Downscaling calculation method and device for earth surface downward short wave radiation and storage medium

Abstract

The invention discloses an earth surface downward short wave radiation downscaling calculation method and device and a storage medium. The earth surface downward short wave radiation downscaling calculation method comprises the following steps: acquiring low spatial resolution data and high spatial resolution data; constructing a cloud layer shielding index based on the rainfall data; constructing a stacking model based on integration of various machine learning models, and training the stacking model by adopting low-spatial-resolution data; inputting the obtained high-spatial-resolution data as an input feature vector into the trained stack model to obtain an initial high-spatial-resolution earth surface downward short wave radiation value of high-spatial-resolution earth surface downward short wave radiation; and applying energy conservation constraint to the obtained initial high-spatial-resolution earth surface downward short wave radiation value to perform consistency correction. According to the method, an energy conservation constraint correction mechanism is introduced, and the spatial resolution, precision and physical consistency of the earth surface downward short wave radiation data under the complex terrain condition are effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of eco-hydrology and meteorological data processing, specifically involving a method for downscaling surface shortwave radiation by integrating multi-source environmental factors and energy conservation constraints. Background Technology

[0002] Shortwave radiation down to the surface is a key meteorological element in land surface energy exchange and eco-hydrological processes, and its spatial distribution directly affects the accuracy of evapotranspiration estimation, ecological process simulation, and land surface model calculations. Influenced by topographic relief, underlying surface differences, and atmospheric conditions, shortwave radiation exhibits significant spatial heterogeneity. However, existing shortwave radiation data mainly come from ground-based observation stations, satellite products, and reanalysis data. Ground-based radiation observation stations are limited in number and spatially unevenly distributed, making it difficult to continuously characterize the spatial variation characteristics of shortwave radiation at the regional scale. Reanalysis and satellite radiation data are limited by spatial resolution, resulting in insufficient ability to represent the fine distribution of shortwave radiation in complex topographic regions.

[0003] To improve the spatial resolution of shortwave radiation data, existing studies typically employ statistical interpolation or empirical downscaling methods to spatially refine low-resolution radiation data. However, these methods struggle to fully account for the combined effects of topographic modulation and atmospheric conditions on shortwave radiation, and they can easily disrupt the consistency of total radiation during the downscaling process. With the application of machine learning methods in meteorological data processing, data-driven models can characterize the nonlinear relationships between multi-source environmental factors and shortwave radiation. However, existing methods often focus on single-model modeling, lacking effective integration of the advantages of multiple models and necessary physical constraints, thus limiting the accuracy and reliability of the downscaling results. Summary of the Invention

[0004] To address the problems of insufficient spatial resolution in existing shortwave radiation data, difficulty in comprehensively characterizing the combined effects of underlying surface differences and local factors such as cloud cover, and the tendency to disrupt the consistency of total radiation during downscaling, this invention provides a surface-downscaling method for shortwave radiation based on multi-source environmental factors and a stacked integrated model.

[0005] To achieve the above objectives, the present invention specifically adopts the following technical solution: This invention first provides a method for downscaling shortwave radiation down from the Earth's surface, comprising the following steps: Step S1: Acquire low spatial resolution data and high spatial resolution data. The low spatial resolution data includes low spatial resolution surface-to-surface shortwave radiation data, as well as low spatial resolution elevation, sky visibility factor, slope, aspect, vegetation index, and precipitation data. The high spatial resolution data includes high spatial resolution elevation, sky visibility factor, slope, aspect, vegetation index, and precipitation data. Step S2: Construct a cloud cover index based on the precipitation data to quantitatively characterize the cloud cover effect on downward shortwave radiation from the ground surface under precipitation conditions. Step S3: Construct a stacked model based on the integration of multiple machine learning models, and train the stacked model using low spatial resolution data; the input feature vector for training the stacked model consists of elevation, sky visibility factor, slope, aspect, vegetation index and cloud cover index under low spatial resolution, and the output feature vector for training the stacked model is the shortwave radiation data down the ground surface. Step S4: The high spatial resolution data obtained in step S1 is used as the input feature vector and input into the trained stacked model to obtain the initial high spatial resolution surface downward shortwave radiation value. Step S5: Apply energy conservation constraints to the obtained initial high spatial resolution surface downward shortwave radiation values ​​for consistency correction, so as to ensure the consistency of the total surface downward shortwave radiation before and after downscaling.

