A method, apparatus, electronic device and storage medium for polar cloud water path inversion

By constructing a polar cloud microphysics training dataset with multi-source data, explicitly calculating features, and utilizing independent coding towers and a shared fusion network, the problems of identifying the boundary between thin clouds and ice sheets and model failure in polar satellite remote sensing technology were solved, achieving high-precision cloud and water path inversion in polar all-weather conditions.

CN123115674APending Publication Date: 2026-11-17AEROSPACE INFORMATION RES INST CAS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611325464.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing satellite remote sensing technology cannot effectively identify the boundary between thin clouds and ice sheets in polar regions, resulting in large estimation uncertainties, the failure of traditional physical models, and the narrow swath and low time revisit rate of active radar observations, which cannot meet the needs of large-scale continuous dynamic monitoring in polar regions.

Method used

By constructing a polar cloud microphysics training dataset with multi-source data, explicitly calculating features such as channel difference, ratio, and polarization ratio, constructing physical constraints, removing background interference from high albedo of ice and snow and the influence of strong surface inversion, and using independent coding towers and shared fusion networks for inversion, independent regression of phase-state clouds is achieved.

Benefits of technology

It can effectively identify the boundary between thin clouds and ice sheets, improve the certainty of estimation, avoid the failure of traditional models, achieve accurate separation of ice clouds, water clouds and mixed clouds, improve the observation swath and time revisit rate, and meet the needs of high-precision all-weather monitoring in the polar regions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115674A_ABST
    Figure CN123115674A_ABST
Patent Text Reader

Abstract

This invention proposes a method, apparatus, electronic device, and storage medium for polar cloud water path inversion. The method includes: performing spatiotemporal matching on all data to construct a polar cloud microphysics training dataset; explicitly calculating channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters based on the polar cloud microphysics training dataset to form multidimensional physical constraint features; adding the multidimensional physical constraint features to the polar cloud microphysics training dataset; dividing and training the new polar cloud microphysics training dataset according to cloud phase classification labels; constructing multiple independent encoding towers based on the trained polar cloud microphysics training dataset; encoding the input features in the polar cloud microphysics training dataset based on the independent encoding towers; and combining a shared fusion network to extract shared representation vectors for training to obtain a trained inversion model, which is then used to perform inversion to obtain the inversion result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of satellite remote sensing and atmospheric environment monitoring, and more specifically, to a method, apparatus, electronic device, and storage medium for polar cloud and water path inversion. Background Technology

[0002] Cloud water path (CWP), including liquid water path and ice water path, is a key parameter characterizing cloud radiation properties and plays a central role in the Earth's atmospheric radiation energy balance and the polar climate amplification effect. Ground-based observation stations are scarce in polar regions; therefore, satellite remote sensing technology has become the primary means of achieving large-scale, continuous monitoring of polar CWP. Currently, CWP retrieval methods based on satellite remote sensing are becoming increasingly mature, mainly forming a technical pattern of parallel active and passive remote sensing (optical and microwave).

[0003] Currently, existing cloud brightness temperature (CWP) inversion methods mainly include passive optical remote sensing inversion schemes, passive microwave remote sensing inversion schemes, and active remote sensing detection schemes. In passive optical remote sensing inversion schemes, the strong reflectivity of clouds in the visible spectrum and the weak absorption characteristics in the short-wave infrared spectrum are utilized for inversion. A dual-channel or multi-channel lookup table method is typically used to jointly invert cloud optical thickness and effective particle radius, and then CWP is calculated under the assumption of cloud vertical profile distribution. In passive microwave remote sensing inversion schemes, the penetrability of microwaves to clouds and the emission or scattering characteristics of cloud water substances in specific frequency bands are utilized to invert the CWP by establishing a physical regression or statistical analytical solution for microwave brightness temperature and path water content. Active remote sensing detection schemes, represented by spaceborne millimeter-wave radar CloudSat and lidar CALIPSO, utilize actively emitted signals to obtain accurate cloud vertical profiles and microphysical structures.

[0004] However, the polar regions experience a polar night lasting up to six months, rendering physical inversion methods that rely entirely on the reflectivity of visible and near-infrared bands completely ineffective. During polar days, the high-reflectivity background of ice and snow is easily confused with cloud signals, causing passive optical sensors to be unable to effectively identify the boundary between thin clouds and ice sheets, resulting in significant uncertainty in estimations. Strong temperature inversions, where cloud top temperatures exceed surface temperatures, frequently occur in the polar regions, causing traditional physical forward radiation transfer models to fail, and thermal infrared algorithms to be unable to accurately separate cloud top and background radiation. Mixed-phase clouds are frequent in the polar regions; due to the drastically different absorption characteristics of ice crystals and water droplets in the infrared band, using a single physical model for fitting can lead to conflicts in bimodal and multimodal nonlinear distributions caused by the mixing of ice and water phases, resulting in non-negligible systemic biases. Furthermore, active radar observations have extremely narrow swaths (only a line along the orbit) and low time revisit rates, with natural orbital holes near the poles (such as after 81.8° latitude), making it impossible to meet the needs of large-scale continuous dynamic monitoring in the polar regions. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and storage medium for polar cloud water path inversion. By explicitly calculating differences, ratios, and polarization ratios using a polar cloud microphysics training dataset constructed from multi-source data, physical constraints are established. This eliminates the interference of high albedo background from polar ice and snow and the influence of strong surface inversion on the radiative transfer solution space, resolving the confusion between ice and snow surface background and cloud signals. It effectively identifies the boundary between thin clouds and ice sheets, improving the certainty of estimation and avoiding the failure of traditional physical forward radiative transfer models. It can accurately separate cloud top and background radiation. Independent regression of phase states—ice clouds, water clouds, and mixed clouds—is achieved, avoiding conflicts in bimodal and multimodal nonlinear distributions and eliminating systematic biases caused by ice-water phase mixing. The application also improves the observation swath and time revisit rate, meeting the needs of large-scale continuous dynamic monitoring in the polar regions even near the poles. Thus, it independently achieves high-precision, all-weather inversion in polar regions.

