A method, system, device, medium and terminal for improving accuracy of a precipitation product

Abstract

The application discloses a kind of precipitation product precision promotion method, system, equipment, medium and terminal.Belong to precipitation product precision technical field, obtain and analyze the characteristic statistical parameters of long sequence monitoring data and short sequence monitoring data, identify the precision of precipitation product, integrate several resampling methods to extract and identify the precipitation product with highest precision in specific area, integrate the deep excavation mechanism of basic specific target point precipitation fusion, the deep excavation mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening and discrimination method coupling and optimization, excavate the model parameter of specific target point, establish the spatial transmission model of relevant parameter by geostatistical method and spatial information regression mode, the purpose of the application is to construct the technical body system that can significantly improve the precision of current international and domestic mainstream high space-time resolution precipitation product by deep excavation multi-source precipitation product and the model conversion and parameter spatial transmission relationship of measured rainfall.

Description

Technical Field

[0001] This invention belongs to the field of precipitation product accuracy technology, specifically relating to methods, systems, equipment, media and terminals for improving the accuracy of precipitation products. Background Technology

[0002] Precipitation is a manifestation of a complex process involving many interactions, exhibiting a high degree of spatiotemporal uncertainty. Against the backdrop of human-driven climate change, this uncertainty is increasing, posing new challenges to sustainable water resource utilization, flood and drought prevention, and emergency management.

[0003] In vast areas of western my country, due to the influence of harsh weather and topographical conditions, hydrological and meteorological monitoring and forecasting are relatively scarce. Many engineering projects rely on data from nearby meteorological stations for design rainstorms or design floods because there is no data available for monitoring. Runoff characteristics are then obtained through methods such as runoff generation and confluence characteristics of similar watersheds or runoff coefficient contour lines. This results in significant deviations in the calculated rainstorms, design rainstorms, and design floods.

[0004] For regions with scarce data, obtaining longer precipitation series is primarily achieved by fusing a range of satellite products, ground-based observation data, and meteorological model reanalysis products, such as TRMM, MSWEP, and CMFD. These products provide high spatiotemporal resolution precipitation series spanning from 1979 to the present, with a spatial resolution of 0.1° and a temporal resolution of up to 3 hours. Given that monitoring stations in these regions were established relatively late, mostly around 2010, and lacked relevant monitoring data in the early stages, the resulting data is significantly longer and has higher spatiotemporal resolution, providing crucial support for related research and analysis.

[0005] A series of precipitation products obtained by coupling satellite products, ground observation data, and meteorological model reanalysis products suffer from significant accuracy issues. While they exhibit high accuracy on a monthly scale, their accuracy is poor on daily and hourly scales. For example, the correlation coefficients of monthly precipitation data from mainstream international and domestic precipitation products such as TRMM, MSWEP, and CMFD in the typical karst region of Guizhou Province average 0.88, 0.91, and 0.86, respectively; on a decadal scale, they are 0.67, 0.64, and 0.72; and on a daily scale, they are 0.61, 0.67, and 0.56. The low accuracy of daily-scale precipitation data hinders the effective implementation of related research. This limits the assessment of the impact of reservoir construction on local precipitation to annual, seasonal, and monthly levels, rather than accurately evaluating daily and hourly precipitation, heavy rainfall, and the potential disaster risks they may induce. Summary of the Invention

[0006] The purpose of this invention is to identify and optimize multi-source precipitation products, deeply explore the multi-scale uncertainty patterns and information mechanisms of precipitation, and couple geostatistical models of key feature parameters spatial transmission to form a technical system that can significantly improve the accuracy of current mainstream high spatiotemporal resolution precipitation products both domestically and internationally.

[0007] A method for improving the accuracy of precipitation products involves acquiring and analyzing the characteristic statistical parameters of long-series and short-series monitoring data to identify the accuracy of precipitation products. Several resampling methods are integrated to extract the precipitation product with the highest accuracy for a specific region. A deep mining mechanism for fusion of precipitation data from specific target points is integrated. This deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods. The method also deeply mines model parameters for specific target points and establishes a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression.