[0006] Downscaling correction for surface-to-surface shortwave radiation: Errors in the accuracy of surface-to-surface shortwave radiation are attributed to the scale limitations of low spatial resolution reanalysis / satellite products, as well as the combined effects of underlying surface differences and local factors such as cloud cover. The downscaling correction is developed based on an explicit characterization of these multi-source influence mechanisms. Specifically, topographic parameters, underlying surface features, and cloud cover indices are used as key driving factors. A stacked model integrating multiple machine learning models is constructed as input features. Low spatial resolution surface-to-surface shortwave radiation data is used as constraints, and high spatial resolution radiation values ​​are used as output targets to obtain initial downscaling results. Based on this, to avoid introducing machine learning system biases and maintain energy consistency at the regional scale, a consistency correction based on energy conservation constraints is performed on the initial results, ultimately yielding a downscaling correction for surface-to-surface shortwave radiation that balances local spatial heterogeneity and regional energy balance characteristics.

[0007] In step S5, the formula for calculating high spatial resolution surface downward shortwave radiation after consistency correction is as follows: (1) In the formula, Mesh at high spatial resolution During the period Corrected surface downwave radiation (W / m) 2 ); Mesh at high spatial resolution During the period Initial surface downward shortwave radiation (W / m 2 ); Mesh at low spatial resolution During the period Energy conservation correction coefficient; Mesh at low spatial resolution The corresponding high spatial resolution subgrid set. The energy conservation correction coefficient is used to adjust the spatial average of the initial downscaling result at high spatial resolution, making it equal to the downward shortwave radiation value of the corresponding low spatial resolution grid, thereby maintaining a constant total radiation while spatially refining. Therefore The calculation is shown in equation (2).

[0008] (2) In the formula, Mesh at low spatial resolution During the period The surface downward shortwave radiation value (W / m 2 ); For the low spatial resolution mesh The number of grids at high spatial resolution within the coverage area; the denominator represents the spatial average of the initial downscaling results at high spatial resolution within the range corresponding to the low spatial resolution grids.

[0009] The initial surface downward shortwave radiation values ​​are output through a stacked model, and their expression is: (3) In the formula, A stacked model built upon the integration of multiple machine learning models; Mesh at high spatial resolution During the period The input feature vector.

[0010] In step S3, the stacked model is trained at low spatial resolution by minimizing the following loss function: (4) In the formula, Train the loss function for the model; The number of time steps for the training samples; This refers to the number of grid cells at low spatial resolution. Mesh at low spatial resolution During the period The input feature vector.

[0011] In step S3, the stacked model consists of an ensemble function of multiple base learning models, and its expression is: (5) In the formula, This is an ensemble function used to fuse the outputs of various base learning models; For integration functions Parameters; For the first Individual base learning model; The number of base-learning models.

[0012] In step S3, the ensemble function used to fuse the outputs of each base learning model adopts a ridge regression form, and its fusion form of the outputs of each base learning model is as follows: (6) In the formula, For the first Weight coefficients of a basic learning model; For bias terms; parameters This is obtained by minimizing the following objective function with a penalty term: (7) In the formula, The regularization parameter is used to control the severity of the penalty. When this happens, the objective function degenerates into ordinary least squares regression.

[0013] In steps S3 and S4, the input feature vector is represented as follows: (8) (9) In the formula, This indicates that data is concatenated along the feature dimension; and Mesh at low and high spatial resolutions, respectively. and During the period The input feature vector; and Mesh at low and high spatial resolutions, respectively. and During the period The underlying surface features are composed of elevation, sky visibility factor, slope, aspect, and vegetation index. and Mesh at low and high spatial resolutions, respectively. and During the period The cloud cover index.

