WO2022104709A1 - 一种耦合伽马与高斯分布的月尺度降水预报校正方法 - Google Patents

一种耦合伽马与高斯分布的月尺度降水预报校正方法 Download PDF

Info

Publication number
WO2022104709A1
WO2022104709A1 PCT/CN2020/130457 CN2020130457W WO2022104709A1 WO 2022104709 A1 WO2022104709 A1 WO 2022104709A1 CN 2020130457 W CN2020130457 W CN 2020130457W WO 2022104709 A1 WO2022104709 A1 WO 2022104709A1
Authority
WO
WIPO (PCT)
Prior art keywords
forecast
distribution
data
gamma
monthly
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2020/130457
Other languages
English (en)
French (fr)
Inventor
赵铜铁钢
黄泽青
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sun Yat Sen University
Original Assignee
Sun Yat Sen University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sun Yat Sen University filed Critical Sun Yat Sen University
Publication of WO2022104709A1 publication Critical patent/WO2022104709A1/zh
Priority to US17/948,240 priority Critical patent/US20230023374A1/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01WMETEOROLOGY
    • G01W1/00Meteorology
    • G01W1/10Devices for predicting weather conditions
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01WMETEOROLOGY
    • G01W1/00Meteorology
    • G01W1/14Rainfall or precipitation gauges
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/18Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02ATECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A10/00TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
    • Y02A10/40Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping

Definitions

  • the invention relates to the field of hydrological forecasting, and more particularly, to a monthly-scale precipitation forecasting correction method coupling gamma and Gaussian distribution.
  • the patent with publication number CN108830419A proposes a joint forecasting method for cascade reservoir group inflow flow based on ECC post-processing, which belongs to the technical field of hydrological forecasting, and discloses a combination of gamma distribution functions based on measured data
  • the systematic error correction is carried out on the aggregated numerical weather forecast data, but there are still random errors, which still have a certain impact on the forecast accuracy.
  • the present invention provides a monthly-scale precipitation forecast correction method coupled with gamma and Gaussian distribution to overcome the defects of the system complexity and random errors affecting the precipitation forecast accuracy described in the prior art.
  • the technical scheme of the present invention is as follows:
  • a monthly-scale precipitation forecast correction method coupling gamma and Gaussian distribution including the following steps:
  • S1 Collect the forecast data of the average monthly precipitation in the basin area and the corresponding observation value of the average precipitation in the basin area as input data;
  • a gamma distribution function is used to fit the forecast data and the observed values, respectively, to obtain the marginal distribution of the original forecast data and the observed values, and the expression formula is as follows:
  • F represents the set of collected K forecast data [f 1 , f 2 ,..., f K ], and O represents the set of collected K observations [o 1 , o 2 ,..., o K ];
  • G( ⁇ ) represents the gamma distribution function, ⁇ f , ⁇ f represent the gamma distribution parameters of the forecast data obtained by fitting, ⁇ o , ⁇ o represent the gamma distribution parameters of the observed values obtained by fitting .
  • the parameters ⁇ f , ⁇ f , ⁇ o , and ⁇ o of the gamma distribution function are respectively calculated by the maximum likelihood estimation method.
  • step S3 the cumulative distribution function of the corresponding gamma distribution is used to calculate the cumulative distribution function value of each forecast data f i and observation value o i in the corresponding gamma distribution, and the expression formula is as follows:
  • the cumulative distribution function value is regarded as the quantile of the standard normal distribution, and the cumulative distribution function value is converted into a variable that obeys the standard normal distribution through the inverse function of the standard normal distribution cumulative distribution function, Its expression formula is as follows:
  • N(0,1 2 ) represents the standard normal distribution.
  • step S5 according to the variable obeying the standard normal distribution and A joint normal distribution is constructed to characterize the correlation between the forecast data and the observed values in the input data, and the expression formula is as follows:
  • represents the variable and correlation.
  • step S6 its specific steps are as follows:
  • the method further includes the following steps: calculating the deviation value and the forecast accuracy according to the corrected forecast result as the forecast check index.
  • the method further includes the following steps: drawing a forecast diagnosis map according to the correction forecast result, the deviation value and the forecast precision.
  • the corrected forecast median value is taken as the x-axis
  • the precipitation forecast distribution interval and the observed value are taken as the y-axis
  • the calculation results of the deviation value and forecast accuracy are inserted into the forecast and diagnosis chart for display.
  • the beneficial effect of the technical solution of the present invention is: the present invention converts the precipitation forecast and observation data into normal distribution through gamma distribution, avoids the complicated data normalization method, and normalizes the data according to the obedience standard.
  • the variables of the normal distribution are constructed with a joint normal distribution to characterize the correlation between the forecast data and the observed values in the input data, and further random sampling of the observed values according to the correlation can effectively quantify the random error, and solve the complexity of the system and the impact of random errors on precipitation.
  • the problem that affects the forecast accuracy can effectively improve the forecast accuracy.
  • FIG. 1 is a flowchart of a method for forecasting and correcting monthly-scale precipitation by coupling gamma and Gaussian distribution according to Embodiment 1.
  • FIG. 1 is a flowchart of a method for forecasting and correcting monthly-scale precipitation by coupling gamma and Gaussian distribution according to Embodiment 1.
  • FIG. 2 is a schematic diagram of input data in Embodiment 2.
  • FIG. 2 is a schematic diagram of input data in Embodiment 2.
  • FIG. 2 is a precipitation observation value and a normal distribution quantile map before conversion in Example 2.
  • FIG. 2 is a precipitation observation value and a normal distribution quantile map before conversion in Example 2.
  • FIG. 3 is a quantile diagram of the converted precipitation observations and normal distribution in Example 2.
  • FIG. 4 is a time series diagram of the original forecast data of Example 2.
  • FIG. 4 is a time series diagram of the original forecast data of Example 2.
  • FIG. 5 is a time-series diagram of correction forecast in Example 2.
  • FIG. 6 is an original prediction diagnosis diagram of Example 2.
  • FIG. 7 is a correction prediction diagnosis diagram of Example 2.
  • This embodiment proposes a monthly-scale precipitation prediction and correction method that couples gamma and Gaussian distribution.
  • FIG. 1 a flowchart of the monthly-scale precipitation prediction and correction method for coupling gamma and Gaussian distribution of this embodiment is shown.
  • S1 Collect the forecast data of the average monthly precipitation in the basin area and the corresponding observation value of the average precipitation in the basin area as input data.
  • the gamma distribution function is used to fit the forecast data and the observed values, respectively, to obtain the marginal distribution of the original forecast data and the observed values.
  • the expression formula is as follows:
  • F represents the set of collected K forecast data [f 1 , f 2 ,..., f K ], and O represents the set of collected K observations [o 1 , o 2 ,..., o K ];
  • G( ⁇ ) represents the gamma distribution function, ⁇ f , ⁇ f represent the gamma distribution parameters of the forecast data obtained by fitting, ⁇ o , ⁇ o represent the gamma distribution parameters of the observed values obtained by fitting .
  • the parameters ⁇ f , ⁇ f , ⁇ o and ⁇ o of the gamma distribution function are calculated by the maximum likelihood estimation method respectively.
  • the cumulative distribution function of the corresponding gamma distribution is used to calculate the cumulative distribution function value of each forecast data f i and observation value o i in the corresponding gamma distribution, and the expression formula is as follows:
