A downscaling rainfall prediction method and system

Abstract

The present invention proposes a downscaling rainfall prediction method and system, which belongs to the technical field of rainfall prediction. By dividing rainfall levels for each station and taking the station level and station spacing into consideration when solving a geographically weighted regression model of the station, after solving the regression coefficient and residual result of each station, interpolation is performed in a grid at high resolution to obtain the regression coefficient and residual result of each grid at high resolution, and further rainfall data is obtained, thereby realizing the prediction of rainfall data at high resolution. Moreover, since the level and distance between stations are simultaneously considered in the weight function, compared with directly interpolating based on the station rainfall data, the accuracy of rainfall data prediction in a sparse station area is guaranteed. Moreover, through comparison, it can be found that the rainfall data estimated for the target area by using the method of the present invention that simultaneously considers the station spacing and the level between stations is richer in details.

Description

Technical Field

[0001] The present invention relates to the technical field of rainfall prediction, and in particular to a downscaled rainfall prediction method and system. Background Art

[0002] Precipitation is one of the important factors in climate change and plays an important role in atmospheric processes at different spatial scales. Accurate prediction of rainfall is of great significance to agriculture, disaster prevention and mitigation, and other fields. High-resolution rainfall data, in particular, has become the key to solving related work in agriculture and disaster areas.

[0003] The traditional method for obtaining high-resolution rainfall data is direct interpolation based on station data. However, this method is limited by factors such as the density and quality of the acquired station data. Direct interpolation has certain limitations in remote areas such as mountainous areas where stations are sparse. Therefore, accurately predicting rainfall data at high resolution has become a key issue in solving problems related to agriculture and disaster relief. Summary of the Invention

[0004] The purpose of the present invention is to provide a downscaled rainfall prediction method and system, which can achieve accurate prediction of high-resolution rainfall data.

[0005] To achieve the above objectives, the present invention provides a downscaling rainfall prediction method, comprising:

[0006] Based on the historical rainfall data of each station in the target area, each station is classified and the rainfall level of each station is determined.

[0007] Based on the cross-validation method, the window width of the geographically weighted regression algorithm is determined according to the rainfall observations of each station in the target area.

[0008] A weight function between any two stations is determined based on the rainfall level of each station, the window width, and the spatial distance between each station.

[0009] For any site, a geographically weighted regression model of the site is established with the rainfall observation value of the site as the dependent variable and multiple auxiliary variables of the site as independent variables; the geographically weighted regression model includes parameters to be determined; the parameters to be determined include the regression coefficient matrix and residual results of the auxiliary variables; the auxiliary variables include parameters used to characterize the geographical location of the site, the terrain of the site or the vegetation coverage of the site.

[0010] For the geographically weighted regression model of any site, the regression coefficient matrix and residual results in the geographically weighted regression model of the site are obtained based on the weight function between the site and each site within the window width range except the site; the window width range is the area with the site as the center and the window width as the radius.

[0011] According to the regression coefficient matrix and residual results in the geographically weighted regression model of each station, interpolation is performed in the target area to obtain the regression coefficient matrix and residual results of each grid at high resolution.

[0012] According to the regression coefficient matrix and residual results of each grid at high resolution, the rainfall of each grid is predicted to obtain the estimated rainfall value of each grid.

[0013] Optionally, the method of predicting the rainfall of each grid based on the regression coefficient matrix and residual results of each grid at high resolution to obtain the rainfall estimate of each grid specifically includes:

[0014] For any grid at high resolution in the target area, a residual-free estimated value of rainfall in the grid is determined based on the regression coefficient matrix of the grid and the auxiliary variable value of the grid.

[0015] The rainfall estimation value of the grid is obtained according to the no-residual estimation value of the rainfall of the grid and the residual result of the grid.

[0016] Optionally, determining a weight function between any two sites based on the rainfall level of each site, the window width, and the spatial distance between any two sites specifically includes:

[0017] According to the rainfall level of each station, the level weight between any two stations is determined.