[0008] A precipitation product accuracy improvement system includes a data source identification module, which is used to analyze the characteristic statistical parameters of long-sequence ground monitoring data and short-sequence ground monitoring data with high spatiotemporal resolution, and extract the precipitation product with the highest accuracy in a specific area through a resampling method to form a new data source;

[0009] The target point-driven mechanism module is used to integrate a deep mining mechanism for precipitation fusion of specific target points. The deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods.

[0010] The spatial transfer module is used to establish a spatial transfer model of relevant parameters by using geostatistical methods and spatial information regression to obtain model parameters obtained after in-depth mining of specific target points.

[0011] In a further preferred embodiment of the present invention, the resampling method includes, but is not limited to, nearest neighbor pixel, bilinear interpolation, cubic convolution, and spline interpolation.

[0012] In a further preferred embodiment of the present invention, the deep mining mechanism includes, but is not limited to, optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods.

[0013] In a further preferred embodiment of the present invention, the geostatistical method includes, but is not limited to, Kriging interpolation and inverse distance weighted interpolation.

[0014] Working principle: (1) Determine the target point or region. Based on the different needs of the user, determine the target location (represented in latitude and longitude) or target region (represented in latitude and longitude of the region boundary) for acquiring long-sequence, high-resolution precipitation data. Its input form is a two-dimensional array. The first column represents the precision, the second column represents the latitude, and each row represents the location of the target location or the location of the control point of the target region boundary.

[0015] (2) Input the collected precipitation data of the target location or surrounding area. The precipitation data input by the user includes two parts: the location information of the characteristic station and the data. The former includes longitude, latitude and elevation, and the latter includes the year, month, day and hour information of historical data and the precipitation of the corresponding period. The format refers to the processing method of the meteorological bureau. The non-precipitated data is assigned the value of 0, forming a 2D array with 5 columns;

[0016] (3) Input data judgment: Based on the location and time period information of the input data, combined with the ground observation data integrated by the system, analyze whether to add the data and recalibrate the parameters, and perform spatial interpolation based on the integrated results. If the relevant data of the target point is already in the system's basic database, proceed directly to step (6); otherwise, proceed to steps (4)-(5).

[0017] (4) Acquisition of fusion parameters of data from newly added monitoring stations: Based on the data from newly added monitoring stations and the precipitation products identified at their corresponding locations, the functional relationship between data at different resolutions and precipitation products at specific time scales (such as year, season, month, etc.) is determined by model regression to obtain the fusion feature parameters, including parameters of multiple linear regression and linear-nonlinear coupling models.

[0018] (5) Parameter update and spatial interpolation: Select different model parameters and spatial interpolation methods to perform spatial interpolation on the corresponding model parameters, and re-obtain the model parameters for different grid points. The interpolation methods include three types: Kriging interpolation, inverse distance weighted interpolation, and direct coupling of the position information of each feature point and the feature information of the underlying surface;

[0019] (6) Select the resampling method for the driving source of precipitation products. Select the resampling method (nearest neighbor pixel, bilinear interpolation, cubic convolution, spline interpolation) and extract the preferred data source data of the target location as input (including MSWEP and CMFD, or only one of them can be selected).

[0020] (7) Integration of new precipitation products: Select characteristic integration methods (including linear regression, linear-nonlinear regression) and parameter space interpolation methods (including Kriging interpolation, inverse distance weighted interpolation and regression of underlying surface feature information), and the program will automatically process and output precipitation products in the specified format; for example, .txt, .xls(x), .nc, .mat, etc.

[0021] (8) Parameter and error evaluation output: Output the corresponding parameters and error analysis results based on different resampling methods, different fusion models, and different spatial interpolation methods for model parameters.

[0022] A computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, performs the following steps: acquiring and analyzing characteristic statistical parameters of long-sequence monitoring data and short-sequence monitoring data; identifying the accuracy of precipitation products; integrating several resampling methods to extract the precipitation product with the highest accuracy for identifying a specific area; integrating a deep mining mechanism for precipitation fusion at specific target points; the deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods; deep mining of model parameters for specific target points; and establishing a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression.

