.

5. The cloud parameter inversion method based on satellite thermal infrared data according to claim 1, characterized in that, The deep learning model includes: an encoder and a decoder; The encoder performs convolution and pooling on the input data to obtain intermediate feature vectors; The decoder performs sampling convolution on the intermediate feature vector to obtain the output vector.

6. The cloud parameter inversion method based on satellite thermal infrared data according to claim 1, characterized in that, During the training of the deep learning model, the model parameters are evaluated using the cross-entropy loss function and / or the squared loss function.

7. The cloud parameter inversion method based on satellite thermal infrared data according to claim 1, characterized in that, Also includes: The accuracy of the migration model is evaluated based on at least one evaluation parameter; The evaluation parameters include one or more of the following: overall accuracy, root mean square error, mean absolute error, mean deviation, and Pearson coefficient.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed to implement the cloud parameter inversion method based on satellite thermal infrared data according to any one of claims 1 to 7.

9. An electronic device, characterized in that, The electronic device includes: Memory, configured to store computer programs; and The processor is configured to invoke the computer program to execute the cloud parameter inversion method based on satellite thermal infrared data according to any one of claims 1 to 7.

Images