[0014] In step S2, the cloud cover index is expressed as: (10) (11) In the formula, This is a mapping function used to quantitatively characterize the cloud cover effect; and Mesh at low and high spatial resolutions, respectively. and During the period Rainfall (mm).

[0015] The mapping function takes the following form: (12) (13) (14) (15) In the formula, This is the baseline shading constant under no-rain or light-rain conditions, set based on regional climate background or historical statistics; Low-intensity precipitation scaling factor (mm) is used to adjust the linear growth rate of the shading index as precipitation increases during the light rain phase. is the saturation scale constant (mm), used to control the saturation rate of the shading index as precipitation increases during the moderate rain stage; This is the saturation bias term, which is a very small positive number, typically taken as 0.05; This represents the cloud cover index value corresponding to the upper limit of light rain intensity.

[0016] The present invention also provides a surface-downward shortwave radiation downscaling device, comprising a processor and a memory; the memory stores a program or instructions, which are loaded and executed by the processor to implement the steps of the above-described surface-downward shortwave radiation downscaling calculation method.

[0017] The present invention also provides a computer-readable storage medium having a program or instructions stored thereon, which, when executed by a processor, implements the steps of the above-described method for calculating downscaling of shortwave radiation down from the Earth's surface.

[0018] The beneficial effects of this invention are: This invention proposes a method for downscaling surface shortwave radiation considering the resolution limitations of remote sensing products, underlying surface characteristics, and the influence of cloud shading. First, it acquires low spatial resolution surface shortwave radiation data, along with elevation, sky visibility factor, slope, aspect, vegetation index, and precipitation data at both low and high resolutions. Then, based on the precipitation data, it constructs a cloud shading index to quantitatively characterize the cloud shading effect on surface shortwave radiation under precipitation conditions. Next, at both low and high spatial resolutions, it constructs a model based on underlying surface characteristics such as elevation, sky visibility factor, slope, aspect, and vegetation index, as well as cloud shading data. The input feature vector, composed of layer shading indices, uses surface-downward shortwave radiation data as the output variable. Based on low spatial resolution data, a stacked model integrating multiple machine learning models is constructed to establish a nonlinear mapping relationship between multi-source environmental factors and surface-downward shortwave radiation. Then, the corresponding input features at high spatial resolution are introduced into the stacked model to obtain the initial downscaling result of surface-downward shortwave radiation at high spatial resolution. Finally, an energy conservation constraint is applied to the initial downscaling result for consistency correction to ensure the consistency of the total surface-downward shortwave radiation before and after downscaling. This invention proposes a surface-downward shortwave radiation downscaling method that considers topographic parameters, underlying surface differences, and cloud shading. By introducing an energy conservation constraint, it solves the problem that traditional methods cannot simultaneously address the expression of local spatial heterogeneity and the consistency of regional energy balance, thus improving the accuracy and reliability of shortwave radiation spatial reconstruction in complex terrain areas. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the calculation method for downscaling correction of shortwave radiation from the ground surface, which takes into account differences in the underlying surface and the effects of cloud cover, provided by the present invention. Figure 2 This is a high spatial resolution cloud occlusion index distribution map in a specific embodiment; Figure 3 This is a specific embodiment of the spatial distribution of low spatial resolution surface-to-surface shortwave radiation; Figure 4 This is the high spatial resolution surface downward shortwave radiation result output by the stacked model in the specific embodiment; Figure 5 This is the high spatial resolution surface downward shortwave radiation correction result after introducing energy conservation constraints in a specific embodiment. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0022] Example 1 like Figure 1 As shown, this invention provides a method for downscaling shortwave radiation down from the Earth's surface, comprising the following steps: Step S1 involves acquiring low spatial resolution surface downwave radiation data, as well as elevation, sky visibility factor, slope, aspect, vegetation index, and precipitation data in both low and high spatial resolutions. Specific steps include: Step S101: Download the low spatial resolution downward shortwave radiation data, as well as the low and high spatial resolution elevation, vegetation index and precipitation data. Step S102: Based on the elevation data, the slope and aspect are calculated using the central difference method of the first-order gradient of the grid, and the horizon occlusion angle in each direction is determined by multi-directional ray tracing. Discrete integration is performed on the sky hemisphere to obtain the sky visibility factor.

