CN113639893B - A Method for Obtaining Near-Earth Weighted Average Temperature Information Based on Multiple Meteorological Factors - Google Patents

A Method for Obtaining Near-Earth Weighted Average Temperature Information Based on Multiple Meteorological Factors Download PDF

Info

Publication number
CN113639893B
CN113639893B CN202110725482.7A CN202110725482A CN113639893B CN 113639893 B CN113639893 B CN 113639893B CN 202110725482 A CN202110725482 A CN 202110725482A CN 113639893 B CN113639893 B CN 113639893B
Authority
CN
China
Prior art keywords
weighted average
average temperature
value
neural network
altitude
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.)
Active
Application number
CN202110725482.7A
Other languages
Chinese (zh)
Other versions
CN113639893A (en
Inventor
高成发
龙凤阳
董彦锋
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.)
Southeast University
Original Assignee
Southeast 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 Southeast University filed Critical Southeast University
Priority to CN202110725482.7A priority Critical patent/CN113639893B/en
Publication of CN113639893A publication Critical patent/CN113639893A/en
Application granted granted Critical
Publication of CN113639893B publication Critical patent/CN113639893B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01K—MEASURING TEMPERATURE; MEASURING QUANTITY OF HEAT; THERMALLY-SENSITIVE ELEMENTS NOT OTHERWISE PROVIDED FOR
    • G01K13/00—Thermometers specially adapted for specific purposes
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01K—MEASURING TEMPERATURE; MEASURING QUANTITY OF HEAT; THERMALLY-SENSITIVE ELEMENTS NOT OTHERWISE PROVIDED FOR
    • G01K13/00—Thermometers specially adapted for specific purposes
    • G01K13/02—Thermometers specially adapted for specific purposes for measuring temperature of moving fluids or granular materials capable of flow
    • G01K13/024—Thermometers specially adapted for specific purposes for measuring temperature of moving fluids or granular materials capable of flow of moving gases
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/08—Adaptations of balloons, missiles, or aircraft for meteorological purposes; Radiosondes
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/10—Devices for predicting weather conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Environmental & Geological Engineering (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Environmental Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Atmospheric Sciences (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Ecology (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a near-earth weighted average temperature information acquisition method based on multiple meteorological factors. The method comprises the following steps: acquiring global radio detection empty station observation data, and extracting approximate truth values of weighted average temperatures on different height surfaces; secondly, acquiring ERA-5 global reanalysis data, and establishing a polynomial model of weighted average temperature, altitude and annual product day of each grid point according to a least square method; calculating a preliminary predicted value of the weighted average temperature of the sounding site by using a polynomial model, and calculating a residual error between the preliminary predicted value and an approximate true value; then, establishing T BP neural network compensation models with different network structures, and training to obtain T groups of neural network model parameters; and finally, calculating T groups of residual prediction values of the weighted average temperature of the target position by utilizing T groups of neural network model parameters, averaging the residual prediction values to obtain a final residual prediction value, and adding a preliminary prediction value calculated by the polynomial model to obtain a final prediction value of the near-earth weighted average temperature.

Description

一种基于多气象因子的近地加权平均温度信息获取方法A Method for Obtaining Near-Earth Weighted Average Temperature Information Based on Multiple Meteorological Factors

技术领域technical field

本发明涉及全球导航卫星系统的气象学应用领域，特别涉及一种基于多个实测气象因子的全球近地加权平均温度信息获取方法。The invention relates to the meteorological application field of a global navigation satellite system, in particular to a method for acquiring global near-earth weighted average temperature information based on a plurality of measured meteorological factors.

背景技术Background technique

利用GNSS反演可降水量(Precipitation Water Vapor,PWV)是卫星导航定位技术在气象学领域的重要应用。利用GNSS反演PWV得益于GNSS卫星信号在穿越大气层时受水汽的影响产生的对流层延迟，对流层延迟映射至天顶方向的部分称为天顶湿延迟(ZenithWet Delay,ZWD)。PWV和和ZWD存在一个近似线性的关系即PWV＝Π·ZWD，其中的ZWD可以通过数据处理软件实时解算出来，但转换系数Π中包含一个待定参数，即加权平均温度(Weighted Mean Temperature,Tm)，因此，及时获取准确的Tm是利用GNSS实时反演实现PWV的关键。目前获取Tm的主要方法包括常数法、数值积分法和模型法。常数法将Tm设置为一个常数，但实际的Tm变化十分复杂，常数法会带来较大的PWV反演误差；数值积分法主要通过实测的大气垂直廓线数据或者再分析数据提取的大气垂直廓线数据通过离散的数值积分公式计算Tm，但实测大气廓线数据和再分析数据不能被普通用户实时获取到，因此也不能用于PWV的实时反演；利用模型法计算加权平均温度是目前应用较为普遍的替代方案。但是当前的大部分加权平均温度模型只能计算地表的加权平均温度值，模型的应用范围十分有限，且对于地形地貌变化复杂的区域，模型精度较低。如传统的Bevis公式没有考虑加权平均温度和气象因子之间关系的位置和季节差异，且没有考虑海拔高度对模型精度的影响，模型精度较差；NN-II模型虽然考虑了海拔高度和温度，但没有考虑加权平均温度与水汽压之间的较强的关联性，模型精度也有待提高。Using GNSS to retrieve Precipitation Water Vapor (PWV) is an important application of satellite navigation and positioning technology in the field of meteorology. The use of GNSS to retrieve PWV benefits from the tropospheric delay caused by the influence of water vapor when GNSS satellite signals pass through the atmosphere. The part of the tropospheric delay mapped to the zenith direction is called the Zenith Wet Delay (ZWD). There is an approximate linear relationship between PWV and ZWD, that is, PWV=Π·ZWD, where ZWD can be calculated in real time by data processing software, but the conversion coefficient Π contains an undetermined parameter, that is, the weighted average temperature (Weighted Mean Temperature, T m ), therefore, obtaining accurate T m in time is the key to realize PWV with GNSS real-time inversion. At present, the main methods for obtaining T m include constant method, numerical integration method and model method. The constant method sets T m as a constant, but the actual T m changes are very complex, and the constant method will bring a large PWV inversion error; the numerical integration method is mainly extracted from the measured atmospheric vertical profile data or reanalysis data. The atmospheric vertical profile data is calculated by discrete numerical integration formula, but the measured atmospheric profile data and reanalysis data cannot be obtained by ordinary users in real time, so they cannot be used for real-time inversion of PWV; the model method is used to calculate the weighted average Temperature is currently the more commonly used alternative. However, most of the current weighted average temperature models can only calculate the weighted average temperature value of the surface, the application scope of the model is very limited, and the model accuracy is low for areas with complex topographic changes. For example, the traditional Bevis formula does not consider the location and seasonal differences in the relationship between the weighted average temperature and meteorological factors, and does not consider the influence of altitude on the model accuracy, so the model accuracy is poor; although the NN-II model considers altitude and temperature, However, the strong correlation between the weighted average temperature and the water vapor pressure is not considered, and the model accuracy needs to be improved.

