Satellite radiation product space downscaling method based on cloud picture

Abstract

The invention discloses a satellite radiation product space downscaling method based on a cloud picture, and relates to the technical field of remote sensing data processing, and the method comprises the following steps: obtaining low-resolution satellite radiation data and high-resolution cloud picture data; upscaling the high-resolution cloud picture data to the resolution which is the same as that of the low-resolution satellite radiation data, and establishing a multivariate statistical mapping relation between the high-resolution cloud picture data and the low-resolution satellite radiation data under the same resolution; applying the mapping relation to high-resolution cloud picture data to generate an initial radiation prediction field; according to the initial radiation prediction field, determining the spatial allocation weight of each high-resolution pixel in the low-resolution pixel to which the high-resolution pixel belongs; and performing spatial reconstruction on the low-resolution satellite radiation data according to the spatial allocation weight to obtain a high-resolution satellite radiation product. According to the method, the problem of insufficient precision caused by neglecting cloud influence and direct resampling in the prior art is effectively solved, and the space detail precision and the physical rationality of a downscaling product under a complex cloud condition are improved.

Description

Technical Field

[0001] This invention relates to the field of remote sensing data processing technology, and more specifically, to a spatial downscaling method for satellite radiometric products based on cloud images. Background Technology

[0002] With the continued growth in demand for renewable energy and the advancement of energy transition, the proportion of photovoltaic power generation in my country's energy structure has been continuously increasing. However, photovoltaic power generation is highly dependent on the spatiotemporal distribution characteristics of incident solar radiation (SSI). Currently, ground-based observation is the most direct and reliable method for obtaining accurate and high temporal resolution solar radiation data. At present, the spatial sampling density of meteorological radiation stations that can directly observe SSI is low, and the spatial resolution of SSI data based on station observations cannot meet the technical requirements of the spatial scale for refined remote sensing applications. To compensate for the shortcomings caused by the sparseness of observation stations, existing technologies generally utilize climatological methods to improve the insufficient accuracy of SSI assessment caused by the sparse radiation stations. Although climatological methods are computationally simple, their estimation results exhibit significant regional biases.

[0003] With the development of meteorological satellite remote sensing technology, acquiring SSI products based on satellite observation data has become an important method, offering advantages such as wide coverage and high temporal resolution. However, the spatial resolution of existing satellite SSI products is relatively low, making it difficult to directly meet the needs of refined applications. Therefore, existing research has proposed spatial downscaling of satellite SSI products to improve their spatial resolution.

[0004] For example, the invention patent with announcement number CN113076865B discloses a method and system for retrieving irradiance based on sky-photographed images and satellite cloud images. This includes establishing a mapping function from low-resolution satellite cloud images to high-resolution sky-images for cloud cover information in the same area; parsing the satellite cloud image data into high-resolution image data using this mapping function; obtaining cloud parameters from the high-resolution image data; obtaining numerical weather prediction data; and substituting the cloud parameters and numerical weather prediction data into a radiative transfer model to obtain irradiance data, including horizontal solar irradiance, direct solar irradiance, and diffuse solar irradiance. This invention uses sky-photographed images and satellite images to establish a super-resolution mapping relationship, thereby obtaining high-resolution satellite cloud image data. The irradiance retrieved from high-resolution satellite cloud image data has higher accuracy than that retrieved directly from ordinary satellite cloud image data, providing a solid foundation for the industrial application of this technology.

[0005] The above-disclosed technical solutions have at least the following technical problems: Existing technologies lack sufficient data coverage and adaptability for large-scale applications. CN113076865B relies on local sky images captured by a ground-based all-sky imager, which can only cover a 180° sky range of a single photovoltaic power station. It cannot achieve continuous cloud information acquisition across regions and over large areas. Furthermore, it requires the deployment of an imager at each power station, resulting in high deployment costs and maintenance difficulties, making it difficult to support the large-scale processing needs of all-domain satellite radiation products.