[0023] Hourly surface shortwave radiation and precipitation data for the Tunxi watershed, with a spatial resolution of 0.1° (approximately 11 km), for 2023-2024 were downloaded from the ERA5-Land reanalysis dataset. The spatial resolution of the precipitation data was then upscaled to 1 km using bilinear interpolation in ArcGIS software. Using the Google Earth Engine (GEE) platform, NASADEM elevation data and MODIS vegetation index data (including NDVI and EVI) at both 1 km and 1 km spatial resolutions for the study area for 2023-2024 were downloaded. Based on the elevation data, slope, aspect, and sky visibility factor data at both 11 km and 1 km spatial resolutions were calculated.

[0024] Step S2 involves constructing a cloud shading index based on precipitation data to quantitatively characterize the shading effect of clouds on downward shortwave radiation from the ground under precipitation conditions. Specific steps include: Step S201: Based on precipitation data, construct a set of basis functions to characterize the cloud cover effect, and perform forward calculations on each basis function to obtain its response value. Construct a mapping function based on the basis function response values. It takes the following form: (1) (2) (3) (4) In the formula, This is a mapping function used to quantitatively characterize the cloud cover effect; and Mesh at low and high spatial resolutions, respectively. and During the period Rainfall (mm); This is the baseline shading constant under no-rain or light-rain conditions, set based on regional climate background or historical statistics; The low-intensity precipitation scaling factor (mm) is used to adjust the linear growth rate of the shading index as precipitation increases during the light rain phase. It is set according to the regional climate background or historical statistics. is the saturation scale constant (mm), used to control the saturation rate of the shading index as precipitation increases during the moderate rain stage; This is the saturation bias term, which is a very small positive number. It is set according to the regional climate background or historical statistics, and is generally taken as 0.05. This represents the cloud cover index value corresponding to the upper limit of light rain intensity. (Grid) For a general grid index format, use a grid. The formulas expressed are applicable to both low and high spatial resolutions.

[0025] Step S202 yields the cloud cover index, expressed as follows: (5) (6) In the formula, and Mesh at low and high spatial resolutions, respectively. and During the period The cloud cover index.

[0026] In this embodiment, the following settings are provided: , , Based on the above parameter settings, a cloud cover index mapping function applicable to various resolutions in the Tunxi River Basin is obtained. This function maps low- and high-spatial-resolution precipitation data in the Tunxi River Basin to corresponding low- and high-spatial-resolution cloud cover indices. Taking 12:00 on August 21, 2024 as an example, the distribution of high-spatial-resolution cloud cover indices during this period is as follows: Figure 2 As shown.

[0027] Step S3: At both low and high spatial resolutions, construct an input feature vector consisting of underlying surface features such as elevation, sky visibility factor, slope, aspect, and vegetation index, as well as cloud cover index. Use surface downwave radiation data as the output variable. Based on the low spatial resolution data, construct a stacked model integrating multiple machine learning models to establish a nonlinear mapping relationship between multi-source environmental factors and surface downwave radiation. Step S301, under both low and high resolutions, the input feature vector constructed using underlying surface features such as elevation, sky visibility factor, slope, aspect, and vegetation index, as well as cloud cover indicators, is represented as follows: (7) (8) In the formula, This indicates that data is concatenated along the feature dimension; and Mesh at low and high spatial resolutions, respectively. and During the period The input feature vector; and Mesh at low and high spatial resolutions, respectively. and During the period The underlying surface features are composed of elevation, sky visibility factor, slope, aspect, and vegetation index.

[0028] Step S302, the fusion form of the outputs of each base learning model is obtained by using ridge regression: (9) In the formula, This is an ensemble function used to fuse the outputs of various base learning models; For the first Individual base learning model; The number of base learning models; For the first Weight coefficients of a basic learning model; For bias terms; parameters By minimizing the following objective function with a penalty term What was obtained: (10) In the formula, Mesh at low spatial resolution During the period The surface downward shortwave radiation value (W / m 2 ); Mesh at low spatial resolution During the period The input feature vector; The number of time steps for the training samples; This refers to the number of grid cells at low spatial resolution. The regularization parameter is used to control the severity of the penalty. When this happens, the objective function degenerates into ordinary least squares regression.