  • the cumulative distribution function value is regarded as the quantile of the standard normal distribution, and the cumulative distribution function value is converted into a variable that obeys the standard normal distribution through the inverse function of the standard normal distribution cumulative distribution function.
  • the expression formula is as follows :
  • N(0,1 2 ) represents the standard normal distribution.
  • represents the variable and correlation.
  • the method further includes the following steps: calculating the deviation value and the prediction precision as the forecast inspection index according to the corrected forecast result, and drawing a forecast diagnosis map according to the corrected forecast result, the deviation value and the forecast precision.
  • the corrected forecast median value is taken as the x-axis
  • the precipitation forecast distribution interval and the observed value are taken as the y-axis
  • the calculation results of the deviation value and forecast accuracy are inserted into the forecast diagnosis chart for display.
  • the monthly-scale precipitation forecast correction method of coupling gamma and Gaussian distribution proposed in this embodiment can be implemented on the open source Python language platform.
  • the read_csv function of the Python open source third-party library Pandas is used to read the precipitation forecast and observation data in the pre-stored file, and the input data to be collected in step S1 is obtained.
  • the mathematical calculation process in steps S2 to S6 is programmed in the Python language platform, mainly using the third-party libraries Numpy and Scipy, and encapsulating them into class functions through class() and def(), then you can call the The class function implements precipitation forecast correction.
  • the deviation value Bias and the forecast accuracy CRPSS forecast inspection index are calculated by Numpy, and then the forecast diagnosis map is drawn by the Python third-party library Matplotlib, and the improvement effect of the corrected forecast result in this embodiment is compared and analyzed.
  • the precipitation forecast and observation data are converted into normal distribution through gamma distribution, which avoids complex data normalization methods, and constructs a joint normal distribution based on variables that obey standard normal distribution to represent the input data.
  • the correlation between the forecast data and the observed value, and further random sampling of the observed value according to the correlation can effectively quantify the random error, solve the problem of the complexity of the system and the impact of random error on the precipitation forecast accuracy, and effectively improve the forecast accuracy.
  • This example proposes a specific implementation process.
  • the monthly-scale precipitation forecast correction method of coupling gamma and Gaussian distribution proposed in Example 1 is applied, and the method is implemented on the Python platform. implement.
  • the forecast data of the average monthly precipitation in the basin area and the corresponding observation value of the average precipitation in the basin area are collected as input data, and stored as csv files, as shown in Tables 1 and 2 below, which are the input data of this embodiment.
  • the precipitation forecast data is the accumulated precipitation with a forecast period of 30 days from January to early December.
  • the original forecast data and observation data that need to be corrected are read through the read_csv function, and the data are stored in the temp_x and temp_y variables, respectively.
  • the gamma distribution is fitted to the mean and observed values of the original forecast data by the stats.gamma.fit function, and the parameters of the gamma distribution function are obtained by using the maximum likelihood estimation method and stored in the para_x and para_y variables;
  • the cumulative distribution function value of the original forecast data and observation data is calculated by the stats.gamma.cdf function;
  • the stats.norm.ppf function is used to convert the cumulative distribution function value into a variable that obeys a normal distribution, so as to normalize the original forecast data and observation value, and separate the normalized data. It is stored in the variables trans_x and trans_y, which is convenient for subsequent modeling.
  • the pyplot function in Matplotlib and the stats.proplot function in Scipy are used to draw the quantile map of precipitation observations before and after transformation to test their normality, such as Figures 2 and 3 show the precipitation observations and normal distribution quantile maps before and after transformation, respectively;
  • the percentile function in Numpy is used to calculate the 10th, 25th, 50th, 75th and 90th quantiles of the original forecast and the corrected forecast, respectively, and the pyplot.
  • plot function in Matplotlib is used to take the year as the x-axis and the precipitation as the y-axis.
  • a time series diagram of precipitation forecast is drawn, as shown in Figures 4 and 5, which are the time series diagrams of the original forecast data and the corrected forecast of this embodiment, respectively.
  • the mean and sum functions in Numpy are used to calculate the bias Bias and the prediction accuracy CRPSS of the original forecast and the corrected forecast.
  • the pyplot.plot function in Matplotlib is used to take the median value of the ensemble forecast as the x-axis, and the distribution interval of the precipitation forecast is related to the observation.
  • the value is the y-axis, and the forecast diagnosis diagram is drawn, and the calculation results of deviation and forecast accuracy are inserted into the diagram with pyplot.text, as shown in Figures 6 and 7, which are the original forecast diagnosis diagram and the corrected forecast diagnosis diagram of this embodiment, respectively.
  • the forecast accuracy CRPSS from January to December in the original forecast is -18.28%, -24.7%, -48.46%, -32.94%, 17.17%, 6.14%, 18.25%, 4.52%, -88.80%, 31.13%, -8.08%, -3.37%.
  • the bias Bias from January to December in the corrected forecast is -1.34%, -1.00%, 0.20%, -0.42%, -0.86%, -0.50%, 0.80%, 0.56%, -0.88%, respectively , 1.04%, -1.46%, -0.26%;
  • the forecast accuracy CRPSS from January to December in the corrected forecast is -2.71%, 1.68%, -4.77%, 27.84%, 15.00%, 7.79%, 19.83%, 3.68%, -8.17%, 31.34%, -0.68 %, 33.41%.
  • class() and def() statements in Python can be used to encapsulate each step of precipitation forecast correction into class functions, namely four functions gamma_fit, trans_norm, back_trans and conditional_distribution and the gamma_gaussian class, which are saved as .py file, you only need to call the class function through the import statement to perform the precipitation forecast correction in the basin.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Environmental & Geological Engineering (AREA)
  • Computational Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Ecology (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Atmospheric Sciences (AREA)
  • Environmental Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Algebra (AREA)
  • Operations Research (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Hydrology & Water Resources (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种耦合伽马与高斯分布的月尺度降水预报校正方法,包括以下步骤:采集流域面平均月尺度降水的预报数据及其对应的流域面平均降水的观测值作为输入数据;将输入数据通过伽马分布函数进行拟合;计算每个输入数据在对应的伽马分布中的累积分布函数值;将累积分布函数值转化为服从标准正态分布的变量;根据服从标准正态分布的变量构建联合正态分布表征输入数据中预报数据与观测值的相关性;根据相关性对观测值进行随机采样,对采集的样本进行逆转换,得到校正预报结果。该月尺度降水预报校正方法能够有效量化随机误差,解决了系统的复杂性和随机误差对降水预报精度造成影响的问题,有效提高预报精度。

Description

一种耦合伽马与高斯分布的月尺度降水预报校正方法 技术领域
本发明涉及水文预报领域,更具体地,涉及一种耦合伽马与高斯分布的月尺度降水预报校正方法。
背景技术
现有的全球气象模型能够提供丰富的降水预报信息,从全球降水预报产品中提取的流域尺度月尺度降水预报,能够为流域水库调度、农业灌溉和防汛抗旱工作提供重要参考。虽然月尺度原始预报数据与观测数据有良好的关联关系,但同时也包含复杂的系统与随机误差,这给降水预报在实际应用中带来了困难,在一定程度上影响了预报精度。
公开号为CN108830419A(公开日2018-11-16)的专利提出了一种基于ECC后处理的梯级水库群入库流量联合预报方法,属于水文预报技术领域,公开了依据实测数据结合伽马分布函数对集合数值天气预报数据进行系统误差校正,但仍然存在随机误差,对预报精度仍存在一定的影响。
发明内容
本发明为克服上述现有技术所述的系统的复杂性和随机误差对降水预报精度造成影响的缺陷,提供一种耦合伽马与高斯分布的月尺度降水预报校正方法。
为解决上述技术问题,本发明的技术方案如下:
一种耦合伽马与高斯分布的月尺度降水预报校正方法,包括以下步骤:
S1:采集流域面平均月尺度降水的预报数据及其对应的流域面平均降水的观测值作为输入数据;
S2:将输入数据通过伽马分布函数进行拟合;
S3:计算每个输入数据在对应的伽马分布中的累积分布函数值;
S4:将累积分布函数值转化为服从标准正态分布的变量;
S5:根据服从标准正态分布的变量构建联合正态分布表征输入数据中预报数据与观测值的相关性;
S6:根据相关性对观测值进行随机采样,对采集的样本进行逆转换,得到校 正预报结果。
优选地,S2步骤中,采用伽马分布函数分别对预报数据和观测值进行拟合,得到原始预报数据和观测值的边缘分布,其表达公式如下:
Figure PCTCN2020130457-appb-000001
式中,F表示采集的K个预报数据的集合[f 1,f 2,...,f K],O表示采集的K个观测值的集合[o 1,o 2,...,o K];G(·)表示伽马分布函数,α f、β f表示通过拟合得到的预报数据的伽马分布参数,α o、β o表示通过拟合得到的观测值的伽马分布参数。
优选地,伽马分布函数的参数α f、β f、α o、β o分别由极大似然估计法推求。
优选地,S3步骤中,采用相应伽马分布的累积分布函数,计算每个预报数据f i和观测值o i在对应的伽马分布中的累积分布函数值,其表达公式如下:
Figure PCTCN2020130457-appb-000002
式中,
Figure PCTCN2020130457-appb-000003
Figure PCTCN2020130457-appb-000004
分别表示对应于第i年的预报数据f i和观测值o i的累积分布函数值;
Figure PCTCN2020130457-appb-000005
Figure PCTCN2020130457-appb-000006
分别表示预报数据f i和观测值o i拟合得到的伽马分布的累积分布函数。
优选地,S4步骤中,将累积分布函数值视为标准正态分布的分位数,通过标准正态分布累积分布函数的反函数,将累积分布函数值转化为服从标准正态分布的变量,其表达公式如下:
Figure PCTCN2020130457-appb-000007
式中,
Figure PCTCN2020130457-appb-000008
表示标准正态分布累积分布函数的反函数,
Figure PCTCN2020130457-appb-000009
Figure PCTCN2020130457-appb-000010
分别是经过正态分位数转换得到的预报数据和观测值;转换后的预报数据
Figure PCTCN2020130457-appb-000011
和转换后的观测值
Figure PCTCN2020130457-appb-000012
均服从正态分布,其表达公式如下:
Figure PCTCN2020130457-appb-000013
式中,N(0,1 2)表示标准正态分布。
优选地,S5步骤中,根据服从标准正态分布的变量
Figure PCTCN2020130457-appb-000014
Figure PCTCN2020130457-appb-000015
构建联合正态分布表征输入数据中的预报数据与观测值的相关性,其表达公式如下:
Figure PCTCN2020130457-appb-000016
式中,ρ表示变量
Figure PCTCN2020130457-appb-000017
Figure PCTCN2020130457-appb-000018
的相关性。
优选地,S6步骤中,其具体步骤如下:
S6.1:将预报数据
Figure PCTCN2020130457-appb-000019
作为预报因子,将各预报数据对应的观测值
Figure PCTCN2020130457-appb-000020
作为预报变量,计算预报变量的条件概率分布,其计算公式如下:
Figure PCTCN2020130457-appb-000021
S6.2:对观测值
Figure PCTCN2020130457-appb-000022
的条件概率分布结果进行随机采样,根据标准正态分布的累积分布函数与由观测值拟合得到的伽马分布累积分布函数的反函数对所采样的样本进行逆转换,得到校正预报结果。
优选地,还包括以下步骤:根据校正预报结果计算偏差值和预报精度作为预报检验指标。
优选地,还包括以下步骤:根据校正预报结果、偏差值和预报精度,绘制预报诊断图。
优选地,预报诊断图中,以校正预报中值作为x轴,以降水预报分布区间与观测值作为y轴,并将偏差值与预报精度计算结果插入在预报诊断图中显示。
与现有技术相比,本发明技术方案的有益效果是:本发明通过伽马分布将降水预报和观测数据转化为正态分布,避开了复杂的数据正态化方法,并根据服从标准正态分布的变量构建联合正态分布表征输入数据中预报数据与观测值的相关性,进一步根据相关性对观测值进行随机采样,能够有效量化随机误差,解决了系统的复杂性和随机误差对降水预报精度造成影响的问题,有效提高预报精度。
附图说明
图1为实施例1的耦合伽马与高斯分布的月尺度降水预报校正方法的流程图。
图2为实施例2的输入数据的示意图。
图2为实施例2的转换前的降水观测值与正态分布分位图。
图3为实施例2的转换后的降水观测值与正态分布分位图。
图4为实施例2的原始预报数据时间序列图。
图5为实施例2的校正预报时间序列图。
图6为实施例2的原始预报诊断图。
图7为实施例2的校正预报诊断图。
具体实施方式
附图仅用于示例性说明,不能理解为对本专利的限制;
对于本领域技术人员来说,附图中某些公知结构及其说明可能省略是可以理解的。
下面结合附图和实施例对本发明的技术方案做进一步的说明。
实施例1
本实施例提出一种耦合伽马与高斯分布的月尺度降水预报校正方法,如图1所示,为本实施例的耦合伽马与高斯分布的月尺度降水预报校正方法的流程图。
本实施例提出的耦合伽马与高斯分布的月尺度降水预报校正方法中,包括以下步骤:
S1:采集流域面平均月尺度降水的预报数据及其对应的流域面平均降水的观测值作为输入数据。
S2:将输入数据通过伽马分布函数进行拟合。
具体的,采用伽马分布函数分别对预报数据和观测值进行拟合,得到原始预报数据和观测值的边缘分布,其表达公式如下:
Figure PCTCN2020130457-appb-000023
式中,F表示采集的K个预报数据的集合[f 1,f 2,...,f K],O表示采集的K个观测值的集合[o 1,o 2,...,o K];G(·)表示伽马分布函数,α f、β f表示通过拟合得到的预报数据的伽马分布参数,α o、β o表示通过拟合得到的观测值的伽马分布参数。
其中,伽马分布函数的参数α f、β f、α o、β o分别由极大似然估计法推求。
S3:计算每个输入数据在对应的伽马分布中的累积分布函数值。
具体的,采用相应伽马分布的累积分布函数,计算每个预报数据f i和观测值o i在对应的伽马分布中的累积分布函数值,其表达公式如下:
Figure PCTCN2020130457-appb-000024
式中,
Figure PCTCN2020130457-appb-000025
Figure PCTCN2020130457-appb-000026
分别表示对应于第i年的预报数据f i和观测值o i的累积分布函数值;
Figure PCTCN2020130457-appb-000027
Figure PCTCN2020130457-appb-000028
分别表示预报数据f i和观测值o i拟合得到的伽马分布的累积分布函数。
S4:将累积分布函数值转化为服从标准正态分布的变量。
具体的,将累积分布函数值视为标准正态分布的分位数,通过标准正态分布累积分布函数的反函数,将累积分布函数值转化为服从标准正态分布的变量,其表达公式如下:
Figure PCTCN2020130457-appb-000029
式中,
Figure PCTCN2020130457-appb-000030
表示标准正态分布累积分布函数的反函数,
Figure PCTCN2020130457-appb-000031
Figure PCTCN2020130457-appb-000032
分别是经过正态分位数转换得到的预报数据和观测值;转换后的预报数据
Figure PCTCN2020130457-appb-000033
和转换后的观测值
Figure PCTCN2020130457-appb-000034
均服从正态分布,其表达公式如下:
Figure PCTCN2020130457-appb-000035
式中,N(0,1 2)表示标准正态分布。
S5:根据服从标准正态分布的变量构建联合正态分布表征输入数据中预报数据与观测值的相关性。
具体的,根据服从标准正态分布的变量
Figure PCTCN2020130457-appb-000036
Figure PCTCN2020130457-appb-000037
构建联合正态分布表征输入数据中的预报数据与观测值的相关性,其表达公式如下:
Figure PCTCN2020130457-appb-000038
式中,ρ表示变量
Figure PCTCN2020130457-appb-000039
Figure PCTCN2020130457-appb-000040
的相关性。
S6:根据相关性对观测值进行随机采样,对采集的样本进行逆转换,得到校正预报结果。其具体步骤如下:
S6.1:将预报数据
Figure PCTCN2020130457-appb-000041
作为预报因子,将各预报数据对应的观测值
Figure PCTCN2020130457-appb-000042
作为预报变量,计算预报变量的条件概率分布,其计算公式如下:
Figure PCTCN2020130457-appb-000043
S6.2:对观测值
Figure PCTCN2020130457-appb-000044
的条件概率分布结果进行随机采样,根据标准正态分布的累积分布函数与由观测值拟合得到的伽马分布累积分布函数的反函数对所采样 的样本进行逆转换,得到校正预报结果。
进一步的,还包括以下步骤:根据校正预报结果计算偏差值和预报精度作为预报检验指标,根据校正预报结果、偏差值和预报精度,绘制预报诊断图。
其中,预报诊断图中,以校正预报中值作为x轴,以降水预报分布区间与观测值作为y轴,并将偏差值与预报精度计算结果插入在预报诊断图中显示。
本实施例提出的耦合伽马与高斯分布的月尺度降水预报校正方法可在开源的Python语言平台上实施。
在具体实施过程中,利用Python开源的第三方库Pandas的read_csv函数读取预存储的文件中的降水预报和观测数据,得到S1步骤需要采集的输入数据。然后将S2~S6步骤中的数学计算过程在Python语言平台中编程实现,主要使用第三方库Numpy和Scipy,并通过class()和def()的方式将其封装成类函数,则可以调用该类函数实现降水预报校正。
在得到校正预报的基础上,通过Numpy计算出偏差值Bias和预报精度CRPSS预报检验指标,再通过Python第三方库Matplotlib进行预报诊断图绘制,对比分析本实施例中校正预报结果的改良效果。
本实施例中,通过伽马分布将降水预报和观测数据转化为正态分布,避开了复杂的数据正态化方法,并根据服从标准正态分布的变量构建联合正态分布表征输入数据中预报数据与观测值的相关性,进一步根据相关性对观测值进行随机采样,能够有效量化随机误差,解决了系统的复杂性和随机误差对降水预报精度造成影响的问题,有效提高预报精度。
实施例2
本实施例提出一种具体实施过程,对珠江流域北江上的ECMWF-S2S月尺度降水预报校正,应用实施例1提出的耦合伽马与高斯分布的月尺度降水预报校正方法,并在Python平台上实施。
首先,采集流域面平均月尺度降水的预报数据及其对应的流域面平均降水的观测值作为输入数据,存储为csv文件,如下表1和表2所示,为本实施例的输入数据。
表1 原始预报
Figure PCTCN2020130457-appb-000045
表2 观测值
Figure PCTCN2020130457-appb-000046
其中,降水预报数据为1-12月月初作出的预见期为30天的累积降水量。
在实施时,通过read_csv函数读取需要进行校正的原始预报数据和观测数据,并将数据分别存储在temp_x和temp_y变量中。
构建耦合伽马分布和高斯分布的月尺度降水预报校正模型,执行实施例1中S2~S6步骤的数学计算过程,主要通过第三方库Scipy进行,主要包括伽马分布拟合、联合正态分布构建和条件概率分布。
具体的,通过stats.gamma.fit函数分别对原始预报数据均值和观测值进行伽马分布的拟合,采用极大似然估计法得到伽马分布函数参数并存储在para_x和 para_y变量中;
根据拟合得到的伽马分布参数,通过stats.gamma.cdf函数计算原始预报数据和观测数据的累积分布函数值;
根据计算得到的累积分布函数值,采用stats.norm.ppf函数将累积分布函数值转化为服从正态分布的变量,从而将原始预报数据和观测值正态化,并将正态化的数据分别存储在变量trans_x和trans_y中,便于后续的建模,同时采用Matplotlib中的pyplot函数和Scipy中的stats.proplot函数,绘制转换前后的降水观测值的分位图,以检验其正态性,如图2、3所示,分别为转换前和转换后的降水观测值与正态分布分位图;
构建联合正态分布模型,通过stats.pearson计算正态转化后的预报数变量trans_x和观测值变量trans_y的相关系数,用于表征两者之间的相关关系;
推求观测值的条件概率分布参数,分别是均值mean与标准差sigma,将mean和sigma作为参数输入到函数stats.norm.rvs中进行随机采样,得到1000个样本;
计算随机采样得到的1000个样本在标准正态分布中的累积分布函数值,最后根据伽马分布参数para_y进行逆变换,得到校正预报结果。
对于每个月的预报,根据上述步骤对原始预报数据进行校正,最终得到该月的一组校正预报结果,依次进行,最后得到12组校正预报结果,然后将12组原始预报数据与校正预报结果一起进行预报检验。
具体的,利用Numpy中的percentile函数分别计算原始预报和校正预报的10、25、50、75和90分位数,采用Matplotlib中的pyplot.plot函数,以年份为x轴,降水量为y轴绘制降水预报时间序列图,如图4、5所示,分别为本实施例的原始预报数据和校正预报的时间序列图。然后利用Numpy中的mean和sum等函数,计算原始预报和校正预报的偏差Bias和预报精度CRPSS,同时采用Matplotlib中的pyplot.plot函数,以集合预报中值为x轴,降水预报分布区间与观测值为y轴,绘制预报诊断图,同时以pyplot.text将偏差与预报精度计算结果插入图中,如图6、7所示,分别为本实施例的原始预报诊断图和校正预报诊断图。
图6中,原始预报中一月至十二月的偏差Bias分别为27.74%、37.04%、22.19%、36.29%、-3.00%、4.00%,17.69%、-0.34%、51.56%、-9.37%、28.97%、54.11%;
原始预报中一月至十二月的预报精度CRPSS分别为-18.28%、-24.7%、-48.46%、-32.94%、17.17%、6.14%、18.25%、4.52%、-88.80%、31.13%、-8.08%、-3.37%。
图7中,校正预报中一月至十二月的偏差Bias分别为-1.34%、-1.00%、0.20%、-0.42%、-0.86%、-0.50%、0.80%、0.56%、-0.88%、1.04%、-1.46%、-0.26%;
校正预报中一月至十二月的预报精度CRPSS分别为-2.71%、1.68%、-4.77%、27.84%、15.00%、7.79%、19.83%、3.68%、-8.17%、31.34%、-0.68%、33.41%。
根据上述偏差Bias和预报精度CRPSS比对可知,采用耦合伽马与高斯分布的月尺度降水预报校正方法后,其校正预报结果与原始预报数据相比显然偏差值Bias有效降低,校正预报中的偏差Bias基本在1.5%内,而预报精度CRPSS更加稳定。
此外,本实施例可以采用Python中的class()和def()语句,将降水预报校正的各个步骤封装成类函数,分别为gamma_fit、trans_norm、back_trans和conditional_distribution四个函数和gamma_gaussian类,并保存为.py文件,使用时只需要通过import语句调用类函数,即可以进行流域降水预报校正。
相同或相似的标号对应相同或相似的部件;
附图中描述位置关系的用语仅用于示例性说明,不能理解为对本专利的限制;
显然,本发明的上述实施例仅仅是为清楚地说明本发明所作的举例,而并非是对本发明的实施方式的限定。对于所属领域的普通技术人员来说,在上述说明的基础上还可以做出其它不同形式的变化或变动。这里无需也无法对所有的实施方式予以穷举。凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等,均应包含在本发明权利要求的保护范围之内。