发明内容SUMMARY OF THE INVENTION

发明目的：本发明的目的是提供一种基于多气象因子的近地加权平均温度信息获取方法，相比多项式模型和单气象因子的NN-II模型，削弱了模型误差，提高了模型精度。Purpose of the invention: The purpose of the present invention is to provide a method for obtaining near-earth weighted average temperature information based on multiple meteorological factors, which weakens the model error and improves the model accuracy compared with the polynomial model and the NN-II model with a single meteorological factor.

技术方案：为实现上述发明目的，本发明采用的技术方案为：一种基于多气象因子的近地加权平均温度信息获取方法，包括以下步骤：Technical solution: In order to achieve the above purpose of the invention, the technical solution adopted in the present invention is: a method for acquiring near-earth weighted average temperature information based on multiple meteorological factors, comprising the following steps:

(1)获取全球分布的无线电探空站观测数据，提取每条大气垂直廓线上的位势高度、水汽压和气温的观测值，并计算探空站点上不同高度面的加权平均温度的近似真值，记为

Figure BDA0003138457900000011
(1) Obtain globally distributed radiosonde station observation data, extract the observed values of geopotential height, water vapor pressure and air temperature on each atmospheric vertical profile, and calculate the approximation of the weighted average temperature at different heights on the radiosonde station true value, denoted as
Figure BDA0003138457900000011

(2)获取ERA-5全球再分析数据，提取各格网点在不同等压面的位势高度、气温和比湿信息，计算各格网点在不同等压面层的加权平均温度值

Figure BDA0003138457900000021
利用最小二乘法建立
Figure BDA0003138457900000022
与海拔高度hg、年积日doy的多项式模型；(2) Obtain the ERA-5 global reanalysis data, extract the geopotential height, air temperature and specific humidity information of each grid point in different isobaric planes, and calculate the weighted average temperature value of each grid point in different isobaric plane layers
Figure BDA0003138457900000021
Created using the least squares method
Figure BDA0003138457900000022
polynomial model with altitude h g , year and day doy;

(3)根据多项式模型计算用于建模的探空站点上不同高度面的加权平均温度的初步预报值，记为

Figure BDA0003138457900000023
并计算初步预报值的残差
Figure BDA0003138457900000024
Figure BDA0003138457900000025
为加权平均温度近似真值
Figure BDA0003138457900000026
与初步预报值
Figure BDA0003138457900000027
的差；(3) Calculate the preliminary forecast value of the weighted average temperature at different heights on the sounding site used for modeling according to the polynomial model, denoted as
Figure BDA0003138457900000023
and calculate the residuals of the preliminary forecast values
Figure BDA0003138457900000024
Figure BDA0003138457900000025
Approximate the true value for the weighted average temperature
Figure BDA0003138457900000026
with preliminary forecast values
Figure BDA0003138457900000027
difference;

(4)建立T个具有不同网络结构的BP神经网络补偿模型，输入参数为用于建模的探空站点的纬度

Figure BDA0003138457900000028
海拔高度hg、年积日doy、相应海拔高度处的气温Ts、水汽压力Pw和多项式模型计算的初步预报值
Figure BDA0003138457900000029
输出参数为初步预报值的残差
Figure BDA00031384579000000210
训练得到T组神经网络模型参数；(4) Establish T BP neural network compensation models with different network structures, and the input parameter is the latitude of the sounding site used for modeling
Figure BDA0003138457900000028
Altitude h g , annual accumulated day doy , air temperature T s at the corresponding altitude T s , water vapor pressure P w and preliminary forecast values calculated by polynomial models
Figure BDA0003138457900000029
The output parameter is the residual of the preliminary forecast value
Figure BDA00031384579000000210
Training to obtain T group neural network model parameters;

(5)采集目标位置的纬度、海拔高度、年积日、气温、水汽压力和多项式模型计算的初步预报值信息，利用步骤(4)中的神经网络模型参数计算得到T组残差预报值，按照等权的原则对其求平均，得到最终的残差预报值

Figure BDA00031384579000000211
Figure BDA00031384579000000212
与多项式模型计算的目标位置加权平均温度的初步预报值相加，即得到近地加权平均温度的最终预报值Tm。(5) Collect the latitude, altitude, annual accumulated days, air temperature, water vapor pressure and the preliminary forecast value information calculated by the polynomial model of the target location, and use the neural network model parameters in step (4) to calculate the residual forecast value of the T group, Average them according to the principle of equal weight to get the final residual forecast value
Figure BDA00031384579000000211
Figure BDA00031384579000000212
Adding with the preliminary forecast value of the weighted average temperature of the target position calculated by the polynomial model, the final forecast value T m of the near-earth weighted average temperature is obtained.

