A rainfall inversion method integrating GNSS water vapor tomography and hydraulic rainfall radar

Abstract

This invention provides a rainfall inversion method that integrates GNSS water vapor tomography and hydrological rainfall radar. The method includes: collecting GNSS data, hydrological rainfall radar data, meteorological data, and geographic information; estimating the atmospheric water vapor content (SWV) along the GNSS slant signal path; obtaining grid water vapor density values ​​through three-dimensional water vapor tomography; identifying inversion factors; and training a neural network to perform rainfall inversion. This rainfall inversion method effectively utilizes GNSS water vapor tomography and hydrological rainfall radar data, comprehensively considering the relationship between atmospheric and liquid water in time, optimizing the rainfall inversion process, and providing new approaches for the refined inversion of rainfall intensity and spatial distribution.

Description

Technical Field

[0001] The present invention relates to the field of water conservancy rainfall inversion, and in particular to a rainfall inversion method integrating GNSS water vapor tomography and water conservancy rainfall radar. Background Art

[0002] Rainfall inversion is extremely important in many fields, including weather forecasting and disaster prevention and mitigation, water resources management, environmental monitoring, and water conservancy project planning. Currently, traditional rainfall inversion methods are mainly based on statistical forecasting methods, numerical weather forecasting methods, and radar echo extrapolation methods. These methods suffer from low inversion accuracy, low temporal resolution, and short rainfall forecast periods. To address these issues with traditional rainfall inversion, recent studies have applied ground-based Global Navigation Satellite System (GNSS) water vapor inversion technology to rainfall inversion. This, combined with remote sensing data, provides more meteorological information for rainfall inversion, effectively improving rainfall inversion accuracy and temporal resolution, and ultimately enhancing the level of rainfall inversion.

[0003] GNSS water vapor retrieval technology offers the advantages of low cost, all-day, all-weather coverage, high temporal and spatial resolution, and high accuracy, making it a powerful complement to traditional water vapor detection methods. GNSS two-dimensional water vapor retrieval can capture the overall water vapor content over a region, typically measured as Precipitable Water Vapor (PWV), the total amount of water vapor contained in a column of air per unit area. However, GNSS two-dimensional water vapor retrieval cannot accurately reflect the distribution of water vapor in three-dimensional space and cannot meet the needs for analyzing three-dimensional water vapor content and temporal and spatial variations.

[0004] The basic principle of GNSS water vapor tomography is to calculate the atmospheric water vapor content (SWV) along the GNSS signal's slant path using a model function. The tomography region is discretized into a three-dimensional grid in three-dimensional space. An integral equation based on the GNSS signal's slant path is formulated and discretized to each 3D grid cell traversed by the GNSS signal, forming the tomography observation equation. Given sufficient observational information, the water vapor density of each 3D grid cell can be solved. Compared to two-dimensional water vapor inversion, water vapor tomography enables three-dimensional reconstruction of the atmospheric water vapor distribution over a region. By revealing the spatiotemporal variations of the three-dimensional water vapor field, it helps capture the accumulation and transport of water vapor before rainfall, providing a more comprehensive and accurate scientific basis for rainfall inversion. However, three-dimensional tomography is computationally complex and subject to numerous constraints, including tomography data selection, tomography region geometry, tomography constraint settings, and GNSS observation network density. Consequently, challenges remain in the spatial resolution, timeliness, reliability, and stability of 3D water vapor tomography results.

[0005] Water conservancy rainfall radars are X-band radars that generate real-time, gridded rainfall data for liquid water in the atmosphere from the ground level to 2 km above the surface. This directly captures rainfall intensity information, facilitating quantitative precipitation estimation and forecasting for the water conservancy industry. However, these radars also have drawbacks. Their small number and limited coverage make radar signals susceptible to obstruction in complex terrain. While radar networking can effectively mitigate blind spots, interference between radars and the complexity of fusing data from overlapping areas can still lead to inaccurate regional rainfall intensity and distribution. Furthermore, radar reflectivity is sensitive to the size and number of precipitation particles, potentially leading to inaccurate rainfall intensity estimates. Summary of the Invention