[0018] The distance weight between any two sites is determined based on the spatial distance between each site and the window width.

[0019] The weight function between any two sites is determined according to the level weight between any two sites and the distance weight between any two sites.

[0020] Optionally, the distance weight between any two sites is determined according to the following formula:

[0021]

[0022] in, is the distance weight between site i and site j, d ij is the spatial distance between site i and site j, and b is the window width.

[0023] The rank weight between any two sites is determined according to the following formula:

[0024]

[0025] in, is the rank weight between site i and site j, C ij is the grade difference between site i and site j, C max is the maximum value of the level difference.

[0026] The weight function between any two sites is determined according to the following formula:

[0027]

[0028] Among them, W ij is the weight function between site i and site j, is the rank weight between site i and site j, is the distance weight between site i and site j.

[0029] Optionally, the natural fracture method is used to classify each station according to the historical rainfall data of each station in the target area to determine the rainfall level of each station.

[0030] Optionally, ordinary kriging is used to perform interpolation in the target area to obtain the regression coefficient matrix and residual results of each grid at high resolution.

[0031] Optionally, the window width of the geographically weighted regression algorithm is determined according to the following formula:

[0032]

[0033] Among them, b best is the window width of the geographically weighted regression algorithm determined based on the cross-validation method, n is the number of sites, and y i is the rainfall observation value at site i, b is the window width, is the estimated rainfall at stations other than station i when the window width b is given.

[0034] Optionally, the geographically weighted regression model of the site is as follows:

[0035]

[0036] in, is the observed rainfall value at station i, β i0 is the intercept term of site i, X ik is the kth auxiliary variable of site i, β ik is the regression coefficient of the kth auxiliary variable of site i, p is the number of auxiliary variables of the site, εi is the residual result.

[0037] Alternatively, the regression coefficient matrix of site i can be obtained by solving the following formula:

[0038]

[0039] in, is the regression coefficient matrix of site i, Includes the intercept term of site i and the regression coefficients of p auxiliary variables, W i is an n*n diagonal matrix, W i Each non-zero element on the diagonal is the weight function between station i and any other station within the window width. X is the auxiliary variable matrix, which includes the auxiliary variables of each station within the window width. Y is the dependent variable matrix, which includes the rainfall observation values ​​of each station within the window width.

[0040] On the other hand, the present invention further provides a downscaled rainfall prediction system, which, when run by a computer, executes the downscaled rainfall prediction method as described above.

[0041] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0042] The present invention provides a downscaling rainfall prediction method and system, which uses the rainfall observation value of the station as the dependent variable and the parameters used to characterize the geographical location, terrain and / or vegetation coverage of the station as the independent variable. Based on the geographically weighted regression algorithm, a geographically weighted regression model is established to characterize the relationship between each parameter and the rainfall observation value. Further, by dividing the rainfall level of each station, a weight function that integrates the station level and station spacing of each station is taken into account in solving the regression coefficient and residual result in the geographically weighted regression model of the station. Then, according to the regression coefficient and residual result at each station obtained by the solution, the regression coefficient and residual result of each grid are interpolated in a high-resolution grid to obtain the regression coefficient and residual result of each grid. Finally, according to the regression coefficient of each grid, the residual result of each grid and the geographically weighted regression model, the rainfall estimation data of each grid is obtained, thereby realizing the prediction of high-resolution rainfall data in the target area. Since the present invention takes into account the level and station spacing of each station in the process of solving the regression coefficient and residual result in the geographically weighted regression model of the station, compared with traditional algorithms, the accuracy of rainfall data prediction in sparse station areas is guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 A flowchart of a downscaling rainfall prediction method provided in Example 1 of the present invention;

[0045] Figure 2 This is a specific flow chart of step S3 in the method provided in Example 1 of the present invention;

[0046] Figure 3 This is a specific flow chart of step S7 in the method provided in Example 1 of the present invention;

[0047] Figure 4 This is a flowchart of a specific example of the method provided in Example 1 of the present invention;

[0048] Figure 5 A comparison chart of visual results using Fujian Province as an example in the method provided in Example 1 of the present invention;

[0049] Figure 6 A comparison chart of data results using Fujian Province as an example in the method provided in Example 1 of the present invention;

[0050] Figure 7 This is a structural diagram of a downscaled rainfall prediction system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] The purpose of the present invention is to provide a downscaled rainfall prediction method and system, which can achieve accurate prediction of high-resolution rainfall data.