优选的，所述步骤(1)中，探空站观测的大气垂直廓线上不同高度面的加权平均温度近似真值的计算公式如下：Preferably, in the step (1), the calculation formula of the approximate true value of the weighted average temperature at different heights on the atmospheric vertical profile observed by the sounding station is as follows:

Figure BDA00031384579000000213
Figure BDA00031384579000000213

式中，

Figure BDA00031384579000000214
Ti、hi和
Figure BDA00031384579000000215
Ti+1、hi+1分别为第i层和第i+1层的水汽压、绝对温度和海拔高度值，N为等压面的层数。In the formula,
Figure BDA00031384579000000214
T i , hi and
Figure BDA00031384579000000215
T i+1 and h i+1 are the water vapor pressure, absolute temperature and altitude of the i-th layer and the i+1-th layer, respectively, and N is the number of layers of the isobaric surface.

优选的，所述步骤(2)中，ERA-5全球再分析数据的格网水平分辨率为1°×1°，包含了从地表至对流层顶的37个等压面的位势高度、气温和比湿信息，其中各等压面层的水汽压力由比湿和大气压换算而来，换算公式如下：Preferably, in the step (2), the horizontal resolution of the grid of the ERA-5 global reanalysis data is 1°×1°, including the geopotential height, air temperature of 37 isobaric surfaces from the surface to the tropopause and specific humidity information, in which the water vapor pressure of each isobaric surface layer is converted from specific humidity and atmospheric pressure, and the conversion formula is as follows:

Figure BDA00031384579000000216
Figure BDA00031384579000000216

式中，Pw,P,R分别为同一等压面上的水汽压、大气压和比湿；where Pw , P, and R are the vapor pressure, atmospheric pressure and specific humidity on the same isobaric surface, respectively;

利用ERA-5全球再分析数据计算各个等压面的加权平均温度值时，起算面是各个等压面所在的海拔高度，而非地表，计算不同等压面的加权平均温度值的公式同式(1)。When using the ERA-5 global reanalysis data to calculate the weighted average temperature value of each isobaric surface, the starting surface is the altitude where each isobaric surface is located, not the surface. The formula for calculating the weighted average temperature value of different isobaric surfaces is the same as the formula (1).

优选的，所述步骤(2)中，各个格网点加权平均温度的多项式模型公式如下：Preferably, in the step (2), the polynomial model formula of the weighted average temperature of each grid point is as follows:

Figure BDA0003138457900000031
Figure BDA0003138457900000031

式中，

Figure BDA0003138457900000032
为格网点位于海平面高度上的加权平均温度值，A0为余项，A1、A2为与年积日有关的三角函数项的系数，d1、d2为年周期和半年周期的初始相位，doy为年积日，
Figure BDA0003138457900000033
为格网点上海拔高度为hg处的加权平均温度值，hg为海拔高度，δ为该格网点处加权平均温度随高度的递减率。In the formula,
Figure BDA0003138457900000032
is the weighted average temperature value of grid points at sea level, A 0 is the remainder, A 1 , A 2 are the coefficients of the trigonometric function terms related to the annual accumulation day, d 1 , d 2 are the annual and semi-annual cycles The initial phase, doy is the day of the year,
Figure BDA0003138457900000033
is the weighted average temperature value at the altitude of h g on the grid point, h g is the altitude, and δ is the decrement rate of the weighted average temperature at the grid point with height.

优选的，所述步骤(3)中，根据多项式模型计算用于建模的探空站点加权平均温度的初步预报值时，找出离该探空站点最近的4个格网点，利用多项式模型分别计算每个格网点上与该探空站点相同海拔高度处的加权平均温度值，再对这4个格网点的加权平均温度值进行双线性内插得到该探空站点的加权平均温度值。Preferably, in the step (3), when calculating the preliminary forecast value of the weighted average temperature of the sounding site for modeling according to the polynomial model, find out the 4 grid points closest to the sounding site, and use the polynomial model to respectively Calculate the weighted average temperature value of each grid point at the same altitude as the sounding site, and then perform bilinear interpolation on the weighted average temperature value of the four grid points to obtain the weighted average temperature value of the sounding site.

优选的，所述步骤(4)中，共构建了T个BP神经网络补偿模型，每个BP神经网络补偿模型为6×P×1的三层网络结构，其中，输入参数包括用于建模的探空站点的纬度

Figure BDA0003138457900000034
海拔高度hg、年积日doy、相应海拔高度处的气温Ts、水汽压力Pw和多项式模型计算的初步预报值
Figure BDA0003138457900000035
输出层为初步预报值的残差
Figure BDA0003138457900000036
对每个BP神经网络补偿模型分别进行训练，得到T组神经网络模型参数。Preferably, in the step (4), a total of T BP neural network compensation models are constructed, and each BP neural network compensation model is a three-layer network structure of 6×P×1, wherein the input parameters include a model for modeling The latitude of the sounding site
Figure BDA0003138457900000034
Altitude h g , annual accumulated day doy , air temperature T s at the corresponding altitude T s , water vapor pressure P w and preliminary forecast values calculated by polynomial models
Figure BDA0003138457900000035
The output layer is the residual of the preliminary forecast value
Figure BDA0003138457900000036
Each BP neural network compensation model is trained separately to obtain T groups of neural network model parameters.

