CN-122021346-A - Surface downward short wave radiation downscaling calculation method, device and storage medium

CN122021346ACN 122021346 ACN122021346 ACN 122021346ACN-122021346-A

Abstract

The invention discloses a surface downward short wave radiation downscaling calculation method, a device and a storage medium. The earth surface downward short wave radiation downscaling calculation method comprises the steps of obtaining low spatial resolution data and high spatial resolution data, constructing cloud layer shielding indexes based on precipitation data, constructing a stacking model based on integration of multiple machine learning models, training the stacking model by using the low spatial resolution data, inputting the obtained high spatial resolution data serving as input feature vectors into the trained stacking model, obtaining initial high spatial resolution earth surface downward short wave radiation values of the 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 values to carry out consistency correction. According to the invention, an energy conservation constraint correction mechanism is introduced, so that the spatial resolution, precision and physical consistency of the surface downward short wave radiation data under the complex terrain condition are effectively improved.

Inventors

ZHANG KE
LIU GUOYAN
LIN HAIYANG
LUO YUNING
WANG YUHAO
ZHANG PENGFEI
Gan Xinjuan
WEN XIAOYAN

Assignees

河海大学

Dates

Publication Date: 20260512
Application Date: 20260408

Claims (10)

1. The surface downward short wave radiation downscaling calculation method is characterized by comprising the following steps of: Step S1, low spatial resolution data and high spatial resolution data are obtained, wherein the low spatial resolution data comprise low spatial resolution surface downward short wave radiation data, and elevation, sky visual factors, gradient, slope direction, vegetation indexes and precipitation data under low spatial resolution; s2, constructing cloud layer shielding indexes based on the precipitation data, and quantitatively characterizing the shielding effect of the cloud layer on the surface downward short wave radiation under the precipitation condition; Step S3, constructing a stacking model based on integration of a plurality of machine learning models, and training the stacking model by adopting low spatial resolution data, wherein an input feature vector trained by the stacking model consists of elevation, sky visual factors, gradient, slope direction, vegetation fingers and cloud layer shielding indexes under low spatial resolution, and an output feature vector trained by the stacking model is surface downward short wave radiation data; S4, taking the high-spatial-resolution data obtained in the step S1 as an input feature vector, and inputting the input feature vector into the trained stacking model to obtain an initial high-spatial-resolution surface downward short wave radiation value of the high-spatial-resolution surface downward short wave radiation; and S5, applying energy conservation constraint to the obtained initial high-spatial-resolution surface downward short wave radiation value to perform consistency correction so as to ensure the consistency of the total amount of the surface downward short wave radiation before and after the downscaling.
2. The method according to claim 1, wherein in step S5, the high spatial resolution surface down short wave radiation after the consistency correction is: In the formula, For grids at high spatial resolution In the time period Is used for correcting the surface downward short wave radiation value; For grids at high spatial resolution In the time period Is determined by the initial high spatial resolution surface down-short wave radiation value; For grids at low spatial resolution In the time period Energy conservation correction coefficients of (2); For grids at low spatial resolution A corresponding set of high spatial resolution sub-grids.
3. The method of calculating the downscaling of surface downward short wave radiation according to claim 2, wherein the energy conservation correction factor is determined according to the following equation: In the formula, For grids at low spatial resolution In the time period Short wave radiation values down the earth's surface; for the low spatial resolution lower grid The denominator represents the spatial average value of the initial downscaling result of the high spatial resolution in the corresponding range of the grid of the low spatial resolution.
4. A method according to claim 3, wherein in step S4, the initial high spatial resolution surface down short wave radiation value is output by the stacking model, and the expression is: In the formula, For a period of time Inner high spatial resolution lower grid Is a high spatial resolution surface down short wave radiation value; A stacking model built for integration based on a plurality of machine learning models; For grids at high spatial resolution In the time period Is a feature vector of the input of the (a); the stacking model is expressed as: In the formula, The method comprises the steps of integrating functions for fusing output results of each base learning model; As an integrated function Parameters of (2); is the first A personal learning model; The number of the basic learning models is given, and X is the input feature vector of the stacking model.
5. The method according to claim 4, wherein in step S3, the stacking model is trained at low spatial resolution by minimizing the following loss function: In the formula, Training a loss function for the model; the number of time steps for training samples; Grid number at low spatial resolution; For grids at low spatial resolution In the time period Is used for the input feature vector of (a).
6. The method for downscaling of surface downward short wave radiation according to claim 4, wherein the integrated function for fusing the output results of the respective base learning models The method comprises the following steps: In the formula, Is the first Weight coefficients of the individual basis learning model; is a bias term, parameter By minimizing the objective function with penalty term The obtained product is as follows: In the formula, Is regularization parameter for controlling punishment force when When this objective function is degraded into a common least squares regression.
7. The method according to claim 4, wherein in step S4, the input eigenvector is expressed as: In the formula, Representing stitching the data in the feature dimension; And Respectively grid at low spatial resolution In the time period Input feature vectors and high spatial resolution grid In the time period Is a feature vector of the input of the (a); And Respectively grid at low spatial resolution In the time period Underlying features and high spatial resolution underlying grid In the time period The underlying surface features of (a) are composed of elevation, sky visual factors, gradient and slope direction, namely vegetation indexes; And Respectively grid at low spatial resolution In the time period Cloud shading index of (c) and grid at high spatial resolution In the time period Is a cloud cover index.
8. The method for calculating the downscaling of the surface downward short wave radiation according to claim 7, wherein in the step S2, the cloud cover index is constructed by precipitation data, which is expressed as: In the formula, Mapping functions for quantitatively characterizing cloud shading effects; And Respectively grid at low spatial resolution In the time period Is a grid with precipitation and high spatial resolution In the time period Is a precipitation amount of (2); In the formula, Setting a baseline shielding constant under the condition of no rain or slight rain according to regional climate background or historical statistics; the low-intensity precipitation scaling factor is used for adjusting the linear increasing rate of the shielding index along with the precipitation enhancement in the small rain stage; the saturation scale constant is used for controlling the saturation rate of the shielding index which is enhanced along with precipitation in the middle rain stage; is a saturation bias term; and the cloud layer shielding index value corresponding to the upper limit of the small rain intensity is obtained.
9. A surface downward short wave radiation downscaling device comprising a processor and a memory, wherein the memory has stored therein a program or instructions that are loaded and executed by the processor to implement the steps of the surface downward short wave radiation downscaling method of any one of claims 1 to 8.
10. A computer readable storage medium, having stored thereon a program or instructions which, when executed by a processor, implement the steps of the surface-down short wave radiation downscaling method of any one of claims 1 to 8.