[0023] A computer-readable medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps: acquiring and analyzing characteristic statistical parameters of long-sequence monitoring data and short-sequence monitoring data; identifying the accuracy of precipitation products; integrating several resampling methods to extract the precipitation product with the highest accuracy in a specific area; integrating a deep mining mechanism for precipitation fusion at specific target points, wherein the deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods; deep mining of model parameters for specific target points; and establishing a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression.

[0024] An information data processing terminal is provided, which is used to implement the system described above.

[0025] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0026] (1) This invention establishes a technology system for fusing long-sequence, high-resolution, and high-precision precipitation products in scarce precipitation data areas by coupling ground observation data, TRMM, MSWEP, CHIRPS and CMFD precipitation products, and deep mining based on resampling optimization identification, multi-scale uncertainty pattern discrimination, fusion of multiple linear regression and linear-nonlinear coupled models, and regression of model parameters by multiple geostatistical methods.

[0027] (2) The accuracy of precipitation products constructed based on this technology system is significantly improved. The correlation coefficient of the fused precipitation products is improved by about 0.1 on the 12-hour scale and by more than 0.2 on the 1-day scale. The median relative deviation is about 20% lower than TRMM and CMFD on the daily scale and below, and is very close to MSWEP. After scale optimization, the relative error of the prediction of the frequency of rainstorms with a rainfall of 50-100 mm / d is reduced from 10%-30% to -5%-5%, and the frequency of rainstorms with a rainfall of 100-200 mm / d is reduced from 10%-40% lower to between -15%-15%. Attached Figure Description

[0028] Figure 1 Multi-scale accuracy evaluation of MSWEP data from different regions of Guizhou Province: correlation coefficient;

[0029] Figure 2 Multi-scale accuracy evaluation of MSWEP data from different regions of Guizhou Province: root mean square error (mm);

[0030] Figure 3 Multi-scale accuracy evaluation of MSWEP data from different regions of Guizhou Province: median relative deviation;

[0031] Figure 4 The selected three typical meteorological stations are considered for optimized scale but with fuzzy probability distribution parameters for precipitation-drought.

[0032] Figure 5 The selected three typical meteorological stations were considered for optimization scale, distinguishing between precipitation and drought but not precipitation order;

[0033] Figure 6 The selection of three typical meteorological stations simultaneously considers the probability distribution-order characteristics of precipitation and the optimization scale.

[0034] Figure 7 It is a comparison of the relative error of the frequency of occurrence of extreme heavy precipitation events (50-100 mm / d rainstorm);

[0035] Figure 8 It is a comparison of the relative error of the frequency of occurrence of extreme heavy precipitation events (100-200 mm / d rainstorm). Detailed Implementation

[0036] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, not all embodiments. All features disclosed in this specification, or steps in all disclosed methods or processes, except for mutually exclusive features and / or steps, can be combined in any way.

[0038] Therefore, the following detailed description of embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0039] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0040] The present invention will now be described in detail.

[0041] Implementation Case 1: A method for improving the accuracy of precipitation products. This method acquires and analyzes the characteristic statistical parameters of long-sequence and short-sequence monitoring data to identify the accuracy of precipitation products. It integrates several resampling methods to extract the precipitation products with the highest accuracy in a specific area. It also integrates a deep mining mechanism for precipitation fusion at specific target points. The deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods. It deeply mines the model parameters of specific target points and establishes a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression.

[0042] Implementation Case 2: A precipitation product accuracy improvement system includes a data source identification module, which is used to analyze the characteristic statistical parameters of long-sequence ground monitoring data and short-sequence ground monitoring data with high spatiotemporal resolution. The system extracts the precipitation product with the highest accuracy in a specific area through resampling methods to form a new data source. The resampling methods include, but are not limited to, nearest neighbor pixel, bilinear interpolation, cubic convolution, and spline interpolation.

[0043] The target point-driven mechanism module is used to integrate a deep mining mechanism for precipitation fusion of specific target points. The deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods.

[0044] The spatial transfer module is used to establish a spatial transfer model of the relevant parameters by means of geostatistical methods and spatial information regression after the model parameters obtained by in-depth mining of specific target points. The geostatistical methods include, but are not limited to, Kriging interpolation and inverse distance weighted interpolation.