优选的，所述步骤(5)中，利用训练好的神经网络模型参数，计算得到目标位置加权平均温度的T组残差预报值，采用求平均的方式计算最终的残差预报值

Figure BDA0003138457900000037
计算公式如下：Preferably, in the step (5), the trained neural network model parameters are used to calculate and obtain the T group residual predicted value of the weighted average temperature of the target location, and the final residual predicted value is calculated by averaging
Figure BDA0003138457900000037
Calculated as follows:

Figure BDA0003138457900000038
Figure BDA0003138457900000038

式中，

Figure BDA0003138457900000039
表示利用第i个BP神经网络模型参数计算得到的残差预报值，T为BP神经网络补偿模型的个数。In the formula,
Figure BDA0003138457900000039
Represents the residual prediction value calculated by the i-th BP neural network model parameter, and T is the number of BP neural network compensation models.

优选的，所述步骤(5)中，目标位置加权平均温度的最终预报值Tm的计算公式如下：Preferably, in the step (5), the calculation formula of the final predicted value T m of the weighted average temperature of the target position is as follows:

Figure BDA00031384579000000310
Figure BDA00031384579000000310

式中，

Figure BDA00031384579000000311
为多项式模型计算的目标位置加权平均温度的初步预报值。In the formula,
Figure BDA00031384579000000311
Preliminary forecast for the weighted average temperature of the target location calculated for the polynomial model.

有益效果：与现有技术相比，本发明的技术方案具有以下有益技术效果：Beneficial effects: compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

本发明公开了一种基于多气象因子的近地加权平均温度信息获取方法，采用多个BP神经网络补偿模型相互融合的方式，提高了模型的泛化性；通过将多个实测气象因子作为模型输入，提高了模型精度。无论是与多项式模型相比，还是与单气象因子的NN-II模型相比，本发明的模型精度都有较大提高。The invention discloses a method for acquiring near-ground weighted average temperature information based on multiple meteorological factors, which adopts the method of mutual fusion of multiple BP neural network compensation models, thereby improving the generalization of the model; by using multiple measured meteorological factors as the model input, which improves model accuracy. Whether compared with the polynomial model or the NN-II model with a single meteorological factor, the model accuracy of the present invention is greatly improved.

附图说明Description of drawings

图1是本发明具体实施方式中用于训练的全球无线电探空站分布图；Fig. 1 is the global radiosonde station distribution map used for training in the specific embodiment of the present invention;

图2是本发明具体实施方式的方法流程图；Fig. 2 is the method flow chart of the specific embodiment of the present invention;

图3是本发明具体实施方式中用于模型测试的全球无线电探空站分布图；Fig. 3 is the global radiosonde station distribution map used for model testing in the specific embodiment of the present invention;

图4是本发明具体实施方式中不同模型在各个高度面的精度对比情况。FIG. 4 is a comparison of the accuracy of different models on various height planes in the specific embodiment of the present invention.

具体实施方式Detailed ways

下面结合附图和具体实施方式对本发明的技术方案作进一步详细说明。The technical solutions of the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

本具体实施方案公开了一种基于多气象因子的近地加权平均温度信息获取方法，本具体实施方案采用了全球分布的209个无线电探空站在2011年至2015年观测的气象资料，图1为各站点的全球分布情况，探空站的观测数据按各站点组织为一个独立的文件，包括观测时间、站点位置信息以及仪器报告的位势高度、气温和水汽压力等信息。本方法流程如图2所示，包括以下步骤：This specific embodiment discloses a method for acquiring near-Earth weighted average temperature information based on multiple meteorological factors. This specific embodiment adopts the meteorological data observed by 209 radiosonde stations distributed globally from 2011 to 2015. Figure 1 For the global distribution of each station, the observation data of the radiosonde stations are organized into an independent file for each station, including the observation time, station location information, and the geopotential height, air temperature and water vapor pressure reported by the instruments. The process flow of the method is shown in Figure 2 and includes the following steps:

(1)获取全球分布的无线电探空站观测数据，提取每条大气垂直廓线上的位势高度、水汽压和气温的观测值，并计算探空站点上不同高度面的加权平均温度的近似真值，记为

Figure BDA0003138457900000041
(1) Obtain globally distributed radiosonde station observation data, extract the observed values of geopotential height, water vapor pressure and air temperature on each atmospheric vertical profile, and calculate the approximation of the weighted average temperature at different heights on the radiosonde station true value, denoted as
Figure BDA0003138457900000041

(2)获取ERA-5全球再分析数据，提取各格网点在不同等压面的位势高度、气温和比湿信息，计算各格网点在不同等压面层的加权平均温度值

Figure BDA0003138457900000042
利用最小二乘法建立
Figure BDA0003138457900000043
与海拔高度hg、年积日doy的多项式模型；(2) Obtain the ERA-5 global reanalysis data, extract the geopotential height, air temperature and specific humidity information of each grid point in different isobaric planes, and calculate the weighted average temperature value of each grid point in different isobaric plane layers
Figure BDA0003138457900000042
Created using the least squares method
Figure BDA0003138457900000043
polynomial model with altitude h g , year and day doy;

(3)根据多项式模型计算用于建模的探空站点上不同高度面的加权平均温度的初步预报值，记为

Figure BDA0003138457900000044
并计算初步预报值的残差
Figure BDA0003138457900000045
Figure BDA0003138457900000046
为加权平均温度近似真值
Figure BDA0003138457900000047
与初步预报值
Figure BDA0003138457900000048
的差；(3) Calculate the preliminary forecast value of the weighted average temperature at different heights on the sounding site used for modeling according to the polynomial model, denoted as
Figure BDA0003138457900000044
and calculate the residuals of the preliminary forecast values
Figure BDA0003138457900000045
Figure BDA0003138457900000046
Approximate the true value for the weighted average temperature
Figure BDA0003138457900000047
with preliminary forecast values
Figure BDA0003138457900000048
difference;