[0006] Traditional scaling methods and cloud information fusion lack sufficient depth, making it difficult to balance physical consistency and detail fidelity. CN113076865B focuses on the single-point irradiance inversion of a single photovoltaic power station, without considering the core physical constraint of satellite radiation product downscaling, namely the consistency of total regional radiation. If directly applied to global downscaling, it will lead to a disconnect between the original low-resolution pixel observations and the total regional radiation of high-resolution products, resulting in physical distortion.

[0007] To address the above problems, this invention proposes a solution. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a spatial downscaling method for satellite radiation products based on cloud images. By establishing and utilizing the statistical mapping relationship between cloud image data and radiation data, the method guides the generation of high-resolution spatial weights and the reconstruction of radiation values, thereby solving the problem that existing downscaling methods lack accuracy under complex cloud conditions due to their failure to effectively integrate the spatial physical information of clouds.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A spatial downscaling method for satellite radiometric products based on cloud imagery includes the following steps: acquiring low-resolution satellite radiometric data and high-resolution cloud imagery data; upscaling the high-resolution cloud imagery data to the same resolution as the low-resolution satellite radiometric data, and establishing a multivariate statistical mapping relationship between the two at the same resolution; applying the mapping relationship to the high-resolution cloud imagery data to generate an initial radiometric prediction field; determining the spatial allocation weight of each high-resolution pixel in its corresponding low-resolution pixel based on the initial radiometric prediction field; and spatially reconstructing the low-resolution satellite radiometric data according to the spatial allocation weights to obtain a high-resolution satellite radiometric product.

[0010] In a preferred embodiment, acquiring low-resolution satellite radiation data and high-resolution cloud image data includes: the spatial resolution of the high-resolution cloud image data is an integer multiple of the low-resolution satellite radiation data; based on the geographical location of the ground radiation observation station and the scale relationship of the integer multiple, extracting paired data regions centered on each station from the satellite radiation data and cloud image data.

[0011] In a preferred embodiment, the step of upscaling the high-resolution cloud image data to the same resolution as the low-resolution satellite radiometric data includes: dividing the high-resolution cloud image pixels into multiple non-overlapping regular grid windows based on the integer multiple scale relationship between the high-resolution cloud image data and the low-resolution satellite radiometric data; performing feature aggregation on all high-resolution pixels within each regular grid window to generate aggregated feature values; and using the aggregated feature values ​​corresponding to each regular grid window as cloud image data values ​​upscaled to the spatial resolution of the low-resolution satellite radiometric data.

[0012] In a preferred embodiment, the characteristic value includes an arithmetic mean, a median, or a weighted average.

[0013] In a preferred embodiment, establishing the multivariate statistical mapping relationship between the two includes: extracting a first data sequence of low-resolution satellite radiometric data and a second data sequence of upscaled high-resolution cloud image data from the paired data region; determining the mapping relationship between the two by using multiple channel values ​​in the second data sequence as independent variables and radiometric values ​​in the first data sequence as dependent variables through statistical modeling, wherein the statistical modeling adopts multivariate linear fitting, and the mapping relationship is characterized by a set of regression coefficients.

[0014] In a preferred embodiment, applying the mapping relationship to high-resolution cloud image data to generate an initial radiation prediction field includes: processing the regression coefficients of the multivariate statistical mapping relationship through spatial interpolation to generate a continuous coefficient distribution field that is spatially consistent with the high-resolution cloud image data; applying the multivariate statistical mapping relationship pixel by pixel based on the coefficient distribution field and the multichannel values ​​of the high-resolution cloud image data to calculate the radiation prediction value of each high-resolution pixel; and spatially combining the radiation prediction values ​​to generate a high-resolution initial radiation prediction field.

[0015] In a preferred embodiment, determining the spatial allocation weight of each high-resolution pixel within its corresponding low-resolution pixel based on the initial radiation prediction field specifically includes: taking multiple high-resolution pixels covered by each low-resolution pixel as a processing unit; and determining the spatial allocation weight within its corresponding low-resolution pixel according to the radiation prediction value of each high-resolution pixel within the processing unit.

