A direct estimation method of surface solar shortwave radiation based on geostationary satellite data

Through feature extraction and radiation prediction models based on stationary satellite data, the nonlinear relationship of short-wave radiation is directly constructed, which solves the problem of slow calculation speed and complex physical quantity dependence in the existing technology, and realizes high-precision and fast surface solar short-wave radiation estimation , suitable for solar energy resource assessment.

CN118886332BActive Publication Date: 2024-11-22SICHUAN UNIVERSITY OF SCIENCE AND ENGINEERING
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411341081.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-25
Publication Date
2024-11-22
Estimated Expiration
2044-09-25

AI Technical Summary

Technical Problem

The prior art has slow calculation speed, low efficiency and poor versatility in short-wave radiation estimation, and reliance on multiple complex physical quantities, resulting in low reliability of calculation results, making it difficult to accurately obtain cloud and short-wave radiation information.

Method used

Using the direct estimation method of surface solar short-wave radiation based on stationary satellite data, by obtaining the atmospheric top albedo data and solar zenith angle data, using feature extraction model and radiation prediction model, the nonlinear relationship between short-wave radiation and its components is directly constructed to avoid Complex physical quantity input and cloud detection process improve calculation speed and accuracy.

Benefits of technology

It realizes high-precision and fast surface solar short-wave radiation estimation, has good versatility and operability, and is suitable for solar resource evaluation and power generation system planning, reducing calculation complexity and error transmission.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure SMS_1
    Figure SMS_1
  • Figure SMS_9
    Figure SMS_9
  • Figure SMS_13
    Figure SMS_13
Patent Text Reader

Abstract

The present invention discloses a method for directly estimating the surface solar shortwave radiation based on geostationary satellite data, and belongs to the field of solar radiation estimation. The estimation method comprises: obtaining the top of atmosphere albedo data of the target area observed by the satellite, the top of atmosphere albedo data comprises a plurality of observation values, and different observation values ​​are the top of atmosphere albedo observed by the satellite in different wavelength ranges at the same time; obtaining a trained feature extraction model, and after calculation, the feature extraction model outputs additional feature data; obtaining a trained radiation prediction model, inputting all observation values, solar zenith angle data and additional feature data into the radiation prediction model, and outputting the surface solar shortwave total radiation value and the surface solar shortwave direct radiation value. The present invention directly starts from the primary data of the satellite and constructs a nonlinear complex relationship with the shortwave radiation and its components. The method has the advantages of high precision, fast calculation speed, and strong versatility.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of solar radiation estimation, and in particular relates to a method for directly estimating surface solar shortwave radiation based on geostationary satellite data. Background Art

[0002] Surface shortwave radiation is a key parameter for calculating surface radiation balance, and is also a basic parameter for studying water cycle, surface evapotranspiration, and surface biochemical processes. Surface shortwave radiation is defined as the integral of the spectral solar radiation reaching the surface in the range of 0.3 to 3.0 μm, with the unit of Wm -2 .

[0003] In terms of obtaining shortwave radiation, there are mainly ground-based observations, model simulations, and satellite remote sensing estimation. Ground-based observations can obtain high-precision shortwave radiation observations, but due to limited stations, it is difficult to obtain spatially continuous results, so it is often used as a reference value to verify the shortwave radiation obtained from other data. Model simulations such as WRF (Weather research and forecasting) rely on various atmospheric profiles and surface information. At the same time, the spatial resolution of the output results is relatively coarse, which is difficult to meet the needs of high spatial resolution and other refinements. In recent years, with the rapid development of satellite remote sensing technology, large-scale, fast and high-precision acquisition of shortwave radiation has become a way. According to the orbit of the satellite, polar-orbiting satellites (such as Terra / MODIS, Aqua / MODIS, NPP / VIIRS) can obtain data information with higher spatial resolution and spectral resolution, and thus obtain shortwave radiation with higher precision. However, polar-orbiting satellites have lower temporal resolution, making it difficult to accurately obtain the daily variation characteristics of shortwave radiation; geostationary meteorological satellites can provide high-temporal resolution observation data, so as to obtain the daily variation characteristics of shortwave radiation. The spatial resolution and spectral resolution of traditional geostationary meteorological satellite data are low (such as the MTSAT series, FY-2 series, etc.), which makes it difficult to accurately obtain cloud and shortwave radiation information. In recent years, with the successful launch of a new generation of geostationary meteorological satellites (GOES-R, Himawari-8 / 9 and FY-4A / B, etc.), the sensors they carry have significantly improved in time, space and spectral resolution, providing an important data source for accurate estimation of shortwave radiation. Many scholars have proposed a variety of shortwave radiation estimation methods from different perspectives, but these methods still have problems such as slow calculation speed, low efficiency and poor versatility, and their calculation accuracy needs to be further improved. Summary of the invention

