A spatiotemporal precipitation prediction method and electronic device

Abstract

The application discloses a precipitation spatio-temporal prediction method and electronic equipment. The method comprises the following steps: obtaining precipitation observation data, obtaining precipitation spatial prediction factors, and obtaining a prior distribution of a spatio-temporal variation parameter; a generalized linear regression model is established according to the precipitation observation data and the precipitation spatial prediction factors; a Gaussian process model is constructed; a Bayesian hierarchical model is constructed according to the generalized linear regression model, the Gaussian process model and the prior distribution of the spatio-temporal variation parameter, and a posterior probability distribution form of spatial point precipitation under an observation condition is determined; and a spatial point precipitation prediction result is determined according to the posterior probability distribution form. The application can effectively represent a precipitation time dynamic change and reflect a spatial variation process of a precipitation related parameter by integrating the precipitation spatial prediction factors and the spatio-temporal variation parameter which can reflect spatial changes into the generalized linear regression model, and constructing the Bayesian hierarchical model by using precipitation observation data of discrete precipitation observation sites.

Description

Technical Field

[0001] This invention relates to the field of spatiotemporal precipitation prediction, and in particular to a spatiotemporal precipitation prediction method and electronic equipment. Background Technology

[0002] The spatial distribution characteristics of precipitation are a major controlling factor influencing runoff simulation and a range of other hydrological issues. Uncertainty in runoff simulation primarily stems from uncertain precipitation input, while spatial variation in precipitation is the most sensitive factor in flood prediction. However, discrete rain gauge observations are insufficient to fully understand the spatiotemporal distribution characteristics of precipitation. Robust spatial interpolation techniques are needed to utilize known data to acquire distributed precipitation data, supporting regional hydrological and water resource analysis, as well as the efficient utilization and management of water resources. Furthermore, high-resolution spatial precipitation fields predicted through interpolation can help identify localized high-intensity precipitation events that could lead to floods and flash floods, facilitating timely issuance of early warnings and the implementation of emergency measures to mitigate flood risks. In addition, spatialized precipitation information is of great significance for ecological environmental protection and governance.

[0003] Currently, spatial interpolation methods are classified in many ways, which can be broadly categorized into deterministic and geostatistical interpolation methods, each with its own applicable conditions. Among deterministic methods, the most representative are the inverse distance weighting method and the Thiessen polygon method. These models typically employ equalization to account for spatial variations in precipitation, and under certain conditions, they perform well in interpolating precipitation within local areas. However, because they ignore the randomness and spatial correlation of precipitation, they struggle to objectively reflect the uneven spatial distribution of precipitation. Geostatistical interpolation methods, on the other hand, consider not only the spatial correlation and dependence of the data but also its statistical characteristics.

[0004] However, there are many "zero values" in daily and hourly precipitation, making spatial interpolation of hourly and daily precipitation very difficult. Static interpolation methods are hard to reflect the temporal dynamics and dependencies of precipitation. Summary of the Invention

[0005] This application provides a method and electronic device for spatiotemporal prediction of precipitation, which can effectively characterize the dynamic changes in precipitation over time and reflect the spatial variation process of precipitation-related parameters, thereby obtaining reliable precipitation prediction results.

[0006] In a first aspect, embodiments of the present invention provide a method for spatiotemporal prediction of precipitation, comprising:

[0007] Acquire precipitation observation data, obtain spatial forecasting factors for precipitation, and obtain the prior distribution of spatiotemporal variation parameters;

[0008] Based on the precipitation observation data and the precipitation spatial forecast factors, a generalized linear regression model is established;

[0009] Construct a Gaussian process model;

[0010] Based on the generalized linear regression model, the Gaussian process model, and the prior distribution of the spatiotemporal variation parameters, a Bayesian hierarchical model is constructed to determine the posterior probability distribution of precipitation at spatial points under observation conditions.

[0011] Based on the aforementioned posterior probability distribution, the spatial point precipitation prediction results are determined.

[0012] In some embodiments, the method further includes: processing precipitation observation data to obtain processed precipitation observation data, specifically including:

[0013] The precipitation observation data is processed using the Box-Cox precipitation transformation method, and the optimal transformation parameter λ is estimated using the maximum likelihood method to obtain the processed precipitation observation data. The Box-Cox precipitation transformation method includes:

[0014]

[0015] Among them, s i Let Z(s) be the spatial location, t be the time, λ be the transformation parameter, and Z(s) be the transformation parameter. i (t) represents the spatial position s at time t. i The precipitation observation data are t∈[1,T],1≤i≤n, where n is the number of precipitation observation stations.

[0016] In some embodiments, the method further includes: screening for precipitation spatial forecasting factors with significant correlation, specifically including:

[0017] The probability value that the spatial forecast factor of precipitation is significantly correlated with the precipitation observation data is calculated by hypothesis testing.

[0018] Set a preset value and compare the probability value with the preset value:

[0019] If the probability value is greater than the preset value, then the precipitation spatial forecast factor and the precipitation observation data are not significantly correlated.

[0020] If the probability value is less than or equal to the preset value, then the precipitation spatial forecast factor is significantly correlated with the precipitation observation data.

[0021] In some embodiments, obtaining the prior distribution of the spatiotemporal variation parameters includes:

[0022] Initialize the spatiotemporal variation parameters and set the spatiotemporal variation parameters within a preset range, wherein the spatiotemporal variation parameters conform to an inverse gamma distribution;

[0023] By specifying the hyperparameters of the inverse gamma distribution of the spatiotemporal variation parameters, the prior distribution of the spatiotemporal variation parameters is obtained.

[0024] In some embodiments, establishing a generalized linear regression model based on the precipitation observation data and the precipitation spatial forecast factor includes:

[0025]

[0026] Y t O is the transpose of the precipitation observation data matrix. t For an average dynamic process, ε t It is white noise, ε t ~N(0, σ ε 2 ), which follows a value with an expected value of 0 and a value with σ ε 2 Let σ be a normal distribution with standard deviation. ε 2 Let α be the first parameter of the spatiotemporal variation, β be the intercept, and α be the first parameter of the j (s i )=(β j (s1), β j (s2), ..., β j (s n ))' represents the spatial variation parameter of the corresponding covariate forecast factor, X jt This is a diagonal matrix, where the elements on the diagonal represent the spatial forecast factor X of precipitation at a certain spatial location. j The value of η is 1≤i≤n, where n is the number of precipitation observation stations, m is the number of precipitation spatial forecasting factors with significant correlation selected, and η is the number of precipitation spatial forecasting factors selected with significant correlation. t This is a spatially correlated random error term.