[0006] To overcome the shortcomings of existing rainfall inversion methods in terms of spatiotemporal resolution, accuracy, and rainfall inversion, the present invention provides a rainfall inversion method that integrates GNSS water vapor tomography and hydraulic rain radar. The method divides the tomographic grid based on the consistency of the GNSS water vapor tomography results and the resolution of the hydraulic rain radar data. This method fully utilizes the advantages of GNSS three-dimensional water vapor tomography, such as its high spatiotemporal resolution and ability to provide atmospheric water vapor content information at different altitudes, as well as the hydraulic rain radar's ability to directly detect rainfall particles. By learning the spatiotemporal characteristics of the two data and rainfall through a neural network, the advantages of the two data are complemented. This method, supplemented by other data, allows for high-precision spatiotemporal rainfall inversion, extending the rainfall forecast period and improving rainfall inversion accuracy. The specific solutions of the present invention are as follows:

[0007] A rainfall inversion method integrating GNSS water vapor tomography and hydraulic rainfall radar comprises the following steps:

[0008] Step 1: Data collection: including GNSS data, water conservancy rainfall radar data, meteorological data and geographic information;

[0009] Step 2: Estimate the atmospheric water vapor content SWV on the GNSS slant signal path;

[0010] Step 3: Obtaining grid water vapor density values ​​using three-dimensional water vapor tomography: The tomography region is discretized and gridded according to the resolution of the hydrological rainfall radar data to obtain n three-dimensional voxel blocks. A GNSS water vapor tomography model is then constructed and solved based on the three-dimensional tomography algorithm to obtain the water vapor density value of each grid. The expression of the GNSS water vapor tomography model is:

[0011]

[0012] Where A is the coefficient matrix, representing the intercept information of the GNSS signal in each voxel block; B is the SWV observation value vector; X is the water vapor density vector; H is the coefficient matrix of horizontal constraints, V is the coefficient matrix of vertical constraints, I is the coefficient matrix of prior constraints, and C is the initial water vapor density value of the tomographic area;

[0013] Step 4: Identify inversion factors: Based on the water vapor density value of each grid, integrate vertically to obtain the atmospheric precipitable water (PWV) for each grid. Use PWV and its related data, as well as water conservancy rainfall radar data, as the main inversion factors. Meteorological data and elevation data are used as auxiliary inversion factors to participate in rainfall inversion. Use linear regression to analyze the correlation between each factor and the measured rainfall data to determine the actual rainfall inversion factor.

[0014] Step 5: Train the neural network and perform rainfall inversion: Use the rainfall inversion factor and actual rainfall data as input layer data, and the predicted rainfall as output layer data to construct a neural network. Train the neural network, learn the spatiotemporal characteristics between the rainfall inversion factor and the measured rainfall, obtain the optimal model parameters, construct a multi-factor rainfall inversion model, and perform rainfall inversion.

[0015] Further optimization, in step 1, the GNSS data includes GNSS raw observation data and IGS precise orbit clock data; the water conservancy rain radar data includes water conservancy rain radar base data Level 2 and product data Level 3, and the water conservancy rain radar base data Level 2 includes reflectivity factor Z, differential reflectivity factor Z DR , differential propagation phase Differential propagation phase shift rate K DP The Level 3 product data includes surface rainfall inversion products, 0-2h surface rainfall forecast products, rainfall-related particle phase classification, and liquid water content data; the meteorological data includes measured rainfall, air pressure, temperature, and relative humidity; and the geographic information includes elevation and longitude and latitude.