[0029] Step S303: Define the stacked model and construct the stacked model structure for subsequent model training; (11) In the formula, A stacked model built on the integration of multiple machine learning models.

[0030] Step S304: Train the constructed stacked model above at low spatial resolution by minimizing the following loss function: (12) In the formula, Use the loss function to train the model.

[0031] In this embodiment, two types of input feature vectors with low and high resolutions were constructed using underlying surface features such as elevation, sky visibility factor, slope, aspect, and vegetation index, as well as cloud cover index in the Tunxi River Basin. Three machine learning models, namely Random Forest, XGBoost, and LightGBM, were used as base learning models to obtain a trained stacked model for subsequent simulations.

[0032] Step S4: Introduce the input features corresponding to the high spatial resolution into the stacked model to obtain the initial downscaling results of the downward shortwave radiation from the Earth's surface at high spatial resolution.

[0033] Step S401, mesh at high spatial resolution according to formula (5) Construction period Input feature vector .

[0034] Step S402, the stacked model trained in step S3 Acting on high spatial resolution input feature vectors We obtained the initial downscaling results for the downward shortwave radiation from the Earth's surface with high spatial resolution: (13) In the formula, A stacked model built upon the integration of multiple machine learning models; For time period Mesh at high spatial resolution Initial surface downward shortwave radiation (W / m 2 ); Mesh at high spatial resolution During the period The input feature vector.

[0035] Taking 12:00 on August 21, 2024 as an example, the spatial distribution of low-resolution surface-downward shortwave radiation during this period is as follows: Figure 3As shown; the high spatial resolution input feature vector of the Tunxi River Basin is introduced into the stacked model trained by S3 to output high spatial resolution surface downward shortwave radiation results. Taking 12:00 on August 21, 2024 as an example, the high spatial resolution surface downward shortwave radiation results for this period are as follows. Figure 4 As shown.

[0036] Step S5: Apply energy conservation constraints to the initial downscaling results for consistency correction to ensure the consistency of the total downward shortwave radiation from the Earth's surface before and after downscaling.

[0037] Step S501: Create a low-resolution mesh With the set of high spatial resolution subgrids within the coverage area of ​​this grid The correspondence, where This represents the number of grid cells within the set at high spatial resolution.

[0038] Step S502: For each high spatial resolution grid in each time period, calculate the energy conservation correction coefficient. : (14) In the formula, Mesh at low spatial resolution During the period Energy conservation correction coefficient; For time period Inner low-resolution mesh The surface downward shortwave radiation value (W / m 2 The denominator represents the spatial average of the initial downscaling result with high spatial resolution within the range corresponding to the low-resolution grid.

[0039] Step S503: Apply the correction coefficients to the high spatial resolution subgrids covered by the corresponding low-resolution grid to obtain the corrected high spatial resolution downlink shortwave radiation: (15) In the formula, For time period Mesh at high spatial resolution Corrected surface downwave radiation (W / m) 2 ); For time period Mesh at high spatial resolution Initial surface downward shortwave radiation (W / m 2 ); For low-resolution grids The corresponding high spatial resolution subgrid set. The energy conservation correction coefficient is used to adjust the spatial average of the initial downscaling result of the high spatial resolution, making it equal to the downward shortwave radiation value of the corresponding low resolution grid, thereby maintaining the total radiation unchanged while spatially refining.

[0040] A low-resolution grid was established based on the initial downscaling results of the Tunxi River Basin. High spatial resolution grid within the coverage area of ​​this grid The correspondence is established, using the radiance value of the low spatial resolution grid as a benchmark, and applying energy conservation correction to the radiance value of the high spatial resolution grid covered by the low spatial resolution grid to obtain the high spatial resolution surface downward shortwave radiation correction result after introducing energy conservation constraints. Taking 12:00 on August 21, 2024 as an example, the high spatial resolution surface downward shortwave radiation correction result for this period is as follows: Figure 5 As shown in the figure, the corrected high spatial resolution surface-downward shortwave radiation range is 49.34–187.46 W / m². 2 Compared to the uncorrected initial results (78.53–224.39 W / m²), 2 The corrected data, in terms of magnitude range, are similar to the original low-resolution radiance values ​​before downscaling (64.99–147.72 W / m²). 2 This is a better fit. This indicates that the correction method can more effectively preserve the macroscopic statistical characteristics of the original low-resolution data.