[0027] In some embodiments, constructing the Gaussian process model includes:

[0028] The parameter η of the generalized linear regression model t and β j (s i The variation of ) follows a Gaussian process, and the spatial variation process expressed as a Gaussian process is as follows:

[0029]

[0030] η t For spatially correlated random error terms, β j (s i ) represents the spatial variation parameter of the corresponding covariate forecast factor, 1≤i≤n, where n is the number of precipitation observation stations, and GP represents a Gaussian process. The second parameter of spatiotemporal variation. The third parameter of spatiotemporal variation, κ(·), is given by the following formula: Among them, s i and s j Let |s| be the spatial position of any two points in space. i -s j || represents the distance between two spatial locations; parameter Φ is the fourth parameter of spatiotemporal variation, parameter ν is the fifth parameter of spatiotemporal variation. When the distance changes, parameter Φ represents the degree of decay of the correlation, and parameter ν represents the smoothness of the random field; Г(ν) is the standard gamma function, and K is the ν-order Bessel function of the second kind.

[0031] In some embodiments, determining the spatial point precipitation prediction result based on the posterior probability distribution includes:

[0032] The posterior probability distribution is sampled using a Markov chain Monte Carlo (MCMC) sampling method to determine the spatial point precipitation prediction results. The Markov chain Monte Carlo (MCMC) sampling method includes:

[0033] The posterior distribution of precipitation prediction for each spatial location is obtained by sampling, the mean statistic of the posterior distribution is calculated, and the mean statistic is determined as the precipitation prediction result for the spatial point.

[0034] In some embodiments, the method further includes: correcting the spatial point precipitation prediction results using the Kriging method, specifically including:

[0035] Obtain spatial information of the study area where precipitation to be predicted is to be obtained;

[0036] The Kriging method is used to calculate the contribution weight of precipitation observation stations within a first range of the spatial location of the predicted precipitation to the precipitation at that location. The probability of precipitation occurring at the predicted location is determined based on whether precipitation occurs at the observation stations within this first range. The probability value is 1 or 0, where 1 represents precipitation occurring and 0 represents no precipitation occurring. This process yields a correction value for the spatial point precipitation prediction result, resulting in a corrected spatial point precipitation prediction result. The Kriging method includes:

[0037]

[0038] Where, λ i P(S) represents the contribution weight of precipitation observation stations within a first range of the spatial location of the predicted precipitation to the precipitation at the spatial location S0 of the predicted precipitation. i) represents the probability of precipitation occurring at the observation station, taking a value of 0 or 1; z represents the number of observation stations within the first range of the spatial location S0 of the precipitation to be predicted; and w represents the threshold for determining whether precipitation has occurred, 0 ≤ w ≤ 1.

[0039] In some embodiments, the method further includes: verifying the prediction performance of the precipitation spatial prediction model, wherein the precipitation spatial adaptation model includes the generalized linear regression model, the Gaussian process model, the prior distribution of the spatiotemporal variability parameters, and the Bayesian hierarchical model, specifically including:

[0040] Based on the corrected spatial precipitation prediction results, a cross-validation method is used to run the simulation process by removing the precipitation data of any reference observation station and using the observation data of the remaining stations.

[0041] Finally, PMCC is calculated. PMCC reflects the prediction performance of the precipitation spatial prediction model. The smaller the PMCC, the better the prediction performance of the precipitation spatial prediction model. The calculation of PMCC includes:

[0042]

[0043] Where n is the number of observation stations, r is the time series, and T is the time point.

[0044] E(Y l (s i ,t) rep -y l (s i ,t)) 2 For mathematical expectation, Var Y l (s i ,t) rep Let Y be the variance. l (s i ,t) rep Let y be the precipitation prediction result of the i-th precipitation observation station at time t. l (s i ,t) represents the precipitation observation data of the i-th precipitation observation station at time t, and PMCC represents the prediction effect of the precipitation spatial prediction model.

[0045] In a second aspect, embodiments of the present invention also provide a computer device, comprising:

[0046] At least one processor; and,

[0047] A memory that is communicatively connected to at least one processor; wherein,

[0048] The memory stores instructions that can be executed by at least one processor, such that the instructions are executed by at least one processor to enable the at least one processor to perform the method of the first aspect.

[0049] Thirdly, embodiments of the present invention also provide a non-volatile computer-readable storage medium storing computer-executable instructions that, when executed by a processor, cause the processor to perform the method of the first aspect.

[0050] The beneficial effects of this invention are as follows: By providing a method for spatiotemporal prediction of precipitation, the method includes: acquiring precipitation observation data, precipitation spatial forecast factors, and the prior distribution of spatiotemporal variation parameters; establishing a generalized linear regression model based on the precipitation observation data and the precipitation spatial forecast factors; constructing a Gaussian process model; constructing a Bayesian hierarchical model based on the generalized linear regression model, the Gaussian process model, and the prior distribution of the spatiotemporal variation parameters to determine the posterior probability distribution of precipitation at spatial points under observation conditions; and determining the predicted precipitation result at spatial points based on the posterior probability distribution.

[0051] By introducing the precipitation spatial forecasting factor, the generalized linear regression model can reflect the spatial variation process of the study area, thereby improving the prediction accuracy, effectively characterizing the dynamic changes in precipitation time, reflecting the spatial variation process of precipitation-related parameters, and obtaining reliable precipitation prediction results. Attached Figure Description

[0052] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0053] Figure 1 This is a schematic diagram illustrating an application scenario of a spatiotemporal precipitation prediction method provided in an embodiment of the present invention.

[0054] Figure 2 This is a flowchart illustrating a spatiotemporal precipitation prediction method provided in an embodiment of the present invention;

[0055] Figure 3 yes Figure 2 A detailed flowchart of step S20 in the process;

[0056] Figure 4 This is a flowchart of a process for processing precipitation observation data provided in an embodiment of the present invention;

[0057] Figure 5 This is a flowchart of a method for screening spatial forecasting factors of precipitation provided in an embodiment of the present invention;

[0058] Figure 6 This is a flowchart of a method for correcting spatial point precipitation prediction results provided in an embodiment of the present invention;

[0059] Figure 7 This is an overall flowchart of a precipitation spatiotemporal prediction method provided by an embodiment of the present invention;

[0060] Figure 8 This is a schematic diagram of the hardware structure of a processing device for a spatiotemporal precipitation prediction method provided in an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] The spatiotemporal precipitation prediction method provided in this invention can be applied to, for example... Figure 1 The application scenarios shown are in Figure 1 The application scenario shown includes a study area 10, which contains multiple precipitation observation stations 11. In this embodiment of the invention, Figure 1 The study area shown can be Beijing, which may have 8 rainfall observation stations.