[0004] In view of the above phenomenon, the present invention provides a method for directly estimating surface solar shortwave radiation based on geostationary satellite data to improve the accuracy and speed of estimation.

[0005] In order to achieve the above object, the solution adopted by the present invention is: a method for directly estimating surface solar shortwave radiation based on geostationary satellite data, comprising the following steps:

[0006] Acquire the top of atmosphere albedo data of the target area observed by the satellite in the wavelength range of 0.3 μm-3.0 μm, wherein the top of atmosphere albedo data includes multiple observation values, and different observation values ​​are the top of atmosphere albedo observed by the satellite in different wavelength ranges at the same time;

[0007] Obtaining a trained feature extraction model, inputting the observed value into the feature extraction model, and after calculation, the feature extraction model outputs additional feature data;

[0008] The trained radiation prediction model is obtained, and all observation values, the solar zenith angle data of the target area and the additional feature data are input into the radiation prediction model. After calculation, the radiation prediction model outputs the total short-wave solar radiation value and the direct short-wave solar radiation value of the surface. All observation values ​​and the solar zenith angle data of the target area are the results observed by the satellite at the same time.

[0009] Furthermore, the additional feature data includes a first feature value and a second feature value. The internal calculation process of the feature extraction model includes the following steps:

[0010] Arrange the observation values ​​in order according to the length of the wavelength to obtain a first array; specifically, the observation values ​​may be arranged in order of increasing wavelength length, or in order of decreasing wavelength length;

[0011] Performing a first convolution and a first activation operation on the first array in sequence to obtain a second array, wherein the length of the first array is equal to the length of the second array;

[0012] The first array is subtracted from the second array to obtain a third array;

[0013] Taking the maximum value of all elements in the third array as the first eigenvalue;

[0014] Multiply the elements of the first array by the elements of the second array correspondingly to obtain a fourth array;

[0015] Concatenate the fourth array with the first array to obtain a fifth array;

[0016] Performing a second convolution and a second activation operation on the fifth array in sequence to obtain a sixth array, wherein the length of the sixth array is less than the length of the fifth array;

[0017] The average value of all elements in the sixth array is calculated to obtain the second eigenvalue.

[0018] The feature extraction model adopts the above calculation process, which not only has a small amount of calculation, but also has relatively good additional feature mining capabilities.

[0019] The feature extraction model and the radiation prediction model can be trained together, and the training method can be the existing conventional method. Furthermore, when the feature extraction model and the radiation prediction model are trained using a training set, the surface shortwave total radiation value and the surface shortwave direct radiation value in the training set can be calculated using many existing methods, and the calculated results are used as label values. As one implementation method, the surface shortwave total radiation value and the surface shortwave direct radiation value in the training set are calculated based on the existing method of the physical model.

[0020] Furthermore, calculating the total shortwave radiation value and the direct shortwave radiation value of the surface in the training set based on the physical model includes the following steps:

[0021] Acquire cloud phases in a preset area, where the cloud phases include clear sky, water clouds, or ice clouds;

[0022] When the cloud phase is clear, a first data set of the preset area is obtained, and the surface shortwave total radiation value and the surface shortwave direct radiation value of the preset area are calculated based on the first physical model and the first data set; the first data set includes the following dimensional data: solar zenith angle, aerosol optical thickness, total water vapor column, total ozone column and surface albedo;

[0023] When the cloud phase is water cloud, a second data set of the preset area is obtained, and the surface shortwave total radiation value and the surface shortwave direct radiation value of the preset area are calculated based on the second physical model and the second data set; the second data set includes the following dimensional data: solar zenith angle, water cloud optical thickness, water cloud particle effective radius and surface albedo;

[0024] When the cloud phase is an ice cloud, a third data set of the preset area is obtained, and the surface shortwave total radiation value and the surface shortwave direct radiation value of the preset area are calculated based on the third physical model and the third data set; the third data set includes the following dimensional data: solar zenith angle, ice cloud optical thickness, ice cloud particle effective radius and surface albedo.

[0025] The first physical model, the second physical model and the third physical model mentioned above can all be directly implemented using existing algorithms.

[0026] Furthermore, the optical thickness of water clouds and the effective radius of water cloud particles in the preset area are obtained by inverting the first cloud microphysical parameter; or / and, the optical thickness of ice clouds and the effective radius of ice cloud particles in the preset area are obtained by inverting the second cloud microphysical parameter.

[0027] Furthermore, the first cloud microphysical parameter inversion comprises the following steps:

[0028] Based on the simulation calculation of the atmospheric radiation transfer model, a first inversion lookup table is constructed for the optical thickness of water clouds and the effective radius of water cloud particles; the first inversion lookup table includes the following dimensional data: solar zenith angle, satellite zenith angle, relative azimuth, solar radiation brightness reflected by water clouds, surface albedo, water cloud spherical albedo, water cloud radiation transmittance, logarithm of water cloud optical thickness and logarithm of water cloud particle effective radius;

[0029] Based on the first inversion lookup table, the first cost function is iterated multiple times until the cost function value converges, the iteration is stopped, and the water cloud optical thickness and the effective radius of the water cloud particles in the preset area are inverted;

[0030] Or / and, the second cloud microphysical parameter inversion comprises the following steps:

[0031] Based on the simulation calculation of the atmospheric radiation transfer model, a second inversion lookup table is constructed for the optical thickness of ice clouds and the effective radius of ice cloud particles; the second inversion lookup table includes the following dimensional data: solar zenith angle, satellite zenith angle, relative azimuth, solar radiation brightness reflected by ice clouds, surface albedo, ice cloud spherical albedo, ice cloud radiation transmittance, logarithm of ice cloud optical thickness and logarithm of ice cloud particle effective radius;

[0032] Based on the second inversion lookup table, the second cost function is iterated multiple times until the cost function value converges, the iteration is stopped, and the ice cloud optical thickness and the effective radius of the ice cloud particles in the preset area are inverted.

[0033] Furthermore, the first cost function and the second cost function are both expressed as the following formulas:

[0034]

[0035] Among them, J is the function value of the cost function, represents the initial value of the iteration, , T represents the transpose of the matrix, , R VIS is the reflectivity matrix observed by the satellite in the visible light channel, R NIR is the reflectivity matrix observed by the satellite in the near-infrared channel, is the observation error covariance matrix, is the prior error covariance matrix, F is the satellite received radiation brightness simulation value calculated based on the atmospheric radiation transfer model; is the parameter to be inverted, , when the above formula is the first cost function, τ cis the optical thickness of water cloud, r e is the effective radius of water cloud particles. When the above formula is the second cost function, τ c is the ice cloud optical thickness, r e is the effective radius of ice cloud particles.

[0036] Furthermore, the calculation formula of F is:

[0037]

[0038] in, is the reflectivity of clouds reflecting solar radiation directly to the satellite, is the surface albedo, S is the spherical albedo of the cloud, t1 is the radiation transmittance in the direction of the sun, and t2 is the radiation transmittance in the direction of the satellite; The calculation formula is:

[0039]

[0040] in, is the brightness of solar radiation reflected by clouds, is the solar radiation flux incident on the top of the atmosphere, is the cosine of the solar zenith angle.

[0041] The radiation prediction model can use existing models such as fully connected neural networks and convolutional neural networks to predict radiation values. As one implementation method, the radiation prediction model is an existing extreme random forest model.

[0042] The beneficial effects of the present invention are:

[0043] (1) The existing methods for estimating surface solar radiation all require cloud optical thickness, cloud particle effective radius, aerosol optical thickness, water vapor, ozone and other physical quantities as the basis for calculation. The calculation process is complicated and may also cause error transmission and accumulation due to errors in the input parameters, resulting in low reliability of the calculation results. The present invention directly starts from the first-level data of the satellite (atmospheric top albedo data and solar zenith angle data) to construct a nonlinear and complex relationship with shortwave radiation and its components, avoiding the input of many complex physical quantities and not requiring complex processes such as cloud detection. Therefore, the method has good operability. The method has the advantages of high accuracy, fast calculation speed and strong versatility, and has important practical application value for the assessment of solar energy resources and the planning and operation of solar power generation systems.

[0044] (2) The present invention sets up a feature extraction model to further fully explore the mapping relationship between the data of each dimension within the satellite primary data (multiple observation values) and generate additional feature data. The additional feature data and the initial satellite observation data are input into the radiation prediction model together, so that the prediction model can better explore the internal logic between the input data and the output data from different perspectives, thereby improving the accuracy of the radiation estimation method and its adaptability to different scenarios. DETAILED DESCRIPTION

[0045] This embodiment provides a method for directly estimating surface solar shortwave radiation based on geostationary satellite data, which specifically includes the following steps:

[0046] Acquire the top of atmosphere albedo data of the target area observed by the satellite in the wavelength range of 0.3 μm-3.0 μm, wherein the top of atmosphere albedo data includes multiple observation values, and different observation values ​​are the top of atmosphere albedo observed by the satellite in different wavelength ranges at the same time;

[0047] Obtain the trained feature extraction model, input the observation value into the feature extraction model, and after calculation, the feature extraction model outputs additional feature data;

[0048] Obtain the trained radiation prediction model, input all observation values, solar zenith angle data of the target area and additional feature data into the radiation prediction model, and after calculation, the radiation prediction model outputs the total surface solar shortwave radiation value and the surface solar shortwave direct radiation value.

[0049] As one implementation method, data of albedo_01, albedo_02, ..., albedo_06 and SOZ channels in the satellite primary data are obtained to obtain seven observation values. Among them, the observation value of the albedo_01 channel represents the albedo of the top of the atmosphere with a central wavelength of 0.47 μm, the observation value of the albedo_02 channel represents the albedo of the top of the atmosphere with a central wavelength of 0.51 μm, the observation value of the albedo_03 channel represents the albedo of the top of the atmosphere with a central wavelength of 0.64 μm, the observation value of the albedo_04 channel represents the albedo of the top of the atmosphere with a central wavelength of 0.86 μm, the observation value of the albedo_05 channel represents the albedo of the top of the atmosphere with a central wavelength of 1.6 μm, the observation value of the albedo_06 channel represents the albedo of the top of the atmosphere with a central wavelength of 2.3 μm, and the data of the SOZ channel represents the solar zenith angle at that time.

[0050] The additional feature data includes a first feature value and a second feature value. The internal calculation process of the feature extraction model includes the following steps:

[0051] Arrange the observations in order according to the length of the wavelength to obtain a first array containing six elements;

[0052] Perform the first convolution and the first activation operations on the first array in sequence to obtain the second array, and the length of the first array is equal to the length of the second array; the convolution kernel size of the first convolution is 1*1*6, and the first activation is implemented using the activation function sigmoid;

[0053] The second array is subtracted from the first array (the elements in the second array are subtracted from the corresponding elements in the first array) to obtain the third array;

[0054] Taking the maximum value of all elements in the third array as the first eigenvalue;

[0055] Multiply the elements of the first array by the elements of the second array correspondingly to obtain a fourth array;

[0056] The fourth array is concatenated with the first array to obtain a fifth array; the lengths of the first array and the fourth array are both 6, and the fifth array obtained after concatenation is a one-dimensional array, so the length of the fifth array is 12;

[0057] The fifth array is sequentially subjected to the second convolution and the second activation operations to obtain the sixth array, so as to realize the fusion of the feature information of the first array and the fourth array. The sixth array contains six elements. The convolution kernel size of the second convolution is 1*1*12, and the second activation is realized by using the activation function sigmoid.

[0058] Calculate the average value of all elements in the sixth array and get the second eigenvalue.

[0059] The calculation of the total shortwave radiation value and the direct shortwave radiation value of the surface in the training set based on the physical model includes the following steps:

[0060] Obtain the cloud phase of a preset area, including clear sky, water cloud or ice cloud;

[0061] When the cloud phase is clear, a first data set of a preset area is obtained, and the surface shortwave total radiation value and the surface shortwave direct radiation value of the preset area are calculated based on the first physical model and the first data set; the first data set includes the following dimensional data: solar zenith angle, aerosol optical thickness, total water vapor column, total ozone column and surface albedo.

[0062] When the cloud phase is water cloud, a second data set of a preset area is obtained, and the surface shortwave total radiation value and the surface shortwave direct radiation value of the preset area are calculated based on the second physical model and the second data set; the second data set includes the following dimensional data: solar zenith angle, water cloud optical thickness, water cloud particle effective radius and surface albedo.

[0063] When the cloud phase is ice cloud, a third data set of a preset area is obtained, and the surface shortwave total radiation value and the surface shortwave direct radiation value of the preset area are calculated based on the third physical model and the third data set; the third data set includes the following dimensional data: solar zenith angle, ice cloud optical thickness, ice cloud particle effective radius and surface albedo.

[0064] In this embodiment, the specific value range of each dimensional data in the first data set is shown in Table 1, the specific value range of each dimensional data in the second data set is shown in Table 2, and the specific value range of each dimensional data in the third data set is shown in Table 3.

[0065]

[0066]

[0067] The optical thickness of water clouds and the effective radius of water cloud particles in the preset area are obtained by inverting the above-mentioned first cloud microphysical parameters; the optical thickness of ice clouds and the effective radius of ice cloud particles in the preset area are obtained by inverting the above-mentioned second cloud microphysical parameters. This embodiment uses the atmospheric radiation transfer model software RSTAR for forward simulation to construct the above-mentioned first inversion lookup table and second inversion lookup table. Specifically, in this embodiment, the values ​​of some dimensional data in the first inversion lookup table are shown in Table 4 below, and the values ​​of some dimensional data in the second inversion lookup table are shown in Table 5 below.

[0068]

[0069] Preferably, the present embodiment uses the Newton iteration method to iteratively calculate the first cost function and the second cost function, and the method can obtain the optimal inversion result through a relatively small number of iterations. When constructing the second inversion lookup table, the shape of the ice crystals in the ice cloud is assumed to be a hexagonal prism, and the solar radiation brightness reflected by the ice cloud, the spherical albedo of the ice cloud, and the ice cloud radiation transmittance are calculated based on the hexagonal prism ice crystal scattering model. The relevant parameters calculated based on the hexagonal prism ice crystal scattering model can be obtained from a public database.

[0070] It should be emphasized that in the prior art, when inverting cloud microphysical parameters, the cloud microphysical parameters in the lookup table are directly the optical thickness of the cloud and the effective radius of the cloud particles, while in the present invention, the data in the inversion lookup table are the logarithm of the cloud optical thickness and the logarithm of the effective radius of the cloud particles. After taking the logarithm of the cloud microphysical parameters, the reflectivity and transmittance have a more linear relationship than the traditional data (especially in the log10 space), making the inversion iteration faster; combined with the relevant parameters of the hexagonal prism ice crystal scattering model, this setting can make the maximum inversion error less than 1%, greatly improving the reliability and accuracy of the results.

[0071] The radiation prediction model of this embodiment adopts the existing Extremely Randomized Trees (ERT), which solves classification and regression problems by constructing multiple decision trees. The Extremely Randomized Forest model can process high-dimensional data and large-scale data sets, can better capture complex patterns in the data, and has a strong fitting ability for data with nonlinear relationships. It also has good generalization performance and effectively reduces the risk of overfitting.

[0072] The method provided by the present invention is used to estimate the surface solar shortwave radiation based on Himawari-8 satellite data and Y-4A satellite data, respectively. From the estimation results, it is concluded that the surface solar shortwave total radiation, direct radiation and diffuse radiation estimated based on the two satellite data have good consistency in spatial distribution, the total radiation and direct radiation values ​​are large under clear skies, and the total radiation and direct radiation values ​​are small under cloudy skies. In addition, since the method directly starts from the first-level satellite data, there is no need for complex cloud identification and cloud parameter inversion processes, so that the estimated surface solar shortwave radiation has very good consistency in space with the first-level product, without obvious data missing, which greatly improves the spatiotemporal availability of radiation products.

[0073] The above-mentioned embodiments only express the specific implementation of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for directly estimating surface solar shortwave radiation based on geostationary satellite data, characterized by: The following steps are involved: Acquire the top of atmosphere albedo data of the target area observed by the satellite in the wavelength range of 0.3 μm-3.0 μm, wherein the top of atmosphere albedo data includes multiple observation values, and different observation values ​​are the top of atmosphere albedo observed by the satellite in different wavelength ranges at the same time; Obtaining a trained feature extraction model, inputting the observed value into the feature extraction model, and after calculation, the feature extraction model outputs additional feature data; Obtaining a trained radiation prediction model, inputting all observation values, the solar zenith angle data of the target area and the additional feature data into the radiation prediction model, and after calculation, the radiation prediction model outputs the total short-wave solar radiation value and the direct short-wave solar radiation value of the surface; The additional characteristic data includes a first characteristic value and a second characteristic value; The internal calculation process of the feature extraction model includes the following steps: Arrange the observed values ​​in order according to the length of the wavelength to obtain a first array; Performing a first convolution and a first activation operation on the first array in sequence to obtain a second array, wherein the length of the first array is equal to the length of the second array; The first array is subtracted from the second array to obtain a third array; Taking the maximum value of all elements in the third array as the first eigenvalue; Multiply the elements of the first array by the elements of the second array correspondingly to obtain a fourth array; Concatenate the fourth array with the first array to obtain a fifth array; Performing a second convolution and a second activation operation on the fifth array in sequence to obtain a sixth array, wherein the length of the sixth array is less than the length of the fifth array; Calculate the average value of all elements in the sixth array to obtain the second eigenvalue; When the feature extraction model and the radiation prediction model are trained using a training set, the surface shortwave total radiation value and the surface shortwave direct radiation value in the training set are calculated based on a physical model; Calculating the total shortwave radiation value and the direct shortwave radiation value of the surface in the training set based on the physical model includes the following steps: Acquire cloud phases in a preset area, where the cloud phases include clear sky, water clouds, or ice clouds; When the cloud phase is clear sky, a first data set of the preset area is obtained, and a surface shortwave total radiation value and a surface shortwave direct radiation value of the preset area are calculated based on a first physical model and the first data set; The first data set includes the following dimensional data: solar zenith angle, aerosol optical thickness, total water vapor column, total ozone column and surface albedo; When the cloud phase is water cloud, a second data set of the preset area is obtained, and a surface shortwave total radiation value and a surface shortwave direct radiation value of the preset area are calculated based on a second physical model and the second data set; The second data set includes the following dimensional data: solar zenith angle, water cloud optical thickness, water cloud particle effective radius and surface albedo; When the cloud phase is an ice cloud, a third data set of the preset area is obtained, and a surface shortwave total radiation value and a surface shortwave direct radiation value of the preset area are calculated based on a third physical model and the third data set; The third data set includes the following dimensional data: solar zenith angle, ice cloud optical thickness, ice cloud particle effective radius and surface albedo.