[0041] Flux tower observation data from the Tunxi River Basin study area were selected as the validation benchmark, covering the period from 2023 to 2024. The initial downscaling results of high-resolution surface-to-surface shortwave radiation output by this invention, along with the results after energy conservation consistency correction, were extracted into simulated sequences at the flux tower locations, and their accuracy was evaluated using synchronous observation sequences from the flux tower. Mean absolute error was used as the benchmark. and root mean square error As an evaluation indicator, its formula is: (16) (17) In the formula, The total number of time steps involved in the evaluation; For the flux tower at time Observations of downward shortwave radiation from the Earth's surface; This is the estimated downward shortwave radiation value output at the corresponding location of the flux tower according to the present invention. The mean absolute error and root mean square error corresponding to the preliminary downscaling results are denoted as follows: and The mean absolute error and root mean square error corresponding to the results after energy conservation consistency correction are denoted as... and .

[0042] In this embodiment, the average measured downward shortwave radiation at the ground surface of this station is 158.96 W / m². 2 Preliminary downscaling results and 45.34 W / m 2 and 82.03 W / m 2 However, after introducing energy conservation consistency correction and 43.06 W / m 2 and 72.67 W / m 2 The calculation results show that the corrected estimated sequence has a reduced overall deviation. Combined with the aforementioned comparison of magnitude ranges, it can be seen that introducing energy conservation consistency correction can effectively suppress numerical drift while maintaining high-resolution spatial details, making the estimated results more consistent with the measured levels and the original low-resolution background values, thereby significantly improving the accuracy and reliability of high-resolution surface-down shortwave radiation products.

[0043] This embodiment provides a surface-downward shortwave radiation downscaling calculation device, including a processor and a memory; the memory stores a program or instructions, which are loaded and executed by the processor to implement the steps of the surface-downward shortwave radiation downscaling calculation method of the above embodiment.

[0044] This embodiment provides a computer-readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the surface-downscaling calculation method for shortwave radiation described in the above embodiment.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any improvements and optimizations made by those skilled in the art without departing from the concept of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for downscaling shortwave radiation down from the Earth's surface, characterized in that, Includes the following steps: Step S1: Acquire low spatial resolution data and high spatial resolution data. The low spatial resolution data includes low spatial resolution surface-to-surface shortwave radiation data, as well as low spatial resolution elevation, sky visibility factor, slope, aspect, vegetation index, and precipitation data. The high spatial resolution data includes high spatial resolution elevation, sky visibility factor, slope, aspect, vegetation index, and precipitation data. Step S2: Construct a cloud cover index based on the precipitation data to quantitatively characterize the cloud cover effect on downward shortwave radiation from the ground surface under precipitation conditions. Step S3: Construct a stacked model based on the integration of multiple machine learning models, and train the stacked model using low spatial resolution data; the input feature vector for training the stacked model consists of elevation, sky visibility factor, slope, aspect, vegetation index and cloud cover index under low spatial resolution, and the output feature vector for training the stacked model is the shortwave radiation data down the ground surface. Step S4: The high spatial resolution data obtained in step S1 is used as the input feature vector and input into the trained stacked model to obtain the initial high spatial resolution surface downward shortwave radiation value. Step S5: Apply energy conservation constraints to the obtained initial high spatial resolution surface downward shortwave radiation values ​​for consistency correction, so as to ensure the consistency of the total surface downward shortwave radiation before and after downscaling.