[0016] In a preferred embodiment, the step of spatially reconstructing low-resolution satellite radiometric data according to spatial allocation weights to obtain high-resolution satellite radiometric products includes: normalizing the spatial allocation weights of all high-resolution pixels contained in the processing unit corresponding to each low-resolution pixel; allocating the observation value of each low-resolution pixel according to the normalized weights of each high-resolution pixel in its processing unit to obtain a high-resolution pixel-level radiometric value; and aggregating the high-resolution pixel-level radiometric values ​​to generate a high-resolution satellite radiometric product.

[0017] In a preferred embodiment, the high-resolution satellite radiometric product satisfies the following condition: the original observation value of any low-resolution pixel is equal to the arithmetic mean of the radiometric values ​​of all corresponding high-resolution pixels within its spatial coverage area.

[0018] In a preferred embodiment, after obtaining the high-resolution satellite radiation product, a correction step is further included: acquiring ground observation data and determining the error between the ground observation data and the satellite radiation product data at the corresponding station location; generating an error correction field consistent with the spatial resolution of the satellite radiation product using a spatial interpolation method; and using the error correction field to spatially correct the satellite radiation product to obtain the final product.

[0019] The technical effects and advantages of the spatial downscaling method for satellite radiometric products based on cloud images proposed in this invention are as follows: 1. This invention establishes a statistical mapping relationship between cloud image data and radiation data based on low resolution, and applies the statistical mapping relationship to high-resolution cloud image data. This breaks through the limitation of traditional downscaling methods that rely solely on numerical resampling, enabling the downscaling process to directly respond to and integrate the cloud variable information that is most critical to solar radiation. In principle, this significantly improves the physical rationality and spatial accuracy of downscaling results under complex cloud conditions (such as cloud edges and broken cloud areas).

[0020] 2. This invention determines the spatial allocation weight of each high-resolution pixel in its corresponding low-resolution pixel by using an initial radiation prediction field, and uses the spatial allocation weight to spatially reconstruct the low-resolution satellite radiation data. This ensures that the downscaled high-resolution product can strictly maintain the consistency with the original satellite observation data in terms of total regional data while introducing fine spatial details. It achieves the unity of detail enhancement and physical fidelity, and effectively avoids the spatial distortion problem caused by simple interpolation. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a spatial downscaling method for satellite radiation products based on cloud images, according to the present invention. Figure 2 This is a result diagram of the FY4B method of the present invention; Figure 3 This is a diagram showing the spatial downscaling results of this invention; Figure 4 This is a diagram showing the site-based correction results of the present invention. Detailed Implementation

[0022] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1, Figure 1 This invention presents a spatial downscaling method for satellite radiation products based on cloud images, comprising the following steps: S1, acquire low-resolution satellite radiation data and high-resolution cloud image data; In this embodiment, acquiring low-resolution satellite radiation data and high-resolution cloud image data specifically includes: From the FY-4B level data products released by the Satellite Meteorological Center, the surface incident solar radiation (SSI) variable and multi-channel scanning imaging radiometer (AGRI) data at the same observation time are obtained from the surface shortwave radiation (SSR) product. The surface incident solar radiation (SSI) variable is used as low-resolution satellite radiation data. In this embodiment, a 15-minute SSI product with a spatial resolution of 4km is selected, that is, each pixel represents an area of ​​about 4km × 4km on the ground. The multi-channel scanning imaging radiometer (AGRI) data is used as high-resolution cloud image data. In this embodiment, L1 level data with a spatial resolution of 1km is selected, and three core spectral channels are extracted to characterize cloud characteristics: channel 1 (0.47µm, visible blue band, sensitive to aerosols and thin clouds), channel 2 (0.65µm, visible red band, sensitive to cloud optical thickness), and channel 3 (0.825µm, near-infrared band, sensitive to cloud particle size).

[0024] The spatial resolution of the high-resolution cloud image data is four times that of the low-resolution satellite radiometric data.

[0025] A ground-based radiation observation station is selected, and the coordinates of that station are located in the 4km satellite radiation data at a 4km distance. A rectangular area containing 50×50 pixels is then cropped from this pixel. This area covers approximately 200km×200km of ground. In the 1km cloud image data, based on the integer multiple relationship (N=4), an area with the exact same geographical extent as the aforementioned 4km area is determined. Since the resolution of 1km is four times that of 4km, a corresponding rectangular area containing 200×200 pixels is cropped. This rectangular area is the paired data area.