Claims (10)

  1. 一种耦合伽马与高斯分布的月尺度降水预报校正方法,其特征在于,包括以下步骤:
    S1:采集流域面平均月尺度降水的预报数据及其对应的流域面平均降水的观测值作为输入数据;
    S2:将所述输入数据通过伽马分布函数进行拟合;
    S3:计算每个输入数据在对应的伽马分布中的累积分布函数值;
    S4:将所述累积分布函数值转化为服从标准正态分布的变量;
    S5:根据所述服从标准正态分布的变量构建联合正态分布表征所述输入数据中预报数据与观测值的相关性;
    S6:根据所述相关性对观测值进行随机采样,对采集的样本进行逆转换,得到校正预报结果。
  2. 根据权利要求1所述的月尺度降水预报校正方法,其特征在于,所述S2步骤中,采用伽马分布函数分别对预报数据和观测值进行拟合,得到原始预报数据和观测值的边缘分布,其表达公式如下:
    Figure PCTCN2020130457-appb-100001
    式中,F表示采集的K个预报数据的集合[f 1,f 2,...,f K],O表示采集的K个观测值的集合[o 1,o 2,...,o K];G(·)表示伽马分布函数,α f、β f表示通过拟合得到的预报数据的伽马分布参数,α o、β o表示通过拟合得到的观测值的伽马分布参数。
  3. 根据权利要求2所述的月尺度降水预报校正方法,其特征在于,所述伽马分布函数的参数α f、β f、α o、β o分别由极大似然估计法推求。
  4. 根据权利要求2所述的月尺度降水预报校正方法,其特征在于,所述S3步骤中,采用相应伽马分布的累积分布函数,计算每个预报数据f i和观测值o i在对应的伽马分布中的累积分布函数值,其表达公式如下:
    Figure PCTCN2020130457-appb-100002
    式中,
    Figure PCTCN2020130457-appb-100003
    Figure PCTCN2020130457-appb-100004
    分别表示对应于第i年的预报数据f i和观测值o i的累积分布 函数值;
    Figure PCTCN2020130457-appb-100005
    Figure PCTCN2020130457-appb-100006
    分别表示预报数据f i和观测值o i拟合得到的伽马分布的累积分布函数。
  5. 根据权利要求4所述的月尺度降水预报校正方法,其特征在于,所述S4步骤中,将所述累积分布函数值视为标准正态分布的分位数,通过标准正态分布累积分布函数的反函数,将所述累积分布函数值转化为服从标准正态分布的变量,其表达公式如下:
    Figure PCTCN2020130457-appb-100007
    式中,
    Figure PCTCN2020130457-appb-100008
    表示标准正态分布累积分布函数的反函数,
    Figure PCTCN2020130457-appb-100009
    Figure PCTCN2020130457-appb-100010
    分别是经过正态分位数转换得到的预报数据和观测值;转换后的预报数据
    Figure PCTCN2020130457-appb-100011
    和转换后的观测值
    Figure PCTCN2020130457-appb-100012
    均服从正态分布,其表达公式如下:
    Figure PCTCN2020130457-appb-100013
    式中,N(0,1 2)表示标准正态分布。
  6. 根据权利要求5所述的月尺度降水预报校正方法,其特征在于,所述S5步骤中,根据所述服从标准正态分布的变量
    Figure PCTCN2020130457-appb-100014
    Figure PCTCN2020130457-appb-100015
    构建联合正态分布表征所述输入数据中的预报数据与观测值的相关性,其表达公式如下:
    Figure PCTCN2020130457-appb-100016
    式中,ρ表示变量
    Figure PCTCN2020130457-appb-100017
    Figure PCTCN2020130457-appb-100018
    的相关性。
  7. 根据权利要求6所述的月尺度降水预报校正方法,其特征在于,所述S6步骤中,其具体步骤如下:
    S6.1:将预报数据
    Figure PCTCN2020130457-appb-100019
    作为预报因子,将各预报数据对应的观测值
    Figure PCTCN2020130457-appb-100020
    作为预报变量,计算预报变量的条件概率分布,其计算公式如下:
    Figure PCTCN2020130457-appb-100021
    S6.2:对观测值
    Figure PCTCN2020130457-appb-100022
    的条件概率分布结果进行随机采样,根据标准正态分布的累积分布函数与由观测值拟合得到的伽马分布累积分布函数的反函数对所采样的样本进行逆转换,得到校正预报结果。
  8. 根据权利要求1~7任一项所述的月尺度降水预报校正方法,其特征在于, 其特征在于,所述方法还包括以下步骤:根据所述校正预报结果计算偏差值和预报精度作为预报检验指标。
  9. 根据权利要求8所述的月尺度降水预报校正方法,其特征在于,所述方法还包括以下步骤:根据所述校正预报结果、偏差值和预报精度,绘制预报诊断图。
  10. 根据权利要求9所述的月尺度降水预报校正方法,其特征在于,所述预报诊断图中,以校正预报中值作为x轴,以降水预报分布区间与观测值作为y轴,并将所述偏差值与预报精度计算结果插入在所述预报诊断图中显示。
PCT/CN2020/130457 2020-11-19 2020-11-20 一种耦合伽马与高斯分布的月尺度降水预报校正方法 Ceased WO2022104709A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/948,240 US20230023374A1 (en) 2020-11-19 2022-09-20 Method for calibrating monthly precipitation forecast by using gamma-gaussian distribution

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011303631.2A CN112415635B (zh) 2020-11-19 2020-11-19 一种耦合伽马与高斯分布的月尺度降水预报校正方法
CN202011303631.2 2020-11-19

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/948,240 Continuation US20230023374A1 (en) 2020-11-19 2022-09-20 Method for calibrating monthly precipitation forecast by using gamma-gaussian distribution

Publications (1)

Publication Number Publication Date
WO2022104709A1 true WO2022104709A1 (zh) 2022-05-27

Family

ID=74773548

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/130457 Ceased WO2022104709A1 (zh) 2020-11-19 2020-11-20 一种耦合伽马与高斯分布的月尺度降水预报校正方法

Country Status (3)

Country Link
US (1) US20230023374A1 (zh)
CN (1) CN112415635B (zh)
WO (1) WO2022104709A1 (zh)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US12105250B2 (en) 2021-04-16 2024-10-01 Sun Yat-Sen University Method for calibrating daily precipitation forecast by using bernoulli-Gamma-Gaussian distribution
CN113095579B (zh) * 2021-04-16 2023-04-18 中山大学 一种耦合伯努利-伽马-高斯分布的日尺度降水预报校正方法
CN115114811B (zh) * 2022-08-30 2022-11-29 水利部交通运输部国家能源局南京水利科学研究院 短时预报降水分类误差和定量误差双重订正方法及系统
CN116011687B (zh) * 2023-03-30 2023-08-11 山东锋士信息技术有限公司 一种基于Copula函数的洪水预报方法、系统及介质
CN116643331A (zh) * 2023-05-17 2023-08-25 昆明思永科技有限公司 基于区域流域的水文信息大数据进行水文预报的方法
CN118050828B (zh) * 2024-04-15 2024-06-25 江西省水利科学院(江西省大坝安全管理中心、江西省水资源管理中心) 一种流域防洪智能优化预报方法
CN119357546B (zh) * 2024-09-30 2025-10-31 中山大学 一种基于分位数映射的预报降水偏差订正方法及系统