2. The method for downscaling shortwave radiation down from the Earth's surface according to claim 1, characterized in that, In step S5, the high spatial resolution downward shortwave radiation from the Earth's surface after consistency correction is: In the formula, Mesh at high spatial resolution During the period The corrected downward shortwave radiation value of the Earth's surface; Mesh at high spatial resolution During the period The initial high spatial resolution downward shortwave radiation values ​​of the Earth's surface; Mesh at low spatial resolution During the period Energy conservation correction coefficient; Mesh at low spatial resolution The corresponding set of high spatial resolution subgrids.

3. The method for downscaling shortwave radiation down from the Earth's surface according to claim 2, characterized in that, The energy conservation correction coefficient is determined according to the following formula: In the formula, Mesh at low spatial resolution During the period The value of shortwave radiation downwards from the Earth's surface; For the low spatial resolution mesh The number of grids at high spatial resolution within the coverage area; the denominator represents the spatial average of the initial downscaling results at high spatial resolution within the area corresponding to the low spatial resolution grids.

4. The method for downscaling shortwave radiation down from the Earth's surface according to claim 3, characterized in that, In step S4, the initial high spatial resolution surface downward shortwave radiation value is output by the stacked model, and its expression is: In the formula, For time period Mesh at high spatial resolution High spatial resolution downward shortwave radiation values ​​of the Earth's surface; A stacked model built upon the integration of multiple machine learning models; Mesh at high spatial resolution During the period The input feature vector; The stacking model is represented as follows: In the formula, This is an ensemble function used to fuse the outputs of various base learning models; For integration functions Parameters; For the first Individual base learning model; X represents the number of base learning models; X is the input feature vector of the stacked model.

5. The method for downscaling shortwave radiation down from the Earth's surface according to claim 4, characterized in that, In step S3, the stacked model is trained at low spatial resolution by minimizing the following loss function: In the formula, Train the loss function for the model; The number of time steps for the training samples; This refers to the number of grid cells at low spatial resolution. Mesh at low spatial resolution During the period The input feature vector.

6. The method for downscaling shortwave radiation down from the Earth's surface according to claim 4, characterized in that, An ensemble function used to fuse the outputs of various base learning models. for: In the formula, For the first Weight coefficients of a basic learning model; For bias terms; parameters By minimizing the following objective function with a penalty term What was obtained: In the formula, Here, is the regularization parameter used to control the severity of the penalty; when When this happens, the objective function degenerates into ordinary least squares regression.

7. The method for downscaling shortwave radiation down from the Earth's surface according to claim 4, characterized in that, In step S4, the input feature vector is represented as: In the formula, This indicates that data is concatenated along the feature dimension; and Mesh at low spatial resolution During the period Input feature vectors and grids at high spatial resolution During the period The input feature vector; and Mesh at low spatial resolution During the period Underlying surface features and high spatial resolution mesh During the period The underlying surface features are composed of elevation, sky visibility factor, slope, aspect, and vegetation index. and Mesh at low spatial resolution During the period Cloud cover index and grid at high spatial resolution During the period The cloud cover index.

8. The method for downscaling shortwave radiation down from the Earth's surface according to claim 7, characterized in that, Step S2, the cloud cover index is constructed from precipitation data, and it is expressed as: In the formula, This is a mapping function used to quantitatively characterize the cloud cover effect; and Mesh at low spatial resolution During the period Precipitation and grid at high spatial resolution During the period The amount of precipitation; In the formula, This is the baseline shading constant under no-rain or light-rain conditions, set based on regional climate background or historical statistics; Low-intensity precipitation scaling factor, used to adjust the linear growth rate of the shading index as precipitation increases during the light rain phase; It is a saturation scale constant used to control the saturation rate of the shading index as precipitation increases during the moderate rain stage; This is the saturation bias term; This represents the cloud cover index value corresponding to the upper limit of light rain intensity.

9. A device for downscaling shortwave radiation from the Earth's surface, characterized in that, It includes a processor and a memory; the memory stores a program or instructions which are loaded and executed by the processor to implement the steps of the method for downscaling shortwave radiation downscaling as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, It stores a program or instruction that, when executed by a processor, implements the steps of the method for calculating the downscaling of shortwave radiation down the surface as described in any one of claims 1 to 8.

Images