(4)建立T个具有不同网络结构的BP神经网络补偿模型，输入参数为用于建模的探空站点的纬度

Figure BDA0003138457900000049
海拔高度hg、年积日doy、相应海拔高度处的气温Ts、水汽压力Pw和多项式模型计算的初步预报值
Figure BDA00031384579000000410
输出参数为初步预报值的残差
Figure BDA00031384579000000411
训练得到T组神经网络模型参数；(4) Establish T BP neural network compensation models with different network structures, and the input parameter is the latitude of the sounding site used for modeling
Figure BDA0003138457900000049
Altitude h g , annual accumulated day doy , air temperature T s at the corresponding altitude T s , water vapor pressure P w and preliminary forecast values calculated by polynomial models
Figure BDA00031384579000000410
The output parameter is the residual of the preliminary forecast value
Figure BDA00031384579000000411
Training to obtain T group neural network model parameters;

(5)采集目标位置的纬度、海拔高度、年积日、气温、水汽压力和多项式模型计算的初步预报值信息，利用步骤(4)中的神经网络模型参数计算得到T组残差预报值，按照等权的原则对其求平均，得到最终的残差预报值

Figure BDA0003138457900000051
Figure BDA0003138457900000052
与多项式模型计算的目标位置加权平均温度的初步预报值相加，即得到近地加权平均温度的最终预报值Tm。(5) Collect the latitude, altitude, annual accumulated days, air temperature, water vapor pressure and the preliminary forecast value information calculated by the polynomial model of the target location, and use the neural network model parameters in step (4) to calculate the residual forecast value of the T group, Average them according to the principle of equal weight to get the final residual forecast value
Figure BDA0003138457900000051
Figure BDA0003138457900000052
Adding with the preliminary forecast value of the weighted average temperature of the target position calculated by the polynomial model, the final forecast value T m of the near-earth weighted average temperature is obtained.

所述步骤(1)中，探空站观测的大气垂直廓线上不同高度面的加权平均温度近似真值的计算公式如下：In the step (1), the calculation formula of the approximate true value of the weighted average temperature at different heights on the atmospheric vertical profile observed by the sounding station is as follows:

Figure BDA0003138457900000053
Figure BDA0003138457900000053

式中，

Figure BDA0003138457900000054
Ti、hi和
Figure BDA0003138457900000055
Ti+1、hi+1分别为第i层和第i+1层的水汽压、绝对温度和海拔高度值，N为等压面的层数。In the formula,
Figure BDA0003138457900000054
T i , hi and
Figure BDA0003138457900000055
T i+1 and h i+1 are the water vapor pressure, absolute temperature and altitude of the i-th layer and the i+1-th layer, respectively, and N is the number of layers of the isobaric surface.

所述步骤(2)中，ERA-5全球再分析数据的格网水平分辨率为1°×1°，包含了从地表至对流层顶的37个等压面的位势高度、气温和比湿信息，其中各等压面层的水汽压力由比湿和大气压换算而来，换算公式如下：In the step (2), the grid horizontal resolution of the ERA-5 global reanalysis data is 1°×1°, including the geopotential height, air temperature and specific humidity of 37 isobaric surfaces from the surface to the tropopause. information, in which the water vapor pressure of each isobaric surface layer is converted from specific humidity and atmospheric pressure, and the conversion formula is as follows:

Figure BDA0003138457900000056
Figure BDA0003138457900000056

式中，Pw,P,R分别为同一等压面上的水汽压、大气压和比湿；where Pw , P, and R are the vapor pressure, atmospheric pressure and specific humidity on the same isobaric surface, respectively;

利用ERA-5全球再分析数据计算各个等压面的加权平均温度值时，起算面是各个等压面所在的海拔高度，而非地表，计算不同等压面的加权平均温度值的公式同式(1)。When using the ERA-5 global reanalysis data to calculate the weighted average temperature value of each isobaric surface, the starting surface is the altitude where each isobaric surface is located, not the surface. The formula for calculating the weighted average temperature value of different isobaric surfaces is the same as the formula (1).

所述步骤(2)中，各个格网点加权平均温度的多项式模型公式如下：In the step (2), the polynomial model formula of the weighted average temperature of each grid point is as follows:

Figure BDA0003138457900000057
Figure BDA0003138457900000057

式中，

Figure BDA0003138457900000058
为格网点位于海平面高度上的加权平均温度值，A0为余项，A1、A2为与年积日有关的三角函数项的系数，d1、d2为年周期和半年周期的初始相位，doy为年积日，
Figure BDA0003138457900000059
为格网点上海拔高度为hg处的加权平均温度值，hg为海拔高度，δ为该格网点处加权平均温度随高度的递减率。In the formula,
Figure BDA0003138457900000058
is the weighted average temperature value of grid points at sea level, A 0 is the remainder, A 1 , A 2 are the coefficients of the trigonometric function terms related to the annual accumulation day, d 1 , d 2 are the annual and semi-annual cycles The initial phase, doy is the day of the year,
Figure BDA0003138457900000059
is the weighted average temperature value at the altitude of h g on the grid point, h g is the altitude, and δ is the decrement rate of the weighted average temperature at the grid point with height.

所述步骤(3)中，根据多项式模型计算用于建模的探空站点加权平均温度的初步预报值时，找出离该探空站点最近的4个格网点，利用多项式模型分别计算每个格网点上与该探空站点相同海拔高度处的加权平均温度值，再对这4个格网点的加权平均温度值进行双线性内插得到该探空站点的加权平均温度值。In the step (3), when calculating the preliminary forecast value of the weighted average temperature of the sounding site for modeling according to the polynomial model, find out the 4 grid points closest to the sounding site, and use the polynomial model to calculate each The weighted average temperature value of the grid point at the same altitude as the sounding site, and then the weighted average temperature value of the four grid points is bilinearly interpolated to obtain the weighted average temperature value of the sounding site.