[0063] The study area 10 is merely illustrative. It refers to a region within a certain geographical area from which precipitation is to be predicted, such as a district, a city, or a geographical area composed of several geographically close cities, similar to the Yangtze River Delta region or North China region. Alternatively, it may refer to a regional concept such as an autonomous region or province. The size and shape of the study area are not limited. The precipitation observation station 11 can be any station or object capable of acquiring local precipitation data. The size of the precipitation observation station 11 is not limited, but it has a certain observation range, and its distribution is arbitrary.

[0064] In this embodiment of the invention, before precipitation prediction, it is necessary to obtain the spatial location information of the study area, including: using the spatial range of the study area as the boundary, setting the spatial resolution, the higher the resolution, the more accurate the obtained spatial distribution; using GIS tools to extract the spatial location of the grid center point corresponding to the spatial resolution, i.e., the geographic coordinates, to obtain the spatial location information, the spatial location information being the grid data of the spatial location, each of the spatial locations having topographic and rainfall information, the grid data including the topographic and rainfall information of each of the spatial locations.

[0065] In the embodiments of the present invention, please refer to Figure 2 , Figure 2 This is a flowchart illustrating a spatiotemporal precipitation prediction method provided in an embodiment of the present invention;

[0066] like Figure 2 As shown, this spatiotemporal precipitation prediction method includes:

[0067] Step S10: Obtain precipitation observation data, obtain precipitation spatial forecast factors, and obtain the prior distribution of spatiotemporal variation parameters;

[0068] In this embodiment of the invention, the precipitation observation data can be obtained through precipitation observation stations. It is understood that the precipitation observation data obtained by the precipitation observation stations is only the precipitation observation data within the observation range of the precipitation observation stations.

[0069] In this embodiment of the invention, the precipitation spatial forecasting factor includes covariates. Covariates are variables that affect precipitation and can reflect spatial changes, such as elevation, slope, aspect, distance of the station from the ocean, and grid data of meteorological radar. Covariates can be obtained through various means, such as publicly available remote sensing data and map data.

[0070] In this embodiment of the invention, the spatiotemporal variation parameter includes a first spatiotemporal variation parameter σ. ε 2 The second parameter of spatiotemporal variation, σ β 2 The third parameter of spatiotemporal variation, σ η 2 The fourth parameter Φ and the fifth parameter v of the spatiotemporal variation, and the first parameter σ of the spatiotemporal variation. ε 2 The second parameter of spatiotemporal variation, σ β 2 The third parameter of spatiotemporal variation, σ η 2 Both the fourth parameter Φ and the fifth parameter v of the spatiotemporal variation conform to an inverse gamma distribution. The prior distribution for obtaining the spatiotemporal variation parameters refers to obtaining the first parameter σ of the spatiotemporal variation. ε2 The second parameter of spatiotemporal variation, σ β 2 The third parameter of spatiotemporal variation, σ η 2 The shapes of the first, second, third, fourth, and fifth inverse gamma functions of the fourth parameter Φ and the fifth parameter v of the spatiotemporal variation.

[0071] In this embodiment of the invention, step S10 further includes obtaining the spatial location information of the study area;

[0072] The acquisition of spatial location information of the study area includes:

[0073] Using the spatial extent of the study area as the boundary, a spatial resolution is set. The higher the resolution, the more accurate the spatial distribution. GIS tools are used to extract the spatial location of the raster center point corresponding to the spatial resolution, thus obtaining spatial location information.

[0074] For details, please refer to [link / reference]. Figure 3 , Figure 3 yes Figure 2 A detailed flowchart of step S10 in the process;

[0075] like Figure 3 The step S10, which involves acquiring precipitation observation data, acquiring precipitation spatial forecast factors, and acquiring the prior distribution of spatiotemporal variation parameters, includes:

[0076] Step S111: Initialize the spatiotemporal variation parameters and set the spatiotemporal variation parameters within a preset range, wherein the spatiotemporal variation parameters conform to an inverse gamma distribution;

[0077] Specifically, the spatiotemporal variation parameters include the first spatiotemporal variation parameter σ. ε 2 The second parameter of spatiotemporal variation, σ β 2 The third parameter of spatiotemporal variation, σ η 2 The fourth parameter Φ and the fifth parameter v of the spatiotemporal variation are set in the first range, the second range, the third range, the fourth range, and the fifth range, respectively.

[0078] Specify the first parameter σ of the spatiotemporal variation respectively ε 2 The second parameter of spatiotemporal variation, σ β 2 The third parameter of spatiotemporal variation, σ η 2The first, second, third, fourth, and fifth hyperparameters of the inverse gamma distribution of the fourth parameter Φ and the fifth parameter v of the spatiotemporal variation.

[0079] In this embodiment of the invention, the first parameter, second parameter, third parameter, and fourth parameter of spatiotemporal variation can be respectively set at σ. ε 2 ∈[0.01,0.1], σ β 2 ∈[0.01,0.1], σ η 2 Within the ranges of ∈[0.1,10] and Φ∈[0.001,0.1], the range of values ​​for the fifth parameter v of spatiotemporal variation can be selected empirically.

[0080] Step S112: Specify the hyperparameters of the inverse gamma distribution of the spatiotemporal variation parameters to obtain the prior distribution of the spatiotemporal variation parameters.

[0081] Specifically, the first parameter σ of the spatiotemporal variation is specified respectively. ε 2 The second parameter of spatiotemporal variation, σ β 2 The third parameter of spatiotemporal variation, σ η 2 The first, second, third, fourth, and fifth hyperparameters of the inverse gamma distribution of the fourth parameter Φ and the fifth parameter v of the spatiotemporal variation are obtained, and the first, second, third, fourth, and fifth prior distributions of the first, second, third, fourth, and fifth parameters of the spatiotemporal variation are obtained.

[0082] In this embodiment of the invention, the first, second, third, fourth, and fifth parameters of spatiotemporal variability all conform to an inverse gamma distribution. The shape of the inverse gamma function is determined by hyperparameters. By setting the hyperparameters, different shapes of the inverse gamma function can be obtained. Based on the different shapes of the inverse gamma function, the first, second, third, fourth, and fifth prior distributions of the first, second, third, fourth, and fifth parameters of spatiotemporal variability are determined. The values ​​of the hyperparameters are determined based on experience and can be adjusted according to experimental results or by referring to published papers or reports.