[0006] In a first aspect, embodiments of this application provide a method for polar cloud water path inversion, the method comprising: Step S1: Obtain multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysics ground truth parameters, and perform spatiotemporal matching on all data to construct a polar cloud microphysics training dataset; wherein, the passive remote sensing radiation observation data includes at least thermal infrared emission band data, geometric parameters, and passive microwave data; Step S2: Based on the polar cloud microphysics training dataset, explicitly calculate the channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters to form multidimensional physical constraint features, and add the multidimensional physical constraint features to the polar cloud microphysics training dataset to obtain a new polar cloud microphysics training dataset. Step S3: Divide and train the new polar cloud microphysics training dataset according to the cloud phase classification labels to obtain the trained polar cloud microphysics training dataset. Step S4: Construct multiple independent encoding towers based on the trained polar cloud microphysics training dataset. Encode the input features in the polar cloud microphysics training dataset based on the independent encoding towers, and combine them with a shared fusion network to extract shared representation vectors for training, thereby obtaining a trained inversion model. Inversion results are obtained by inverting through the inversion model.

[0007] In one possible implementation, in step S1, the thermal infrared emission band data includes thermal infrared channel calibration radiation intensity; the step of performing spatiotemporal matching on all data to construct a polar cloud microphysics training dataset includes: Using preset cloud radar data as ground truth, spatiotemporal matching is performed to construct a polar cloud microphysics training dataset. The calibrated radiation intensity of the thermal infrared channel is converted into a brightness temperature within the target range using a preset Planck blackbody radiation formula.

[0008] In one possible implementation, in step S2, the passive microwave data includes the horizontal and vertical polarization brightness temperatures of the first and second frequency channels; the explicit calculation of channel difference characteristics, channel ratio characteristics, polarization ratio characteristics, and environmental thermodynamic layering parameters includes: The channel difference characteristics are obtained by calculating the split window difference, the gas and water vapor absorption difference, the short-wave infrared-window difference, and the deep background and environmental constraint characteristics. Based on the brightness temperature, the brightness temperature ratio between different channels is calculated to obtain the channel ratio feature; wherein, the channel ratio feature is used to eliminate the multiplicative scaling drift factor and characterize the relative slope change of the radiation curve. Based on the passive microwave data, the ratio of the vertical and horizontal polarization temperature difference between the first and second frequency channels in the advanced microwave scanning radiometer is calculated to obtain the polarization ratio feature; wherein, the polarization ratio feature characterizes the vertical integral total amount of water material inside the deep cloud column. The difference between the surface temperature and the atmospheric temperature at the target air pressure is calculated to obtain the environmental thermodynamic stratification parameters; wherein, the environmental thermodynamic stratification parameters are used to correct the radiative transfer solution space under inversion conditions.

[0009] In one possible implementation, step S3, which involves dividing and training the new polar cloud microphysics training dataset based on cloud phase classification labels, includes: The input features in the polar cloud microphysics training dataset are divided according to the cloud phase classification labels to construct an independent physical feature set containing water clouds, ice clouds and mixed clouds; The feature vectors in the independent physical feature set are split into three independent regression subnetworks dedicated to water cloud, ice cloud and mixed cloud to perform nonlinear mapping and regression training of cloud-water paths in the corresponding phase.

[0010] In one possible implementation, in step S4, encoding the input features in the polar cloud microphysics training dataset based on the independent encoding tower and obtaining a shared representation vector by combining the shared fusion network includes: Based on the independent coding tower, the input features in the polar cloud microphysics training dataset are independently and nonlinearly fused and encoded using linear layers, batch normalization, and linear rectified activation functions. The features are concatenated and merged in the horizontal dimension to form a multimodal fusion feature vector, which is then fed into a shared fusion network to extract a shared representation vector.

[0011] In one possible implementation, step S4, which involves training the shared representation vector extracted from the shared fusion network, includes: A year discriminator branch is set at the output of the shared fusion network, and the year is classified based on the year discriminator branch of the shared representation vector. A gradient reversal layer is set at the entry point of the year discriminator branch, so that the shared representation vector is passed identically during forward propagation, while the year classification cross-entropy loss gradient is multiplied by a negative adjustment coefficient and then passed back to the shared fusion network during back propagation, so that the shared fusion network outputs a pure and universal representation that is immune to scaling decay.

[0012] In one possible implementation, the method further includes: The cloud water path target value in the true parameters of the active radar cloud microphysics is subjected to square root transformation and normalized on an intra-annual scale. Perform global standardization on the normalized cloud water path target values ​​for the year.

[0013] Secondly, embodiments of this application also provide a polar cloud water path inversion device, the device comprising: A construction module is used to acquire multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysics ground truth parameters, and to perform spatiotemporal matching on all data to construct a polar cloud microphysics training dataset; wherein, the passive remote sensing radiation observation data includes at least thermal infrared emission band data, geometric parameters, and passive microwave data; The acquisition module is used to explicitly calculate channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters based on the polar cloud microphysics training dataset, form multidimensional physical constraint features, and add the multidimensional physical constraint features to the polar cloud microphysics training dataset to obtain a new polar cloud microphysics training dataset. The training module is used to divide and train the new polar cloud microphysics training dataset according to the cloud phase classification labels to obtain the trained polar cloud microphysics training dataset. The inversion module is used to construct multiple independent encoding towers based on the trained polar cloud microphysics training dataset, encode the input features in the polar cloud microphysics training dataset based on the independent encoding towers, and combine them with a shared fusion network to extract shared representation vectors for training, thereby obtaining a trained inversion model, and then using the inversion model to perform inversion to obtain the inversion result.

[0014] In one possible implementation, the thermal infrared emission band data includes thermal infrared channel calibration radiation intensity; the construction module is specifically used for: Using preset cloud radar data as ground truth, spatiotemporal matching is performed to construct a polar cloud microphysics training dataset. The calibrated radiation intensity of the thermal infrared channel is converted into a brightness temperature within the target range using a preset Planck blackbody radiation formula.