[0026] Within the extracted paired data area, each low-resolution (4km) radiative pixel spatially and uniquely covers the area. A continuous high-resolution (1km) cloud image element.

[0027] S2, upscale the high-resolution cloud image data to the same resolution as the low-resolution satellite radiometric data, and establish a multivariate statistical mapping relationship between the two at the same resolution; In this embodiment, the step of upscaling the high-resolution cloud image data to the same resolution as the low-resolution satellite radiometric data specifically involves: S21, based on the integer multiple scale relationship between high-resolution cloud image data and low-resolution satellite radiometric data, the high-resolution cloud image pixels are divided into multiple non-overlapping regular grid windows, specifically as follows: The entire 200×200 high-resolution cloud image data matrix is ​​considered as consisting of 50×50 non-overlapping regular grid windows, where each window contains N×N (i.e., 4×4) consecutive high-resolution pixels. Each regular grid window strictly corresponds to the position of a specific low-resolution pixel in the low-resolution paired data region (50×50 low-resolution pixels).

[0028] S22, for each 4×4 regular grid window, calculate the aggregated feature value of all 16 high-resolution cloud image pixels within it. The feature value is specifically the arithmetic mean, median, or weighted average. In this embodiment, as a preferred and stable implementation, the feature value is the arithmetic mean. That is, for each regular grid window, the 16 pixel values ​​for its three channels (0.47μm, 0.65μm, 0.825μm) are calculated independently:

[0029] in, This represents the upscaled contour value of the regular grid window (i.e., the corresponding low-resolution location) on channel k. This represents the value of the i-th high-resolution pixel in channel k within the regular grid window.

[0030] S23. The calculated feature values ​​of each regular grid window (for three-channel data, each window yields three average values) are assigned to the unique low-resolution pixel location corresponding to that window. After traversing and calculating all 50×50 windows, a new data product is finally generated, namely, the upscaled high-resolution cloud image data: its spatial grid is completely consistent with the original low-resolution radiometric data (50×50 4km pixels), and the value on each pixel carries the aggregated information from the corresponding 16 high-resolution cloud image pixels.

[0031] It should be noted that the aforementioned methods for calculating eigenvalues ​​such as the median or weighted average can also be used in this step, achieving the goal of aggregating high-resolution information to a low-resolution scale. For example, using the median can enhance robustness to outliers; using a weighted average based on the distance from the pixel to the window center can reflect spatial weights. In this embodiment, the arithmetic mean is used. The core purpose of the upscaling step is to provide unbiased and robust regional representative values ​​for the subsequent establishment of multivariate statistical mapping relationships. As the optimal unbiased estimator describing the central tendency of the data set, the arithmetic mean can effectively filter out random noise between pixels in the high-resolution cloud image and evenly reflect the overall average state of cloud conditions within the regular grid window, thereby ensuring that the "cloud-radiation" statistical relationship learned at the low-resolution scale has the best generalization ability and physical universality.

[0032] In this embodiment, establishing the multivariate statistical mapping relationship between the two specifically involves: S24. Arrange all 2500 pixel values ​​from the 50×50 low-resolution satellite radiometric data into a vector of length 2500 in spatial order (e.g., row priority). This serves as the first data sequence (radiance value sequence). 2500 pixels at the same spatial location within the upscaled high-resolution cloud image data block, each containing 3 channel values, are arranged into a 2500-row × 3-column matrix. As the second data sequence (cloud map value sequence), each row The three channels of upscaling value corresponding to a pixel.

[0033] S25. Using the second data sequence as the independent variable (C) and the first data sequence as the dependent variable (R), construct a multiple linear regression model:

[0034] in, Let be the radiation value of the j-th pixel. , and These represent the upscaling contour values ​​of the three channels corresponding to that pixel. It is the intercept. , and Let be the regression coefficients to be determined. It represents the residual.

[0035] This embodiment uses the least squares method to solve the above multiple linear regression model, aiming to minimize the sum of squared residuals. A set of optimal regression coefficients is obtained through this solution. , , and .