[0083] In this embodiment of the invention, the first parameter of spatiotemporal variation is σ. ε 2The variance of white noise, also known as the nugget variance, reflects the variability and measurement error of variables below the smallest sampling scale. The second parameter of spatiotemporal variability is σ. β 2 For β j (s i The hyperparameters of β j (s i ) represents the spatial variation parameter of the corresponding covariate forecast factor, and σ is the third parameter of spatiotemporal variability. η 2 Let Φ be the site-invariant variance of the spatial random process, Φ be the fourth parameter of the spatiotemporal variation and the spatial smoothing parameter, and v be the fifth parameter of the spatiotemporal variation and the spatial decay parameter.

[0084] By obtaining the prior distribution of the spatiotemporal variation parameters, the present invention can utilize the prior distribution of the spatiotemporal variation parameters to construct a generalized linear regression model and a Gaussian process model.

[0085] After acquiring precipitation observation data, the method further includes processing the precipitation observation data.

[0086] For details, please refer to [link / reference]. Figure 4 , Figure 4 This is a flowchart of a process for processing precipitation observation data provided in an embodiment of the present invention;

[0087] Step S121: Process the precipitation observation data using the Box-Cox transformation method, specifically including:

[0088] The Box-Cox transformation method was used to process precipitation observation data Z(s). i The precipitation distribution exhibits a significant skewness, necessitating data transformation to obtain more reliable simulation results for hourly and daily precipitation. In this embodiment of the invention, the processing method is the Box-Cox transformation method, with the transformation formula as follows:

[0089]

[0090] Among them, s i Let Z(s) be the spatial location, t be the time, λ be the transformation parameter, and Z(s) be the transformation parameter. i (t) represents the spatial position s at time t. i The precipitation observation data is t∈[1,T], where T is the time of the precipitation observation data. For example, if the precipitation observation data is the precipitation observation data of a certain observation station in the study area within the past 20 days, then T is 20, 1≤i≤n, where n is the number of precipitation observation stations. It can be understood that n and i are positive integers.

[0091] Step S122: Estimate the optimal transformation parameter λ using the maximum likelihood method. *The processed precipitation observation data obtained includes:

[0092] Based on existing precipitation observation data samples, find the transformation parameter values ​​that maximize the joint probability density function of the samples. Assuming the samples follow a log-normal distribution, substitute all samples into the probability density function. The product of these probability density functions is a function of λ. Determine the probability density function that maximizes the product. Calculate the partial derivative of this probability density function with respect to λ. A partial derivative of 0 indicates that the probability density function has reached an extreme value, and the corresponding λ is the optimal transformation parameter λ. * Finally, based on the optimal transformation parameter λ * Obtain the processed precipitation observation data.

[0093] Precipitation distribution exhibits a significant skewness. By processing the precipitation observation data using the Box-Cox transformation method, this invention can obtain more reliable simulation results of the precipitation observation data.

[0094] After obtaining the spatial forecasting factors for precipitation, the method further includes: screening the spatial forecasting factors for precipitation;

[0095] For details, please refer to [link / reference]. Figure 5 , Figure 5 This is a flowchart of a method for screening spatial forecasting factors of precipitation provided in an embodiment of the present invention;

[0096] like Figure 5 As shown, the screening of precipitation spatial forecasting factors includes:

[0097] Step S131: Calculate the probability value that the spatial forecast factor for precipitation is significantly correlated with the observed precipitation data using hypothesis testing, specifically including:

[0098] Hypothesis testing is performed between the spatial forecasting factors of precipitation and the observed precipitation data to obtain the P-value. The P-value is the probability that the spatial forecasting factors of precipitation and the observed precipitation data are significantly correlated. The P-value represents the statistical difference between the covariate and the processed observed precipitation data. The smaller the P-value, the more significant the statistical difference between the covariate and the processed observed precipitation data, and the more accurate the spatial precipitation prediction results obtained from subsequent processing.

[0099] Step S132: Set a preset value, which is used to compare the magnitude with the probability value. In this embodiment of the invention, the preset value can be 0.05.

[0100] Step S133: Compare the preset value with the probability value. The comparison between the preset value and the probability value can determine whether the precipitation spatial forecast factor and the precipitation observation data are significantly correlated.

[0101] Step S134: If the probability value is greater than the preset value, then the precipitation spatial forecast factor and the precipitation observation data are not significantly correlated.

[0102] Step S135: If the probability value is less than or equal to the preset value, then the precipitation spatial forecast factor is significantly correlated with the precipitation observation data.

[0103] By screening precipitation spatial forecasting factors, this invention can obtain precipitation spatial forecasting factors that are significantly correlated with the precipitation observation data, and the significantly correlated precipitation spatial forecasting factors have a significant impact on precipitation in the study area.

[0104] Step S20: Based on the precipitation observation data and the precipitation spatial forecasting factors, establish a generalized linear regression model;

[0105] The generalized linear regression model is as follows:

[0106]

[0107] Y t O is the transpose of the precipitation observation data matrix. t For an average dynamic process, ε t It is white noise, ε t ~N(0, σ ε 2 ), which follows a value with an expected value of 0 and a value with σ ε 2 Let σ be a normal distribution with standard deviation. ε 2 Let α be the first parameter of the spatiotemporal variation, β be the intercept, and α be the first parameter of the j (s i )=(β j (s1), β j (s2), ..., β j (s n ))′ represents the spatial variation parameter of the corresponding covariate forecast factor, X jt This is a diagonal matrix, where the elements on the diagonal represent the spatial forecast factor X of precipitation at a certain spatial location. j The value of η is 1≤i≤n, where n is the number of precipitation observation stations, m is the number of precipitation spatial forecasting factors with significant correlation selected, and η is the number of precipitation spatial forecasting factors selected with significant correlation. t This is a spatially correlated random error term.

[0108] in, O t =(O(s1,t),O(s2,t),…,O(s n ,t))',η t=(n(s1,t),n(s2,t),...,n(s) n ,t))' is the spatially correlated random error term that follows a Gaussian distribution, i.e., η(s,t)~N(0,∑ η ), which follows a value with an expected value of 0 and a value with respect to ∑ η Let be a normal distribution with standard deviation, where ∑ η =σ η 2 S, σ η 2 The third parameter of spatiotemporal variability is S=κ(Φ,ν), which represents the coefficient of spatial correlation and conforms to the Matern covariance model.