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] Example 1:

[0055] This embodiment provides a downscaling rainfall prediction method, such as Figure 1As shown in the flowchart, the downscaling rainfall prediction method includes the following steps:

[0056] S1. Site classification: Classify each site based on the historical rainfall data of each site in the target area to determine the rainfall level of each site. In this embodiment, the natural fracture method is used to classify each site based on the historical rainfall data of each site in the target area to determine the rainfall level of each site.

[0057] S2. Determine the bandwidth of the geographically weighted regression algorithm: Based on the cross-validation method, the window width of the geographically weighted regression algorithm is determined according to the rainfall observation values ​​of each station in the target area. Specifically, in this embodiment, the window width of the geographically weighted regression algorithm is determined according to the following formula:

[0058]

[0059] Among them, b best is the window width of the geographically weighted regression algorithm determined based on the cross-validation method, n is the number of sites, and y i is the rainfall observation value at site i, b is the window width, is the estimated rainfall at stations other than station i when the window width b is given.

[0060] S3. Determination of weight function between stations: Based on the rainfall level, window width and spatial distance between each station, the weight function between any two stations is determined; e.g. Figure 2 As shown, step S3 specifically includes:

[0061] S31. Determine the level weight between any two sites based on the rainfall level of each site. In this embodiment, the level weight between any two sites is determined according to the following formula:

[0062]

[0063] in, is the rank weight between site i and site j, C ij is the grade difference between site i and site j, C max is the maximum value of the level difference.

[0064] S32. Determine the distance weight between any two sites based on the spatial distance between each site and the window width; determine the distance weight between any two sites based on the following formula:

[0065]

[0066] in, is the distance weight between site i and site j, d ijis the spatial distance between site i and site j, and b is the window width.

[0067] S33. Determine a weight function between any two sites based on the level weight between the two sites and the distance weight between the two sites; determine the weight function between any two sites based on the following formula:

[0068]

[0069] Among them, W ij is the weight function between site i and site j, is the rank weight between site i and site j, is the distance weight between site i and site j.

[0070] S4. Establishing a Geographically Weighted Regression Model: For any station, establish a geographically weighted regression model for the station using the observed rainfall at the station as the dependent variable and multiple auxiliary variables of the station as independent variables. The geographically weighted regression model includes parameters to be determined, including the regression coefficient matrix and residual results of the auxiliary variables. The auxiliary variables include parameters used to characterize the geographic location of the station, the topography of the station, or the vegetation coverage of the station, such as the station's elevation, slope, longitude and latitude, and normalized difference vegetation index.

[0071] Specifically, in this embodiment, the geographically weighted regression model of the site is shown as follows:

[0072]

[0073] in, is the observed rainfall value at station i, β i0 is the intercept term of site i, X ik is the kth auxiliary variable of site i, β ik is the regression coefficient of the kth auxiliary variable of site i, p is the number of auxiliary variables of the site, ε i is the residual result.

[0074] S5. Calculation of the geographically weighted regression model: For the geographically weighted regression model of any site, the regression coefficient matrix and residual results in the geographically weighted regression model of the site are solved based on the weight function between the site and each site within the window width range except the site; the window width range is the area centered on the site and with the window width as the radius.