所述步骤(4)中，共构建了T个BP神经网络补偿模型，每个BP神经网络补偿模型为6×P×1的三层网络结构，其中，输入参数包括用于建模的探空站点的纬度

Figure BDA0003138457900000061
海拔高度hg、年积日doy、相应海拔高度处的气温Ts、水汽压力Pw和多项式模型计算的初步预报值
Figure BDA0003138457900000062
输出层为初步预报值的残差
Figure BDA0003138457900000063
对每个BP神经网络补偿模型分别进行训练，得到T组神经网络模型参数。In the step (4), a total of T BP neural network compensation models are constructed, and each BP neural network compensation model is a three-layer network structure of 6×P×1, wherein the input parameters include the sounding used for modeling. Latitude of the site
Figure BDA0003138457900000061
Altitude h g , annual accumulated day doy , air temperature T s at the corresponding altitude T s , water vapor pressure P w and preliminary forecast values calculated by polynomial models
Figure BDA0003138457900000062
The output layer is the residual of the preliminary forecast value
Figure BDA0003138457900000063
Each BP neural network compensation model is trained separately to obtain T groups of neural network model parameters.

所述步骤(5)中，利用训练好的神经网络模型参数，计算得到目标位置加权平均温度的T组残差预报值，采用求平均的方式计算最终的残差预报值

Figure BDA0003138457900000064
计算公式如下：In the step (5), the trained neural network model parameters are used to calculate and obtain the T group residual predicted value of the weighted average temperature of the target position, and the final residual predicted value is calculated by averaging.
Figure BDA0003138457900000064
Calculated as follows:

Figure BDA0003138457900000065
Figure BDA0003138457900000065

式中，

Figure BDA0003138457900000066
表示利用第i个BP神经网络模型参数计算得到的残差预报值，T为BP神经网络补偿模型的个数。In the formula,
Figure BDA0003138457900000066
Represents the residual prediction value calculated by the i-th BP neural network model parameter, and T is the number of BP neural network compensation models.

所述步骤(5)中，目标位置加权平均温度的最终预报值Tm的计算公式如下：In the step (5), the calculation formula of the final predicted value T m of the weighted average temperature of the target position is as follows:

Figure BDA0003138457900000067
Figure BDA0003138457900000067

式中，

Figure BDA0003138457900000068
为多项式模型计算的目标位置加权平均温度的初步预报值。In the formula,
Figure BDA0003138457900000068
Preliminary forecast for the weighted average temperature of the target location calculated for the polynomial model.

为了比较多项式模型，单气象因子的NN-II模型和多气象因子模型的预测性能，采用2016年至2018年全球分布的576个探空站观测的气象数据进行验证，探空站的分布情况如图3所示，采用平均偏差MD和均方根误差RMSE作为模型评价指标，其中MD表示准确度，即模型值与真值的偏离程度；RMSE表示精度，用来衡量模型的可靠性和稳定性。它们的计算公式分别为：In order to compare the prediction performance of the polynomial model, the NN-II model with a single meteorological factor and the multi-meteorological factor model, the meteorological data observed from 576 sounding stations distributed globally from 2016 to 2018 were used for verification. The distribution of sounding stations is as follows As shown in Figure 3, the average deviation MD and the root mean square error RMSE are used as model evaluation indicators, where MD represents the accuracy, that is, the degree of deviation between the model value and the true value; RMSE represents the accuracy, which is used to measure the reliability and stability of the model . Their calculation formulas are:

Figure BDA0003138457900000069
Figure BDA0003138457900000069

Figure BDA00031384579000000610
Figure BDA00031384579000000610

式中，

Figure BDA00031384579000000611
和
Figure BDA00031384579000000612
分别表示模型值和真值，n为测试样本的个数。表1给出了不同模型的精度对比情况。In the formula,
Figure BDA00031384579000000611
and
Figure BDA00031384579000000612
represent the model value and the true value, respectively, and n is the number of test samples. Table 1 shows the accuracy comparison of different models.

表1 2016-2018年不同模型的MD和RMSE比较Table 1 MD and RMSE comparison of different models in 2016-2018

Figure BDA00031384579000000613
Figure BDA00031384579000000613

Figure BDA0003138457900000071
Figure BDA0003138457900000071

为了进一步比较三种模型在不同高度的预测效果，做出三种模型的RMSE随高度变化的情况，如图4所示。从表1和图4可以看出：In order to further compare the prediction effects of the three models at different heights, the RMSE of the three models varies with height, as shown in Figure 4. It can be seen from Table 1 and Figure 4 that:

(1)相比无实测气象因子的多项式模型和仅采用单个气象因子作为输入的NN-II模型，基于多气象因子的融合模型的预测精度最高，比前两者分别提高了约43％和18％；(1) Compared with the polynomial model without measured meteorological factors and the NN-II model with only a single meteorological factor as input, the fusion model based on multiple meteorological factors has the highest prediction accuracy, which is about 43% and 18% higher than the former two, respectively. %;

(2)相比无实测气象因子的多项式模型和仅采用单个气象因子作为输入的NN-II模型，基于多气象因子的融合模型在预测性能上的优势主要表现在高程方向上，即在不同海拔高度面上的精度都优于其它两种模型。(2) Compared with the polynomial model without measured meteorological factors and the NN-II model with only a single meteorological factor as input, the advantage of the fusion model based on multiple meteorological factors in prediction performance is mainly in the elevation direction, that is, at different altitudes. The accuracy on the height plane is better than the other two models.