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101672810B1 (ko) * 2015-08-24 2016-11-04 부경대학교 산학협력단 지형효과를 고려한 레이더 초단기 강수 예측과 수치모델 강우 예측 병합 방법
JP2017003416A (ja) * 2015-06-10 2017-01-05 古野電気株式会社 降水予測システム
US20170351963A1 (en) * 2016-06-02 2017-12-07 The Climate Corporation Estimating confidence bounds for rainfall adjustment values
CN108830419A (zh) * 2018-06-15 2018-11-16 武汉大学 一种基于ecc后处理的梯级水库群入库流量联合预报方法
CN109814178A (zh) * 2018-12-25 2019-05-28 河海大学 基于Copula-模型条件处理器的水文概率预报方法
CN110110339A (zh) * 2018-01-29 2019-08-09 中国电力科学研究院有限公司 一种日前水文预报误差校正方法及系统
JP2020134300A (ja) * 2019-02-19 2020-08-31 富士通株式会社 予測方法、予測プログラム及び情報処理装置
CN111831966A (zh) * 2020-05-21 2020-10-27 中山大学 一种基于高维概率分布函数的组合河道水位预报方法
CN111898831A (zh) * 2020-08-06 2020-11-06 长江水利委员会水文局 一种实时洪水概率预报实用化方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3966139B2 (ja) * 2002-09-27 2007-08-29 株式会社日立製作所 気象物理量の推定方法
US10545263B2 (en) * 2015-07-13 2020-01-28 The Climate Corporation Systems and methods for generating computer-based representations of probabilities of precipitation occurrences and intensities
CN105425319B (zh) * 2015-09-16 2017-10-13 河海大学 基于地面测量数据校正的降雨卫星暴雨同化方法
CN105808948B (zh) * 2016-03-08 2017-02-15 中国水利水电科学研究院 一种自动修正的多模式数值降雨集合预报方法
US11144835B2 (en) * 2016-07-15 2021-10-12 University Of Connecticut Systems and methods for outage prediction
CN107703564B (zh) * 2017-10-13 2020-04-14 中国科学院深圳先进技术研究院 一种降雨预测方法、系统及电子设备
CN110263293A (zh) * 2019-05-13 2019-09-20 中山大学 一种基于小波变换和联合概率分布的水文预报方法

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2017003416A (ja) * 2015-06-10 2017-01-05 古野電気株式会社 降水予測システム
KR101672810B1 (ko) * 2015-08-24 2016-11-04 부경대학교 산학협력단 지형효과를 고려한 레이더 초단기 강수 예측과 수치모델 강우 예측 병합 방법
US20170351963A1 (en) * 2016-06-02 2017-12-07 The Climate Corporation Estimating confidence bounds for rainfall adjustment values
CN110110339A (zh) * 2018-01-29 2019-08-09 中国电力科学研究院有限公司 一种日前水文预报误差校正方法及系统
CN108830419A (zh) * 2018-06-15 2018-11-16 武汉大学 一种基于ecc后处理的梯级水库群入库流量联合预报方法
CN109814178A (zh) * 2018-12-25 2019-05-28 河海大学 基于Copula-模型条件处理器的水文概率预报方法
JP2020134300A (ja) * 2019-02-19 2020-08-31 富士通株式会社 予測方法、予測プログラム及び情報処理装置
CN111831966A (zh) * 2020-05-21 2020-10-27 中山大学 一种基于高维概率分布函数的组合河道水位预报方法
CN111898831A (zh) * 2020-08-06 2020-11-06 长江水利委员会水文局 一种实时洪水概率预报实用化方法

Also Published As

Publication number Publication date
CN112415635B (zh) 2022-03-29
US20230023374A1 (en) 2023-01-26
CN112415635A (zh) 2021-02-26

Similar Documents

Publication Publication Date Title
WO2022104709A1 (zh) 一种耦合伽马与高斯分布的月尺度降水预报校正方法
CN116451879B (zh) 一种干旱风险预测方法、系统及电子设备
Odening et al. Analysis of rainfall derivatives using daily precipitation models: Opportunities and pitfalls
Ballesteros et al. FORETo: New software for reference evapotranspiration forecasting
CN106227998B (zh) 一种基于优化时间窗口的风资源评估方法
CN110619291B (zh) 一种植被覆盖度与气候因子非线性响应关系的识别方法
CN115495991A (zh) 一种基于时间卷积网络的降水区间预测方法
CN114357719B (zh) 利用卡尔曼滤波精确校正土壤侵蚀理论值的方法及系统
CN111291945B (zh) 一种牧业旱灾损失动态评估方法
CN107944636B (zh) 一种流域生态干旱评估与预报方法
CN107084709A (zh) 一种多弹性对径流变化驱动因素的定量分割方法
CN114819260A (zh) 一种水文时间序列预测模型动态生成方法
CN119227527A (zh) 一种耦合机器学习和物理过程模型的径流过程校正方法
CN113793228B (zh) 现状防御条件下的不同干旱频率农业因旱减产率确定方法
CN113095579B (zh) 一种耦合伯努利-伽马-高斯分布的日尺度降水预报校正方法
CN107038501B (zh) 一种基于r语言的小麦生育期特征参数估算方法
CN113487069A (zh) 一种基于grace日降尺度和新型dwsdi指数的区域洪涝灾害风险评估方法
CN108009398A (zh) 一种考虑逐日数据波动特征的gcm校正方法
CN116777290A (zh) 变化环境下水文要素演变特征及径流变化归因分析方法
CN116485035A (zh) 一种基于自适应整体性能优化的风电功率超短期概率预测方法
CN107944466B (zh) 一种基于分段思想的降雨偏差纠正方法
CN115422840B (zh) 一种基于物理模型混合深度学习模型的日尺度径流估算方法
US12105250B2 (en) Method for calibrating daily precipitation forecast by using bernoulli-Gamma-Gaussian distribution
CN115454984A (zh) 一种窗口自调节的卫星降水数据订正方法
CN108108860A (zh) 一种四步耦合中长期水文预报方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20962000

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20962000

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 31.01.2024)

122 Ep: pct application non-entry in european phase

Ref document number: 20962000

Country of ref document: EP

Kind code of ref document: A1