[0109] The formula for κ(·) is as follows:

[0110]

[0111] Among them, s i and s j Let |s| be the spatial position of any two points in space. i -s j || represents the distance between two spatial locations; parameter Φ is the fourth parameter of spatiotemporal variation, parameter ν is the fifth parameter of spatiotemporal variation. When the distance changes, parameter Φ represents the degree of decay of the correlation, and parameter ν represents the smoothness of the random field; Γ(ν) is the standard gamma function, and K is the ν-order Bessel function of the second kind.

[0112] By incorporating the aforementioned significantly correlated spatial forecasting factors of precipitation into a generalized linear regression model, the generalized linear regression model established in this invention can reflect the spatiotemporal distribution characteristics of precipitation.

[0113] Step S30: Construct a Gaussian process model;

[0114] Assume the spatial variation parameter β of the covariate j (s i Spatial-related random error term η t If it follows a Gaussian process, then we can construct the Gaussian process model as follows:

[0115]

[0116] η t For spatially correlated random error terms, β j (s i ) represents the spatial variation parameter of the corresponding covariate forecast factor, 1≤i≤n, where n is the number of precipitation observation stations, and GP represents a Gaussian process. The second parameter of spatiotemporal variation. The third parameter of spatiotemporal variation is κ(·), and the formula is the same as that in step S30.

[0117] By constructing a Gaussian process model, this invention can describe the potential Gaussian processes of spatial precipitation and address the problem of high-frequency censoring of precipitation observation data in precipitation time series.

[0118] Step S40: Based on the generalized linear regression model, the Gaussian process model, and the prior distribution of the spatiotemporal variation parameters, construct a Bayesian hierarchical model to determine the posterior probability distribution of precipitation at spatial points under observation conditions.

[0119] Specifically, based on the generalized linear regression model, the Gaussian process model, and the prior distribution of the spatiotemporal variation parameters, the data model is determined as follows: Process model: 0~GP(X) j ,β j (s i ), σ η 2 κ(·)), parametric model: prior distribution of θ:

[0120] Where Y follows a mathematical expectation of O, and Let I be a normal distribution with standard deviation, and let I be the identity matrix. Let O be the first parameter of the spacetime variation, and let O follow a constant with respect to X. jt β j (s i Let σ be the mathematical expectation. η 2 κ(·) is a normal distribution with standard deviation, X jt This is a diagonal matrix, where the elements on the diagonal represent the spatial forecast factor X of precipitation at a certain spatial location. j The value of σ η 2 β is the third parameter of spatiotemporal variation. j (s i )=(β j (s1), β j (s2), ..., β j (s n ))' represents the spatial variation parameter of the corresponding covariate forecast factor, where θ includes the first parameter of spatial-temporal variability σ. ε 2 The second parameter of spatiotemporal variation, σ β 2 The third parameter of spatiotemporal variation, σ η 2 The formulas for the fourth parameter Φ and the fifth parameter v,κ(·) of the spacetime variation are as follows:

[0121]

[0122] Based on the data model, the process model, and the prior distribution of the spatiotemporal variation parameters, a Bayesian hierarchical model is constructed to obtain the posterior probability distribution of precipitation at spatial points under precipitation observation conditions, as follows:

[0123]

[0124] Where N = n*T, n is the number of precipitation observation stations, T is the maximum value of the precipitation observation data at any given time, y is the precipitation observation data, and y* is the dataset of spatial locations outside the stations that have no observations, i.e., the precipitation dataset to be predicted. Additionally, θ includes the first parameter σ of the spatial-temporal variability. ε 2 The second parameter of spatiotemporal variation, σ β 2 The third parameter of spatiotemporal variation, σ η 2 And the fourth parameter Φ of spacetime variation and the fifth parameter v of spacetime variation, X jt This is a diagonal matrix, where the elements on the diagonal represent the spatial forecast factor X of precipitation at a certain spatial location. j The value of β j (s i )=(β j (s1), β j (s2), ..., β j (s n ))' represents the spatial variation parameter of the corresponding covariate forecast factor, 1≤i≤n, where n is the number of precipitation observation stations; π(σ η 2 ), π(σ) ε 2 ), π(σ) β 2 ), π(Φ) and π(v) represent the prior distributions of the spatiotemporal variation parameters, β j (s i The coefficients are regression coefficients, whose prior distributions also follow the inverse gamma function distribution. Assume their hyperparameters are a and b, with corresponding mean and variance a / b and a / b, respectively. 2 By specifying appropriate hyperparameters, the regression coefficient β can be determined. j (s i The distribution of ) is given by S, an n×n vector, representing the covariance matrix among all observation stations, with the i*j-th element being κ(s). i ,s j ;Φ.ν) represents the covariance between observation station i and observation station j. When the distance changes, parameter Φ is the degree of decay of the correlation, while parameter ν is the smoothness of the random field.

[0125] Next, we will obtain the posterior probability distribution of precipitation at spatial points under the precipitation observation conditions:

[0126] For any point s0 in space, the predicted precipitation distribution at time t is: Y(s0,t)~N(O(s0,t),σ ε 2 ), σ ε 2 Let O(s0,t) be the first parameter of the spatiotemporal variation, and its sequence is as follows:

[0127]

[0128] According to the given Y(s0,t)~N(O(s0,t),σ ε 2 The sequence of O(s0,t) and the spatial point precipitation probability distribution under precipitation observation conditions are further obtained as follows:

[0129]

[0130] Where, β j (s0)|θ~N(S 12 S -1 β j (s),σ β 2 (1-S 12 S -1 S 21 )) represents β j (s0) obeys S 12 S -1 β j (s) is the expectation, and σ is the expectation. β 2 (1-S 12 S -1 S 21 The standard deviation of O(s0,t) follows a normal distribution and is related to the parameter θ. Furthermore, the predicted conditional distribution of O(s0,t) is in the form of O(s0,t)|β. j (s0), The mean and variance are calculated as follows:

[0131]

[0132] Among them, S 12 S represents the covariance matrix between the predicted point and all observation stations. 12 S is a 1×n vector. 21 For vector S 12 The transpose of S 12 =S 21 ', where n is the number of observation stations, and the i-th element is κ(s) i,s0;Φ.ν),s i Let s0 be the spatial location of the i-th observation station (1≤i≤n), and s0 be the spatial location of the predicted point.

[0133] By constructing a Bayesian hierarchical model, this invention can obtain the posterior probability distribution of precipitation at spatial points under observation conditions, and the prediction result of precipitation at spatial points can be determined based on the posterior probability distribution.

[0134] Furthermore, the method further includes estimating the posterior distribution of the spatiotemporal variation parameters using the MCMC sampling method based on the prior distribution of the spatiotemporal variation parameters. The MCMC sampling method specifically includes:

[0135] By constructing a Markov chain and performing a random walk based on the Markov chain, a sequence of samples is generated, and samples with a stationary distribution are obtained. Then, the stationary distribution samples are used for approximate numerical calculations. The MCMC sampling method requires multiple iterations before the sequence approaches the target distribution, and a warm-up period needs to be set. In this embodiment of the invention, the warm-up period can be set to 4000 times, and the maximum number of iterations can be set to 10000 times.

[0136] Step S50: Based on the aforementioned posterior probability distribution, determine the spatial point precipitation prediction result, specifically including:

[0137] The posterior distribution π(Z(s0,t)|z) of precipitation prediction at each spatial location was estimated by sampling using the Markov chain Monte Carlo (MCMC) sampling method. The mean statistic of the posterior distribution was determined as the precipitation prediction result at the spatial point. The MCMC sampling method specifically includes:

[0138] By constructing a Markov chain and performing a random walk based on the Markov chain, a sequence of samples is generated to obtain a stationary distribution of samples. Then, the stationary distribution of samples is used to perform approximate numerical calculations. The MCMC sampling method requires multiple iterations before the sequence approaches the target distribution, and a warm-up period needs to be set. In this embodiment of the invention, the warm-up period can optionally be set to 4000 times, and the maximum number of iterations can be set to 10000 times.

[0139] Please refer to the following: Figure 6 , Figure 6 This is a flowchart of a method for correcting spatial point precipitation prediction results provided in an embodiment of the present invention;

[0140] Step S60: Correct the spatial point precipitation prediction results using the Kriging method, specifically including:

[0141] The contribution weight λ of each precipitation observation station to the precipitation at spatial location s0 was calculated using the Kriging method. iBased on whether precipitation occurred at each precipitation observation station, the probability value P(s0,t) of precipitation occurring at spatial location s0 is determined (1: precipitation occurred; 0: precipitation did not occur), thereby obtaining the spatial point precipitation prediction result after precipitation correction. The Kriging interpolation method is shown in the following formula:

[0142]

[0143] Where z is the number of observation stations within the first range of the spatial location S0 of the precipitation to be predicted, P(S i ) represents the precipitation observation data of the i-th precipitation observation station, and 0≤w≤1 is the threshold for determining whether precipitation has occurred. In this embodiment of the invention, the recommended range of w is [0.2, 0.5].

[0144] The final predicted spatial precipitation value is The corrected spatial point precipitation prediction result is a precipitation distribution time series with a certain spatial resolution.

[0145] In this embodiment of the invention, the calculation weight λ i The function can be a semivariance function, and further, the parameter value of the parameter v, which represents the smoothness of the random field, can be adjusted according to the correction value of the spatial prediction result of precipitation.

[0146] By correcting spatial precipitation prediction results using the Kriging method, this invention overcomes the problem of numerous "zero values" in daily and hourly precipitation sequences, improves the rationality of prediction results for cases where precipitation does not occur, and realizes the correction of precipitation prediction values.

[0147] In this embodiment of the invention, before correcting the spatial precipitation prediction results using the Kriging method, it is necessary to transform the spatial precipitation prediction results using the inverse BOX-COX transformation to obtain the spatial precipitation prediction results after the inverse BOX-COX transformation. Then, the spatial precipitation prediction results after the inverse BOX-COX transformation are corrected using the Kriging method to obtain the final spatial precipitation prediction results. The inverse BOX-COX transformation can be implemented using tools such as Python and MATLAB.

[0148] Step S70: Verify the prediction performance of the spatial prediction model for precipitation;

[0149] In this embodiment of the invention, the precipitation spatial adaptation model includes the generalized linear regression model, the Gaussian process model, the prior distribution of the spatiotemporal variation parameters, and the Bayesian hierarchical model.

[0150] In this embodiment of the invention, the verification method can be a cross-validation method, which removes the precipitation observation data of any precipitation observation station and runs the simulation process using the precipitation observation data of the remaining precipitation observation stations. Each time, the precipitation observation data of one precipitation observation station is removed without repetition. There are a total of n precipitation observation stations, so the simulation process is run n times to obtain n simulation results.

[0151] After obtaining the simulation results, PMCC is calculated. PMCC reflects the prediction performance of the precipitation spatial prediction model. The PMCC formula is as follows:

[0152]

[0153] Where n is the number of observation stations, r is the time series, T is the time point, and E(Y) l (s i ,t) rep -y l (s i ,t)) 2 For mathematical expectation, Var Y l (s i ,t) rep Let Y be the variance. l (s i ,t) rep Let y be the precipitation prediction result of the i-th precipitation observation station at time t. l (s i ,t) represents the precipitation observation data of the i-th precipitation observation station at time t, and PMCC represents the prediction effect of the precipitation spatial prediction model.

[0154] In this embodiment of the invention, relevant evaluation indicators such as root mean square error, mean absolute error, precipitation false alarm rate, and coefficient of determination can be further obtained to verify the model prediction effect.

[0155] By verifying the precipitation spatial prediction model using a verification method, the present invention can determine the prediction effect of the model. Furthermore, the model parameter values ​​can be modified based on the prediction effect.

[0156] Please refer to the following: Figure 7 , Figure 7 This is an overall flowchart of a precipitation spatiotemporal prediction method provided by an embodiment of the present invention;

[0157] like Figure 7 As shown, the overall process of the precipitation spatiotemporal prediction method includes:

[0158] Step S811: Obtain precipitation observation data.

[0159] Step S812: Box-Cox conversion of precipitation observation data;

[0160] The purpose of Box-Cox transformation of precipitation observation data is to address the problem of significant skewness in precipitation distribution and to obtain more reliable simulation results of hourly and daily precipitation data through precipitation data transformation.

[0161] Step S813: Obtain and screen relevant spatial forecasting factors for precipitation;

[0162] Specifically, the spatial forecasting factors for precipitation with significant correlations were screened using hypothesis testing.

[0163] Step S814: Construct the generalized linear regression model and the Gaussian process model;

[0164] Specifically, a generalized linear regression model and a Gaussian process model are constructed to incorporate the effects of influencing factors into spatial precipitation prediction.

[0165] Step S815: Initialize the spatiotemporal variation parameters;

[0166] Specifically, initializing the spatiotemporal variation parameter includes: the functional form of the spatiotemporal variation parameter conforms to the inverse gamma distribution, and the prior distribution of the spatiotemporal variation parameter is obtained by specifying the hyperparameter of the inverse gamma distribution function of the spatiotemporal variation parameter.

[0167] Step S816: Obtain the posterior distribution of the spatiotemporal variation parameters through Markov Monte Carlo simulation;

[0168] Specifically, the posterior distribution of spatiotemporal variation parameters is obtained through Markov Monte Carlo simulation, and this posterior distribution is used to obtain the posterior distribution of precipitation at spatial points under observation conditions.

[0169] Step S817: Set the spatial sampling resolution;

[0170] Specifically, setting the spatial sampling rate resolution includes: using the spatial range of the study area as the boundary, setting the spatial resolution, and using GIS tools to extract the spatial location of the raster center point corresponding to the spatial resolution, i.e., the geographic coordinates.

[0171] Step S818: Obtain spatial point location information;

[0172] Specifically, spatial point location information is obtained by extracting the spatial location of the center point of the raster with the corresponding spatial resolution using GIS tools. The generated spatial location information is then input into the precipitation spatial prediction adaptation model to calculate and obtain the time series map of precipitation spatial distribution within the study area. This allows for in-depth analysis and research on precipitation spatiotemporal patterns.

[0173] Step S819: Obtain the predicted sequence of point precipitation O(s0,t);

[0174] Specifically, the predicted sequence of point precipitation O(s0,t) is obtained. The predicted sequence of O(s0,t) is related to the spatial point location information. Step S40 has already been described and will not be repeated here.

[0175] Step S820: Construct a Bayesian hierarchical model, including: data model, process model, and parameter model.

[0176] Step S821: Obtain the spatial prediction adaptation model for precipitation;

[0177] Specifically, a spatial prediction model for precipitation is obtained, which includes: a generalized linear regression model, a Gaussian process model, a prior distribution of spatiotemporal variation parameters, and a Bayesian hierarchical model.

[0178] Step S822: Obtain the posterior distribution of precipitation at spatial points under the observation conditions;

[0179] Specifically, the posterior distribution of precipitation at spatial points under observation conditions is obtained. This posterior distribution is determined by a precipitation spatial adaptation model, which utilizes the basic principles of hierarchical Bayesian methods to deduce predicted values ​​through layer-by-layer conditional distributions.

[0180] Step S823: Obtain spatial point precipitation prediction results through Markov Monte Carlo simulation;

[0181] Specifically, spatial precipitation prediction results are obtained through Markov Monte Carlo simulation. This step estimates spatial precipitation using the Markov Monte Carlo simulation method.

[0182] Step S824: Convert the spatial point precipitation prediction results through BOX-COX inverse transform;

[0183] Specifically, the BOX-COX inverse transform can be implemented using tools such as Python and MATLAB.

[0184] Step S825: Correct the spatial precipitation prediction results using the Kriging method;

[0185] Specifically, the spatial precipitation prediction results after the BOX-COX inverse transformation are corrected by using the Kriging method to obtain more accurate prediction results.

[0186] Step S826: Obtain the time series map of the spatial distribution of precipitation;

[0187] Specifically, a time series map of the spatial distribution of precipitation is obtained, which is a time series of precipitation distribution with a certain spatial resolution.

[0188] Step S827: Model validation and forecast performance evaluation (PMCC, RMSE, R)2 (etc.), the validation methods include cross-validation, and the evaluation methods include calculating PMCC.

[0189] The spatiotemporal precipitation prediction method provided in this invention establishes a spatiotemporal variability parameter model to describe the potential Gaussian process of spatial precipitation, addresses the censoring problem of high-frequency precipitation-free observation data in precipitation time series, introduces a generalized linear regression model to facilitate the inclusion of spatial precipitation forecasting factors and improve precipitation prediction accuracy, and combines Bayesian hierarchical inference and Monte Carlo-Markov sampling algorithms to solve the complex Gaussian prediction process and probability calculations, obtaining a high-precision time series of spatial precipitation fields. Furthermore, the model method proposed in this invention can further incorporate time dynamic parameters and consider the hysteresis effect of dynamic parameters in the model hierarchical structure.

[0190] Please see Figure 8 , Figure 8 This is a schematic diagram of the electronic device of the present invention.

[0191] like Figure 8 As shown, the electronic device 80 includes: one or more processors 81 and a memory 82. Figure 8 Take the 81 processor as an example.

[0192] Processor 81 and memory 82 can be connected via a bus or other means. Figure 8 Taking the example of a connection between China and Israel via a bus.

[0193] Processor 81 is used to acquire precipitation observation data, acquire precipitation spatial forecast factors, and acquire the prior distribution of spatiotemporal variation parameters; establish a generalized linear regression model based on the precipitation observation data and the precipitation spatial forecast factors; construct a Gaussian process model; construct a Bayesian hierarchical model based on the generalized linear regression model, the Gaussian process model, and the prior distribution of the spatiotemporal variation parameters to determine the posterior probability distribution of precipitation at spatial points under observation conditions; and determine the spatial point precipitation prediction result based on the posterior probability distribution.

[0194] By introducing the precipitation spatial forecasting factor, the generalized linear regression model can reflect the spatial variation process of the study area, thereby improving the prediction accuracy, effectively characterizing the dynamic changes in precipitation time, reflecting the spatial variation process of precipitation-related parameters, and obtaining reliable precipitation prediction results.

[0195] The memory 82, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules of the precipitation spatiotemporal prediction method in the embodiments of this application. The processor 82 executes various functional applications and data processing of the controller by running the non-volatile software programs, instructions, and modules stored in the memory 82, thereby implementing the precipitation spatiotemporal prediction method of the above-described method embodiments.

[0196] The memory 82 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the controller. Furthermore, the memory 82 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 82 may optionally include memory remotely located relative to the processor 81, and these remote memories may be connected to the controller via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0197] The one or more modules are stored in the memory 82. When executed by the one or more processors 81, they perform the precipitation spatiotemporal prediction method in any of the above method embodiments, for example, the method described above. Figure 2 Method steps S10 to S60.

[0198] It should be noted that the above-described product can execute the method provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects of executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of this application.

[0199] This application provides a non-volatile computer-readable storage medium storing computer-executable instructions that are executed by one or more processors, for example... Figure 8 One of the processors 81 can enable the above-described one or more processors to execute the precipitation spatiotemporal prediction method in any of the above method embodiments, and to perform the above-described... Figure 2 Method steps S10 to S60.

[0200] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0201] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; under the concept of the present invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the present invention as described above, which are not provided in detail for the sake of brevity; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for spatiotemporal prediction of precipitation, characterized in that, The method includes: Obtain the prior distribution of precipitation observation data, precipitation spatial forecast factors, and spatiotemporal variability parameters; Based on the precipitation observation data and the precipitation spatial forecast factors, a generalized linear regression model is established; Construct a Gaussian process model; Based on the generalized linear regression model, the Gaussian process model, and the prior distribution of the spatiotemporal variation parameters, a Bayesian hierarchical model is constructed to determine the posterior probability distribution of precipitation at spatial points under observation conditions. Based on the aforementioned posterior probability distribution, the spatial point precipitation prediction results are determined; The acquisition of the prior distribution of the spatiotemporal variation parameters includes: initializing the spatiotemporal variation parameters and setting the spatiotemporal variation parameters within a preset range, wherein the spatiotemporal variation parameters conform to an inverse gamma distribution; specifying the hyperparameters of the inverse gamma distribution of the spatiotemporal variation parameters to obtain the prior distribution of the spatiotemporal variation parameters. The method further includes: screening out precipitation spatial forecasting factors with significant correlation; The step of establishing a generalized linear regression model based on the precipitation observation data and the precipitation spatial forecast factors includes: Based on the precipitation observation data and the significantly correlated precipitation spatial forecasting factors, a generalized linear regression model is established, including: This is the transpose of the precipitation observation data matrix. For the average dynamic process, It is white noise. It follows a value with an expected value of 0 and a value with σ. ε 2 Let σ be a normal distribution with standard deviation. ε 2 Let α be the first parameter of the spatiotemporal variation, and let α be the intercept. For the spatial variation parameters of the corresponding covariate forecast factors, This is a diagonal matrix, where the elements on the diagonal represent spatial forecast factors for precipitation at a given location. The value of , 1≤i≤n, where n is the number of precipitation observation stations, and m is the number of precipitation spatial forecasting factors with significant correlation selected. This is a spatially correlated random error term; The construction of the Gaussian process model includes: The parameters of the generalized linear regression model and The variation follows a Gaussian process, and the spatial variation process expressed as a Gaussian process is as follows: For spatially correlated random error terms, The parameters represent the spatial variation of the corresponding covariate forecast factors, 1 ≤ i ≤ n, where n is the number of precipitation observation stations, and GP represents a Gaussian process. The second parameter of spatiotemporal variation. The third parameter of spatiotemporal variation. The formula is as follows Among them, s i and s j Let |s| be the spatial position of any two points in space. i - s j || represents the distance between two spatial locations; parameter Φ is the fourth parameter of spatiotemporal variation, parameter ν is the fifth parameter of spatiotemporal variation. When the distance changes, parameter Φ represents the degree of decay of the correlation, and parameter ν represents the smoothness of the random field; Г(ν) is the standard gamma function, and K is the ν-order Bessel function of the second kind; The step of determining the spatial point precipitation prediction result based on the posterior probability distribution includes: The posterior probability distribution is sampled using the Markov Chain Monte Carlo (MCMC) sampling method to determine the spatial point precipitation prediction result. The Markov Chain Monte Carlo (MCMC) sampling method includes: sampling to obtain the posterior distribution of precipitation prediction for each spatial location, calculating the mean statistic of the posterior distribution, and determining the mean statistic as the spatial point precipitation prediction result.

2. The method according to claim 1, characterized in that, The method further includes: processing precipitation observation data to obtain processed precipitation observation data, specifically including: The precipitation observation data is processed using the Box-Cox precipitation transformation method, and the optimal transformation parameter λ is estimated using the maximum likelihood method to obtain the processed precipitation observation data. The Box-Cox precipitation transformation method includes: in, Let Z(s) be the spatial location, t be the time, λ be the transformation parameter, and Z(s) be the transformation parameter. i (t) represents the spatial position at time t. The precipitation observation data is t∈[1,T], where T is the time of the precipitation observation data, 1≤i≤n, and n is the number of precipitation observation stations.

3. The method according to claim 1, characterized in that, The selected precipitation spatial forecasting factors with significant correlation include: The probability value that the spatial forecast factor of precipitation is significantly correlated with the precipitation observation data is calculated by hypothesis testing. Set a preset value and compare the probability value with the preset value: If the probability value is greater than the preset value, then the precipitation spatial forecast factor and the precipitation observation data are not significantly correlated. If the probability value is less than or equal to the preset value, then the precipitation spatial forecast factor is significantly correlated with the precipitation observation data.

4. The method according to claim 1, characterized in that, The method further includes: correcting the spatial point precipitation prediction results using the Kriging method, specifically including: Obtain spatial information of the study area where precipitation to be predicted is to be obtained; The Kriging method is used to calculate the contribution weight of precipitation observation stations within a first range of the spatial location of the predicted precipitation to the precipitation at that location. The probability of precipitation occurring at the predicted location is determined based on whether precipitation occurs at the observation stations within this first range. The probability value is 1 or 0, where 1 represents precipitation occurring and 0 represents no precipitation occurring. This process yields a correction value for the spatial point precipitation prediction result, resulting in a corrected spatial point precipitation prediction result. The Kriging method includes: Where, λ i For precipitation observation stations within a first range of the spatial location of the precipitation to be predicted, the spatial location of the precipitation to be predicted is... The contribution weight of precipitation, P( The probability of precipitation occurring at the observation station is represented by 0 or 1, and z represents the spatial location of the precipitation to be predicted. The number of observation stations within the first range, where w is the threshold for determining whether precipitation has occurred, 0 ≤ w ≤ 1.

5. The method according to claim 4, characterized in that, The method further includes: The prediction performance of the precipitation spatial prediction model was validated. The precipitation spatial adaptation model includes the generalized linear regression model, the Gaussian process model, the prior distribution of the spatiotemporal variability parameters, and the Bayesian hierarchical model, specifically including: Based on the corrected spatial precipitation prediction results, a cross-validation method is used to run the simulation process by removing the precipitation data of any reference observation station and using the observation data of the remaining stations. Finally, PMCC is calculated, wherein the calculation of PMCC includes: Where n is the number of observation stations, r is the time series, and T is the time point. For mathematical expectation, For variance, Let be the precipitation prediction result for the i-th precipitation observation station at time t. Let t represent the precipitation observation data of the i-th precipitation observation station at time t, and PMCC represent the prediction result of the precipitation spatial prediction model.

6. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the spatiotemporal precipitation prediction method according to any one of claims 1-5.

Images