Claims (7)

1. A method for acquiring near-earth weighted average temperature information based on multiple meteorological factors is characterized by comprising the following steps:
(1) acquiring observation data of the globally distributed radio detection stations, extracting observation values of potential height, vapor pressure and air temperature on each atmospheric vertical profile, calculating approximate truth values of weighted average temperatures of different height surfaces on the detection stations, and recording the approximate truth values as
Figure FDA0003663727520000011
(2) Acquiring ERA-5 global reanalysis data, and extracting potential height and gas of each grid point on different equal pressure surfacesThe temperature and the specific humidity information are used for calculating the weighted average temperature value of each grid point on different isobaric surface layers
Figure FDA0003663727520000012
Establishing by least square method
Figure FDA0003663727520000013
Altitude h g A polynomial model of year and day doy;
(3) calculating the initial prediction values of the weighted average temperatures of different height surfaces on the exploration station for modeling according to the polynomial model, and recording the initial prediction values as
Figure FDA0003663727520000014
And calculating the residual error of the preliminary predicted value
Figure FDA0003663727520000015
Figure FDA0003663727520000016
Approximate truth value for weighted average temperature
Figure FDA0003663727520000017
And preliminary predicted value
Figure FDA0003663727520000018
A difference of (d);
(4) establishing T BP neural network compensation models with different network structures, and inputting parameters into latitude of a sounding site for modeling
Figure FDA0003663727520000019
Altitude h g Annual accumulation day doy, air temperature T at corresponding altitude s Water vapor pressure P w Preliminary prediction values for sum polynomial model calculations
Figure FDA00036637275200000110
The output parameter isResidual error of preliminary predicted value
Figure FDA00036637275200000111
Training to obtain T groups of neural network model parameters;
(5) acquiring the latitude, altitude, accumulated days, air temperature, water vapor pressure and preliminary predicted value information calculated by a polynomial model of a target position, calculating by utilizing the parameters of the neural network model in the step (4) to obtain T groups of residual predicted values, averaging the residual predicted values according to the equal weight principle to obtain the final residual predicted value
Figure FDA00036637275200000112
Figure FDA00036637275200000113
Adding the initial predicted value of the weighted average temperature of the target position calculated by the polynomial model to obtain the final predicted value T of the weighted average temperature of the ground m ；
In the step (1), a calculation formula of weighted average temperature approximate truth values of different height surfaces on the atmospheric vertical profile observed by the sounding station is as follows:
Figure FDA00036637275200000114
in the formula (I), the compound is shown in the specification,
Figure FDA00036637275200000115
T i 、h i and
Figure FDA00036637275200000116
T i+1 、h i+1 the vapor pressure, absolute temperature and altitude value of the ith layer and the (i + 1) th layer are respectively, and N is the layer number of the isobaric surface.
2. The method for obtaining near-earth weighted average temperature information based on multiple meteorological factors according to claim 1, wherein in the step (2), the horizontal resolution of the grid of ERA-5 global reanalysis data is 1 ° × 1 °, potential heights, air temperatures and specific humidity information of 37 isobaric surfaces from the earth surface to the top of the convection layer are included, wherein the water vapor pressure of each isobaric surface layer is converted from the specific humidity and the atmospheric pressure, and the conversion formula is as follows:
Figure FDA0003663727520000021
in the formula, P w P and R are respectively the water vapor pressure, the atmospheric pressure and the specific humidity on the same equal pressure surface;
when the weighted average temperature value of each isobaric surface is calculated by utilizing ERA-5 global reanalysis data, the calculation surface is the altitude of each isobaric surface, but not the ground surface, and the formula for calculating the weighted average temperature values of different isobaric surfaces is the same as the formula (1).
3. The multi-meteorological-factor based near-earth weighted average temperature information acquisition method according to claim 1, wherein in the step (2), the polynomial model formula of the weighted average temperature of each grid point is as follows:
Figure FDA0003663727520000022
in the formula (I), the compound is shown in the specification,
Figure FDA0003663727520000023
is a weighted average temperature value, A, of the grid points at sea level height 0 For the remainder, A 1 、A 2 As a coefficient of a trigonometric function term related to the chronological order, d 1 、d 2 Is the initial phase of the annual and semiannual cycles, doy is the product of the year,
Figure FDA0003663727520000024
altitude of grid is h g Weighted average temperature value of (h) g Is the altitude, δ is the gridThe rate of decrease of the weighted average temperature with height at the mesh point.
4. The method according to claim 1, wherein in step (3), when calculating the preliminary predicted value of the weighted average temperature of the sounding site for modeling according to the polynomial model, 4 grid points closest to the sounding site are found, the polynomial model is used to calculate the weighted average temperature value at the same altitude as the sounding site on each grid point, and then the weighted average temperature values of the 4 grid points are bilinearly interpolated to obtain the weighted average temperature value of the sounding site.
5. The method for obtaining the multiple meteorological factor-based near-earth weighted average temperature information according to claim 1, wherein in the step (4), T BP neural network compensation models are constructed in total, each BP neural network compensation model is a three-layer network structure of 6 × P × 1, and the input parameters include latitudes of sounding sites for modeling
Figure FDA0003663727520000025
Altitude h g Annual accumulation day doy, air temperature T at corresponding altitude s Water vapor pressure P w Preliminary prediction values for sum polynomial model calculations
Figure FDA0003663727520000026
Residual error with output layer as preliminary predicted value
Figure FDA0003663727520000027
And respectively training each BP neural network compensation model to obtain T groups of neural network model parameters.
6. The method for obtaining the multiple meteorological factor-based near-earth weighted average temperature information according to claim 1, wherein in the step (5), the trained neural network model parameters are used for calculationCalculating the final residual prediction value by averaging the T groups of residual prediction values of the weighted average temperature of the target position
Figure FDA0003663727520000028
The calculation formula is as follows:
Figure FDA0003663727520000031
in the formula (I), the compound is shown in the specification,
Figure FDA0003663727520000032
and (4) representing a residual prediction value obtained by calculation by using the ith BP neural network model parameter, wherein T is the number of BP neural network compensation models.
7. The multiple meteorological factor-based approximate weighted average temperature information acquisition method according to claim 6, wherein in the step (5), the final predicted value T of the target position weighted average temperature m The calculation formula of (a) is as follows:
Figure FDA0003663727520000033
in the formula (I), the compound is shown in the specification,
Figure FDA0003663727520000034
a preliminary prediction of the weighted average temperature of the target location calculated for the polynomial model.
CN202110725482.7A 2021-06-29 2021-06-29 A Method for Obtaining Near-Earth Weighted Average Temperature Information Based on Multiple Meteorological Factors Active CN113639893B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110725482.7A CN113639893B (en) 2021-06-29 2021-06-29 A Method for Obtaining Near-Earth Weighted Average Temperature Information Based on Multiple Meteorological Factors