[0036] The set of regression coefficients obtained by solving , , and The multivariate statistical mapping relationship between low-resolution satellite radiometric data and upscaled high-resolution cloud image data was determined.

[0037] By traversing all M ground radiation observation stations, the local multivariate statistical mapping relationship of the M stations can be obtained.

[0038] S3, apply the mapping relationship to the high-resolution cloud image data to generate an initial radiation prediction field; In this embodiment, applying the mapping relationship to high-resolution cloud image data to generate an initial radiation prediction field specifically involves: The regression coefficients are spatially interpolated onto the same spatial grid as the high-resolution cloud map data to generate a continuous regression coefficient field. This embodiment uses ordinary kriging interpolation, interpolating the four regression coefficients separately. , , and Independent interpolation calculations were performed to obtain four two-dimensional raster layers that were perfectly aligned with the 1km geographic grid of the target area. , , and These together constitute a spatially continuous regression coefficient field. For any 1km pixel within the target area, its location corresponds to a specific set of regression coefficients. .

[0039] It should be noted that Kriging interpolation is an existing technique, and its detailed process will not be elaborated here. Based on geostatistical principles, Kriging calculates the optimal coefficient estimate for each target grid point within the target area, taking into account the spatial autocorrelation of the site. This method provides unbiased, optimal linear predictions and generates smooth, continuous surfaces.

[0040] For each high-resolution pixel (i.e., a 1km pixel), from the regression coefficient field , , and In the process, four coefficient values ​​corresponding to the pixel location are read. From the original high-resolution cloud image data, three channel observation values ​​of the pixel at the same time are read. The extracted coefficient values ​​and the three channel observation values ​​are substituted into the multiple linear regression model to directly calculate the radiometric prediction value of the high-resolution pixel. It iterates through and calculates the predicted radiation value for each 1km pixel within the target area. Then, all the results are arranged and combined according to their original geospatial coordinates to obtain a high-resolution initial radiation prediction field.

[0041] S4. Based on the initial radiation prediction field, determine the spatial allocation weight of each high-resolution pixel in its corresponding low-resolution pixel. In this embodiment, determining the spatial allocation weight of each high-resolution pixel to its corresponding low-resolution pixel based on the initial radiation prediction field specifically involves: The processing unit consists of an original low-resolution (4km) radiative pixel and all the high-resolution (1km) pixels that it completely covers in space. Since the resolution has an integer multiple relationship of 4, each low-resolution (4km) radiative pixel corresponds precisely to 16 consecutive 1km×1km high-resolution pixels in space.

[0042] The step of determining the spatial allocation weight within its respective low-resolution pixel according to proportion specifically involves: extracting the radiation prediction values ​​of the 16 high-resolution pixels belonging to the processing unit from the high-resolution initial radiation prediction field, denoted as... Calculate the predicted radiometric value of the k-th high-resolution pixel. The proportion of the total radiometric prediction values ​​of all 16 high-resolution pixels in this processing unit :

[0043] proportion The spatial weights for this high-resolution pixel are determined.

[0044] S5 reconstructs low-resolution satellite radiation data spatially based on spatial allocation weights to obtain high-resolution satellite radiation products.

[0045] In this embodiment, the step of spatially reconstructing low-resolution satellite radiometric data according to spatial allocation weights to obtain high-resolution satellite radiometric products specifically involves: For each processing unit, the spatial allocation weights of all high-resolution pixels within it are normalized to obtain the normalized weights. Read the raw, true satellite radiation observation value of the unique low-resolution (4km) radiation pixel corresponding to that processing unit. ,Will According to the normalized weights The radiance value ultimately assigned to the k-th high-resolution pixel is one of the 16 high-resolution pixels it covers. :

[0046] Traversing all processing units within the target area, each high-resolution (1km) pixel receives an assignment value from its corresponding low-resolution (4km) pixel. After all distribution The values ​​are rearranged and integrated according to their inherent 1km geospatial coordinates to generate high-resolution satellite radiation products covering the target area, with spatial resolution consistent with the high-resolution cloud image data.