[0016] Furthermore, the estimation of the atmospheric water vapor content SWV on the GNSS slant signal path in step 2 includes the following steps:

[0017] Step 2-1: Solve the GNSS data based on the precise point positioning technology, use the VMF1 model as the zenith mapping function model, and obtain the zenith tropospheric total delay (ZTD);

[0018] Step 2-2: Calculate the zenith static delay ZHD using the Saastamoinen model. The expression is:

[0019]

[0020] Among them, P s is the air pressure value at the underlying surface of the measuring station, hPa; θ is the latitude of the measuring station; H is the height of the measuring station, m;

[0021] Step 2-3: Calculate the zenith wet delay ZWD in mm using the expression:

[0022] ZED=ZTD-ZHD

[0023] Step 2-4: Calculate the wet delay (SWD) of the oblique path in mm using the following expression:

[0024] SWD=M w (ε)×ZWD+M g (ε)×[G NS cos(α)+G EW sin(α)]+R

[0025] Where ε and α represent the satellite cutoff elevation angle and azimuth angle respectively; M w and M g Represent the wet mapping function and gradient mapping function respectively; G NS and G EW denote the moist horizontal gradient terms in the north-south and east-west directions respectively; R denotes the residual term that is not modeled;

[0026] Step 2-5: Calculate the atmospheric water vapor content SWV on the oblique signal path in mm using the expression:

[0027] SWV=Π×SWD

[0028] Among them, П is the atmospheric water vapor conversion coefficient, which is a dimensionless value and its calculation formula is:

[0029]

[0030] Where ρ is the density of liquid water, kg / m 3 ; R v is the water vapor gas constant, which is 461.495 J / (kg·K); k'2 and k3 are atmospheric refractive index constants, which are 16.48 K / hPa and 3.776×10 5 K 2 / hPa; T m is the weighted mean temperature of the atmosphere, K;

[0031] Furthermore, in step three, the tomography area is discretized and gridded based on the resolution of the water conservancy rain measuring radar data: including determining the layer height, horizontal resolution and vertical resolution of the tomography area, discretizing the geometric space and regional water vapor elements of the tomography area; setting the heights of the bottom and top of the grid to 0km and 11km respectively; the horizontal range tomography grid division needs to ensure that the horizontal resolution of the GNSS water vapor tomography is consistent with the horizontal resolution of the water conservancy rain measuring radar; the vertical direction tomography area grid division adopts an uneven division method of dense at the bottom and sparse at the top based on the distribution characteristics of water vapor in the vertical direction, and the grid division of the tomography area below 2km is encrypted to ensure that the vertical resolution of the GNSS water vapor tomography is consistent with the vertical resolution of the water conservancy rain measuring radar.

[0032] Furthermore, in step three, the horizontal constraint assumes that the water vapor value of a voxel block in the horizontal direction is the weighted average of the water vapor parameters of several adjacent voxel blocks in the same layer based on the Gaussian weighted function; the vertical constraint uses the exponentially decreasing change characteristics of water vapor in the vertical direction to construct the vertical constraint relationship between two adjacent layers; the prior constraint obtains the initial field of atmospheric water vapor as a constraint condition.

[0033] Furthermore, in step 4, the PWV-related data includes: PWV change and PWV change rate; the water conservancy rain measuring radar data includes the basic data Level 2 and product data Level 3 of the water conservancy rain measuring radar.

[0034] Furthermore, in step five, the neural network: combines the convolutional neural network CNN and the long short-term memory neural network LSTM model structure, extracts the spatial features between each rainfall inversion factor and the actual rainfall based on CNN, and combines the LSTM model to capture the temporal features between each rainfall inversion factor and the actual rainfall, so as to enhance the ability of rainfall spatiotemporal inversion.

[0035] Beneficial effects of the present invention:

[0036] The present invention proposes a rainfall inversion method that integrates GNSS water vapor tomography and hydraulic rain radar, which accurately captures the water vapor accumulation and transport process in the three-dimensional space before rainfall, comprehensively considers the relationship between spatial and temporal air water and liquid water, enhances the correlation between water vapor and rainfall, and combines the direct characterization of rainfall by hydraulic rain radar to optimize the rainfall inversion process, improve the accuracy and spatiotemporal resolution of rainfall inversion, enhance the ability of rainfall forecasting, make up for the shortcomings of single data, extend the rainfall forecast period, and provide new ideas for the refined inversion of rainfall intensity and spatial distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flowchart of rainfall inversion using integrated GNSS water vapor tomography and hydraulic rainfall radar;

[0038] Figure 2Schematic diagram of the rainfall inversion model based on GNSS water vapor tomography and water conservancy rainfall radar. DETAILED DESCRIPTION

[0039] The present invention is further described in detail below with reference to the accompanying drawings:

[0040] Example 1

[0041] like Figure 1 The process shown is a rainfall inversion method that integrates GNSS water vapor tomography and hydraulic rainfall radar, including the following steps:

[0042] Step 1: Data collection: including GNSS data, water conservancy rainfall radar data, meteorological data and geographic information; including the following operations:

[0043] (1.1) GNSS data collection

[0044] Obtain GNSS raw observation data and IGS precise orbit and clock error data.

[0045] (1.2) Water conservancy rainfall radar data collection

[0046] Obtain water conservancy rainfall radar base data (Level 2) including reflectivity factor (Z), differential reflectivity factor (Z DR ), differential propagation phase Differential propagation phase shift rate (K DP ), product data (Level 3) includes surface rainfall inversion products, 0-2h surface rainfall forecast products, rainfall-related particle phase classification, liquid water content and other data.

[0047] (1.3) Other data collection

[0048] Obtain measured meteorological data such as rainfall, air pressure, temperature, relative humidity, etc.; obtain regional elevation information, longitude and latitude, and other geographic information.

[0049] Step 2: Estimate the atmospheric water vapor content (SWV) along the GNSS slant signal path, including the following operations:

[0050] (2.1) Based on the precise point positioning technology, the VMF1 model is used as the zenith mapping function model to obtain the zenith tropospheric total delay (ZTD);

[0051] (2.2) The zenith static delay ZHD is calculated using the Saastamoinen model, which is expressed as:

[0052]

[0053] Among them, P s is the air pressure value at the underlying surface of the measuring station, hPa; θ is the latitude of the measuring station; H is the height of the measuring station, m.

[0054] (2.3) Calculate the zenith wet delay ZWD in mm using the expression:

[0055] ZWD=ZTD-ZHD

[0056] (2.4) The slant path wet delay SWD can be recovered from the ZWD and horizontal wet gradient delay terms, as shown in the following expression:

[0057] SWD=M w (ε)×ZWD+M g (ε)×[G NS cos(α)+G EW sin(α)]+R

[0058] Where ε and α represent the satellite cutoff elevation angle and azimuth angle respectively; M w and M g Represent the wet mapping function and gradient mapping function respectively; G NS and G EW represent the moist horizontal gradient terms in the north-south and east-west directions respectively; R represents the residual term that is not modeled.

[0059] (2.5) Calculate the atmospheric water vapor content SWV in the oblique path in mm using the expression:

[0060] SWV=II×SWD

[0061] Among them, П is the atmospheric water vapor conversion coefficient, which is a dimensionless value and its calculation formula is:

[0062]

[0063] Where ρ is the density of liquid water, kg / m 3 ; R v is the water vapor gas constant, which is 461.495 J / (kg·K); k2' and k3 are atmospheric refractive index constants, which are 16.48 K / hPa and 3.776×10 5 K 2 / hPa; T m is the weighted mean temperature of the atmosphere, K.

[0064] There is an obvious linear relationship between the atmospheric weighted mean temperature and the surface temperature, which can be expressed as:

[0065] T m =a×T s b

[0066] The coefficients a and b in the formula can be calculated based on the long-term surface temperature T of the study area. sand the corresponding atmospheric weighted average temperature, determine its value by the least square method, and establish the region T m Model.

[0067] Step 3: Obtain grid water vapor density values ​​using 3D water vapor tomography: Based on the resolution of the hydro-rainfall radar data, the tomography region is discretized and gridded to obtain n 3D voxel blocks. A GNSS water vapor tomography model is then constructed and solved based on the 3D tomography algorithm to obtain the water vapor density value for each grid. The specific steps are as follows:

[0068] (3.1) The horizontal area of ​​the tomography should be slightly larger than the GNSS station network to ensure that the GNSS signals of each time period are included in the tomography area and that the GNSS signals pass through the top of the layer. Based on the empirical method, the block method is used to perform regional tomography modeling for the area below 11 km in the troposphere covered by the GNSS signals.