[0015] In one possible implementation, the passive microwave data includes the horizontal and vertical polarization brightness temperatures of a first frequency channel and a second frequency channel; the acquisition module is specifically used for: The channel difference characteristics are obtained by calculating the split window difference, the gas and water vapor absorption difference, the short-wave infrared-window difference, and the deep background and environmental constraint characteristics. Based on the brightness temperature, the brightness temperature ratio between different channels is calculated to obtain the channel ratio feature; wherein, the channel ratio feature is used to eliminate the multiplicative scaling drift factor and characterize the relative slope change of the radiation curve. Based on the passive microwave data, the ratio of the vertical and horizontal polarization temperature difference between the first and second frequency channels in the advanced microwave scanning radiometer is calculated to obtain the polarization ratio feature; wherein, the polarization ratio feature characterizes the vertical integral total amount of water material inside the deep cloud column. The difference between the surface temperature and the atmospheric temperature at the target air pressure is calculated to obtain the environmental thermodynamic stratification parameters; wherein, the environmental thermodynamic stratification parameters are used to correct the radiative transfer solution space under inversion conditions.

[0016] In one possible implementation, the training module is specifically used for: The input features in the polar cloud microphysics training dataset are divided according to the cloud phase classification labels to construct an independent physical feature set containing water clouds, ice clouds and mixed clouds; The feature vectors in the independent physical feature set are split into three independent regression subnetworks dedicated to water cloud, ice cloud and mixed cloud to perform nonlinear mapping and regression training of cloud-water paths in the corresponding phase.

[0017] In one possible implementation, the inversion module is specifically used for: Based on the independent coding tower, the input features in the polar cloud microphysics training dataset are independently and nonlinearly fused and encoded using linear layers, batch normalization, and linear rectified activation functions. The features are concatenated and merged in the horizontal dimension to form a multimodal fusion feature vector, which is then fed into a shared fusion network to extract a shared representation vector.

[0018] In one possible implementation, the inversion module is specifically used for: A year discriminator branch is set at the output of the shared fusion network, and the year is classified based on the year discriminator branch of the shared representation vector. A gradient reversal layer is set at the entry point of the year discriminator branch, so that the shared representation vector is passed identically during forward propagation, while the year classification cross-entropy loss gradient is multiplied by a negative adjustment coefficient and then passed back to the shared fusion network during back propagation, so that the shared fusion network outputs a pure and universal representation that is immune to scaling decay.

[0019] In one possible implementation, the device further includes: The processing module is used to perform square root transformation on the cloud water path target value in the cloud microphysical true value parameters of the active radar cloud and normalize it on the intra-annual scale. The execution module is used to perform global standardization on the normalized cloud water path target values ​​for the year.

[0020] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the polar cloud water path inversion method described in any of the first aspects.

[0021] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, performs the steps of the polar cloud water path inversion method described in any one of the first aspects.

[0022] This application provides the aforementioned polar cloud water path inversion method, apparatus, electronic equipment, and storage medium. It acquires multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysical ground truth parameters. All data are spatiotemporally matched to construct a polar cloud microphysical training dataset. Based on this training dataset, channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters are explicitly calculated to form multidimensional physical constraint features. These multidimensional physical constraint features are then added to the polar cloud microphysical training dataset to obtain a new training dataset. This new training dataset is then divided and trained according to cloud phase classification labels to obtain a trained polar cloud microphysical training dataset. Multiple independent encoding towers are constructed based on the trained dataset. The input features in the training dataset are encoded using these independent encoding towers, and a shared representation vector is extracted and trained using a shared fusion network to obtain a trained inversion model. The inversion model is then used to obtain the inversion result. This application constructs physical constraints by explicitly calculating differences, ratios, and polarization ratios using a polar cloud microphysics training dataset built from multi-source data. This eliminates the interference of high albedo background from polar ice and snow and the influence of strong surface inversion on the radiative transfer solution space, resolving the confusion between ice and snow surface background and cloud signals. It effectively identifies the boundary between thin clouds and ice sheets, improving the certainty of estimation and avoiding the failure of traditional physical forward radiative transfer models. It accurately separates cloud top and background radiation. Furthermore, it achieves independent regression of phase states—ice clouds, water clouds, and mixed clouds—avoiding conflicts in bimodal and multimodal nonlinear distributions and eliminating systematic biases caused by ice-water phase mixing. It also improves the observation swath and time revisit rate, meeting the needs of large-scale continuous dynamic monitoring in the polar regions even near the poles. Thus, it independently achieves high-precision inversion across all weather conditions in the polar regions. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart of the first polar cloud water path inversion method provided in this application embodiment is shown; Figure 2 This paper shows a schematic diagram of the structure of a polar cloud water path inversion device provided in an embodiment of this application; Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0026] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0027] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0028] Given that cloud water path (CWP) includes liquid water path and ice water path, it is a key parameter characterizing cloud radiation properties and plays a central role in the Earth's atmospheric radiation energy balance and the polar climate amplification effect. Ground-based observation stations are scarce in polar regions, therefore satellite remote sensing technology has become the primary means of achieving large-scale, continuous monitoring of polar CWP. Currently, CWP retrieval methods based on satellite remote sensing are becoming increasingly mature, mainly forming a technical pattern of parallel active and passive remote sensing (optical and microwave).

[0029] Currently, existing cloud brightness temperature (CWP) inversion methods mainly include passive optical remote sensing inversion schemes, passive microwave remote sensing inversion schemes, and active remote sensing detection schemes. In passive optical remote sensing inversion schemes, the strong reflectivity of clouds in the visible spectrum and the weak absorption characteristics in the short-wave infrared spectrum are utilized for inversion. A dual-channel or multi-channel lookup table method is typically used to jointly invert cloud optical thickness and effective particle radius, and then CWP is calculated under the assumption of cloud vertical profile distribution. In passive microwave remote sensing inversion schemes, the penetrability of microwaves to clouds and the emission or scattering characteristics of cloud water substances in specific frequency bands are utilized to invert the CWP by establishing a physical regression or statistical analytical solution for microwave brightness temperature and path water content. Active remote sensing detection schemes, represented by spaceborne millimeter-wave radar CloudSat and lidar CALIPSO, utilize actively emitted signals to obtain accurate cloud vertical profiles and microphysical structures.

[0030] However, the polar regions experience a polar night lasting up to six months, rendering physical inversion methods that rely entirely on the reflectivity of visible and near-infrared bands completely ineffective. During polar days, the high-reflectivity background of ice and snow is easily confused with cloud signals, causing passive optical sensors to be unable to effectively identify the boundary between thin clouds and ice sheets, resulting in significant uncertainty in estimations. Strong temperature inversions, where cloud top temperatures exceed surface temperatures, frequently occur in the polar regions, causing traditional physical forward radiation transfer models to fail, and thermal infrared algorithms to be unable to accurately separate cloud top and background radiation. Mixed-phase clouds are frequent in the polar regions; due to the drastically different absorption characteristics of ice crystals and water droplets in the infrared band, using a single physical model for fitting can lead to conflicts in bimodal and multimodal nonlinear distributions caused by the mixing of ice and water phases, resulting in non-negligible systemic biases. Furthermore, active radar observations have extremely narrow swaths (only a line along the orbit) and low time revisit rates, with natural orbital holes near the poles (such as after 81.8° latitude), making it impossible to meet the needs of large-scale continuous dynamic monitoring in the polar regions.

[0031] Based on this, embodiments of this application provide a polar cloud water path inversion method, apparatus, electronic device, and storage medium. By explicitly calculating the differences, ratios, and polarization ratios of a polar cloud microphysics training dataset constructed from multi-source data, physical constraints are established. This removes the interference of high albedo background from polar ice and snow and the influence of strong surface inversion on the radiative transfer solution space, solving the problem of confusion between ice and snow surface background and cloud signals. It can effectively identify the boundary between thin clouds and ice sheets, improve the certainty of estimation, avoid the failure of traditional physical forward radiative transfer models, and accurately separate cloud top and background radiation. It achieves independent regression of phase states, namely ice clouds, water clouds, and mixed clouds, avoiding conflicts in bimodal and multimodal nonlinear distributions and eliminating systematic biases caused by ice-water phase mixing. It improves the observation swath and time revisit rate, and can still meet the needs of large-scale continuous dynamic monitoring in the polar region near the poles, thereby independently realizing high-precision inversion of polar all-weather conditions.

[0032] like Figure 1 As shown in the figure, this application provides a method for polar cloud water path inversion, the method comprising: S101. Acquire multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysical ground truth parameters, and perform spatiotemporal matching on all data to construct a polar cloud microphysical training dataset.

[0033] S102. Based on the polar cloud microphysics training dataset, explicitly calculate the channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters to form multidimensional physical constraint features. Then, add the multidimensional physical constraint features to the polar cloud microphysics training dataset to obtain a new polar cloud microphysics training dataset.

[0034] S103. Divide and train the new polar cloud microphysics training dataset according to the cloud phase classification label to obtain the trained polar cloud microphysics training dataset.

[0035] S104. Construct multiple independent encoding towers based on the trained polar cloud microphysics training dataset. Encode the input features in the polar cloud microphysics training dataset based on the independent encoding towers, and combine them with a shared fusion network to extract shared representation vectors for training, so as to obtain the trained inversion model. Inversion results are obtained by inversion through the inversion model.

[0036] The aforementioned polar cloud-water path inversion method explicitly calculates features such as differences, ratios, and polarization ratios using a polar cloud microphysics training dataset constructed from multi-source data. This establishes physical constraints, removes background interference from high albedo of polar ice and snow and the influence of strong surface inversion on the radiative transfer solution space, and solves the problem of confusion between ice and snow surface background and cloud signals. It can effectively identify the boundary between thin clouds and ice sheets, improve the certainty of estimation, avoid the failure of traditional physical forward radiative transfer models, and accurately separate cloud top and background radiation. It achieves independent regression of phase states, namely ice clouds, water clouds, and mixed clouds, avoids conflicts in bimodal and multimodal nonlinear distributions, and eliminates systematic biases caused by the mixing of ice and water phases. It improves the observation swath and time revisit rate, and can still meet the needs of large-scale continuous dynamic monitoring in the polar regions near the poles. Thus, it independently achieves high-precision inversion of polar all-weather conditions.

[0037] The exemplary embodiments described above will be explained below: S101. Acquire multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysical ground truth parameters, and perform spatiotemporal matching on all data to construct a polar cloud microphysical training dataset.

[0038] In this embodiment, the passive remote sensing radiation observation data includes at least thermal infrared emission band data, geometric parameters, and passive microwave data. The thermal infrared emission band data includes thermal infrared channel calibration radiation intensity; the geometric parameters include at least solar zenith angle and elevation; the passive microwave data includes horizontal and vertical polarization brightness temperatures of the first and second frequency channels (e.g., 89.0 GHz and 36.5 GHz). Spatiotemporal matching is performed on the acquired multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysical ground truth parameters to construct a polar cloud microphysical training dataset for subsequent processing. For example, MODIS satellite thermal infrared emission band (Band 20-25, 27-36) data, geometric parameters (solar zenith angle, elevation, etc.), AMSR (Advanced Microwave Scanning Radiometer) passive microwave data (horizontal and vertical polarization brightness temperatures of 89.0 GHz and 36.5 GHz), and ERA5 meteorological reanalysis data (surface temperature and 800 hPa atmospheric temperature) are acquired. See Table 1 below: Table 1 Summary of Data Sources and Preprocessing Parameters

[0039] Optionally, when performing spatiotemporal matching on all data to construct a polar cloud microphysics training dataset, a preset cloud radar data is used as the ground truth to perform spatiotemporal matching and construct the polar cloud microphysics training dataset; the thermal infrared channel calibration radiation intensity is converted into a brightness temperature within the target range using a preset Planck blackbody radiation formula.

[0040] For example, a polar cloud microphysics training dataset is constructed by using pre-defined third-party CloudSat cloud radar data (characterizing cloud water paths and cloud ice paths) as ground truth for spatiotemporal matching. The acquired thermal infrared channel calibration radiation intensity data is converted into brightness temperature using Planck's blackbody radiation formula and limited to the target range of 150 to 400 K.

[0041] It should be noted that the thermal infrared and microwave sensors mentioned in this application are not limited to MODIS and AMSR. Wide-swath passive thermal infrared and microwave radiometers mounted on other meteorological satellites (such as the Fengyun (FY) series polar-orbiting meteorological satellites, the Suomi-NPP / NOAA series (mounted on VIIRS), the MetOp series, and next-generation geostationary meteorological satellites (such as Himawari-8 / 9 and GOES-R) can also be used to acquire observational data for the corresponding channels. The auxiliary data can be further fused with reanalysis data such as atmospheric boundary layer vertical wind fields and humidity profiles at higher spatiotemporal resolution to enhance the ability to capture and analyze complex micrometeorological processes and ultrathin low-level cloud systems in polar regions.

[0042] S102. Based on the polar cloud microphysics training dataset, explicitly calculate the channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters to form multidimensional physical constraint features. Then, add the multidimensional physical constraint features to the polar cloud microphysics training dataset to obtain a new polar cloud microphysics training dataset.

[0043] In this embodiment of the application, the channel difference characteristics, channel ratio characteristics, polarization ratio characteristics, and environmental thermodynamic stratification parameters, etc., are explicitly calculated based on the polar cloud microphysics training dataset in step S101 and added to the polar cloud microphysics training dataset. This is to remove the background interference of high albedo of polar ice and snow and the influence of strong surface inversion on the radiative transfer solution space, thereby avoiding the misjudgment of cloud top height caused by the failure of assumptions in traditional physical models.

[0044] Optionally, when explicitly calculating channel difference characteristics, channel ratio characteristics, polarization ratio characteristics, and environmental thermodynamic stratification parameters, the following calculations can be made: split window difference, gas and water vapor absorption difference, shortwave infrared-window difference, and deep background and environmental constraint characteristics to obtain channel difference characteristics; brightness temperature ratio between different channels can be calculated based on brightness temperature to obtain channel ratio characteristics; the ratio of vertical and horizontal polarization temperature difference between the first and second frequency channels in an advanced microwave scanning radiometer can be calculated based on passive microwave data to obtain polarization ratio characteristics; and the difference between surface temperature and atmospheric temperature at the target air pressure value can be calculated to obtain environmental thermodynamic stratification parameters.

[0045] Among them, the channel ratio feature is used to eliminate the multiplicative scaling drift factor and characterize the relative slope change of the radiation curve; the polarization ratio feature characterizes the vertical integral total amount of water material inside the deep cloud column; and the environmental thermodynamic stratification parameter is used to correct the radiation transfer solution space under inversion conditions.

[0046] Specifically, a multidimensional derived feature matrix is ​​explicitly constructed as an input model, including: channel difference features (e.g., split window difference, gas and water vapor absorption difference, e.g., 6.7-7.3μm, shortwave infrared-window difference, e.g., 3.7-11μm), and deep background and environmental constraint features); channel ratio features (calculating the brightness-temperature ratio between different channels, e.g., b20 / b31, eliminating multiplicative scaling drift factors, and characterizing the relative slope changes of the radiation curve); polarization ratio features (PR) (calculating the ratio of vertical and horizontal polarization temperature differences between 89GHz and 36GHz, used to characterize the vertical integral total amount of water material inside deep cloud columns); and reanalyzed environmental thermodynamic stratification parameters (calculating the difference between surface temperature and 800hPa atmospheric temperature, directly inputting them into the network to correct the radiative transfer solution space under inversion conditions).

[0047] S103. Divide and train the new polar cloud microphysics training dataset according to the cloud phase classification label to obtain the trained polar cloud microphysics training dataset.

[0048] In this embodiment of the application, the cloud phase classification labels include water clouds, ice clouds and mixed clouds; the new polar cloud microphysics training dataset in step S102 is divided and trained according to the cloud phase classification labels to obtain the trained polar cloud microphysics training dataset for subsequent processing.

[0049] In some implementations, the input features in the polar cloud microphysics training dataset are divided according to the cloud phase classification labels to construct an independent physical feature set containing water clouds, ice clouds, and mixed clouds; the feature vectors in the independent physical feature set are then distributed to three independent regression subnetworks dedicated to water clouds, ice clouds, and mixed clouds for nonlinear mapping and regression training of cloud-water paths for the corresponding phases.

[0050] Specifically, based on the cloud phase classification labels, the input features are accurately diverted to independent regression subnetworks dedicated to liquid clouds, ice clouds, and mixed clouds for forward inference, i.e., nonlinear mapping and regression training, so as to learn the radiation and microphysical mapping laws specific to their own phases and eliminate the distribution deviation noise caused by the mixing of ice and water.

[0051] S104. Construct multiple independent encoding towers based on the trained polar cloud microphysics training dataset. Encode the input features in the polar cloud microphysics training dataset based on the independent encoding towers, and combine them with a shared fusion network to extract shared representation vectors for training, so as to obtain the trained inversion model. Inversion results are obtained by inversion through the inversion model.

[0052] In this embodiment, the independent coding towers include infrared coding towers, microwave coding towers, optical and geometric coding towers, and engineering feature towers. Multiple independent coding towers are constructed based on the polar cloud microphysics training dataset trained in step S103. Each independent coding tower encodes the input features in the polar cloud microphysics training dataset and combines them with a shared fusion network to extract shared representation vectors for training to obtain an inversion model. Subsequently, the cloud water path inversion can be performed using the inversion model to obtain the inversion result.

[0053] Optionally, when encoding the input features in the polar cloud microphysics training dataset based on the independent encoding tower and obtaining the shared representation vector by combining the shared fusion network, the input features in the polar cloud microphysics training dataset are independently and nonlinearly fused and encoded based on the independent encoding tower through linear layers, batch normalization and linear rectified activation functions; the features are then concatenated in the horizontal dimension to form a multimodal fusion feature vector, which is then fed into the shared fusion network to extract the shared representation vector.

[0054] Specifically, four independent coding towers are constructed (infrared coding tower, microwave coding tower, optical and geometric coding tower, and engineering feature tower). Each independent tower independently performs nonlinear fusion coding of the input features through linear layers, batch normalization, and linear rectified activation functions to avoid confusion of shallow features. Subsequently, the features are concatenated in the horizontal dimension to form a multimodal fusion feature vector, which is then fed into a shared fusion network to extract a shared representation vector.

[0055] Optionally, when training by combining the shared representation vector extracted by the shared fusion network, a year discriminator branch is set at the output of the shared fusion network, and the year is classified based on the year discriminator branch; a gradient reversal layer is set at the entrance of the year discriminator branch, so that the shared representation vector is passed identically during forward propagation, while the year classification cross-entropy loss gradient is multiplied by a negative adjustment coefficient and then passed back to the shared fusion network during back propagation, so that the output of the shared fusion network is a pure and universal representation that is immune to scaling decay.

[0056] Furthermore, the method includes performing a square root transformation on the cloud-water path target values ​​in the true parameters of the active radar cloud microphysics, and normalizing them on an intra-annual scale; then performing global standardization on the intra-annual normalized cloud-water path target values. This improves the training balance under long-tail distributions, thereby enabling the collaborative domain adversarial mechanism to suppress inter-annual drift artifacts.

[0057] Specifically, this application optimizes training based on a domain adversarial mechanism. A year discriminator branch is introduced after the shared representation layer of the shared fusion network, with a gradient reversal layer at the entry point. An identity transformation is performed during forward propagation, and during backpropagation, the cross-entropy loss gradient of the year classification is multiplied by a negative adjustment coefficient and fed back to the shared network. This forces the network to strip away the year label information and sensor drift behavior carried in the features, extracting a pure and universal representation immune to calibration decay. Simultaneously, square root transformation and double normalization are applied to the cloud water path target values ​​(true parameters of active radar cloud microphysics)—intra-year normalization and global normalization—to improve the training balance of the long-tailed distribution.

[0058] This application provides a polar cloud water path inversion method, which acquires multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysical ground truth parameters. All data are spatiotemporally matched to construct a polar cloud microphysical training dataset. Based on this training dataset, channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters are explicitly calculated to form multidimensional physical constraint features. These multidimensional physical constraint features are then added to the polar cloud microphysical training dataset to obtain a new training dataset. This new training dataset is then divided and trained according to cloud phase classification labels to obtain a trained polar cloud microphysical training dataset. Based on this trained dataset, multiple independent encoding towers are constructed. The input features in the training dataset are encoded using these independent encoding towers, and a shared representation vector is extracted and trained using a shared fusion network to obtain a trained inversion model. The inversion model is then used to obtain the inversion result. The polar cloud-water path inversion method proposed in this application explicitly calculates features such as differences, ratios, and polarization ratios using a polar cloud microphysics training dataset constructed from multi-source data. This establishes physical constraints, removes background interference from high albedo of polar ice and snow and the influence of strong surface inversion on the radiative transfer solution space, and solves the problem of confusion between ice and snow surface background and cloud signals. It can effectively identify the boundary between thin clouds and ice sheets, improve the certainty of estimation, avoid the failure of traditional physical forward radiative transfer models, and accurately separate cloud top and background radiation. It achieves independent regression of phase states, namely ice clouds, water clouds, and mixed clouds, avoids conflicts in bimodal and multimodal nonlinear distributions, and eliminates systematic biases caused by the mixing of ice and water phases. It improves the observation swath and time revisit rate, and can still meet the needs of large-scale continuous dynamic monitoring in the polar regions near the poles. Thus, it independently achieves high-precision inversion of polar all-weather conditions.

[0059] It should be noted that the polar cloud water path inversion method proposed in this application can be directly applied to operational meteorological and remote sensing systems to produce and publish a high-fidelity, all-weather, and spatially continuous remote sensing product set of polar cloud microphysical parameters. This mainly includes: High spatiotemporal resolution cloud water path products: Based on wide-swath data transmitted in real-time from satellite transits, high spatial resolution liquid water and ice water path inversion products are generated. A planar panoramic gridded climate dataset: Overcoming the data gaps of traditional passive optical products during the polar night and the orbital spatial blind spots of active radar in the deep polar regions (81.8°~90°), a large-scale, seamless gridded daily, monthly, quarterly, and annual global / polar regional climate standard dataset is generated.

[0060] Based on the same inventive concept, this application also provides a polar cloud and water path inversion device corresponding to the polar cloud and water path inversion method. Since the principle of the device in this application is similar to the polar cloud and water path inversion method described above in this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0061] Reference Figure 2 As shown, this is a polar cloud water path inversion device provided in an embodiment of this application. The device includes: Module 201 is used to acquire multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysical ground truth parameters, and to perform spatiotemporal matching on all data to construct a polar cloud microphysical training dataset; among which, the passive remote sensing radiation observation data includes at least thermal infrared emission band data, geometric parameters, and passive microwave data.

[0062] The acquisition module 202 is used to explicitly calculate channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters based on the polar cloud microphysics training dataset, form multidimensional physical constraint features, and add the multidimensional physical constraint features to the polar cloud microphysics training dataset to obtain a new polar cloud microphysics training dataset.

[0063] Training module 203 is used to divide and train the new polar cloud microphysics training dataset according to the cloud phase classification label, so as to obtain the trained polar cloud microphysics training dataset. The inversion module 204 is used to construct multiple independent encoding towers based on the trained polar cloud microphysics training dataset. The input features in the polar cloud microphysics training dataset are encoded based on the independent encoding towers, and the shared representation vector is extracted by the shared fusion network for training to obtain the trained inversion model. The inversion result is obtained by inversion through the inversion model.

[0064] In one possible implementation, the thermal infrared emission band data includes thermal infrared channel calibration radiation intensity; the construction module is specifically used for: Using preset cloud radar data as ground truth, spatiotemporal matching is performed to construct a polar cloud microphysics training dataset. The thermal infrared channel calibration radiation intensity is converted into a brightness temperature within the target range using a preset Planck blackbody radiation formula.

[0065] In one possible implementation, the passive microwave data includes the horizontal and vertical polarization brightness temperatures of a first frequency channel and a second frequency channel; the acquisition module is specifically used for: The channel difference characteristics are obtained by calculating the split window difference, the gas and water vapor absorption difference, the short-wave infrared-window difference, and the deep background and environmental constraint characteristics. Based on the brightness temperature, the brightness temperature ratio between different channels is calculated to obtain the channel ratio characteristics; among them, the channel ratio characteristics are used to eliminate the multiplicative scaling drift factor and characterize the relative slope change of the radiation curve. The polarization ratio characteristic is obtained by calculating the ratio of the vertical and horizontal polarization temperature difference between the first and second frequency channels in an advanced microwave scanning radiometer based on passive microwave data; the polarization ratio characteristic represents the vertical integral total amount of water material inside the deep cloud column. The difference between the surface temperature and the atmospheric temperature at the target air pressure is calculated to obtain the environmental thermodynamic stratification parameters; among them, the environmental thermodynamic stratification parameters are used to correct the radiative transfer solution space under inversion conditions.

[0066] In one possible implementation, the training module is specifically used for: The input features in the polar cloud microphysics training dataset are divided according to the cloud phase classification labels to construct an independent physical feature set containing water clouds, ice clouds and mixed clouds; The feature vectors in the independent physical feature set are split into three independent regression subnetworks dedicated to water cloud, ice cloud and mixed cloud to perform nonlinear mapping and regression training of cloud-water paths in the corresponding phase.

[0067] In one possible implementation, the inversion module is specifically used for: Based on the independent coding tower, the input features in the polar cloud microphysics training dataset are independently and nonlinearly fused and encoded using linear layers, batch normalization, and linear rectified activation functions. The features are concatenated and merged in the horizontal dimension to form a multimodal fusion feature vector, which is then fed into a shared fusion network to extract a shared representation vector.

[0068] In one possible implementation, the inversion module is specifically used for: A year discriminator branch is set at the output of the shared fusion network, and the year is classified based on the year discriminator branch of the shared representation vector. A gradient reversal layer is set at the entry point of the year discriminator branch, so that the shared representation vector is passed identically during forward propagation, while the year classification cross-entropy loss gradient is multiplied by a negative adjustment coefficient and then passed back to the shared fusion network during back propagation, so that the shared fusion network outputs a pure and universal representation that is immune to scaling decay.

[0069] In one possible implementation, the device further includes: The processing module is used to perform square root transformation on the cloud water path target values ​​in the true parameters of cloud microphysics of active radar and normalize them on an intra-annual scale. The execution module is used to perform global standardization on the normalized cloud water path target values ​​for the year.

[0070] This application provides the aforementioned polar cloud water path inversion device, which acquires multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysical ground truth parameters. It performs spatiotemporal matching on all data to construct a polar cloud microphysical training dataset. Based on this training dataset, it explicitly calculates channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters to form multidimensional physical constraint features. These multidimensional physical constraint features are then added to the training dataset to obtain a new training dataset. The new training dataset is then divided and trained according to cloud phase classification labels to obtain a trained training dataset. Multiple independent encoding towers are constructed based on this training dataset. The input features in the training dataset are encoded using these independent encoding towers, and a shared representation vector is extracted and trained using a shared fusion network to obtain a trained inversion model. The inversion model is then used to obtain the inversion result. The polar cloud-water path inversion device of this application explicitly calculates the differences, ratios, and polarization ratios of features in a polar cloud microphysics training dataset constructed from multi-source data, thereby establishing physical constraints. This eliminates the interference of high albedo background from polar ice and snow and the influence of strong surface inversion on the radiative transfer solution space, resolving the confusion between the ice and snow surface background and cloud signals. It can effectively identify the boundary between thin clouds and ice sheets, improving the certainty of estimation and avoiding the failure of traditional physical forward radiative transfer models. It can accurately separate cloud top and background radiation; achieve independent regression of phase states, namely ice clouds, water clouds, and mixed clouds, avoiding conflicts in bimodal and multimodal nonlinear distributions, and eliminating systematic biases caused by the mixing of ice and water phases; improves the observation swath and time revisit rate, and can still meet the needs of large-scale continuous dynamic monitoring in the polar regions near the poles, thus independently achieving high-precision inversion in polar all-weather conditions.

[0071] like Figure 3 As shown in the embodiment of this application, an electronic device 300 includes a processor 301, a memory 302, and a bus. The memory 302 stores machine-readable instructions that can be executed by the processor 301. When the electronic device is running, the processor 301 communicates with the memory 302 via the bus. The processor 301 executes the machine-readable instructions to perform the steps of the polar cloud water path inversion method described above.

[0072] Specifically, the memory 302 and processor 301 mentioned above can be general-purpose memory and processor, without any specific limitations. When the processor 301 runs the computer program stored in the memory 302, it can execute the above-mentioned polar cloud water path inversion method.

[0073] Corresponding to the above-described polar cloud water path inversion method, this application embodiment also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the above-described polar cloud water path inversion method.

[0074] The electronic devices and storage media provided in this application, through explicit calculation of differences, ratios, and polarization ratios in a polar cloud microphysics training dataset constructed from multi-source data, establish physical constraints, eliminate the interference of high albedo background from polar ice and snow and the influence of strong surface inversion on the radiative transfer solution space, solve the problem of confusion between ice and snow surface background and cloud signals, effectively identify the boundary between thin clouds and ice sheets, improve the certainty of estimation, avoid the failure of traditional physical forward radiative transfer models, and accurately separate cloud top and background radiation; achieve independent regression of phase states, namely ice clouds, water clouds, and mixed clouds, avoid conflicts in bimodal and multimodal nonlinear distributions, and eliminate systematic biases caused by the mixing of ice and water phases; improve the observation swath and time revisit rate, and still meet the needs of large-scale continuous dynamic monitoring in the polar regions near the poles, thereby independently realizing high-precision inversion in polar regions under all weather conditions.

[0075] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0076] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0077] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0078] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0079] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for inverting cloud water paths in polar regions, characterized in that, The method includes: Step S1: Obtain multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysics ground truth parameters, and perform spatiotemporal matching on all data to construct a polar cloud microphysics training dataset; wherein, the passive remote sensing radiation observation data includes at least thermal infrared emission band data, geometric parameters, and passive microwave data; Step S2: Based on the polar cloud microphysics training dataset, explicitly calculate the channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters to form multidimensional physical constraint features, and add the multidimensional physical constraint features to the polar cloud microphysics training dataset to obtain a new polar cloud microphysics training dataset. Step S3: Divide and train the new polar cloud microphysics training dataset according to the cloud phase classification label to obtain the trained polar cloud microphysics training dataset. Step S4: Construct multiple independent encoding towers based on the trained polar cloud microphysics training dataset. Encode the input features in the polar cloud microphysics training dataset based on the independent encoding towers, and combine them with a shared fusion network to extract shared representation vectors for training, thereby obtaining a trained inversion model. Inversion results are obtained by inverting through the inversion model.

2. The method according to claim 1, characterized in that, In step S1, the thermal infrared emission band data includes thermal infrared channel calibration radiation intensity; the step of performing spatiotemporal matching on all data to construct a polar cloud microphysics training dataset includes: Using preset cloud radar data as ground truth, spatiotemporal matching is performed to construct a polar cloud microphysics training dataset. The calibrated radiation intensity of the thermal infrared channel is converted into a brightness temperature within the target range using a preset Planck blackbody radiation formula.

3. The method according to claim 2, characterized in that, In step S2, the passive microwave data includes the horizontal and vertical polarization brightness temperatures of the first and second frequency channels; The explicit calculation of channel difference characteristics, channel ratio characteristics, polarization ratio characteristics, and environmental thermodynamic layering parameters includes: The channel difference characteristics are obtained by calculating the split window difference, the gas and water vapor absorption difference, the short-wave infrared-window difference, and the deep background and environmental constraint characteristics. Based on the brightness temperature, the brightness temperature ratio between different channels is calculated to obtain the channel ratio feature; wherein, the channel ratio feature is used to eliminate the multiplicative scaling drift factor and characterize the relative slope change of the radiation curve. Based on the passive microwave data, the ratio of the vertical and horizontal polarization temperature difference between the first and second frequency channels in the advanced microwave scanning radiometer is calculated to obtain the polarization ratio feature; wherein, the polarization ratio feature characterizes the vertical integral total amount of water material inside the deep cloud column. The difference between the surface temperature and the atmospheric temperature at the target air pressure is calculated to obtain the environmental thermodynamic stratification parameters; wherein, the environmental thermodynamic stratification parameters are used to correct the radiative transfer solution space under inversion conditions.

4. The method according to claim 1, characterized in that, In step S3, the process of dividing and training the new polar cloud microphysics training dataset based on cloud phase classification labels includes: The input features in the polar cloud microphysics training dataset are divided according to the cloud phase classification labels to construct an independent physical feature set containing water clouds, ice clouds and mixed clouds; The feature vectors in the independent physical feature set are split into three independent regression subnetworks dedicated to water cloud, ice cloud and mixed cloud to perform nonlinear mapping and regression training of cloud-water paths in the corresponding phase.

5. The method according to claim 1, characterized in that, In step S4, encoding the input features in the polar cloud microphysics training dataset based on the independent encoding tower and obtaining the shared representation vector by combining the shared fusion network includes: Based on the independent coding tower, the input features in the polar cloud microphysics training dataset are independently and nonlinearly fused and encoded using linear layers, batch normalization, and linear rectified activation functions. The features are concatenated and merged in the horizontal dimension to form a multimodal fusion feature vector, which is then fed into a shared fusion network to extract a shared representation vector.

6. The method according to claim 1, characterized in that, In step S4, the step of combining the shared fusion network to extract shared representation vectors for training includes: A year discriminator branch is set at the output of the shared fusion network, and the year is classified based on the year discriminator branch of the shared representation vector. A gradient reversal layer is set at the entry point of the year discriminator branch, so that the shared representation vector is passed identically during forward propagation, while the year classification cross-entropy loss gradient is multiplied by a negative adjustment coefficient and then passed back to the shared fusion network during back propagation, so that the shared fusion network outputs a pure and universal representation that is immune to scaling decay.

7. The method according to claim 6, characterized in that, The method further includes: The cloud water path target value in the true parameters of the active radar cloud microphysics is subjected to square root transformation and normalized on an intra-annual scale. Perform global standardization on the normalized cloud water path target values ​​for the year.

8. A polar cloud water path inversion device, characterized in that, The device includes: A construction module is used to acquire multi-source satellite passive remote sensing radiation observation data, meteorological reanalysis data, and active radar cloud microphysics ground truth parameters, and to perform spatiotemporal matching on all data to construct a polar cloud microphysics training dataset; wherein, the passive remote sensing radiation observation data includes at least thermal infrared emission band data, geometric parameters, and passive microwave data; The acquisition module is used to explicitly calculate channel difference features, channel ratio features, polarization ratio features, and environmental thermodynamic stratification parameters based on the polar cloud microphysics training dataset, form multidimensional physical constraint features, and add the multidimensional physical constraint features to the polar cloud microphysics training dataset to obtain a new polar cloud microphysics training dataset. The training module is used to divide and train the new polar cloud microphysics training dataset according to the cloud phase classification labels, so as to obtain the trained polar cloud microphysics training dataset. The inversion module is used to construct multiple independent encoding towers based on the trained polar cloud microphysics training dataset, encode the input features in the polar cloud microphysics training dataset based on the independent encoding towers, and combine them with a shared fusion network to extract shared representation vectors for training, thereby obtaining a trained inversion model, and then using the inversion model to perform inversion to obtain the inversion result.

9. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the polar cloud water path inversion method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the polar cloud water path inversion method as described in any one of claims 1 to 7.