[0047] The high-resolution satellite radiation product satisfies the following condition: the original observation value of any low-resolution pixel is equal to the arithmetic mean of the radiation values ​​of all corresponding high-resolution pixels within its spatial coverage area.

[0048] In this embodiment, after obtaining the high-resolution satellite radiation product, a correction step is further included, specifically: Pixel values ​​of high-resolution satellite radiation products corresponding to the geographical locations of ground radiation observation stations are extracted. Observational data (i.e., measured ground radiation values) of ground radiation observation stations are acquired simultaneously. The error between the high-resolution satellite radiation product data and the observation data at each station is calculated. The error values ​​of discrete ground radiation observation stations are extended to the same resolution as the high-resolution satellite radiation products through spatial interpolation (such as ordinary kriging) to form an error correction field. The error correction field is then added pixel by pixel to the spatially downscaled high-resolution satellite radiation products to obtain the final product after station correction.

[0049] Example 2: This example describes the spatial downscaling and correction of satellite radiation data for a certain region.

[0050] like Figure 2 As shown, Figure 2The data is raw 4km×4km low-resolution FY-4B satellite radiometric data. The color blocks are distributed smoothly on a large scale. For example, the northern region is entirely orange (800~900W / m²), while the southeastern region is a low-value blue-green area (300~500W / m²), with no small-scale textural variations. This reflects the inherent limitations of low-resolution satellite radiometric data, namely, the inability to capture surface radiometric heterogeneity at the 1km scale. Small-scale radiometric gradients are completely masked, and only large-scale regional radiometric patterns can be provided. This data cannot meet the high-resolution requirements for solar energy resource assessment, agricultural ecological monitoring, and other applications.

[0051] like Figure 3 As shown, Figure 3 Yes Figure 2 The resulting image after spatial downscaling improves spatial resolution to 1km. Color texture changes from large-scale smooth blocks to fine patches. For example, small-scale gradient changes in yellow and orange appear in the northern region, and the blue-green texture in the low-value area of ​​the southeast also shows a more refined patch distribution. Optimizing from 4km to 1km improves spatial detail recognition capability by 4 times, capturing small-scale radiometric heterogeneity that was masked in the original data. While adding small-scale details, it still retains the large-scale radiometric distribution patterns of the original 4km data (such as the overall pattern of high in the north and low in the south), without any distortion of spatial distribution.

[0052] like Figure 4 As shown, Figure 4 To be based on ground-based radiation observation stations Figure 3 Corrected results map. At a high resolution of 1 km, significant red and purplish-red high-value areas (approaching 1100 W / m²) appear in the northern and eastern regions. This feature... Figure 2 and Figure 3 The data was not fully reflected in the image; meanwhile, the blue-green texture in the low-value area in the southeast better matches the solar radiation distribution observed on the ground. By combining corrections with ground station data, the systematic bias of the satellite inversion was corrected, making the radiation values ​​in the high-resolution product closer to actual observations. For example, Figure 4 The red high-value areas in the data are highly consistent with the measured surface radiation values ​​at noon on a sunny day in the northern part of the region, which solves the problem of underestimation of high values ​​caused by smoothing of the original data.

[0053] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0054] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0055] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0056] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0057] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0058] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A spatial downscaling method for satellite radiation products based on cloud images, characterized in that, Includes the following steps: Acquire low-resolution satellite radiometric data and high-resolution cloud imagery data; High-resolution cloud image data is upscaled to the same resolution as low-resolution satellite radiometric data, and a multivariate statistical mapping relationship between the two is established at the same resolution. The mapping relationship is applied to high-resolution cloud image data to generate an initial radiation prediction field; Based on the initial radiation prediction field, determine the spatial allocation weight of each high-resolution pixel in its corresponding low-resolution pixel. Low-resolution satellite radiation data is spatially reconstructed based on spatial allocation weights to obtain high-resolution satellite radiation products.

2. The spatial downscaling method for satellite radiation products based on cloud images according to claim 1, characterized in that, The acquisition of low-resolution satellite radiometric data and high-resolution cloud image data includes: The spatial resolution of the high-resolution cloud image data is an integer multiple of that of the low-resolution satellite radiometric data. Based on the geographical location of the ground radiation observation stations and the integer multiple scale relationship, paired data regions centered on each station are extracted from the satellite radiation data and cloud image data.

3. The spatial downscaling method for satellite radiation products based on cloud images according to claim 2, characterized in that, Upscaling high-resolution cloud image data to the same resolution as low-resolution satellite radiometric data includes: Based on the integer multiple scale relationship between high-resolution cloud image data and low-resolution satellite radiometric data, high-resolution cloud image pixels are divided into multiple non-overlapping regular grid windows. Feature aggregation is performed on all high-resolution pixels within each of the regular grid windows to generate aggregated feature values; The aggregated feature value corresponding to each regular grid window is used as the cloud image data value upscaled to the spatial resolution of the low-resolution satellite radiometric data.

4. The spatial downscaling method for satellite radiation products based on cloud images according to claim 3, characterized in that, The characteristic values ​​include the arithmetic mean, median, or weighted average.

5. The spatial downscaling method for satellite radiation products based on cloud images according to claim 4, characterized in that, The establishment of the multivariate statistical mapping relationship between the two includes: From the paired data regions, extract the first data sequence of low-resolution satellite radiometric data and the second data sequence of upscaled high-resolution cloud image data; Using multiple channel values ​​in the second data sequence as independent variables and radiation values ​​in the first data sequence as dependent variables, the mapping relationship between the two is determined through statistical modeling. The statistical modeling adopts multiple linear fitting, and the mapping relationship is characterized by a set of regression coefficients.

6. The spatial downscaling method for satellite radiation products based on cloud images according to claim 5, characterized in that, The step of applying the mapping relationship to high-resolution cloud image data to generate an initial radiation prediction field includes: The regression coefficients of the multivariate statistical mapping relationship are processed by spatial interpolation to generate a continuous coefficient distribution field consistent with the high-resolution cloud map data space. Based on the coefficient distribution field and the multi-channel values ​​of the high-resolution cloud map data, the multivariate statistical mapping relationship is applied pixel by pixel to calculate the radiation prediction value of each high-resolution pixel. The predicted radiation values ​​are spatially combined to generate a high-resolution initial radiation prediction field.

7. The spatial downscaling method for satellite radiation products based on cloud images according to claim 6, characterized in that, The step of determining the spatial allocation weight of each high-resolution pixel to its corresponding low-resolution pixel based on the initial radiation prediction field specifically includes: Each low-resolution pixel covers multiple high-resolution pixels as a processing unit; Based on the radiometric prediction value of each high-resolution pixel within the processed unit, its spatial allocation weight within its corresponding low-resolution pixel is determined proportionally.

8. The spatial downscaling method for satellite radiation products based on cloud images according to claim 7, characterized in that, The process of spatially reconstructing low-resolution satellite radiometric data based on spatial allocation weights to obtain high-resolution satellite radiometric products includes: For each processing unit corresponding to a low-resolution pixel, the spatial allocation weights of all high-resolution pixels contained therein are normalized. The observation value of each low-resolution pixel is allocated according to the normalized weight of each high-resolution pixel in its processing unit to obtain the high-resolution pixel-level radiometric value. By aggregating the high-resolution pixel-level radiation values, a high-resolution satellite radiation product is generated.

9. The spatial downscaling method for satellite radiation products based on cloud images according to claim 8, characterized in that, The high-resolution satellite radiation product satisfies the following condition: the original observation value of any low-resolution pixel is equal to the arithmetic mean of the radiation values ​​of all corresponding high-resolution pixels within its spatial coverage area.

10. The spatial downscaling method for satellite radiation products based on cloud images according to claim 9, characterized in that, After obtaining the high-resolution satellite radiation product, a correction step is also included: Acquire ground-based observation data and determine the error between it and satellite radiation product data at corresponding station locations; The error is then used to generate an error correction field consistent with the spatial resolution of the satellite radiometric product using a spatial interpolation method. The satellite radiation product is spatially corrected using the error correction field to obtain the final product.

Images