[0045] Implementation Case 3: A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, performs the following steps: acquiring and analyzing characteristic statistical parameters of long-sequence monitoring data and short-sequence monitoring data, identifying the accuracy of precipitation products, integrating several resampling methods to extract the precipitation product with the highest accuracy in a specific area, integrating a deep mining mechanism for precipitation fusion of specific target points, wherein the deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods, deep mining of model parameters of specific target points, and establishing a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression.

[0046] Implementation Case 4: A computer-readable medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps: acquiring and analyzing characteristic statistical parameters of long-sequence monitoring data and short-sequence monitoring data; identifying the accuracy of precipitation products; integrating several resampling methods to extract the precipitation product with the highest accuracy in a specific area; integrating a deep mining mechanism for precipitation fusion of specific target points, wherein the deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods; deep mining of model parameters of specific target points; and establishing a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression.

[0047] Implementation Case 5: (1) Test data: The overall level of product accuracy improvement was evaluated using hourly meteorological observation data from 88 meteorological stations in Guizhou Province from 2016 to 2020 (from the meteorological data sharing network). The data years for TRMM, MSWEP, and CHIRPS were 2016-2020, and the data for CMDF were 2016-2018.

[0048] (2) Resampling method

[0049] (a) The nearest neighbor value method is used as follows:

[0050] The distances from a rain gauge station to its four adjacent grid points are d1, d2, d3, and d4, respectively, and the rainfall values ​​corresponding to a certain precipitation product at each grid point are P1, P2, P3, and P4. Then, the rainfall value corresponding to the grid point with the minimum value among d1, d2, d3, and d4 is selected.

[0051] (b) Using bilinear interpolation as follows:

[0052] The coordinates of the rain gauge locations are denoted as (X0, Y0). The coordinates of the four adjacent grid points, starting from the top left and proceeding clockwise, are: (X1, Y1), (X2, Y1), (X2, Y2), (X1, Y2), with values ​​P1, P2, P3, and P4 respectively. Linear interpolation is used to obtain the precipitation amounts P01 and P02 at longitude X1 and X2 with latitude Y0. The calculation formula is as follows:

[0053] (1)

[0054] Then, through linear interpolation, the rainfall value at coordinates (X0, Y0) is obtained:

[0055] (2)

[0056] (c) The formula for cubic convolution is as follows:

[0057] The coordinates of the rain gauge location are represented by (X0, Y0). The 16 grid points of the two adjacent grid layers form a 4*4 matrix, and the precipitation at each grid point is recorded as follows:

[0058] P = [ P 11 , P 12 , P 13 , P 14 P 21 , P 22 , P 23 , P 24 P 31 , P 32 , P 33 , P 34 P 41 , P 42 , P 43 , P 44 ] (3)

[0059] The coordinates of each grid point are recorded as: (X11, Y11), (X12, Y12), ..., (X44, Y44). The precipitation P0 at coordinate (X0, Y0) is:

[0060] (4)

[0061] Where Wx and WY are respectively:

[0062] W X = [ W X 1 , W X 2 , W X 3 , W X 4 ] W Y = [ W Y 1 , W Y 2 , W Y 3 , W Y 4 ] (5)

[0063] (6)

[0064] (7)

[0065] in:

[0066] (8)

[0067] (d) The cubic splines are as follows:

[0068] The coordinates of the rain gauge location are represented by (X0, Y0), and the precipitation at each grid point in the adjacent grid cells is denoted as:

[0069] P = [ P 11 , P 12 ,..., P 1 n P 21 , P 22 ,..., P 2 n ... P m 1 , P m 2 ,..., P mn ] (9)

[0070] The X and Y coordinates corresponding to each grid point are:

[0071] X = [ X 1 , X 2 ,..., X n X 1 , X 2 ,..., X n ... X 1 , X 2 ,..., X n ] (10)

[0072] Y = [ Y 1 , Y 1 ,..., Y n Y 2 , Y 2 ,..., Y 2 ... Y m , Y m ,..., Y m ] (11)

[0073] Combining formulas (13)-(15), and following the idea of ​​bilinear interpolation, cubic splines are used for interpolation along the X and Y directions respectively. The cubic spline function is:

[0074] (12)

[0075] (3) Evaluation indicators:

[0076] The evaluation indicators selected are the correlation coefficient, root mean square error, and median relative deviation.

[0077] (a) Correlation coefficient

[0078] (13)

[0079] In the formula, R represents the correlation coefficient between a precipitation sample from a certain monitoring station and precipitation products from the same period. Let O and S represent their respective standard deviations, and Cov(O,S) represent their covariance. The covariance Cov(O,S) is calculated using unbiased estimators of the sample values:

[0080] (14)

[0081] In the formula, Oi represents the observed precipitation value for each unit time period. Let represent the mean of all observed precipitation samples, and Si represent the precipitation amount of the precipitation product for the corresponding time period. This represents the average precipitation for the precipitation product, and N represents the number of precipitation sample points.

[0082] (b) Root mean square error

[0083] (15)

[0084] In the formula, RMSE represents the root mean square error, and Oi, Si and N have the same meaning as in (2).

[0085] (c) Relative average deviation

[0086] (16)

[0087] In the formula, MRE represents the relative average deviation, and Oi, Si and N have the same meaning as in (2).

[0088] Evaluation results: The accuracy and spatial distribution characteristics of precipitation products were evaluated based on the MSWEP nearest neighbor value resampling results. The average levels of other products are shown in Tables 1-4.

[0089] (1) Correlation coefficient

[0090] The spatial distribution characteristics of the correlation coefficients between MSWEP precipitation products at different scales and ground observations are shown in the figure below. Figure 1 As shown, the correlation coefficients of MSWEP data in different regions of Guizhou Province are evaluated for multi-scale accuracy. The results show that, except for 24h and 1 month, the correlation coefficients of the southeast region are higher, especially the 10-day scale, where the correlation coefficients mostly exceed 0.9. For the 1-day and 1-month scales, the correlation coefficients are relatively evenly distributed, with higher values ​​in the central region.

[0091] (2) Root mean square error

[0092] The spatial distribution characteristics of the root mean square error of MSWEP precipitation products are as follows: Figure 2 As shown.

[0093] like Figure 2 As shown, the root mean square error (RMSE) of MSWEP data in different regions of Guizhou Province exhibits a spatial distribution that differs significantly from the correlation coefficient: the RMSE is larger in the southwest region, while it is smaller in other regions (northern, central, northeastern, etc.). This characteristic is consistent with the spatial distribution of TRMM data, indicating that precipitation in the southwest region has high uncertainty and is difficult to retrieve.

[0094] (3) Median of relative deviation

[0095] The spatial distribution characteristics of the median relative deviation of MSWEP precipitation products are shown in the figure below.

[0096] Its spatial distribution exhibits characteristics similar to the correlation coefficient: the relative deviation is lower in the southeast region, and higher in the western, northern, central, and northeastern regions. This shows good spatial consistency with TRMM data at 10-day and monthly scales, but overall, the continuity of its spatial evolution is not very good.

[0097] Table 1 shows the error analysis between the observation values ​​of typical stations in Guizhou Province and their nearest neighbor unit TRMM data.

[0098]

[0099] Table 2 shows the error analysis between the observation values ​​of typical stations in Guizhou Province and their nearest neighbor unit MSWEP data.

[0100]

[0101] Table 3 shows the error analysis between the observation values ​​of typical stations in Guizhou Province and their nearest neighbor cell Chirps data.

[0102]

[0103] Table 4 shows the error analysis between the observation values ​​of typical stations in Guizhou Province and their nearest neighbor CMFD data.

[0104]

[0105] Deep fusion system and feature comparison

[0106] (1) Single-point depth mining

[0107] (a) Multi-source linear regression

[0108] The high-quality source data for a certain rain gauge's precipitation product obtained through resampling are P1, P2, ..., Pn, and the measured data is Py. Ignoring their interaction terms, its multiple regression model is:

[0109] (17)

[0110] (b) Linear-nonlinear coupling

[0111] Firstly, based on the selected precipitation products, power functions are used for coupling to identify their optimal fractal characteristics:

[0112] (18)

[0113] (19)

[0114] (20)

[0115] (twenty one)

[0116] (2) Parameter space transfer model

[0117] (a) Inverse distance squared interpolation

[0118] Given that the coordinates of the target point to be interpolated are (x, y), and the coordinates of a known information point are (xi, yi) with a precipitation value of Pi, calculate the distances from this point to all other points:

[0119] (twenty two)

[0120] Their respective weights are:

[0121] (twenty three)

[0122] By inverse distance interpolation:

[0123] (twenty four)

[0124] (b) Kriging interpolation

[0125] The semivariance of all paired points in space that are h apart is:

[0126] γ *( h ) = 1 2 N h ∑ i = 1 N h [ z ( x i ) − z ( x i + h )] 2 (25)

[0127] In the formula: γ*(h) is the semivariance value of points separated by a distance h; Z(xi) and Z(xi+h) are the values ​​of two points separated by a distance h, respectively; N(h) is the number of pairs of all points separated by a distance h.

[0128] Kriging interpolation is obtained through calculation using an empirical model and fitting of parameters to a theoretical model. The fitting models used are exponential and hemispherical models.

[0129] Exponential model:

[0130] (26)

[0131] Hemispherical model:

[0132] (27)

[0133] In the formula, γ represents the semivariance, h represents the lag distance, a represents the nugget value, and r represents the range.

[0134] (3) Judgment of fusion results

[0135] In addition to the above evaluation indicators, a multi-scale uncertainty model of fused sequences and measured precipitation sequences is added for discrimination. This indicator is determined by multi-scale sample entropy and the frequency and intensity of extreme precipitation.

[0136] (a) Multi-scale sample entropy discrimination

[0137] A precipitation sequence of length n is represented as: x(i) = {x1, x2, ..., xn}. The dimension of this precipitation sequence is 1. Based on this, a new sequence Xm of dimension m is constructed:

[0138] (28)

[0139] In the formula, Xm(j) is the j-th column of the new sequence, which is a vector of length m, and 0 <j<N-m+1。

[0140] Define D(j, k) as the distance between any two vectors in the new sequence Xm, whose value is represented by the largest absolute difference between the corresponding elements of the two vectors, which can be expressed as:

[0141] (29)

[0142] Calculate the distance between each vector in the sequence Xm and all other vectors to form a new matrix Dm with dimension (N-m+1)×(N-m+1).

[0143] (30)

[0144] The distances between the first vector and all other elements form the first row of Dm, the distances between the second vector and all other elements form the second row of Dm, and so on. Due to the symmetry of the sequence and the fact that the diagonal elements of the vectors are 0, it is also possible to calculate only the upper or lower triangular portion of matrix Dm.

[0145] Based on a certain threshold R = r * STD, filter all elements in the matrix that are less than R and count their number Nm. Then calculate the ratio Bm(r) of Nm to (N-m+1)(Nm). Since the diagonal of the matrix is ​​a vector and its difference from itself is 0, it needs to be excluded. Therefore, the denominator is not (N-m+1)².

[0146] Increase the dimension of the sequence to m+1, and repeat steps (1) to (4) to obtain Bm+1(r). When the length N of the sequence is a large finite value, the sample entropy can be obtained by estimation:

[0147] (31)

[0148] The sequence is scaled coarsened, and the above process is repeated to calculate the sample entropy at different scales, thereby obtaining its multi-scale uncertainty mode.

[0149] (b) Determination of extreme precipitation frequency and intensity

[0150] If the frequency and intensity of an extreme rainstorm of a specific intensity at a certain station are F0 and I0, and the frequency of its occurrence in the simulation or fusion results is Fs and Is, then their relative deviations are as follows:

[0151] (31)

[0152] (31)

[0153] (4) Optimal scale identification

[0154] By selecting monthly, quarterly, and annual scales respectively, and following the above method, the differences in relevant indicators are analyzed, and the most suitable scale is selected for fusion.

[0155] Fusion data accuracy assessment

[0156] a. Overall accuracy level comparison

[0157] The accuracy of the integrated products will be improved according to the new technology and compared with the current typical international and domestic mainstream products. The results are shown in Table 5.

[0158] Table 5. Accuracy of Precipitation Products Based on Multi-Source Data Fusion and Comparison with Mainstream Products

[0159]

[0160] b. Multiscale uncertainty mode discrimination based on different fusion methods

[0161] like Figure 4 , Figure 5 and Figure 6 As shown, three typical meteorological stations were selected to analyze multi-scale uncertainty models of daily precipitation sequences based on different deep fusion mechanisms. The results are as follows. Figure 4 As shown, the results indicate that, considering probability distribution, order characteristics, and the optimal scale, the multi-scale entropy difference of the model precipitation sequence is minimized.

[0162] c. Comparison of the frequency of extreme precipitation

[0163] like Figure 7 and Figure 8 As shown, 312 daily precipitation series from 1951 to 2011 in southern China were selected. The prediction biases of extreme rainfall frequency and flood walls were analyzed by comparing the current mainstream international methods with the newly integrated method. The results show that after scale optimization, the relative error of the predicted frequency of rainfall events with 50-100 mm / d decreased from 10%-30% to -5%-5%, and the predicted frequency of rainfall events with 100-200 mm / d decreased from 10%-40% underestimation to between -15%-15%.

[0164] Although the invention has been described herein with reference to several illustrative embodiments, it should be understood that many other modifications and implementations can be devised by those skilled in the art, which will fall within the scope and spirit of the principles disclosed herein. More specifically, various variations and modifications can be made to the components and / or layout of the subject matter combination within the scope of the disclosure and claims. Besides variations and modifications to the components and / or layout, other uses will be apparent to those skilled in the art.

Claims

1. A method for improving the accuracy of precipitation products, characterized in that, The system acquires and analyzes the characteristic statistical parameters of long-sequence and short-sequence monitoring data, identifies the accuracy of precipitation products, integrates several resampling methods to extract the precipitation products with the highest accuracy in specific areas, and integrates a deep mining mechanism for precipitation fusion of specific target points. The deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods. The system also mines the model parameters of specific target points and establishes a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression. The resampling methods include, but are not limited to, nearest neighbor pixel value, bilinear interpolation, cubic convolution, and spline interpolation; The geostatistical methods mentioned include, but are not limited to, Kriging interpolation and inverse distance weighted interpolation.

2. A precipitation product accuracy improvement system, characterized in that, It includes a data source identification module, which is used to analyze the characteristic statistical parameters of long-sequence ground monitoring data and short-sequence ground monitoring data with high spatiotemporal resolution, and extract the precipitation products with the highest accuracy in identifying specific areas through resampling methods to form a new data source; The target point-driven mechanism module is used to integrate a deep mining mechanism for precipitation fusion of specific target points. The spatial transfer module is used to establish a spatial transfer model of relevant parameters by using geostatistical methods and spatial information regression to obtain model parameters after in-depth mining of specific target points. The deep mining mechanism includes, but is not limited to, optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods.

3. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, it performs the following steps: acquiring and analyzing the characteristic statistical parameters of long-sequence monitoring data and short-sequence monitoring data; identifying the accuracy of precipitation products; integrating several resampling methods to extract the precipitation product with the highest accuracy in a specific area; integrating a deep mining mechanism for precipitation fusion of specific target points; the deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods; deep mining of model parameters of specific target points; and establishing a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression. The resampling methods include, but are not limited to, nearest neighbor pixel value, bilinear interpolation, cubic convolution, and spline interpolation; The geostatistical methods mentioned include, but are not limited to, Kriging interpolation and inverse distance weighted interpolation.

4. A computer-readable storage medium, characterized in that, The system stores a computer program, which, when executed by a processor, causes the processor to perform the following steps: acquire and analyze the characteristic statistical parameters of long-sequence monitoring data and short-sequence monitoring data; identify the accuracy of precipitation products; integrate several resampling methods to extract the precipitation product with the highest accuracy in a specific area; integrate a deep mining mechanism for precipitation fusion of specific target points, wherein the deep mining mechanism includes optimal probability distribution identification, fusion scale optimization, order feature screening, and coupling and optimization of discrimination methods; deeply mine the model parameters of specific target points; and establish a spatial transfer model of relevant parameters through geostatistical methods and spatial information regression. The resampling methods include, but are not limited to, nearest neighbor pixel value, bilinear interpolation, cubic convolution, and spline interpolation; The geostatistical methods mentioned include, but are not limited to, Kriging interpolation and inverse distance weighted interpolation.

5. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the system as described in claim 2.

Images