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110725482.7A CN113639893B (en) 2021-06-29 2021-06-29 A Method for Obtaining Near-Earth Weighted Average Temperature Information Based on Multiple Meteorological Factors

Publications (2)

Publication Number Publication Date
CN113639893A CN113639893A (en) 2021-11-12
CN113639893B true CN113639893B (en) 2022-09-30

Family

ID=78416254

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110725482.7A Active CN113639893B (en) 2021-06-29 2021-06-29 A Method for Obtaining Near-Earth Weighted Average Temperature Information Based on Multiple Meteorological Factors

Country Status (1)

Country Link
CN (1) CN113639893B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114662665A (en) * 2022-03-28 2022-06-24 神华北电胜利能源有限公司 Mining area atmospheric water vapor density obtaining method, device and equipment
CN116027459B (en) * 2022-12-30 2024-05-14 桂林理工大学 Calculation method of atmospheric weighted mean temperature based on numerical weather forecast data
CN117390875B (en) * 2023-10-27 2024-03-12 长安大学 Construction method of atmosphere weighted average temperature model
CN118897932B (en) * 2024-09-30 2025-01-21 杭州电子科技大学 A typhoon monitoring method based on PWV and multi-parameter comprehensive weighting strategy

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102156007B (en) * 2011-02-14 2013-02-20 国家粮食局科学研究院 Grain condition prediction system, grain condition control system and grain bulk safety state judging method
CN103033833B (en) * 2012-12-13 2014-12-10 中国航天科工信息技术研究院 Method of correcting troposphere delaying errors
CN109145344A (en) * 2018-03-06 2019-01-04 东南大学 A kind of experience ZTD model refinement method based on sounding data
CN108680268B (en) * 2018-03-20 2020-07-31 东南大学 An improved method of Bevis model for regional weighted average temperature based on sounding data
CN109902346B (en) * 2019-01-24 2023-04-07 东南大学 Neural network-based regional weighted average temperature information acquisition method
CN110378540B (en) * 2019-08-02 2023-05-09 桂林理工大学 A Calculation Method of Atmospheric Weighted Average Temperature Applicable to Beibu Gulf Area of Guangxi
CN111539109B (en) * 2020-04-24 2021-05-21 桂林理工大学 Real-time high-precision global multi-dimensional troposphere zenith delay grid model construction method
CN112034490B (en) * 2020-10-10 2022-03-22 山东科技大学 An improved method for NWP inversion of tropospheric delay

Also Published As

Publication number Publication date
CN113639893A (en) 2021-11-12

Similar Documents

Publication Publication Date Title
CN113639893B (en) A Method for Obtaining Near-Earth Weighted Average Temperature Information Based on Multiple Meteorological Factors
CN109001845B (en) A Prediction Method of Rainfall
CN109902346B (en) Neural network-based regional weighted average temperature information acquisition method
CN108983320A (en) A kind of numerical weather forecast-artificial intelligence coupling prediction method of coastal typhoon Maximum wind speed
CN112034490B (en) An improved method for NWP inversion of tropospheric delay
CN109978275B (en) Extreme strong wind speed prediction method and system based on mixed CFD and deep learning
CN108680268B (en) An improved method of Bevis model for regional weighted average temperature based on sounding data
CN111126466B (en) Multi-source PWV data fusion method
WO2021164480A1 (en) Spatial location-based weighted mean temperature calculation method
CN111382507B (en) Global troposphere delay modeling method based on deep learning
CN111539109A (en) Real-time high-precision global multi-dimensional troposphere zenith delay grid model construction method
CN113534194A (en) A tropospheric temperature and humidity profile retrieval method based on combined GNSS and wind Lidar
CN117669171A (en) An improved GNSS tropospheric delay real-time grid model construction method and system
CN115755103A (en) Robust self-adaptive GNSS (Global navigation satellite System) water vapor chromatography method
CN115203625A (en) A method for imputing missing values of drought and flood index data and its imputation device
CN118897932B (en) A typhoon monitoring method based on PWV and multi-parameter comprehensive weighting strategy
CN105868443A (en) Construction method for microtopographical micrometeorological element field
CN111753408A (en) A Weather-Considering GNSS Atmospheric Weighted Average Temperature Calculation Method
CN111274707A (en) Weighted average temperature calculation method based on reanalysis data and wireless sounding data
CN111538943B (en) A new method for constructing a grid model of global ZTD vertical profile with high spatial and temporal resolution
CN106546958A (en) A kind of radar data assimilation method of optimization
CN115993668B (en) Polynomial correction and neural network-based PWV reconstruction method and system
CN117113828A (en) A numerical prediction correction method based on ship-based traveling observations
CN115982589A (en) Method and equipment for acquiring region weighted average temperature information based on LSTM model
CN114528708A (en) Basin erosion sand production index simulation method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant