Method for analyzing optical properties of cloud layers by fusing radiative transfer models with meteorological data

Abstract

The application provides a cloud optical property analysis method fusing a radiation transmission model and meteorological data, relates to the technical field of atmospheric radiation transmission, and comprises the following steps: acquiring multi-source meteorological data of a target area; performing fusion analysis on the multi-source meteorological data, calculating a mode determination coefficient, and determining a radiation transmission geometric mode according to the mode determination coefficient; for the planar parallel mode, calling a scalar radiation transmission model, solving an atmospheric scalar radiation transmission equation by using a CDISORT method, and calculating a first optical property parameter; for the spherical mode, calling a vector radiation transmission model, solving an atmospheric vector radiation transmission equation by using a reverse Monte Carlo method, and calculating a second optical property parameter; and outputting the first optical property parameter or the second optical property parameter as a cloud optical property analysis result. The reflectivity and linear polarization degree are accurately simulated.

Description

Technical Field

[0001] This invention relates to the field of atmospheric radiation transfer technology, and in particular to a method for analyzing the optical properties of clouds by fusing radiation transfer models and meteorological data. Background Technology

[0002] Atmospheric radiative transfer theory studies the interactions (absorption, scattering, and emission) of light radiation with atmospheric molecules, aerosols, and cloud particles as it travels through the atmospheric medium. It is widely used in fields such as laser communication, remote sensing imaging, climate modeling, and weather forecasting. Among these, ice crystal clouds have a particularly significant impact on laser transmission; their non-spherical shape, complex orientation distribution, and strong forward scattering characteristics pose a significant challenge to accurate calculations.

[0003] However, existing methods for calculating radiative transfer face challenges such as the difficulty of adapting the planar parallel assumption to the model of the actual spherical geometry, the imbalance between numerical calculation efficiency and accuracy caused by strong forward scattering, and the disconnect between meteorological observation data and radiative transfer models. These problems make it difficult to meet the needs of accurate simulation of laser transmission characteristics under complex conditions such as heterogeneous clouds, double-layered clouds, and cloud overlap in practical applications. Summary of the Invention

[0004] This invention addresses the limitations of existing radiative transfer calculation methods in terms of model adaptation, computational accuracy and efficiency, polarization effect simulation, and meteorological data utilization caused by the strong forward scattering, non-spherical shape, and complex spatial distribution of ice crystal particles in clouds. It provides a cloud optical characteristic analysis method that integrates radiative transfer models and meteorological data.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This invention provides a method for analyzing the optical properties of clouds by fusing radiative transfer models and meteorological data, including: Acquire multi-source meteorological data for the target area, wherein the multi-source meteorological data includes microphysical parameters of ice crystal particle clouds, geometric structure parameters of ice crystal particle clouds, and solar zenith angle, and the microphysical parameters include ice water content and effective particle radius; The multi-source meteorological data are fused and analyzed to calculate the model determination coefficients, and the radiative transfer geometric model is determined based on the model determination coefficients. The radiative transfer geometric model includes a planar parallel model or a spherical model. For the aforementioned planar parallel mode, the scalar radiative transfer model is invoked, and the CDISORT method is used to solve the atmospheric scalar radiative transfer equation to calculate the first optical characteristic parameters. For the spherical mode, the vector radiative transfer model is invoked, and the atmospheric vector radiative transfer equation is solved using the inverse Monte Carlo method to calculate the second optical characteristic parameters; Output the first optical characteristic parameter or the second optical characteristic parameter as the result of cloud optical characteristic analysis.

[0006] The beneficial effects of this invention are: Compared to existing technologies, this application acquires and fuses multi-source meteorological data to calculate model determination coefficients, thereby adaptively determining the radiative transfer geometry model. This achieves accurate description of the true characteristics of ice crystal particles in clouds and intelligent matching of model selection, avoiding calculation deviations caused by model misjudgment in traditional fixed threshold methods. For planar parallel models, an improved CDISORT method is used to solve the scalar radiative transfer equation, overcoming the computational bottleneck caused by strong forward scattering of ice crystal particles and improving the accuracy and efficiency of transmittance and reflectance calculations. For spherical models, an inverse Monte Carlo method is used to solve the vector radiative transfer equation, breaking through the technical difficulty of coupled solution of spherical geometry and polarization effects, and achieving accurate simulation of reflectance and linear polarization.

[0007] Through the above technical solution, this application realizes the deep integration of radiative transfer model and meteorological data, and can adaptively select the most suitable radiative transfer geometric mode and solution method according to real-time meteorological conditions. It effectively solves the problems of poor model adaptability, low accuracy of strong forward scattering calculation, difficulty in simulating polarization effect, and insufficient ability to describe heterogeneous clouds in the prior art. It provides accurate and reliable cloud optical characteristic analysis results for fields such as laser communication link budget, atmospheric correction of remote sensing images, assimilation of meteorological forecast data, and assessment of climate radiative forcing. Attached Figure Description

[0008] Figure 1 A flowchart illustrating the cloud optical property analysis method based on the fusion of radiative transfer model and meteorological data provided by this invention; Figure 2 This is a flowchart illustrating the calculation of the model determination coefficient in the cloud optical characteristic analysis method based on the fused radiative transfer model and meteorological data provided by the present invention. Detailed Implementation

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0010] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0011] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0012] Examples, such as Figure 1 As shown, this embodiment of the invention provides a method for analyzing the optical properties of clouds by fusing radiative transfer models and meteorological data, including: S10: Acquire multi-source meteorological data for the target area, wherein the multi-source meteorological data includes microphysical parameters of the ice crystal particle cloud layer, geometric structure parameters of the ice crystal particle cloud layer, and solar zenith angle, and the microphysical parameters include ice water content and effective particle radius.

[0013] Traditional radiative transfer calculations typically rely on idealized cloud assumptions or limited input parameters, making it difficult to accurately reflect the complex influence of ice crystal cloud layers on laser transmission in the real atmosphere.

[0014] Meanwhile, ice crystal particle clouds have unique characteristics such as non-spherical shape, complex orientation distribution, strong forward scattering, and non-uniform distribution in the horizontal and vertical directions. These characteristics directly determine the energy attenuation and polarization state change of lasers when passing through the clouds.

[0015] To address the aforementioned issues, this application acquires multi-source meteorological data for the target area. This multi-source meteorological data includes microphysical parameters of ice crystal particle clouds, geometric structural parameters of ice crystal particle clouds, and solar zenith angle. The microphysical parameters include ice-water content and effective particle radius, providing comprehensive physical input for subsequent fusion analysis, thereby overcoming the computational accuracy bottleneck caused by insufficient input information in traditional methods.

[0016] Specifically, step S10 in the method includes: Obtain information such as ice and water content, effective particle radius, cloud cover, cloud height, cloud geometric thickness, cloud vertical overlap pattern, ice crystal particle morphology, particle orientation distribution, and solar zenith angle from ground-based remote sensing equipment, spaceborne remote sensing equipment, or meteorological reanalysis data. The optical thickness is calculated based on the ice water content and the effective radius of the particles using Mie scattering theory or the T-matrix method. The ice-water content, the effective radius of the particles, the optical thickness, the morphology of the ice crystal particles, and the orientation distribution of the particles are used as the microphysical parameters of the ice crystal particle cloud. The cloud cover, cloud height, cloud geometric thickness, and cloud vertical overlap pattern are used as the geometric structural parameters of the ice crystal particle cloud layer.

[0017] In this embodiment, ice-water content, effective particle radius, cloud cover, cloud height, cloud geometric thickness, cloud vertical overlap pattern, ice crystal particle morphology, particle orientation distribution, and solar zenith angle are first obtained from ground-based remote sensing equipment, spaceborne remote sensing equipment, or meteorological reanalysis data. Ground-based remote sensing equipment includes ground-based observation equipment such as microwave radiometers, lidar, and solar photometers; spaceborne remote sensing equipment includes satellite-borne observation equipment such as spaceborne lidar and spaceborne imaging spectrometers; meteorological reanalysis data refers to gridded meteorological datasets generated by assimilating historical observation data through numerical weather prediction models, such as the ERA5 reanalysis data from the European Centre for Medium-Range Weather Forecasts.

[0018] Ice water content refers to the mass of ice crystal particles contained in a unit volume of ice crystal particle cloud, usually expressed in g / m³. 3 The unit is: effective radius of ice crystal particles, usually in μm; cloud cover refers to the proportion of the sky covered by ice crystal particles, ranging from 0 to 1, where 0 represents clear sky and 1 represents completely cloudy sky; cloud height refers to the vertical distance from the bottom of the ice crystal particle cloud layer to the ground, usually in km; and cloud geometric thickness refers to the vertical distance between the top and bottom of the ice crystal particle cloud layer, usually in km.

[0019] The vertical overlap pattern of clouds refers to the arrangement of multiple layers of ice crystal particles in the vertical direction, including the maximum overlap pattern, the random overlap pattern, or a mixed overlap pattern between the two: the maximum overlap pattern means that the ice crystal particle cloud layers are completely aligned in the horizontal direction, which is suitable for describing stratiform clouds formed by large-scale weather systems; the random overlap pattern means that the ice crystal particle cloud layers are independently and randomly distributed in the horizontal direction, which is suitable for describing convective clouds or fragmented clouds; the mixed overlap pattern is between the two, with partial overlap and partial randomness. Among them, the morphology of ice crystal particles refers to the geometric shape of ice crystal particles, such as solid columnar, hollow columnar, plate-like, polymer, etc. Different morphologies of ice crystal particles have significantly different scattering characteristics.

[0020] Among them, particle orientation distribution refers to the orientation distribution pattern of ice crystal particles in space, such as random orientation or horizontal orientation: random orientation means that the orientation of ice crystal particles is completely random, which is suitable for describing ice crystal particle clouds with sufficient turbulent mixing; horizontal orientation means that the main axis of ice crystal particles tends to be horizontal, which is suitable for describing ice crystal particles falling slowly under weak turbulent conditions.

[0021] The solar zenith angle is the angle between the sun's rays and the zenith direction.

[0022] For example, ice-water content, effective particle radius, cloud height, and cloud geometric thickness can be retrieved from CALIOP spaceborne lidar data. The intensity of the backscattered signal received by CALIOP is related to the ice-water content; the depolarization ratio can be used to distinguish ice crystal particles from water clouds and infer the morphology of ice crystal particles; and the signal attenuation rate can be used to calculate the optical thickness, which, combined with the ice-water content, can then be used to estimate the effective particle radius. Cloud cover, cloud vertical overlap pattern, and solar zenith angle can be directly obtained from ERA5 meteorological reanalysis data.

[0023] Secondly, the optical thickness is calculated using Mie scattering theory or the T-matrix method based on the ice-water content and the effective radius of the particles. Optical thickness refers to the attenuation ability of the ice crystal cloud to radiation; it is a comprehensive reflection of the ice-water content, effective particle radius, and cloud geometric thickness, and is dimensionless. Specifically, Mie scattering theory is suitable for calculating the optical properties of spherical particles, while the T-matrix method is suitable for calculating the optical properties of non-spherical particles. Since ice crystal particles are usually non-spherical, the T-matrix method is preferred for calculating the optical thickness.

[0024] For example, the calculation process of Mie scattering theory is as follows: Consider a laser beam with wavelength λ incident on a spherical particle with diameter d and refractive index n. _p The refractive index of the surrounding medium is n _med First, calculate the dimensional parameter x = πdn. _med / λ and relative refractive index m=n _p / n _med Then, the scattering coefficient α is solved using the Mie scattering algorithm. _n and b _n Thus, the scattering efficiency factor Q is obtained. _sca Absorption efficiency factor Q _abs and extinction efficiency factor Q _ext =Q _sca +Q _abs For particle swarm optimization, the number of particles per unit volume is needed, which is called the number density. The extinction cross section of a single particle is The volume extinction coefficient Optical thickness , where L is the geometric thickness of the cloud layer.

[0025] For example, for a polystyrene microsphere with a diameter of 0.304 μm (refractive index 1.5721) suspended in water (refractive index 1.3316), with an incident wavelength of 0.6328 μm, the size parameter x≈2.01 and the relative refractive index m≈1.18 can be calculated. The scattering efficiency factor Q can then be obtained using the Mie algorithm. _sca ≈0.197, and parameters such as the extinction coefficient can then be calculated.

[0026] For example, the T-matrix method is an accurate numerical method applicable to non-spherical particles. Its basic principle is to expand the incident and scattered fields using vector spherical harmonic functions, and establish a linear transformation relationship between the incident and scattered field expansion coefficients by solving the particle's T-matrix. The T-matrix depends only on the particle's own properties (shape, size, refractive index, orientation) and the incident wavelength, and is independent of the incident field direction. Therefore, it is particularly efficient for calculating the ensemble-averaged scattering characteristics of randomly oriented particles. For ice crystal particles, the morphology of the ice crystal (e.g., solid column, hollow column, plate), characteristic dimensions (e.g., column length, diameter), the complex refractive index of the ice (varying with wavelength), the incident wavelength, and the assumed orientation distribution of the particles are required. After obtaining the T-matrix by solving the integral equations of Maxwell's equations under particle boundary conditions, the scattering phase function, Mueller matrix, extinction cross section, scattering cross section, and single-scattering albedo can be calculated for incident in any direction.

[0027] For example, for an ice crystal particle cloud with an effective particle radius of 30 μm, if the ice water content is 0.1 g / m³ 3 The geometric thickness of the cloud layer is 1 km, and the optical thickness can be calculated to be approximately 5.2 km using the T-matrix method.

[0028] Furthermore, ice water content, effective particle radius, optical thickness, ice crystal particle morphology, and particle orientation distribution were used as microphysical parameters of the ice crystal particle cloud. These parameters collectively describe the number, size, shape, and spatial orientation of ice crystal particles, and are the basis for determining its scattering characteristics.

[0029] Finally, cloud cover, cloud height, cloud geometric thickness, and vertical overlap pattern were used as geometric structural parameters of the ice crystal cloud layer. These parameters collectively describe the distribution characteristics of the ice crystal cloud layer in three-dimensional space and are important criteria for determining the geometric pattern of radiative transfer.

[0030] In summary, compared to existing technologies, this application acquires multi-source meteorological data for the target area. This multi-source meteorological data includes microphysical parameters of the ice crystal cloud layer, geometric structural parameters of the ice crystal cloud layer, and the solar zenith angle. The microphysical parameters include ice-water content and effective particle radius. This provides comprehensive and accurate input parameters for subsequent fusion analysis, accurately reflecting the microphysical characteristics, spatial distribution, and incident geometry of the ice crystal cloud layer, thereby overcoming the computational accuracy bottleneck caused by insufficient input information in traditional methods.

[0031] S20: Perform fusion analysis on the multi-source meteorological data, calculate the model determination coefficient, and determine the radiative transfer geometric model based on the model determination coefficient, wherein the radiative transfer geometric model includes a planar parallel model or a spherical model.

[0032] In traditional radiative transfer calculations, the choice between planar parallel mode and spherical mode usually depends on a fixed threshold, such as using spherical mode when the solar zenith angle is greater than 60°. This mechanical selection method cannot adapt to the complex and varied non-uniform structure of ice crystal particle clouds in the real atmosphere.

[0033] Meanwhile, factors such as cloud cover, vertical overlap patterns of clouds, and internal heterogeneity of clouds have a significant impact on radiative transmission, and relying solely on a single angle threshold to determine the pattern will result in significant errors.

[0034] To address the aforementioned issues, this application performs fusion analysis on the multi-source meteorological data, calculates the model determination coefficient, and determines the radiative transfer geometric model based on the model determination coefficient. The radiative transfer geometric model includes a planar parallel model or a spherical model, thereby overcoming the limitations of the traditional fixed threshold method and achieving adaptive matching with the actual cloud features.

[0035] Specifically, such as Figure 2 As shown, step S20 in the method includes: The microphysical parameters, geometric parameters, and solar zenith angle of the ice crystal particle cloud layer are input into a pre-trained cloud inhomogeneity discrimination model, and the cloud inhomogeneity index is output. The solar zenith angle is normalized to obtain the standard solar zenith angle; The model determination coefficient is calculated by weighting and fusing the standard solar zenith angle, cloud cover, and cloud inhomogeneity index.

[0036] In this embodiment, the microphysical parameters, geometric parameters, and solar zenith angle of the ice crystal particle cloud layer are first input into a pre-trained cloud inhomogeneity discrimination model, which then outputs a cloud inhomogeneity index. The cloud inhomogeneity discrimination model, trained on a large number of samples, can automatically learn the relationship between the microphysical parameters, geometric parameters, and solar zenith angle of the ice crystal particle cloud layer and the complexity of its internal structure, thus outputting a cloud inhomogeneity index characterizing the degree of inhomogeneity of the ice crystal particle cloud layer.

[0037] The cloud heterogeneity index is a value ranging from 0 to 1. The larger the value, the more heterogeneous the internal structure of the ice crystal particles in the cloud and the more significant the horizontal radiative transfer effect.

[0038] For example, for a horizontally uniform single-layer ice crystal particle cloud, the cloud inhomogeneity index is close to 0.1; for a horizontally uniform double-layer ice crystal particle cloud, the cloud inhomogeneity index is about 0.3; and for a broken convective cloud, the cloud inhomogeneity index can reach more than 0.7.

[0039] Secondly, the solar zenith angle is normalized to obtain the standard solar zenith angle. Specifically, the degree of the solar zenith angle is divided by 90° to obtain the standard solar zenith angle with a value ranging from 0 to 1.

[0040] For example, if the solar zenith angle is 30°, the normalized standard solar zenith angle = 30 / 90 = 0.333.

[0041] Finally, the model determination coefficient was calculated by weighting and fusing the standard solar zenith angle, cloud cover, and cloud inhomogeneity index. The model determination coefficient is a comprehensive index; a higher value indicates a stronger preference for a spherical model, while a lower value indicates a stronger preference for a planar parallel model.

[0042] For example, the weighted fusion can adopt a linear weighted formula: mode determination coefficient = α × standard solar zenith angle + β × (1 - cloud cover) + γ × cloud inhomogeneity index, where α, β, and γ are weighting coefficients, and α + β + γ = 1. The weighting coefficients can be dynamically set according to the type of ice crystal particles in the cloud. Preferably, for cirrus clouds, due to their high altitude and large influence range, the weight of the solar geometric position is relatively high, and α = 0.5, β = 0.2, and γ = 0.3 can be taken; for cumulus clouds, due to their small horizontal scale and strong inhomogeneity, the weight of the cloud inhomogeneity index is relatively high, and α = 0.3, β = 0.2, and γ = 0.5 can be taken.

[0043] For example, if the standard solar zenith angle is 0.333, the cloud cover is 0.8, the cloud heterogeneity index is 0.4, and the ice crystal particle cloud type is cirrus, then the cirrus weighting coefficients α=0.5, β=0.2, and γ=0.3 are adopted, and the model determination coefficient = 0.5×0.667+0.2×(1-0.8)+0.3×0.4=0.4935.

[0044] Specifically, the construction process of the "cloud non-uniformity discrimination model" includes: An initial cloud non-uniformity discrimination model was constructed based on a deep learning algorithm; Multiple sets of sample data are collected from satellite observation data or high-resolution cloud simulation data. Each set of sample data includes the microphysical parameters of the ice crystal particle cloud, the geometric structure parameters of the ice crystal particle cloud, and the solar zenith angle. For each set of sample data, a three-dimensional radiative transfer model is used to calculate the corresponding reference inhomogeneity index, and the reference inhomogeneity index is used as the training label. Using the sample data as input and the training labels as supervision signals, supervised training is performed on the initial cloud inhomogeneity discrimination model until verification convergence is achieved, resulting in a trained cloud inhomogeneity discrimination model.

[0045] In this embodiment, an initial cloud unevenness discrimination model is first constructed based on a deep learning algorithm. For example, a convolutional neural network can be used to construct the initial cloud unevenness discrimination model. This model mainly consists of an input layer, multiple convolutional blocks, a flattening layer, and a fully connected output layer. The model composition and parameter configuration are as follows: 1. Regarding model composition, firstly, multi-source meteorological data is received through the input layer. The multi-source meteorological data is organized into a multi-channel two-dimensional grid. Each grid point contains values ​​for parameters such as ice water content, effective particle radius, cloud cover, and cloud height. The solar zenith angle is used as a global parameter input to the subsequent fully connected layer. Subsequently, feature extraction is performed through multiple convolutional blocks. Each convolutional block... The block sequentially contains convolutional layers, activation functions, and pooling layers. The convolutional layers extract local spatial features of the cloud field, such as cloud texture, edges, and clustering patterns, through sliding convolutional kernels. The activation functions introduce nonlinear transformations, and the ReLU function can be used to accelerate convergence and alleviate the gradient vanishing problem. The pooling layers use max pooling to reduce the dimension of the feature map, reduce computation, and enhance the translation invariance of the model. Finally, the flattening layer converts the multidimensional feature map output by the pooling layer into a one-dimensional feature vector, which is input to the output module composed of multiple fully connected layers. Finally, a single neuron outputs a cloud inhomogeneity index in the range of 0 to 1.

[0046] 2. Regarding parameter configuration, the following configuration can be used as an example: The feature map received by the input layer is 64×64 grid size, containing 8 input channels (corresponding to ice water content, effective particle radius, cloud cover, cloud height, cloud geometric thickness, cloud vertical overlap pattern encoding, ice crystal particle morphology encoding, and particle orientation distribution encoding, respectively). The first convolutional block is configured with 32 3×3 convolutional kernels with a stride of 1, followed by a ReLU activation function and a 2×2 max pooling layer (with a stride of 2); the second convolutional block is configured with 64 3×3 convolutional kernels, also followed by a ReLU activation and a 2×2 max pooling layer; the third convolutional block is configured with 128 3×3 convolutional kernels, followed by a ReLU activation and a global average pooling layer. After flattening, two fully connected layers are connected. The first fully connected layer contains 256 neurons and is equipped with ReLU activation and Dropout (dropout rate 0.5) to prevent overfitting; the output layer is a fully connected layer with 1 neuron, using a Sigmoid activation function to output a cloud inhomogeneity index of 0~1. In addition, a batch normalization layer can be added after each convolutional layer to accelerate training convergence and enhance model stability.

[0047] Secondly, multiple sets of sample data are collected from satellite observation data or high-resolution cloud simulation data. Each set of sample data includes the microphysical parameters of the ice crystal particle cloud layer, the geometric structure parameters of the ice crystal particle cloud layer, and the solar zenith angle. The sample data comes from at least two main sources: the first source is satellite observation data. For example, a large number of real ice crystal particle cloud layer parameter samples can be obtained from satellite products such as the Medium Resolution Imaging Spectroradiometer (MODIS) or the Cloud-Aerosol LiDAR (CALIOP). Each set of samples includes: latitude and longitude location, observation time, ice-water content, effective particle radius, cloud cover, cloud height, cloud geometric thickness, cloud vertical overlap pattern, and solar zenith angle. The second source is high-resolution cloud simulation data. For example, a large eddy simulation model (e.g., WRF-LES) is used to generate high-resolution ice crystal particle cloud scenes with a spatial resolution of up to 100 meters and a temporal resolution of up to minutes. A large number of ice crystal particle cloud samples with different degrees of inhomogeneity can be extracted from the simulation results, covering the complete spectral range from uniform stratiform clouds to broken convective clouds.

[0048] Next, for each set of sample data, a corresponding reference inhomogeneity index is calculated using a three-dimensional radiative transfer model, and this reference inhomogeneity index is used as the training label. The three-dimensional radiative transfer model can be either an SHDOM or a three-dimensional Monte Carlo model. Specifically, each set of sample data is used as input to construct a three-dimensional cloud field, and the radiative transfer results incident from multiple directions are calculated. Then, by analyzing the degree of variation of the radiation field at different horizontal positions, the inhomogeneity of the ice crystal particle cloud layer is quantified.

[0049] For example, the specific operation steps are as follows: First, each set of sample data is used as input to construct the corresponding ice crystal particle cloud scene in three-dimensional space. This includes setting the horizontal grid resolution, such as 1km×1km, vertical stratification, and microphysical parameters such as ice-water content and effective particle radius within each grid. The three-dimensional distribution of the cloud layer is then generated based on cloud cover and the vertical overlap pattern of the cloud layers. Second, the radiative transfer calculation parameters are set, including the incident solar direction (preferably, a fixed zenith angle, such as 30°, can be used to eliminate geometric influences), wavelength (e.g., 0.55μm for visible light), and underlying surface boundary conditions (e.g., assuming a Lambertian reflectance of 0). A three-dimensional radiative transfer model is called for forward simulation to calculate the radiative intensity of each grid point in the entire three-dimensional cloud field in different directions. This is then integrated to obtain the downward or upward radiative flux at each horizontal grid position (x, y). Then, the radiative flux values ​​at all horizontal positions are extracted to form a two-dimensional array, and the mean and standard deviation of this two-dimensional array are calculated. The inhomogeneity index is defined as the coefficient of variation = standard deviation / mean. The coefficient of variation reflects the degree of dispersion of radiation flux in the horizontal direction. The larger the value, the more uneven the internal structure of the cloud and the more significant the horizontal radiative transfer effect.

[0050] For example, suppose a 3D cloud field constructed from a sample contains 100×100 horizontal grids. The downward radiative flux of each grid is calculated using SHDOM, and the average value is found to be 320 W / m. 2 Standard deviation = 48 W / m 2 Therefore, the reference non-uniformity index = 48 / 320 = 0.15. The reference non-uniformity index will be used as a training label to supervise the learning of the cloud non-uniformity discrimination model.

[0051] Finally, using the sample data as input and the training labels as supervision signals, supervised training is performed on the initial cloud inhomogeneity discrimination model until validation convergence, resulting in the trained cloud inhomogeneity discrimination model. For example, the following technical path can be used to train the cloud inhomogeneity discrimination model: 1. Data preparation: The collected sample data and corresponding training labels are randomly divided into training, validation, and test sets in an 8:1:1 ratio. The training set is used for model parameter updates, the validation set is used to monitor the training process and adjust hyperparameters, and the test set is used for final model performance evaluation.

[0052] 2. Model Training: Mean squared error is used as the loss function to calculate the difference between the predicted cloud inhomogeneity index output by the model and the reference inhomogeneity index label. The Adam optimizer is selected, with an initial learning rate set to 0.001. The Adam optimizer can adaptively adjust the learning rate, accelerating convergence and improving training stability. The batch size can be set to 32 or 64 based on the GPU memory size, meaning that a batch of data is randomly selected from the training set for gradient calculation and parameter updates in each iteration. The maximum number of training epochs is set to 200, with each epoch representing a complete traversal of the training set. After each training epoch, the loss value of the current model on the validation set is calculated. If the validation loss does not decrease for 10 consecutive epochs, training is stopped early to prevent overfitting. If the validation loss does not improve within 5 consecutive epochs, the learning rate is multiplied by 0.5 for decay, allowing the model to converge more finely. After each epoch, the model parameters with the minimum validation loss are saved as the final trained cloud inhomogeneity discrimination model. Training terminates when the early stopping condition is met or the maximum number of training rounds is reached. The best model saved at this time is the cloud inhomogeneity discrimination model that has been trained and can be used to calculate the subsequent mode determination coefficients.

[0053] Furthermore, step S20 of the method also includes: The mode determination coefficient, optical thickness, and solar zenith angle are input into the fuzzy logic decision system. The fuzzy logic decision system includes a predefined membership function and a fuzzy rule base. The membership function is used to convert the mode determination coefficient, the optical thickness, and the solar zenith angle into corresponding fuzzy set membership degrees. The fuzzy rule base contains multiple fuzzy rules based on the physical principle of radiative transfer. The fuzzy set is obtained by performing rule activation and conclusion aggregation on the membership degree of the fuzzy set through fuzzy reasoning; The output fuzzy set is defuzzified, and the mode selection confidence score is calculated. Based on the selected confidence level, the radiative transfer geometry mode is determined to be either a planar parallel mode or a spherical mode.

[0054] In this embodiment, the mode determination coefficient, optical thickness, and solar zenith angle are first input into the fuzzy logic decision system. The fuzzy logic decision system is a decision system built on fuzzy logic theory, used to handle decision problems with fuzzy boundaries, such as mode selection.

[0055] Secondly, the fuzzy logic decision system includes predefined membership functions and a fuzzy rule base. The membership functions are used to convert mode determination coefficients, optical thickness, and solar zenith angle into corresponding fuzzy set membership degrees. The fuzzy rule base contains multiple fuzzy rules based on the physical principles of radiative transfer. Specifically, for mode determination coefficients (hereinafter denoted by D), three fuzzy sets are defined: a small set, a medium set, and a large set. Each set corresponds to a membership function. Based on this membership function, specific mode determination coefficients are mapped to the fuzzy set membership degrees of that set. For example, a trapezoidal membership function can be defined as follows: ; ; In the formula, , , Let represent the membership functions of the small set, medium set, and large set, respectively, and D be the pattern determination coefficient. By substituting the membership functions of the small set, medium set, and large set respectively, the corresponding fuzzy set membership can be calculated.

[0056] The reason and logic behind the above parameter settings are as follows: the model determination coefficient D is a comprehensive index obtained by weighted fusion of solar zenith angle, cloud cover, and cloud inhomogeneity index, and its value range has been normalized to [0,1]. The thresholds of 0.3 and 0.7 are selected because: when D≤0.3, the deviation between the calculation results of the planar parallel model and the spherical model is usually less than the acceptable engineering error, thus it can be clearly classified as a small set; when D≥0.7, the deviation has significantly exceeded the tolerance range, and the spherical model must be used, hence it is classified as a large set. The middle range of 0.3~0.7 is the fuzzy transition zone, where 0.5 is the peak value of the middle set, representing the most fuzzy state of decision-making. The design of a linear transition interval width of 0.2 ensures the smoothness of membership degree changes and avoids abrupt changes in model selection due to small fluctuations in D.

[0057] Among them, optical thickness (hereinafter referred to as...) (Representation), defining three fuzzy sets: thin set, medium set, and thick set. For example, the membership function can be designed based on the typical range of optical thickness of ice crystal particle clouds: ; ; In the formula, , , These represent the membership functions for thin, medium, and thick sets, respectively. For optical thickness, the optical thickness By substituting the membership functions of the thin set, medium set, and thick set respectively, the corresponding fuzzy set membership can be calculated.

[0058] The reason and logic behind the above parameter settings is: optical thickness It is a key parameter that determines the intensity of multiple scattering. According to the theory of radiative transfer, when At this time, single scattering dominates, while multiple scattering contributes less. Planar parallel modes are usually accurate enough, therefore... Defined as a thin set; when At that time, multiple scattering was already extremely significant, the effective optical path increased dramatically, and the spherical geometric effects could not be ignored, therefore... Defined as a thick set; the middle This is a transition zone, and other factors need to be considered in conjunction with it for comprehensive judgment. The thresholds of 22 and 10 are because the optical thickness of most cirrus clouds is concentrated between 0.1 and 10, and these are typical dividing points derived from multiple scattering theory analysis. The width of the linear transition interval is determined based on the difference between adjacent thresholds: a thin-to-medium transition bandwidth of 3 and a medium-to-thick transition bandwidth of 5, ensuring continuous change in membership degree.

[0059] For the solar zenith angle (hereinafter referred to as θ), three fuzzy sets are defined: a small set, a medium set, and a large set. For example, each set corresponds to a membership function: ; ; In the formula, , , These represent the membership functions for the small set, medium set, and large set, respectively. The solar zenith angle is the angle of the sun. By substituting the membership functions of the small set, medium set, and large set respectively, the corresponding fuzzy set membership can be calculated.

[0060] The reason and logic behind the above parameter settings are as follows: the solar zenith angle θ directly affects the path length of light passing through the atmosphere and the degree of influence of the Earth's curvature. When θ ≤ 30°, the error between the planar parallel approximation and the exact spherical solution is usually within 5%, which can be considered a small set; when θ ≥ 60°, the error increases sharply, and a spherical model must be used, hence it is defined as a large set; the middle 30°~60° is the transition zone, with 45° serving as the symmetry center of the medium set. The selection of a linear transition bandwidth of 15° balances smoothness and physical distinguishability, ensuring that the membership degree changes gradually within the typical zenith angle range.

[0061] The fuzzy rule base contains multiple fuzzy rules, each in the form of "if (precondition) then (conclusion)". The precondition is composed of the fuzzy state of the input parameters, and the conclusion is the radiative transfer geometric mode.

[0062] Thus, by using a predefined membership function, the precise values ​​of the input parameters can be converted into the membership degrees of the corresponding fuzzy sets, thereby enabling the fuzzy description of continuous physical quantities such as mode determination coefficients, optical thickness, and solar zenith angle. Through the fuzzy rule base, expert knowledge based on the physical principles of radiative transfer can be formalized into multiple fuzzy rules, thereby realizing the logical mapping from the fuzzy state of the input parameters to the geometric mode decision of radiative transfer during the fuzzy inference process.

[0063] Next, fuzzy inference is used to activate the membership degrees of the fuzzy set through rule activation and conclusion aggregation, resulting in the output fuzzy set. The fuzzy inference process includes two main steps: rule activation and conclusion aggregation. Rule activation refers to calculating the activation strength of each rule based on the membership function of the input parameters. For a rule, its preconditions consist of multiple sub-conditions, and the activation strength is usually the minimum value or product of the membership degrees of each sub-condition. For example, in a certain scenario... =0.6, =0.4, =0.8, then the activation intensity is min( =0.6, =0.4, =0.8)=0.4, if the calculated activation strength is greater than 0, then the rule is activated.

[0064] Conclusion aggregation refers to combining the conclusions of all activated rules to form an output fuzzy set. In a fuzzy logic system, the conclusion of each rule is itself a fuzzy set, such as "planar parallel mode" or "spherical mode". The sum of the activation intensities of rules pointing to different conclusions is counted to obtain the output fuzzy set.

[0065] For example, suppose three rules are activated in a certain reasoning: Rule 1 (planar parallel mode), activation intensity α1=0.6, Rule 2 (spherical mode), activation intensity α2=0.8, Rule 3 (spherical mode), activation intensity α3=0.4. The sum of the activation intensities of the rules that lead to the conclusion of the planar parallel mode is S_planar=0.6, and the sum of the activation intensities of the rules that lead to the conclusion of the spherical mode is S_spherical=0.8+0.4=1.2. The integrated result is the output fuzzy set: (planar parallel mode: 0.6, spherical mode: 1.2).

[0066] Furthermore, the output fuzzy set is defuzzified to calculate the mode selection confidence score. For example, the mode selection confidence score can be calculated using a normalization method: S_sphere / (S_plane + S_sphere), where the mode selection confidence score is a value ranging from [0,1], used to quantitatively represent the degree of confidence that the spherical mode should be used under the current input conditions. A mode selection confidence score of 0 indicates a complete preference for the planar parallel mode, a mode selection confidence score of 1 indicates a complete preference for the spherical mode, and a mode selection confidence score of 0.5 indicates that the two modes have equal preference and are in a completely fuzzy boundary state.

[0067] For example, suppose the output fuzzy set in a certain inference is: (planar parallel mode: 0.6, spherical mode: 1.2), then the mode selection confidence = 1.2 / (0.6 + 1.2) ≈ 0.67.

[0068] Finally, the radiative transfer geometry mode is determined to be either a planar parallel mode or a spherical mode based on the mode selection confidence level. Since a mode selection confidence level of 0.5 indicates that the two modes have equal propensity and is the critical point for decision-making, 0.5 is used as the decision threshold: if the mode selection confidence level is ≥0.5, it is determined to be a spherical mode; otherwise, if the mode selection confidence level is <0.5, it is determined to be a planar parallel mode.

[0069] For example, the calculated mode selection confidence level is 0.67 ≥ 0.5, which means that the confidence level of adopting the spherical mode is higher than that of the planar parallel mode. Therefore, the current radiative transfer geometry mode is determined to be the spherical mode.

[0070] For example, based on the above explanation, the following is a complete example of determining the radiative transfer geometry mode: assuming the mode determination coefficient D = 0.45, and the optical thickness... 1. Solar zenith angle θ = 50°. First, substitute the values ​​into the predefined membership function to calculate the membership degree: According to the aforementioned membership function, for the pattern determination coefficient D = 0.45: when belonging to a small set: =(0.5-0.45) / 0.2=0.25, when it belongs to the middle set: =(0.45-0.3) / 0.2=0.75, when it belongs to a large set: =0. For optical thickness When it belongs to a thin set: =0, when it belongs to the middle set: =(10-8) / 5=0.4, when it belongs to a thick set: =(8-5) / 5=0.6. For a solar zenith angle θ=50°: when belonging to a small set: =0, when it belongs to the middle set: =(60-50) / 15=0.67, when it belongs to a large set: =(50-45) / 15=0.33.

[0071] For example, the calculated membership degree is then input into the fuzzy rule base. For instance, for the rule "First rule, if the mode determination coefficient belongs to a small set, the optical thickness belongs to a thin set, and the solar zenith angle belongs to a small set, then the radiative transfer geometry mode is a planar parallel mode", the activation intensity is min( =0.25, =0, If the activation strength is 0, then the first rule is not activated. Similarly, for the second rule, the activation strength is min( =0, =0, If (=0.33)=0, then the second rule is not activated. For the third rule, the activation strength = min( =0.75, If the value is 0.6, then the third rule is activated. For the fourth rule, the activation strength is min( =0.25, =0.6)=0.25, then the fourth rule is activated. Regarding the fifth rule, if the mode determination coefficients belong to a large set, the optical thickness belongs to a thin set, and the solar zenith angle belongs to a small set, then the radiative transfer geometry mode is a planar parallel mode, and the activation intensity = min( =0, =0, If (=0)=0, then the fifth rule is not activated.

[0072] For example, since both the activated third and fourth rules point to the spherical mode, the output fuzzy set obtained through conclusion aggregation is (planar parallel mode: 0, spherical mode: 0.85). Further defuzzification of the output fuzzy set yields a mode selection confidence score of 0.85 / (0+0.85) = 1, indicating that under the current input conditions, the confidence level for the spherical mode reaches its maximum, completely favoring the spherical mode. Finally, based on a mode selection confidence score ≥ 0.5, the radiative transfer geometry mode is determined to be the spherical mode. Moreover, this determination is consistent with physical expectations: although the solar zenith angle is at a moderate level (50°), the optical thickness is relatively thick (…). Furthermore, the mode determination coefficients are in the middle set, and due to a combination of factors, a spherical mode must be used to accurately describe the radiative transfer process.

[0073] Specifically, the "fuzzy rule base" includes at least one of the following rules: The first rule states that if the mode determination coefficients belong to a small set, the optical thickness belongs to a thin set, and the solar zenith angle belongs to a small set, then the radiative transfer geometry mode is a planar parallel mode. The second rule states that if the mode determination coefficients belong to a large set, the optical thickness belongs to a thick set, and the solar zenith angle belongs to a large set, then the radiative transfer geometry mode is a spherical mode. The third rule states that if the mode determination coefficients belong to the middle set and the optical thickness belongs to the thick set, then the radiative transfer geometric mode is a spherical mode. The fourth rule states that if the mode determination coefficients belong to a small set and the optical thickness belongs to a thick set, then the radiative transfer geometry mode is a spherical mode. The fifth rule states that if the mode determination coefficients belong to a large set, the optical thickness belongs to a thin set, and the solar zenith angle belongs to a small set, then the radiative transfer geometry mode is a planar parallel mode.

[0074] In this embodiment, the first rule stipulates that if the mode determination coefficient belongs to a small set, the optical thickness belongs to a thin set, and the solar zenith angle belongs to a small set, then the radiative transfer geometry mode is a planar parallel mode. The physical basis of this rule is that when the mode determination coefficient is small (indicating that the overall characteristics of the ice crystal cloud layer tend to be a planar mode), the optical thickness is thin (indicating weak multiple scattering), and the solar zenith angle is small (indicating that the light path is close to perpendicular), the assumptions of a planar parallel mode are fully satisfied, therefore a planar parallel mode should be adopted.

[0075] In this embodiment of the application, the second rule stipulates that if the mode determination coefficient belongs to a large set, the optical thickness belongs to a thick set, and the solar zenith angle belongs to a large set, then the radiative transfer geometry mode is a spherical mode. The physical basis of this rule is that when the mode determination coefficient is large (indicating that the overall characteristics of the ice crystal cloud tend to be spherical), the optical thickness is thick (indicating strong multiple scattering), and the solar zenith angle is large (indicating significant curvature of the light path), the curvature of the Earth and horizontal radiative transfer must be considered, therefore a spherical mode should be adopted.

[0076] In this embodiment, the third rule stipulates that if the mode determination coefficient belongs to the middle set and the optical thickness belongs to the thick set, then the radiative transfer geometry mode is a spherical mode. The physical basis of this rule is that when the mode determination coefficient is at a middle level, it indicates that the overall trend of factors such as solar zenith angle, cloud cover, and cloud inhomogeneity is not clear, and a single indicator cannot dominate mode selection. However, if the optical thickness belongs to the thick set, it means that the ice crystal cloud layer has a strong multiple scattering effect. Even if the solar zenith angle is not large or the cloud layer is relatively uniform, the effective propagation path of light within the cloud layer will be significantly increased due to multiple scattering, thus producing a non-negligible spherical geometric influence. Therefore, in this case, a spherical mode must be used to accurately describe the radiative transfer process.

[0077] In this embodiment of the application, the fourth rule stipulates that if the mode determination coefficient belongs to a small set and the optical thickness belongs to a thick set, then the radiative transfer geometry mode is a spherical mode. The physical basis of this rule is that even if the mode determination coefficient is small (indicating that the overall characteristics tend to be planar), if the optical thickness is thick (indicating that the ice crystal particle cloud is very thick), multiple scattering will lead to a significant increase in the effective optical path, producing a spherical effect, therefore a spherical mode should be adopted.

[0078] In this embodiment of the application, the fifth rule stipulates that if the mode determination coefficient belongs to a large set, the optical thickness belongs to a thin set, and the solar zenith angle belongs to a small set, then the radiative transfer geometry mode is a planar parallel mode. The physical basis of this rule is that even if the mode determination coefficient is large (indicating that the overall characteristics tend to be spherical), if the optical thickness is thin (indicating that the ice crystal cloud layer is very thin) and the solar zenith angle is small (indicating that the light path is close to perpendicular), multiple scattering is weak, and the planar approximation is still acceptable. Therefore, a planar parallel mode can be adopted.

[0079] It should be noted that these five rules can be used individually or in combination. When multiple rules are activated at the same time, the fuzzy logic decision system will integrate the suggestions of all rules to obtain the final decision result.

[0080] In summary, compared to existing technologies, this application performs fusion analysis on the multi-source meteorological data to calculate model determination coefficients, and determines the radiative transfer geometric model based on these coefficients. The radiative transfer geometric model includes a planar parallel model or a spherical model. Thus, by calculating model determination coefficients through fusion analysis of multi-source meteorological data and determining the radiative transfer geometric model accordingly, adaptive matching of the radiative transfer model selection with the true characteristics of ice crystal cloud layers is achieved, avoiding the calculation deviations caused by model misjudgment in traditional fixed-threshold methods.

[0081] S30: For the aforementioned planar parallel mode, the scalar radiative transfer model is invoked, and the CDISORT method is used to solve the atmospheric scalar radiative transfer equation to calculate the first optical characteristic parameters.

[0082] In scenarios with a small solar zenith angle and relatively uniform cloud layers, the planar parallel model is the basic assumption for radiative transfer calculation. However, the strong forward scattering characteristic unique to ice crystal cloud layers makes the traditional FDISORT method suffer from bottlenecks such as slow calculation speed, poor numerical stability, and insufficient accuracy, making it difficult to meet the needs of high-precision simulation of complex media such as ice crystal clouds.

[0083] To address the aforementioned issues, this application, for the aforementioned planar parallel mode, invokes the scalar radiative transfer model and employs the CDISORT method to solve the atmospheric scalar radiative transfer equation, calculating the first optical characteristic parameters. This effectively overcomes the shortcomings of traditional methods and achieves accurate, stable, and efficient solutions to the radiative transfer equation of ice crystal particle clouds under the assumption of planar parallelism.

[0084] Specifically, step S30 in the method includes: The ice water content and the effective particle radius are used as input parameters and substituted into the scalar radiative transfer model. Based on the ice water content and the effective radius of the particles, the volume extinction coefficient, single scattering albedo, and scattering phase function of the ice crystal particle cloud are calculated using Mie scattering theory or the T matrix method. The CDISORT method is used to numerically solve the atmospheric scalar radiative transfer equation in the plane parallel mode. The CDISORT method preprocesses the scattering phase function by introducing the δ-M scaling technique, approximating the strong forward scattering peak of the scattering phase function as a transmission term, thereby reducing the Legendre polynomial order required to expand the scattering phase function. Substitute the preprocessed scattering phase function into the atmospheric scalar radiative transfer equation, and expand the atmospheric scalar radiative transfer equation in multiple discrete directions to obtain a set of radiative intensity equations. By solving the eigenvalues ​​and eigenvectors of the radiation intensity equations, the distribution of radiation intensity at different optical depths and in different directions can be obtained; Based on the distribution of radiation intensity, the transmittance and reflectance of the ice crystal particle cloud during laser transmission are calculated, and the transmittance and reflectance are used as the first optical characteristic parameters.

[0085] In this embodiment, the ice-water content and effective particle radius are first used as input parameters and substituted into the scalar radiative transfer model. The scalar radiative transfer model is a mathematical model used to solve for the radiative intensity distribution in a plane-parallel mode, and its core equation is the atmospheric radiative transfer equation. This model assumes that the atmosphere is infinitely uniform in the horizontal direction, and that the radiative intensity varies only with the optical depth and direction in the vertical direction.

[0086] The atmospheric radiative transfer equation is a fundamental equation describing the changes in radiation intensity caused by absorption, scattering, and emission as radiation propagates through a medium. Under the assumption of a plane-parallel atmosphere, neglecting horizontal variations and considering the effects of multiple scattering, its standard form can be written as: In the formula, Optical depth (optical thickness) is a dimensionless quantity representing the vertical extinction integral measured downwards from the top of the atmosphere. The zenith angle θ represents the top of the atmosphere; μ is the cosine of the zenith angle θ, i.e., μ = cosθ. μ > 0 indicates that the radiation propagates upward (towards the top of the atmosphere), and μ < 0 indicates that the radiation propagates downward (towards the ground). It is the azimuth angle; To achieve optical depth Location, direction Radiation intensity on; The source function describes the enhancement contribution of the medium itself to radiation, including single scattering source functions, multiple scattering source functions, and thermal emission source functions.

[0087] The specific expression of the source function J is as follows: ,in: This refers to the single-scattering albedo. Let be the scattering phase function, describing the radiation from the incident direction. Scattered to the outgoing direction The probability of B is given by the equation; B(T) is the Planck blackbody radiation function, representing the thermal emission contribution of the medium under local thermodynamic equilibrium. This equation is a complex integral-differential equation, which usually cannot be solved analytically and can only be solved numerically.

[0088] Secondly, based on the ice-water content and the effective radius of the particles, the volume extinction coefficient, single-scattering albedo, and scattering phase function of the ice crystal particle cloud are calculated using Mie scattering theory or the T-matrix method. Mie scattering theory is suitable for calculating the optical properties of spherical particles, while the T-matrix method is suitable for calculating the optical properties of non-spherical particles. Since ice crystal particles are typically non-spherical, the T-matrix method is preferred.

[0089] Among them, the volume extinction coefficient (using β) ext (represented by) refers to the attenuation capacity of a unit volume of ice crystal particle cloud for radiation, including both scattering and absorption contributions, and is measured in km. -1 or m -1 According to the Beer-Lambert law, the decrease in radiation intensity with propagation distance can be expressed as I = I0e^(-I / I ... -βext·s , where s represents the geometric distance of radiation propagation in the medium, in km or m, I0 represents the incident radiation intensity, i.e. the radiation intensity at s=0, and I represents the radiation intensity after propagation over a distance s.

[0090] Among them, single-scattering albedo (using Albedo (or single-scattering albedo) refers to the proportion of scattering to total attenuation, ranging from 0 to 1. A single-scattering albedo of 1 indicates pure scattering with no absorption, while a single-scattering albedo of 0 indicates complete absorption. The single-scattering albedo of ice crystal particles is typically close to 1 in the visible and near-infrared bands, but absorption is enhanced in the mid-infrared and far-infrared bands.

[0091] Among them, the scattering phase function (using (represented by) refers to the probability distribution function of radiation scattered by ice crystal particles in different directions, satisfying the normalization condition. ,in, The scattering angle is the angle between the incident direction and the scattering direction. The scattering phase function of ice crystal particles is typically in the range of... =0° (forward) has an extremely sharp peak, which can reach 10 times the isotropic value. 4 The calculation of radiation transfer in ice crystal particle clouds is more than twice as high, which is the main challenge.

[0092] Furthermore, the CDISORT method is used to numerically solve the atmospheric scalar radiative transfer equation in the plane parallel mode. The CDISORT method preprocesses the scattering phase function by introducing the δ-M scaling technique, approximating the strong forward scattering peak of the scattering phase function as a transmission term to reduce the Legendre polynomial order required for expanding the scattering phase function. The CDISORT method is an improvement on the traditional FDISORT method, primarily optimized for the strong forward scattering characteristics of ice crystal cloud layers. The basic principle of the δ-M scaling technique is to decompose the scattering phase function into a sharp forward peak and a relatively smooth background portion. The forward peak is approximated by the Dirac delta function, indicating that this part of the radiation is not actually scattered but continues to propagate along its original direction; the background portion is expanded using a finite-order Legendre polynomial. Mathematically, the scattering phase function is expressed as: Where f is the proportion of the forward peak. It is the smoothed scattering phase function.

[0093] Specifically, substituting this decomposition into the radiative transfer equation leaves the equation form unchanged, but the optical thickness and single-scattering albedo are rescaled as follows: ( This transformation effectively transfers the strong forward scattering effect to the increase in optical thickness, thus improving the residual scattering phase function. It can be accurately represented using Legendre polynomials of lower order.

[0094] For example, the original ice crystal scattering phase function may need to be expanded to a Legendre polynomial of order 2048 to accurately describe the forward peak, resulting in a huge computational burden. After processing with the δ-M scaling technique, the smoothed scattering phase function may only require a Legendre polynomial of order 16 or 32 to accurately describe it, reducing the computational burden by several orders of magnitude.

[0095] Furthermore, the preprocessed scattering phase function is substituted into the atmospheric scalar radiative transfer equation, and the atmospheric scalar radiative transfer equation is expanded in multiple discrete directions to obtain a set of radiative intensity equations. These discrete directions are usually selected with Gaussian quadrature nodes, called discrete ordinates, and each discrete direction corresponds to an unknown radiative intensity, thus transforming the original integral-differential equations into a set of ordinary differential equations.

[0096] For example, the 16-stream approximation is used, which involves selecting 8 incident directions and 8 exit directions (a total of 16 directions) in the zenith angle direction, with each direction having a corresponding weighting coefficient. The radiation intensity equation set contains 16 equations, each describing the variation of radiation intensity in a direction with optical depth.

[0097] Furthermore, by solving the eigenvalues ​​and eigenvectors of the radiation intensity equations, the distribution of radiation intensity at different optical depths and directions can be obtained. The radiation intensity equations can be written in matrix form: In this equation, I is the radiation intensity vector, A is the coefficient matrix, and S is the source term (including contributions from thermal radiation and multiple scattering). By solving for the eigenvalues ​​and eigenvectors of the coefficient matrix A, a homogeneous solution can be obtained. Combining this with the particular solution (considering the source term), a general solution is obtained. Finally, by utilizing boundary conditions, such as no downward radiation from the top of the atmosphere and surface reflection, the undetermined coefficients are determined, and the specific numerical distribution of radiation intensity at different optical depths and in different directions is obtained.

[0098] Finally, based on the distribution of radiation intensity, the transmittance and reflectance of the ice crystal cloud layer during laser transmission were calculated, and these transmittance and reflectance were used as the first optical characteristic parameters. Transmittance is defined as the ratio of outgoing radiation intensity to incident radiation intensity. For solar radiation, the ratio of downward radiative flux at the cloud base to the cloud top is typically calculated. In the formula, Indicates the top of the cloud layer (optical depth) The downward radiation flux at a point is the solar radiation energy incident on the top of the cloud. Indicates the bottom of the cloud layer (optical depth) The downward radiation flux at a point is the radiation energy emitted after penetrating the entire cloud layer.

[0099] Reflectivity is defined as the ratio of upward incident radiant flux to downward incident radiant flux. In the formula, This represents the upward radiative flux at the top of the cloud layer, that is, the radiative energy reflected back into space by the cloud layer. This represents the downward incident radiation flux at the top of the cloud layer. The downward radiation flux F... down and upward radiative flux F up The following can be obtained by multiplying the radiation intensity in each direction by the direction cosine and integrating: , Where θ is the zenith angle. It is the azimuth angle. To achieve optical depth Location, direction The radiation intensity on it.

[0100] For example, suppose the optical depth of a certain ice crystal particle cloud is calculated using the CDISORT method. (Genting) and The radiation intensity distribution at the (cloud base) is as follows: At the cloud top, the downward radiation intensity is mainly concentrated in the direction with a smaller zenith angle. Integrating, we obtain the downward radiative flux. (Assuming the solar incident radiation is 1000 W / m) 2 (partially reflected and absorbed by clouds); upward radiative flux At the cloud base, the downward radiation intensity, after being attenuated by the cloud layer, can be integrated to obtain the downward radiative flux. Based on the above definitions, the following calculations are obtained: transmittance T = 400 / 800 = 0.5, indicating that 50% of the radiation energy incident on the cloud top penetrates the ice crystal cloud layer and reaches the cloud base; reflectance R = 200 / 800 = 0.25, indicating that 25% of the radiation energy incident on the cloud top is reflected back into space by the cloud layer. The remaining 1 - TR = 0.25 represents the proportion of radiation energy absorbed by the cloud layer. These three parameters together describe the scattering, absorption, and transmission characteristics of incident radiation by the ice crystal cloud layer.

[0101] Furthermore, the numerical solution of the atmospheric scalar radiative transfer equations in the plane parallel mode using the CDISORT method also includes: Obtain the optical thickness of the ice crystal particle cloud and the scattering asymmetry factor of the scattering phase function; Based on the optical thickness and the scattering asymmetry factor, the number of discrete ordinates and the angle discretization scheme are adaptively selected. When the optical thickness is greater than a preset optical thickness threshold or the scattering asymmetry factor is greater than a preset asymmetry factor threshold, the number of discrete ordinates is increased, and the angle grid is densified in the forward scattering peak region. When the optical thickness is less than or equal to the preset optical thickness threshold and the scattering asymmetry factor is less than or equal to the preset asymmetry factor threshold, the number of discrete ordinates is reduced and a uniform angle grid is used. The adaptive selection of the number of discrete ordinates is achieved through a pre-established lookup table. The lookup table takes the optical thickness and the scattering asymmetry factor as inputs and the optimal number of discrete ordinates as output. The lookup table is constructed through offline radiative transfer simulation experiments, and the criterion for determining the optimal number of discrete ordinates is that the rate of change of the radiative transfer result when the number of discrete ordinates increases is less than a preset rate of change threshold.

[0102] In this embodiment, the optical thickness of the ice crystal particle cloud and the scattering asymmetry factor of the scattering phase function are first obtained. The scattering asymmetry factor (denoted by g) is a parameter characterizing the forward skewness of the scattering phase function, defined as the average value of the cosine of the scattering angle. The value ranges from -1 to 1, where g=1 indicates complete forward scattering, g=-1 indicates complete backscattering, and g=0 indicates isotropic scattering.

[0103] For example, the scattering asymmetry factor in the visible light band is about 0.85 for solid columnar ice crystal particles with an effective particle radius of 50 μm; for plate-shaped ice crystal particles, it may be slightly lower, about 0.75.

[0104] Secondly, based on the optical thickness and scattering asymmetry factor, the number of discrete ordinates and the angle discretization scheme are adaptively selected. The number of discrete ordinates determines the number of directions in which the radiative transfer equation is expanded, while the angle discretization scheme determines the distribution of these directions in the angular space.

[0105] Furthermore, when the optical thickness exceeds a preset optical thickness threshold or the scattering asymmetry factor exceeds a preset asymmetry factor threshold, the number of discrete ordinates is increased, and the angular grid is refined in the forward scattering peak region. Specifically, when the optical thickness exceeds a preset optical thickness threshold or the scattering asymmetry factor exceeds a preset asymmetry factor threshold, it indicates that the cloud optical thickness is large or the forward scattering is extremely strong, requiring higher computational accuracy. In this case, the number of discrete ordinates is increased, for example, from 16 streams to 32 or 64 streams, and the angular grid is refined in the forward scattering peak region (the direction where the scattering angle is close to 0°) to more accurately describe the forward scattering.

[0106] The preset optical thickness threshold and the preset asymmetry factor threshold can be dynamically set according to the actual situation. Preferably, the preset optical thickness threshold can be set to 10 and the preset asymmetry factor threshold can be set to 0.8.

[0107] Furthermore, when the optical thickness is less than or equal to a preset optical thickness threshold and the scattering asymmetry factor is less than or equal to a preset asymmetry factor threshold, the number of discrete ordinates is reduced, and a uniform angle grid is used. Specifically, when the optical thickness is less than or equal to a preset optical thickness threshold and the scattering asymmetry factor is less than or equal to a preset asymmetry factor threshold, it indicates that the cloud optical thickness is not large and the forward scattering is relatively weak. In this case, a smaller number of discrete ordinates can be used, such as 8 or 16 streams, and a uniform angle grid can be used to save computation time.

[0108] The adaptive selection of the number of discrete ordinates is achieved through a pre-established lookup table. The lookup table takes optical thickness and scattering asymmetry factor as input and the optimal number of discrete ordinates as output. The lookup table is constructed through offline radiative transfer simulation experiments, with the criterion that the rate of change of the radiative transfer result as the number of discrete ordinates increases is less than a preset rate of change threshold. For example, the steps for constructing the lookup table are as follows: Selecting the optical thickness... Typical values ​​for scattering asymmetry factor g include 1, 2, 5, 10, 20, 50, and 100. Typical values ​​for g include 0.6, 0.7, 0.75, 0.8, 0.85, and 0.9. For each group... Combine the data to generate the corresponding scattering phase function (assuming a certain ice crystal particle morphology distribution); gradually increase the number of discrete ordinates N from low to high, such as 4, 8, 16, 32, 64, 128, and calculate the radiative transfer results, such as reflectivity, respectively; calculate the rate of change of the radiative transfer results as the number of discrete ordinates increases; when the rate of change is less than a preset rate of change threshold, it is considered that the current number of discrete ordinates N is accurate enough, and the current number of discrete ordinates N is stored as the optimal number of discrete ordinates in the lookup table.

[0109] For example, for The possible results for this combination are: R=0.45 when N=8, R=0.52 when N=16, with a rate of change of 15.6%; from N=16 to N=32, the rate of change is 3.2%; from N=32 to N=64, the rate of change is 0.8%. Since 0.8% < 1%, 32 flows are taken as the optimal number of discrete ordinates for this combination.

[0110] In summary, compared to existing technologies, this application, for the aforementioned planar parallel mode, invokes a scalar radiative transfer model and employs the CDISORT method to solve the atmospheric scalar radiative transfer equation, calculating the first optical characteristic parameters. Thus, by using the improved CDISORT method to solve the radiative transfer equation in the planar parallel mode, the shortcomings of the traditional FDISORT method—slow calculation speed, poor stability, and low accuracy—in handling strong forward scattering from ice crystal particles are overcome, achieving accurate, stable, and efficient solutions for the transmittance and reflectance of ice crystal cloud layers.

[0111] S40: For the spherical mode, call the vector radiative transfer model and use the inverse Monte Carlo method to solve the atmospheric vector radiative transfer equation to calculate the second optical characteristic parameters.

[0112] When the solar zenith angle is large, such as greater than 60°, or when there is significant horizontal non-uniformity in the cloud layer, the planar parallel assumption no longer holds, and a spherical model must be used to accurately describe the radiative transfer process.

[0113] However, solving the vector radiative transfer equation containing polarization information under spherical geometry is extremely difficult. Traditional numerical methods face problems of excessive computational complexity and difficulty in convergence when dealing with complex media such as ice crystal particle clouds. While the forward Monte Carlo method can handle complex geometry, it is computationally inefficient.

[0114] To address the aforementioned issues, this application, for the spherical model, invokes the vector radiative transfer model and employs the inverse Monte Carlo method to solve the atmospheric vector radiative transfer equation, thereby calculating the second optical characteristic parameters. This achieves efficient and accurate solutions for the reflectivity and linear polarization degree of ice crystal particle clouds in the spherical model.

[0115] Specifically, step S40 in the method includes: The ice water content, effective particle radius, cloud cover, and cloud vertical overlap pattern are used as input parameters and substituted into the vector radiative transfer model. The coverage rate of the ice crystal particle cloud layer in the horizontal direction is determined based on the cloud amount, and the arrangement of the multi-layer ice crystal particle cloud layer in the vertical direction is determined based on the vertical overlap pattern of the cloud layer. The three-dimensional geometric structure of the heterogeneous ice crystal particle cloud layer is constructed based on the coverage rate and the arrangement. The atmospheric vector radiative transfer equation in spherical mode is solved using the reverse Monte Carlo method to simulate the propagation path of photons starting from the detector position and tracing through the heterogeneous ice crystal particle cloud in the reverse direction. During the reverse tracing process, based on the ice water content and the effective radius of the particles, the single scattering characteristics of the ice crystal particles are calculated using Mie scattering theory or the T matrix method. The single scattering characteristics include the scattering phase function and the Mueller matrix. When a photon scatters with an ice crystal particle, the Mueller matrix is ​​multiplied by the Stokes vector currently carried by the photon, based on the scattering angle and azimuth angle of the current scattering event, to obtain the scattered Stokes vector. When the reverse tracing path traces back to the direction of the solar source and the photon weight is greater than a preset weight threshold, the contribution of the path is added to the detector's received signal. The reflectivity and linear polarization of the ice crystal cloud during laser transmission are obtained by averaging the reverse tracking results of multiple photons, and the reflectivity and linear polarization are used as the second optical characteristic parameters.

[0116] In this embodiment, ice-water content, effective particle radius, cloud cover, and cloud vertical overlap pattern are first used as input parameters and substituted into the vector radiative transfer model. The vector radiative transfer model is a mathematical model used to solve the radiative transfer equation considering polarization effects in a spherical mode. Its core is the vector radiative transfer equation incorporating the Stokes vector. Unlike the aforementioned scalar radiative transfer model, which only considers radiation intensity, the vector radiative transfer model fully describes the intensity and polarization state of radiation through the Stokes vector and describes the transformation of the polarization state by scattering events through the Mueller matrix.

[0117] The specific mathematical form of the vector radiative transfer model is as follows: Under the assumption of a plane parallel atmosphere, the vector radiative transfer equation describing polarized radiative transfer can be written as: In the formula, =[I,Q,U,V] T It is a Stokes vector, which contains four components: I is the total radiation intensity, Q and U describe linear polarization, and V describes circular polarization; θ is the optical depth, representing the degree of attenuation of radiation propagating in the medium; μ = cosθ is the cosine of the zenith angle θ. It is the azimuth angle; The single-scattering albedo represents the proportion of scattering to the total attenuation. The scattering phase matrix is ​​a 4×4 matrix describing the radiation from the incident direction. Scattered to the outgoing direction When, the transformation effect on the Stokes vector.

[0118] The essential difference between the vector radiative transfer model and the scalar radiative transfer equation lies in the fact that the scalar equation only solves for the radiation intensity I, while the vector equation, through a 4×1 Stokes vector and a 4×4 scattering phase matrix, fully describes the influence of scattering events on the radiation polarization state. When considering thermal radiation source terms or surface boundary conditions, corresponding source terms need to be added to the right side of the equation.

[0119] It should be noted that the scattering phase matrix P depends on the shape, size and orientation distribution of the particles in the medium. For non-spherical particles such as ice crystals, their variation with the scattering angle can be pre-calculated and stored using numerical methods such as the T matrix method.

[0120] Secondly, the horizontal coverage of the ice crystal particle cloud layer is determined based on cloud cover, and the vertical arrangement of the multi-layered ice crystal particle cloud layer is determined based on the vertical overlap pattern of the cloud layer. Based on the coverage and arrangement, a three-dimensional geometric structure of the heterogeneous ice crystal particle cloud layer is constructed. Specifically, the horizontal coverage of the ice crystal particle cloud layer is determined based on cloud cover. For example, cloud cover = 0.6 means that on the horizontal plane, 60% of the area is randomly selected as cloudy and 40% of the area is clear, generating a two-dimensional grid. Each grid point is randomly assigned a cloudy or clear state, such that the proportion of cloudy grid points to the total number of grid points equals the cloud cover.

[0121] For example, the vertical arrangement of multiple ice crystal particle clouds is determined based on the vertical overlap pattern. Assume there are two ice crystal particle cloud layers with heights h1 and h2, and geometric thicknesses Δh1 and Δh2, respectively. For the maximum overlap pattern: the horizontal positions of the two ice crystal particle cloud layers are perfectly aligned; that is, if a horizontal position has clouds in the first layer, it also has clouds in the second layer, and vice versa. For the random overlap pattern: the horizontal positions of the two ice crystal particle cloud layers are independently and randomly distributed. The horizontal distribution of the second ice crystal particle cloud layer is independent of the first layer, and each layer randomly generates its coverage area based on cloud cover. For the mixed overlap pattern: an overlap coefficient α (0 ≤ α ≤ 1) is defined to represent the maximum overlap ratio. The first ice crystal particle cloud layer is randomly generated based on cloud cover; in the second ice crystal particle cloud layer, the region with an α ratio has maximum overlap with the first layer, and the remaining regions with a 1-α ratio are randomly and independently distributed. Finally, a three-dimensional geometric structure is constructed based on the coverage and arrangement: In three-dimensional space, each grid point (x, y, z) is marked as "inside the cloud" or "clear sky" according to its horizontal and vertical positions. The position inside the cloud needs to be further determined based on the ice water content and the effective radius of the particles to determine the optical properties of the point.

[0122] Next, the inverse Monte Carlo method is used to solve the atmospheric vector radiative transfer equation in the spherical model, simulating the propagation path of a photon starting from the detector position and tracing through a heterogeneous ice crystal cloud in the reverse direction. The inverse Monte Carlo method traces the light in the opposite direction of propagation from the detector position. The specific tracing process includes: 1. Initializing the photon: A photon is initialized at the detector position (usually located at the top of the atmosphere or above the Earth's surface), setting its initial direction to the observation direction. The photon carries an initial Stokes vector, typically (1,0,0,0), representing unpolarized light per unit intensity, with an initial weight of 1. 2. Determining the first-step path: Based on the atmospheric density distribution and geometric relationships, the distance at which the photon first encounters the ice crystal cloud or the Earth's surface while propagating in the reverse direction is calculated. 3. Propagation within the ice crystal cloud: After the photon enters the ice crystal cloud, the distance at which the next scattering event occurs is sampled based on the optical properties (extinction coefficient) of the ice crystal cloud at the current location. The scattering distance follows an exponential distribution: p(s) = β ext ×e -βext·s 4. Scattering event occurs: Scattering calculations are performed at the scattering point. 5. Tracing to the boundary: Repeat steps 3-5 until the photon passes through the top of the atmosphere (towards the sun) or reaches the Earth's surface. 6. Reaching the sun's direction: When the photon's direction points towards the sun and its current altitude exceeds the top of the atmosphere, the path is considered to have traced back to the solar source, and the contribution of this path is added to the detector signal. 7. Reaching the Earth's surface: When the photon reaches the Earth's surface, it is processed according to the surface reflection characteristics, the photon direction and Stokes vector are updated, and the reverse tracing continues.

[0123] Furthermore, during the reverse tracing process, based on the ice-water content and the effective particle radius, the single-scattering characteristics of the ice crystal particles are calculated using Mie scattering theory or the T-matrix method. These single-scattering characteristics include the scattering phase function and the Mueller matrix. Specifically, the single-scattering characteristics include the scattering phase function... Mueller matrix M: scattering phase function The Mueller matrix M is a 4×4 matrix that describes the probability distribution of scattered light in different directions and represents the transformation of the scattering event with respect to the Stokes vector. For ice crystal particles, the Mueller matrix depends on the scattering angle. Azimuth In addition to the morphology of ice crystal particles, the effective size of particles, and the distribution of particle orientation.

[0124] Furthermore, when a photon scatters with an ice crystal particle, the Mueller matrix is ​​multiplied by the Stokes vector currently carried by the photon, based on the scattering angle and azimuth angle of the current scattering event, to obtain the scattered Stokes vector. Specifically, when a photon scatters, the scattering angle is determined by sampling... and azimuth Extract the corresponding matrix from the calculated Mueller matrix. Multiplying this by the Stokes vector S currently carried by the photon yields the scattered Stokes vector: Stokes vector S = (I, Q, U, V) T It is a complete representation of light intensity and polarization state. Here, I is the total intensity, Q represents linear polarization in the horizontal / vertical direction, U represents linear polarization in the 45° / 135° direction, and V describes circular polarization (left-handed / right-handed). This process is equivalent to updating the polarization state of the light with each scattering.

[0125] Furthermore, when the reverse tracing path returns to the direction of the solar source and the photon weight is greater than a preset weight threshold, the path contribution is added to the detector's received signal. The contribution value is the photon's current Stokes vector multiplied by the photon weight: ΔS det =W·S end Where W is the photon accumulation weight, and S end Let be the Stokes vector at the end of the path. The method for determining the preset weight threshold is as follows: those skilled in the art can preset this threshold through convergence testing based on the balance requirements of computational accuracy and efficiency. Specifically, the weight threshold is usually set to a small positive number, such as 10. -6 Up to 10 -3 The purpose of this is to eliminate photon paths that have become extremely low in weight after multiple scatterings and whose contribution to the final detection signal is negligible, thereby avoiding invalid calculations and improving computational efficiency.

[0126] For example, in a typical Monte Carlo simulation of atmospheric radiative transfer, multiple sets of comparative tests with different weight thresholds can be performed in advance: using 10... -4 10 -5 10 -6 10 -7 This is used as a threshold in the simulation. When the threshold decreases to a certain value, such as from 10... -6 Reduced to 10 -7 If the change in the calculated result is less than the preset tolerance error, such as 0.1%, then the threshold is considered sufficiently low. Further reducing the threshold offers limited improvement in accuracy and only increases the computational burden. In this case, the threshold can be used as the preset weight threshold for subsequent calculations. Preferably, for the radiative transport problem of ice crystal particles in clouds, the preset weight threshold can be set to 10. -6 .

[0127] Finally, the average of the reverse tracking results of multiple photons is calculated to obtain the reflectivity and linear polarization degree of the ice crystal cloud during laser transmission. Reflectivity and linear polarization degree are then used as the second optical characteristic parameter. The above tracking process is repeated for a large number of photons, with each photon contributing a path. The contributions of all paths are averaged to obtain the Stokes vector received by the detector. The reflectivity is defined as the ratio of the upward incident radiant flux to the downward incident radiant flux, which can be expressed by S. det The I component is calculated; the degree of linear polarization (DoLP) is defined as the proportion of the linearly polarized component to the total intensity. In the formula, Q represents linear polarization in the horizontal / vertical direction, and U represents linear polarization in the 45° / 135° direction.

[0128] For example, tracking 10 7 After 1 photon, S was obtained statistically. det =(0.35,0.08,0.03,0.00), then the reflectivity is 0.35, and the linear polarization degree is... ≈0.243. The calculated reflectivity and linear polarization degree are used as the second optical characteristic parameters.

[0129] Furthermore, solving the atmospheric vector radiative transfer equation in the spherical model using the inverse Monte Carlo method also includes: A database of optical properties of ice crystal particles is pre-constructed. The database calculates and stores the corresponding Mueller matrix variation data with scattering angle and azimuth angle for different ice crystal particle morphologies, effective particle radii, and orientation distributions using the T-matrix method or the discrete dipole approximation method. During the reverse tracking process, based on the ice crystal particle morphology, effective radius, and orientation distribution corresponding to the current photon location, the corresponding Mueller matrix is ​​interpolated from the ice crystal particle optical property database to calculate the transformation of the scattering event on the Stokes vector. An importance sampling strategy is used to sample the photon scattering direction, and the sampling probability density function is set to be proportional to the scattering phase function; Based on the Stokes vector currently carried by the photon, the sampling probability density function is adaptively adjusted so that the sampling direction tends to the direction that contributes more to the signal received by the detector. The contribution of the photon path is weighted and corrected to obtain the unbiased contribution of the photon path to the signal received by the detector. The weighting factor for the weight correction is the ratio of the actual sampling probability to the target probability.

[0130] In this embodiment, a database of optical properties of ice crystal particles is first pre-constructed. This database pre-calculates and stores the Mueller matrix variations with scattering angle and azimuth angle for different ice crystal particle morphologies, effective radii, and orientation distributions using the T-matrix method or the discrete dipole approximation method. The database is a pre-calculated lookup table designed to avoid repeatedly calculating complex Mueller matrices during Monte Carlo tracing. The dimensions of the database include: 1. Ice crystal particle morphology: such as solid pillars, hollow pillars, plates, polymers, etc. 2. Effective particle radius. 3. Particle orientation distribution. 4. Scattering angle. 5. Azimuth For each combination, the scattering characteristics are calculated using the T-matrix method or the discrete dipole approximation method, including: extinction efficiency factor Qext, scattering efficiency factor Qsca, absorption efficiency factor Qabs, and scattering phase function. Muller matrix elements After the calculation is completed, the data is stored as a binary file or HDF5 format for easy reading and interpolation.

[0131] Secondly, during the reverse tracing process, based on the morphology, effective radius, and orientation distribution of the ice crystal particles corresponding to the current photon location, the corresponding Mueller matrix is ​​interpolated from the ice crystal particle optical property database to calculate the transformation of the scattering event on the Stokes vector. For example, if the current particle's effective radius is 28 μm, and there are two neighboring values ​​in the database, 25 μm and 30 μm, then the Mueller matrices corresponding to the two particle effective radii are linearly interpolated, with the weights inversely proportional to the distance: M = (30-28) / (30-25) × M25 + (28-25) / (30-25)M30.

[0132] Secondly, an importance sampling strategy is employed to sample photon scattering directions, with the sampling probability density function set to be proportional to the scattering phase function. Importance sampling is a variance reduction technique that, by altering the sampling probability distribution, concentrates sampling on regions that contribute significantly to the results, and then weight adjustments ensure unbiasedness. In traditional methods, photon scattering directions are determined according to the scattering phase function. Sampling, i.e., probability density function While this sampling method is physically correct, it is inefficient because a large number of photons are scattered forward, contributing very little to the detector. Importance sampling sets the sampling probability density function to... In this step, it is set to be proportional to the scattering phase function, that is... This allows the sampling direction to be more inclined towards the direction with a higher scattering probability, thereby improving sampling efficiency.

[0133] Furthermore, based on the Stokes vector currently carried by the photon, the sampling probability density function is adaptively adjusted to favor directions that contribute more to the signal received by the detector. The basic principle is that certain scattering directions contribute more to specific polarization components. For example, if the current photon has a high degree of linear polarization, certain scattering angles, such as around 90°, may significantly alter the polarization state, and these directions may contribute more to the final result. Therefore, the sampling probability of these directions can be increased.

[0134] Finally, the contribution of the photon path is weighted and corrected to obtain the unbiased contribution of the photon path to the detector received signal. The weighting factor for this correction is the ratio of the actual sampling probability to the target probability. Because the sampling probability has changed, the contribution of the photon path needs to be weighted to ensure the unbiasedness of the statistical results. The weighting correction factor is the ratio of the actual sampling probability to the target probability, and the target probability density is set to... The actual sampling probability density used is Then the weight correction factor for each sampling is: The cumulative weight of photons W = ∏w i , where w i This is the weighting correction factor for each scattering event. Ultimately, the contribution of this photon path to the detector signal is W·S. end S end It is the Stokes vector at the end of the path. This weighting adjustment ensures that the expected value of the final statistical result is the same as the result of sampling by physical probability (unbiasedness), and at the same time, due to the more concentrated sampling, the variance is reduced and the convergence is faster.

[0135] For example, suppose in a certain scattering event, according to physical probability... Sampling by importance Therefore, the weight correction factor w = 0.1 / 0.5 = 0.2. The current weight of a photon is 0.8, so the updated weight is 0.8 × 0.2 = 0.16. Although the weight has decreased, the probability of this sampling is higher, thus improving overall efficiency.

[0136] In summary, compared to existing technologies, this application, for the aforementioned spherical mode, invokes the vector radiative transfer model and employs the inverse Monte Carlo method to solve the atmospheric vector radiative transfer equation, thereby calculating the second optical characteristic parameters. Thus, by using the inverse Monte Carlo method to solve the vector radiative transfer equation in the spherical mode, it overcomes the technical bottleneck of traditional methods in the coupled solution of spherical geometry and polarization effects, achieving efficient and accurate calculation of the reflectivity and linear polarization degree of ice crystal particle clouds, and providing an effective means for studying the polarization characteristics of non-uniform cloud atmospheres.

[0137] S50: Output the first optical characteristic parameter or the second optical characteristic parameter as the result of cloud optical characteristic analysis.

[0138] This application outputs either a first optical characteristic parameter or a second optical characteristic parameter as the result of cloud optical characteristic analysis. Specifically, if a planar parallel mode is used, the transmittance and reflectance calculated using the CDISORT method are output; if a spherical mode is used, the reflectance and degree of linear polarization calculated using the inverse Monte Carlo method are output. These parameters comprehensively characterize the energy attenuation and polarization state change characteristics of laser propagation in ice crystal cloud layers. They can be directly applied to practical scenarios such as laser communication link budget assessment, atmospheric correction of remote sensing images, assimilation of weather forecast data, and assessment of climate radiative forcing, providing crucial input data support for related engineering applications and scientific research.

[0139] In summary, the embodiments of this application have at least the following technical effects: Compared with existing technologies, this application first obtains multi-source meteorological data of the target area, providing comprehensive and accurate input parameters for subsequent fusion analysis. This data can truly reflect the microphysical characteristics, spatial distribution, and incident geometry of ice crystal particle clouds, thus breaking through the computational accuracy bottleneck caused by insufficient input information in traditional methods.

[0140] Secondly, this application performs fusion analysis on multi-source meteorological data to calculate model determination coefficients, and determines the radiative transfer geometric model based on these coefficients. This achieves adaptive matching between the radiative transfer model selection and the actual characteristics of ice crystal cloud layers, effectively avoiding calculation errors caused by model misjudgment in traditional fixed-threshold methods.

[0141] Furthermore, this application, for the planar parallel mode, invokes the scalar radiative transfer model and employs the CDISORT method to solve the atmospheric scalar radiative transfer equation, calculating the first optical characteristic parameters. Thus, by using the improved CDISORT method to solve the radiative transfer equation in the planar parallel mode, the shortcomings of the traditional FDISORT method—slow calculation speed, poor stability, and low accuracy—in handling strong forward scattering from ice crystal particles are overcome, achieving accurate, stable, and efficient solutions for the transmittance and reflectance of ice crystal cloud layers.

[0142] Furthermore, this application, for the spherical model, invokes the vector radiative transfer model and employs the inverse Monte Carlo method to solve the atmospheric vector radiative transfer equation, thereby calculating the second optical characteristic parameter. Thus, by using the inverse Monte Carlo method to solve the vector radiative transfer equation in the spherical model, it overcomes the technical bottleneck of traditional methods in the coupled solution of spherical geometry and polarization effects, achieving efficient and accurate calculation of the reflectivity and linear polarization degree of ice crystal particle clouds, providing an effective means for studying the polarization characteristics of non-uniform cloud atmospheres.

[0143] Finally, this application outputs the first or second optical characteristic parameters as the results of cloud optical characteristic analysis, providing key input data support for practical applications such as laser communication link budgeting, atmospheric correction of remote sensing images, assimilation of meteorological forecast data, and assessment of climate radiative forcing.

[0144] Through the above technical solution, this application realizes the deep fusion and adaptive matching of radiative transfer model and meteorological data, which improves the accuracy and efficiency of optical characteristic analysis of ice crystal particles in clouds. It can be widely used in fields such as laser communication link budgeting, atmospheric correction of remote sensing images, assimilation of meteorological forecast data and assessment of climate radiative forcing.

[0145] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0146] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0147] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0148] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0149] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0150] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0151] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for analyzing the optical properties of clouds by integrating radiative transfer models and meteorological data, characterized in that, The method includes: Acquire multi-source meteorological data for the target area, wherein the multi-source meteorological data includes microphysical parameters of ice crystal particle clouds, geometric structure parameters of ice crystal particle clouds, and solar zenith angle, and the microphysical parameters include ice water content and effective particle radius; The multi-source meteorological data are fused and analyzed to calculate the model determination coefficients, and the radiative transfer geometric model is determined based on the model determination coefficients. The radiative transfer geometric model includes a planar parallel model or a spherical model. For the aforementioned planar parallel mode, the scalar radiative transfer model is invoked, and the CDISORT method is used to solve the atmospheric scalar radiative transfer equation to calculate the first optical characteristic parameters. For the spherical mode, the vector radiative transfer model is invoked, and the atmospheric vector radiative transfer equation is solved using the inverse Monte Carlo method to calculate the second optical characteristic parameters; Output the first optical characteristic parameter or the second optical characteristic parameter as the result of cloud optical characteristic analysis.

2. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 1, characterized in that, Acquire multi-source meteorological data for the target area, including: Obtain information such as ice and water content, effective particle radius, cloud cover, cloud height, cloud geometric thickness, cloud vertical overlap pattern, ice crystal particle morphology, particle orientation distribution, and solar zenith angle from ground-based remote sensing equipment, spaceborne remote sensing equipment, or meteorological reanalysis data. The optical thickness is calculated based on the ice water content and the effective radius of the particles using Mie scattering theory or the T-matrix method. The ice-water content, the effective radius of the particles, the optical thickness, the morphology of the ice crystal particles, and the orientation distribution of the particles are used as the microphysical parameters of the ice crystal particle cloud. The cloud cover, cloud height, cloud geometric thickness, and cloud vertical overlap pattern are used as the geometric structural parameters of the ice crystal particle cloud layer.

3. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 1, characterized in that, The multi-source meteorological data are fused and analyzed to calculate the model determination coefficient, including: The microphysical parameters, geometric parameters, and solar zenith angle of the ice crystal particle cloud layer are input into a pre-trained cloud inhomogeneity discrimination model, and the cloud inhomogeneity index is output. The solar zenith angle is normalized to obtain the standard solar zenith angle; The model determination coefficient is calculated by weighting and fusing the standard solar zenith angle, cloud cover, and cloud inhomogeneity index.

4. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 3, characterized in that, The construction process of the cloud inhomogeneity discrimination model includes: An initial cloud non-uniformity discrimination model was constructed based on a deep learning algorithm; Multiple sets of sample data are collected from satellite observation data or high-resolution cloud simulation data. Each set of sample data includes the microphysical parameters of the ice crystal particle cloud, the geometric structure parameters of the ice crystal particle cloud, and the solar zenith angle. For each set of sample data, a three-dimensional radiative transfer model is used to calculate the corresponding reference inhomogeneity index, and the reference inhomogeneity index is used as the training label. Using the sample data as input and the training labels as supervision signals, supervised training is performed on the initial cloud inhomogeneity discrimination model until verification convergence is achieved, resulting in a trained cloud inhomogeneity discrimination model.

5. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 1, characterized in that, Determining the radiative transfer geometric mode based on the mode determination coefficients includes: The mode determination coefficient, optical thickness, and solar zenith angle are input into the fuzzy logic decision system. The fuzzy logic decision system includes a predefined membership function and a fuzzy rule base. The membership function is used to convert the mode determination coefficient, the optical thickness, and the solar zenith angle into corresponding fuzzy set membership degrees. The fuzzy rule base contains multiple fuzzy rules based on the physical principle of radiative transfer. The fuzzy set is obtained by performing rule activation and conclusion aggregation on the membership degree of the fuzzy set through fuzzy reasoning; The output fuzzy set is defuzzified, and the mode selection confidence score is calculated. Based on the selected confidence level, the radiative transfer geometry mode is determined to be either a planar parallel mode or a spherical mode.

6. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 5, characterized in that, The fuzzy rule base contains at least one of the following rules: The first rule states that if the mode determination coefficients belong to a small set, the optical thickness belongs to a thin set, and the solar zenith angle belongs to a small set, then the radiative transfer geometry mode is a planar parallel mode. The second rule states that if the mode determination coefficients belong to a large set, the optical thickness belongs to a thick set, and the solar zenith angle belongs to a large set, then the radiative transfer geometry mode is a spherical mode. The third rule states that if the mode determination coefficients belong to the middle set and the optical thickness belongs to the thick set, then the radiative transfer geometric mode is a spherical mode. The fourth rule states that if the mode determination coefficients belong to a small set and the optical thickness belongs to a thick set, then the radiative transfer geometry mode is a spherical mode. The fifth rule states that if the mode determination coefficients belong to a large set, the optical thickness belongs to a thin set, and the solar zenith angle belongs to a small set, then the radiative transfer geometry mode is a planar parallel mode.

7. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 1, characterized in that, For the aforementioned planar parallel mode, the scalar radiative transfer model is invoked, and the CDISORT method is used to solve the atmospheric scalar radiative transfer equation, calculating the first optical characteristic parameters, including: The ice water content and the effective particle radius are used as input parameters and substituted into the scalar radiative transfer model. Based on the ice water content and the effective radius of the particles, the volume extinction coefficient, single scattering albedo, and scattering phase function of the ice crystal particle cloud are calculated using Mie scattering theory or the T matrix method. The CDISORT method is used to numerically solve the atmospheric scalar radiative transfer equation in the plane parallel mode. The CDISORT method preprocesses the scattering phase function by introducing the δ-M scaling technique, approximating the strong forward scattering peak of the scattering phase function as a transmission term, thereby reducing the Legendre polynomial order required to expand the scattering phase function. Substitute the preprocessed scattering phase function into the atmospheric scalar radiative transfer equation, and expand the atmospheric scalar radiative transfer equation in multiple discrete directions to obtain a set of radiative intensity equations. By solving the eigenvalues ​​and eigenvectors of the radiation intensity equations, the distribution of radiation intensity at different optical depths and in different directions can be obtained; Based on the distribution of radiation intensity, the transmittance and reflectance of the ice crystal particle cloud during laser transmission are calculated, and the transmittance and reflectance are used as the first optical characteristic parameters.

8. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 7, characterized in that, The CDISORT method is used to numerically solve the atmospheric scalar radiative transfer equations in the plane parallel mode, and the method also includes: Obtain the optical thickness of the ice crystal particle cloud and the scattering asymmetry factor of the scattering phase function; Based on the optical thickness and the scattering asymmetry factor, the number of discrete ordinates and the angle discretization scheme are adaptively selected. When the optical thickness is greater than a preset optical thickness threshold or the scattering asymmetry factor is greater than a preset asymmetry factor threshold, the number of discrete ordinates is increased, and the angle grid is densified in the forward scattering peak region. When the optical thickness is less than or equal to the preset optical thickness threshold and the scattering asymmetry factor is less than or equal to the preset asymmetry factor threshold, the number of discrete ordinates is reduced and a uniform angle grid is used. The adaptive selection of the number of discrete ordinates is achieved through a pre-established lookup table. The lookup table takes the optical thickness and the scattering asymmetry factor as inputs and the optimal number of discrete ordinates as output. The lookup table is constructed through offline radiative transfer simulation experiments, and the criterion for determining the optimal number of discrete ordinates is that the rate of change of the radiative transfer result when the number of discrete ordinates increases is less than a preset rate of change threshold.

9. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 1, characterized in that, For the spherical mode, the vector radiative transfer model is invoked, and the atmospheric vector radiative transfer equation is solved using the inverse Monte Carlo method to calculate the second optical characteristic parameters, including: The ice water content, effective particle radius, cloud cover, and cloud vertical overlap pattern are used as input parameters and substituted into the vector radiative transfer model. The coverage rate of the ice crystal particle cloud layer in the horizontal direction is determined based on the cloud amount, and the arrangement of the multi-layer ice crystal particle cloud layer in the vertical direction is determined based on the vertical overlap pattern of the cloud layer. The three-dimensional geometric structure of the heterogeneous ice crystal particle cloud layer is constructed based on the coverage rate and the arrangement. The atmospheric vector radiative transfer equation in spherical mode is solved using the reverse Monte Carlo method to simulate the propagation path of photons starting from the detector position and tracing through the heterogeneous ice crystal particle cloud in the reverse direction. During the reverse tracing process, based on the ice water content and the effective radius of the particles, the single scattering characteristics of the ice crystal particles are calculated using Mie scattering theory or the T matrix method. The single scattering characteristics include the scattering phase function and the Mueller matrix. When a photon scatters with an ice crystal particle, the Mueller matrix is ​​multiplied by the Stokes vector currently carried by the photon, based on the scattering angle and azimuth angle of the current scattering event, to obtain the scattered Stokes vector. When the reverse tracing path traces back to the direction of the solar source and the photon weight is greater than a preset weight threshold, the contribution of the path is added to the detector's received signal. The reflectivity and linear polarization of the ice crystal cloud during laser transmission are obtained by averaging the reverse tracking results of multiple photons, and the reflectivity and linear polarization are used as the second optical characteristic parameters.

10. The cloud optical characteristic analysis method based on the fusion of radiative transfer model and meteorological data according to claim 9, characterized in that, Solving the atmospheric vector radiative transfer equations in the spherical model using the inverse Monte Carlo method also includes: A database of optical properties of ice crystal particles is pre-constructed. The database calculates and stores the corresponding Mueller matrix variation data with scattering angle and azimuth angle for different ice crystal particle morphologies, effective particle radii, and orientation distributions using the T-matrix method or the discrete dipole approximation method. During the reverse tracking process, based on the ice crystal particle morphology, effective radius, and orientation distribution corresponding to the current photon location, the corresponding Mueller matrix is ​​interpolated from the ice crystal particle optical property database to calculate the transformation of the scattering event on the Stokes vector. An importance sampling strategy is used to sample the photon scattering direction, and the sampling probability density function is set to be proportional to the scattering phase function; Based on the Stokes vector currently carried by the photon, the sampling probability density function is adaptively adjusted so that the sampling direction tends to the direction that contributes more to the signal received by the detector. The contribution of the photon path is weighted and corrected to obtain the unbiased contribution of the photon path to the signal received by the detector. The weighting factor for the weight correction is the ratio of the actual sampling probability to the target probability.

Images