[0075] The regression coefficient matrix of site i is obtained by solving the following formula:

[0076]

[0077] in, is the regression coefficient matrix of site i, Includes the intercept term of site i and the regression coefficients of p auxiliary variables, W i is an n*n diagonal matrix, W i Each non-zero element on the diagonal is the weight function between station i and any other station within the window width. X is the auxiliary variable matrix, which includes the auxiliary variables of each station within the window width. Y is the dependent variable matrix, which includes the rainfall observation values ​​of each station within the window width.

[0078] S6. Interpolation, downscaling and refinement of data within the target area: Based on the regression coefficient matrix in the geographically weighted regression model of each station and the residual results of each station, interpolation is performed within the target area to obtain the regression coefficient matrix and residual results of each grid at high resolution. In this embodiment, ordinary kriging is used to interpolate within the target area to obtain the regression coefficient matrix and residual results of each grid at high resolution.

[0079] S7. Grid rainfall prediction: Based on the regression coefficient matrix and residual results of each grid at high resolution, the rainfall of each grid is predicted to obtain the estimated rainfall value of each grid; Figure 3 As shown, step S7 specifically includes:

[0080] S71. For any grid in the target area at high resolution, determine a residual-free estimated value of rainfall in the grid based on the regression coefficient matrix and auxiliary variable values ​​of the grid.

[0081] S72. Obtain a rainfall estimation value for the grid according to the rainfall estimation value without residuals for the grid and the residual result of the grid.

[0082] The downscaled rainfall prediction method provided in this embodiment is described below with reference to a specific example. The downscaled rainfall prediction method includes the following steps:

[0083] (1) Site level classification

[0084] According to the rainfall data of meteorological stations, the stations are divided into five levels based on the average annual rainfall value, and the grade value of each station is obtained.

[0085] (2) Determine the independent and dependent variables

[0086] The dependent variable in this regression algorithm is the station rainfall value, and the independent variables are the corresponding station elevation, slope, latitude and longitude, normalized difference vegetation index NDVI and IMERG rainfall reanalysis data.

[0087] (3) Constructing a geographically weighted regression model based on site level

[0088] a. Determine the window width

[0089] The optimal window width was determined using a cross-validation (CV) method. The specific steps of the cross-validation algorithm are as follows: The sites were divided into 10 groups (hereafter referred to as "folds"), with each fold containing 10% of the total sites. To ensure a roughly even distribution of sites within each fold, all sites were first divided into N / 10 categories using K-means clustering. Then, one site was randomly selected from each category without replacement, resulting in the N / 10 sites required for one fold. This selected fold served as the validation set, and the remaining (NN / 10) sites served as the training set. This process was repeated ten times, allowing each fold to serve as the validation set for validation.

[0090] The formula for cross-validation CV is:

[0091]

[0092] Among them, b best is the window width of the geographically weighted regression algorithm determined based on the cross-validation method, n is the number of sites, and y i is the rainfall observation value at site i, b is the window width, is the estimated rainfall at stations other than station i when the window width b is given.

[0093] b. Determine the distance weight function

[0094] This algorithm uses the Gaussian kernel function method to calculate the distance weight. It uses a continuous monotonic decay function to represent the continuous monotonic decrease between weight and distance, which can overcome the disadvantage of the discontinuous spatial weight function. Commonly used distance functions include the distance threshold method and the inverse distance method. In this embodiment, the Gaussian function is used to determine the distance weight. Its function form is as follows:

[0095]

[0096] Where b is the determined bandwidth, d ij is the distance between the jth site and the central site i within the bandwidth, That is the corresponding determined weight.

[0097] c. Determine the grade weight function

[0098] In addition to considering the distance between stations, the rainfall level calculated at each station will also be considered, and the rainfall level of each station will be expressed as follows:

[0099]

[0100] in Represents each level value, and the corresponding site level gap can be expressed as follows:

[0101]

[0102] The greater the grade difference between sites, the greater the difference between the two sites, and the smaller the corresponding grade weight. Therefore, the grade weight function can be expressed as follows:

[0103]

[0104] here, This represents the rank weights of sites i and j when the regression center is i. Based on the defined rank values, the maximum rank difference between sites is 4 (5 minus 1), and the rank weights range from 0 to 1.