2. The method for directly estimating surface solar shortwave radiation based on geostationary satellite data according to claim 1 is characterized in that: The optical thickness of water clouds and the effective radius of water cloud particles in the preset area are obtained by inverting the first cloud microphysical parameter; Or / and, the ice cloud optical thickness and the ice cloud particle effective radius of the preset area are obtained by inverting the second cloud microphysical parameters.

3. The method for directly estimating surface solar shortwave radiation based on geostationary satellite data according to claim 2 is characterized by: The first cloud microphysical parameter inversion comprises the following steps: Based on the simulation calculation of the atmospheric radiation transfer model, a first inversion lookup table is constructed for the optical thickness of water clouds and the effective radius of water cloud particles; the first inversion lookup table includes the following dimensional data: solar zenith angle, satellite zenith angle, relative azimuth, solar radiation brightness reflected by water clouds, surface albedo, water cloud spherical albedo, water cloud radiation transmittance, logarithm of water cloud optical thickness and logarithm of water cloud particle effective radius; Based on the first inversion lookup table, performing multiple iterative calculations on the first cost function to invert and obtain the water cloud optical thickness and the effective radius of water cloud particles in the preset area; Or / and, the second cloud microphysical parameter inversion comprises the following steps: Based on the simulation calculation of the atmospheric radiation transfer model, a second inversion lookup table is constructed for the optical thickness of ice clouds and the effective radius of ice cloud particles; the second inversion lookup table includes the following dimensional data: solar zenith angle, satellite zenith angle, relative azimuth, solar radiation brightness reflected by ice clouds, surface albedo, ice cloud spherical albedo, ice cloud radiation transmittance, logarithm of ice cloud optical thickness and logarithm of ice cloud particle effective radius; Based on the second inversion lookup table, the second cost function is iterated multiple times to invert and obtain the ice cloud optical thickness and the ice cloud particle effective radius of the preset area.

4. The method for estimating surface solar radiation according to claim 3, characterized in that: The first cost function and the second cost function are both expressed as the following formula: Among them, J is the function value of the cost function, X a represents the initial value of the iteration, T represents the transpose of the matrix, R VIS is the reflectivity matrix observed by the satellite in the visible light channel, R NIR is the reflectivity matrix observed by the satellite in the near-infrared channel, S y is the observation error covariance matrix, S a is the prior error covariance matrix, F is the satellite received radiation brightness simulation value calculated based on the atmospheric radiation transfer model; X is the parameter to be inverted, When the above formula is the first cost function, τ c is the optical thickness of water cloud, r e is the effective radius of water cloud particles. When the above formula is the second cost function, τ c is the ice cloud optical thickness, r e is the effective radius of ice cloud particles.

5. The method for estimating surface solar radiation according to claim 4, characterized in that: The calculation formula of F is: Among them, R c is the reflectivity of the cloud reflecting solar radiation directly to the satellite, A g is the surface albedo, S is the spherical albedo of the cloud, t1 is the radiation transmittance from the sun, and t2 is the radiation transmittance from the satellite; R C The calculation formula is: Among them, L is the brightness of solar radiation reflected by clouds, F0 is the solar radiation flux incident on the top of the atmosphere, and μ0 is the cosine value of the solar zenith angle.

6. The method for directly estimating surface solar shortwave radiation based on geostationary satellite data according to claim 1, characterized in that: The radiation prediction model is an extreme random forest model.