Description

Surface downward short wave radiation downscaling calculation method, device and storage medium Technical Field The invention belongs to the field of ecological hydrology and meteorological data processing, and particularly relates to a surface downward short wave radiation downscaling method integrating multisource environmental factors and energy conservation constraint. Background The surface downward short wave radiation is a key meteorological element in the land energy exchange and ecological hydrologic process, and the spatial distribution characteristics of the surface downward short wave radiation directly influence the calculation accuracy of the evapotranspiration estimation, the ecological process simulation and the land model. Under the combined influence of topography relief, underlying surface differences and atmospheric condition variations, short wave radiation has significant spatial non-uniformities. However, the existing short-wave radiation data mainly originate from ground observation stations, satellite products and analysis data, wherein the number of the ground radiation observation stations is limited, the space distribution is uneven, the space variation characteristics of the short-wave radiation with regional scale are difficult to continuously describe, the analysis and satellite radiation data are limited by space resolution, and the characterization capability of the short-wave radiation fine distribution in a complex terrain area is insufficient. In order to improve the spatial resolution of the short-wave radiation data, a statistical interpolation or empirical downscaling method is generally adopted for carrying out spatial refinement on the low-spatial resolution radiation data in the prior research, but the method is difficult to fully consider the comprehensive influence of the terrain modulation effect and the atmospheric condition on the short-wave radiation, and the consistency of the total radiation quantity is easily destroyed in the downscaling process. With the application of the machine learning method in meteorological data processing, the data driving model can describe the nonlinear relation between the multisource environmental factors and the short wave radiation, but the existing method focuses on modeling of a single model in multiple ways, lacks effective fusion of multimode advantages and necessary physical constraint, and limits the accuracy and reliability of the downscaling result. Disclosure of Invention The invention provides a surface downward short-wave radiation downscaling method based on a multi-source environment factor and a stacked integrated model, which aims to solve the problems that the existing short-wave radiation data has insufficient spatial resolution, is difficult to comprehensively describe the comprehensive influence of local factors such as underlying surface difference and cloud layer shielding and the like, and is easy to damage the consistency of the total radiation amount in the downscaling process. In order to achieve the above purpose, the present invention adopts the following technical scheme: The invention firstly provides a surface downward short wave radiation downscaling calculation method, which comprises the following steps: Step S1, low spatial resolution data and high spatial resolution data are obtained, wherein the low spatial resolution data comprise low spatial resolution surface downward short wave radiation data, and elevation, sky visual factors, gradient, slope direction, vegetation indexes and precipitation data under low spatial resolution; s2, constructing cloud layer shielding indexes based on the precipitation data, and quantitatively characterizing the shielding effect of the cloud layer on the surface downward short wave radiation under the precipitation condition; Step S3, constructing a stacking model based on integration of a plurality of machine learning models, and training the stacking model by adopting low spatial resolution data, wherein an input feature vector trained by the stacking model consists of elevation, sky visual factors, gradient, slope direction, vegetation fingers and cloud layer shielding indexes under low spatial resolution, and an output feature vector trained by the stacking model is surface downward short wave radiation data; S4, taking the high-spatial-resolution data obtained in the step S1 as an input feature vector, and inputting the input feature vector into the trained stacking model to obtain an initial high-spatial-resolution surface downward short wave radiation value of the high-spatial-resolution surface downward short wave radiation; and S5, applying energy conservation constraint to the obtained initial high-spatial-resolution surface downward short wave radiation value to perform consistency correction so as to ensure the consistency of the total amount of the surface downward short wave radiation before and after the downscaling. And (3) reducing the scale