[0105] d. Calculate the total weight function

[0106] By combining the distance weight and the level weight, the final weight function is:

[0107]

[0108] In addition to considering the distance factor, the downscaling rainfall prediction method proposed in this embodiment also considers the impact of the differences in the nature of the site itself, that is, the differences in rainfall levels at the site. ij Finally, the above formula is the total weight of distance and level. When the site level difference is 0 (C ij =0), the level weight is 1, and the total weight degenerates into the distance weight in the general formula.

[0109] (4) Performing a geographically weighted regression process based on site level

[0110] Taking any station as the center and window width b as the radius, all stations within the calculation range (window width) are included in the calculation. The closer to the central station, the greater the weight; the smaller the level difference from the central station, the greater the weight. The regression equation of the central station is obtained, including the regression coefficient and residual results.

[0111] (5) End the geographically weighted regression process based on site level

[0112] Repeat step (4) until all stations in the region have been regressed, and obtain the regression coefficient matrix and residual results of the regression model of each station, thus ending the geographically weighted regression process based on the station level.

[0113] (6) Interpolation is performed based on the regression coefficients and residual results of each station; in this embodiment, high resolution refers to a regional grid with a resolution of 1 km*1 km.

[0114] The residual results and regression coefficients are interpolated into a grid with a resolution of 1 km*1 km in the target area.

[0115] (7) Rainfall result estimation

[0116] The rainfall estimate at 1km*1km resolution was calculated based on the interpolated 1km*1km resolution regression coefficient.

[0117] (8) Calculation of fusion downscaling results

[0118] The final rainfall data with a resolution of 1 km*1 km can be obtained by adding the obtained 1 km*1 km resolution rainfall estimate to the 1 km*1 km resolution residual result.

[0119] Combine Figure 4 , where Step 1 mainly describes the process of classifying the stations, corresponding to (1) in the previous article: the classification operation represents the classification of meteorological stations; NNA is the nearest neighbor domain specification algorithm, which assigns the level of the pixels of the raster images at different resolutions in the region according to the distance to the nearest meteorological station. Step 2 describes the entire regression process and the interpolation process after obtaining the residual results and regression coefficients; F represents four independent variables, namely elevation, slope, latitude and longitude, and normalized vegetation index. The image resolution of these four variables is 1*1km; IMERG represents IMERG remote sensing reanalysis rainfall image, with a resolution of (0.1°*0.1°, converted to km, approximately 111.11*111.11km); GUAGE is the meteorological station data, with different colors and different levels; RGWR represents the geographically weighted regression process based on station level. The rightmost figure shows that in the regression process, not only the distance weight is considered, but also the level weight between different stations. Figure 4 The whole process represented is mainly as follows:

[0120] First, the F and IMERG values ​​corresponding to each station are extracted as independent variables; then, the observed values ​​of the meteorological station are used as the true values ​​to perform geographically weighted regression; the weights in the regression process use the weight values ​​of the comprehensive level and distance, and then the residual value results (residuals) and regression coefficient results (Para_F) corresponding to each meteorological station are obtained; the obtained residual results are interpolated to 1km*1km, and Para_F is assigned according to the nearest neighbor domain assignment algorithm; note that the Para_F obtained at this time is a series of independent variable parameter results, that is, each independent variable (elevation, slope, longitude and latitude, etc.) has a 1km*1km image result obtained after the NNA algorithm; EPRE H It refers to the result of multiplying and adding these regression coefficients with the corresponding F and IMERG, that is, the estimated value that does not include the regression residual; MPRE H It refers to the final result after adding the regression residuals.

[0121] Figure 5 The algorithm results of the present invention are shown using Fujian Province as an example. From top to bottom, each horizontal row shows the rainfall results for 2001, 2008, and 2018, respectively. From left to right, they are: the results of the IMERG rainfall reanalysis data itself (0.1°*0.1°), the 1km*1km results obtained by direct interpolation of meteorological station data, the results calculated by the regression method of the present invention, and the results of the geographically weighted regression algorithm using only distance weights without grade weights in the traditional algorithm. Compared with the results obtained by direct interpolation in the second column, the results obtained by geographically weighted regression in the third and fourth columns contain richer details because they include the characteristics of both the station data and the IMERG data. Further comparing the third and fourth columns, we can see that in the black box area of ​​the figure, the details in the third column are richer than those in the fourth column, and are closer to the results in the first and second columns. In other words, the results using grade weights are better than those calculated using only distance weights.

[0122] Figure 6 The superiority of the method provided in this embodiment is illustrated from the data level. From top to bottom, the first row is an analysis of the 1km*1km rainfall results obtained by this patent and the actual values ​​of the station. The second row is an analysis of the results obtained using only the distance weight and the actual values ​​of the station. The third row is an analysis of the original IMERG rainfall reanalysis data and the actual values ​​of the station. The results in each image mainly have four values: RMSE (root mean square error), R2 (correlation coefficient), MAE (mean absolute error), and BIAS (Bayesian bias). It can be seen that the results in the first row are better than those in the second and third rows. That is to say, the 1km*1km rainfall data product obtained by the method of this patent is not only improved in spatial resolution compared to the original rainfall reanalysis data, but also better in data quality than the results obtained by the traditional geographically weighted regression algorithm using only distance weights.

[0123] In this embodiment, each station is divided into rainfall levels, and the station level and station spacing of each station are taken into consideration in solving the geographically weighted regression model of the station. After the regression coefficient and residual result of each station are obtained, interpolation is performed in the grid at high resolution to obtain the regression coefficient and residual result of each grid at high resolution. The regression coefficient and residual result of each grid are further brought into the geographically weighted regression model to obtain the rainfall data of each grid, thereby realizing the prediction of high-resolution rainfall data in the target area.

[0124] Example 2:

[0125] The method of embodiment 1 of the present invention can also be used by Figure 7The downscaled rainfall prediction system architecture is shown in the figure. Figure 7 As shown, the downscaling rainfall prediction system may include a site classification module, a window width determination module, a weight function determination module, a regression model establishment module, and an interpolation downscaling module; some modules may also have subunits for implementing their functions, such as a distance weight determination unit and a level weight determination unit in the weight function determination module; and a regression coefficient matrix interpolation unit and a residual result interpolation unit in the interpolation downscaling module. Of course, Figure 7 The architecture shown is only exemplary. In some implementations, other units may be added to some modules. In addition, when different functions need to be implemented, the modules may be omitted according to actual needs. Figure 7 One or at least two components of the system shown.

[0126] Specific examples are used herein, but the above descriptions are merely illustrative of the principles and implementation methods of the present invention. The descriptions of the above embodiments are intended only to help understand the methods and core concepts of the present invention. Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, thereby storing them in a storage device and executing them by the computing device, or making them into separate integrated circuit modules, or making multiple modules or steps into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0127] At the same time, for those skilled in the art, according to the concept of the present invention, there will be changes in the specific implementation and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A downscaling rainfall prediction method, characterized in that: The downscaling rainfall prediction method comprises: Based on the historical rainfall data of each station in the target area, each station is classified into different levels to determine the rainfall level of each station; Based on the cross-validation method, the window width of the geographically weighted regression algorithm is determined according to the rainfall observation values ​​of each station in the target area; Determining a weight function between any two stations based on the rainfall level of each station, the window width, and the spatial distance between each station; For any station, a geographically weighted regression model of the station is established with the rainfall observation value of the station as the dependent variable and multiple auxiliary variables of the station as independent variables; the geographically weighted regression model includes parameters to be determined; the parameters to be determined include the regression coefficient matrix and residual results of the auxiliary variables; the auxiliary variables include parameters used to characterize the geographical location of the station, the terrain of the station, or the vegetation coverage of the station; For a geographically weighted regression model of any site, a regression coefficient matrix and residual results in the geographically weighted regression model of the site are obtained based on a weight function between the site and each site within a window width range other than the site; the window width range is an area centered on the site and with the window width as a radius; Based on the regression coefficient matrix and residual results of the geographically weighted regression model of each station, interpolation is performed in the target area to obtain the regression coefficient matrix and residual results of each grid at high resolution; According to the regression coefficient matrix and residual results of each grid at high resolution, the rainfall of each grid is predicted to obtain the estimated rainfall value of each grid.

2. The downscaling rainfall prediction method according to claim 1, characterized in that: The method of predicting the rainfall of each grid based on the regression coefficient matrix and residual results of each grid at high resolution to obtain the rainfall estimate of each grid specifically includes: For any grid at high resolution in the target area, determine a residual-free estimated value of rainfall in the grid based on the regression coefficient matrix of the grid and the auxiliary variable value of the grid; The rainfall estimation value of the grid is obtained according to the no-residual estimation value of the rainfall of the grid and the residual result of the grid.

3. The downscaling rainfall prediction method according to claim 1, characterized in that: The weight function between any two sites is determined based on the rainfall level of each site, the window width, and the spatial distance between each site, specifically including: According to the rainfall level of each station, the level weight between any two stations is determined; Determine the distance weight between any two sites based on the spatial distance between each site and the window width; The weight function between any two sites is determined according to the level weight between any two sites and the distance weight between any two sites.

4. The downscaling rainfall prediction method according to claim 3, characterized in that: The distance weight between any two stations is determined according to the following formula: in, is the distance weight between site i and site j, d ij is the spatial distance between site i and site j, b is the window width; The rank weight between any two sites is determined according to the following formula: in, is the rank weight between site i and site j, C ij is the grade difference between site i and site j, C max is the maximum value of the grade difference; The weight function between any two sites is determined according to the following formula: Among them, W ij is the weight function between site i and site j, is the rank weight between site i and site j, is the distance weight between site i and site j.

5. The downscaling rainfall prediction method according to claim 1, wherein: The natural fracture method is used to classify each station according to the historical rainfall data of each station in the target area to determine the rainfall level of each station.

6. The downscaling rainfall prediction method according to claim 1, characterized in that: Ordinary Kriging is used to perform interpolation in the target area to obtain the regression coefficient matrix and residual results of each grid at high resolution.

7. The downscaling rainfall prediction method according to claim 1, characterized in that: The window width of the geographically weighted regression algorithm is determined according to the following formula: Among them, b best is the window width of the geographically weighted regression algorithm determined based on the cross-validation method, n is the number of sites, and y i is the rainfall observation value at site i, b is the window width, is the estimated rainfall at stations other than station i when the window width b is given.

8. The downscaling rainfall prediction method according to claim 1, characterized in that: The geographically weighted regression model for the site is as follows: in, is the observed rainfall value at station i, β i0 is the intercept term of site i, X ik is the kth auxiliary variable of site i, β ik is the regression coefficient of the kth auxiliary variable of site i, p is the number of auxiliary variables of the site, ε i is the residual result.

9. The downscaling rainfall prediction method according to claim 1, characterized in that: The regression coefficient matrix of site i is obtained by solving the following formula: in, is the regression coefficient matrix of site i, Includes the intercept term of site i and the regression coefficients of p auxiliary variables, W i is an n*n diagonal matrix, W i Each non-zero element on the diagonal is the weight function between station i and any other station within the window width. X is the auxiliary variable matrix, which includes the auxiliary variables of each station within the window width. Y is the dependent variable matrix, which includes the rainfall observation values ​​of each station within the window width.

10. A downscaled rainfall prediction system, characterized in that: When the downscaled rainfall prediction system is run by a computer, the downscaled rainfall prediction method according to any one of claims 1 to 9 is executed.

Images