[0069] (3.2) Discretization of the tomographic region mainly involves determining the tomographic region layer height, horizontal resolution, and vertical resolution, discretizing the tomographic region geometric space and regional water vapor elements. The grid bottom and top heights are set to 0 km and 11 km, respectively. The tomographic region grid division strategy varies depending on the hydro-rainfall radar data used. When using phased array rainfall radar data, the horizontal tomographic region grid division adopts a uniform division method of 30×30 m along longitude and latitude; when using mechanical rainfall radar data, a uniform division method of 75×75 m along longitude and latitude is used. The horizontal tomographic grid division must ensure that the horizontal resolution of the GNSS water vapor tomography is consistent with that of the hydro-rainfall radar. The vertical tomographic region grid division adopts a non-uniform division method with denser bottom and sparser top based on the vertical distribution characteristics of water vapor. The grid division of the tomographic region below 2 km is denser to ensure that the vertical resolution of the GNSS water vapor tomography is consistent with that of the hydro-rainfall radar.

[0070] (3.3) According to the discretization principle of the tomographic region, n three-dimensional voxel blocks can be obtained. Based on the voxel blocks, the functional relationship between water vapor parameters and SWV observation values ​​is established, and the tomographic observation equations are constructed:

[0071] AX=B

[0072] Where A is the coefficient matrix, representing the intercept information of the GNSS signal within each voxel block; B is the SWV observation value vector; and X is the water vapor density vector.

[0073] Due to the lack of sufficient observation data and the uneven spatial distribution of GNSS signal slant paths, observation equations constructed directly using GNSS signals suffer from non-unique solutions, instability, and other inadequacies. Physical or a priori constraint equations are needed to constrain horizontal and vertical directions, as well as boundary water vapor information or initial water vapor density, to eliminate these inadequacies.

[0074] (3.4) Construct tomographic constraint equations, including horizontal constraints, vertical constraints, and prior constraints, and solve the water vapor tomography observation equations. Horizontal constraints are usually based on the Gaussian weighting function, assuming that the water vapor value of a certain voxel block in the horizontal direction is the weighted average of the water vapor parameters of several adjacent voxel blocks in the same layer; vertical constraints use the exponentially decreasing change characteristics of water vapor in the vertical direction to construct the vertical constraint relationship between two adjacent layers; prior constraints generally use radiosondes, microwave radiometers, and other methods to obtain the initial atmospheric water vapor field as a constraint condition. Construct a GNSS water vapor tomography model, expressed as:

[0075]

[0076] Where H is the coefficient matrix of horizontal constraints, V is the coefficient matrix of vertical constraints, I is the coefficient matrix of prior constraints, and C is the initial water vapor density value of the tomographic area obtained by methods such as radiosonde and microwave radiometer.

[0077] Step 4: Identify the inversion factor:

[0078] The water vapor density data of each grid obtained by three-dimensional water vapor tomography is integrated in the vertical direction to obtain the PWV data of each grid. The PWV, PWV change, PWV change rate, water conservancy rainfall radar reflectivity factor (Z), differential propagation phase shift rate (K) of each grid are determined. DP ) as the necessary rainfall inversion factor, combined with the basic data (Level 2) and product data (Level 3) of other water conservancy rainfall radars as the main inversion factors, and coordinated with air pressure, temperature, longitude, latitude, elevation and other data as auxiliary factors to participate in rainfall inversion. The linear regression method is used to analyze the correlation between each factor and the measured rainfall data to determine the actual rainfall inversion factor.

[0079] Step 5: Train the neural network and perform rainfall inversion:

[0080] (5.1) Neural network design, such as Figure 2 As shown, rainfall inversion factors and actual rainfall data serve as input layer data, and predicted rainfall amounts serve as output layer data. For example, a convolutional neural network (CNN) and a long short-term memory (LSTM) neural network model can be combined. The CNN can be used to extract spatial features between each rainfall inversion factor and actual rainfall, while the LSTM model can be used to capture temporal features between each rainfall inversion factor and actual rainfall, thereby enhancing the spatiotemporal rainfall inversion capability. Furthermore, other neural network structures and model algorithms can be explored based on this foundation and flexibly selected based on actual needs.

[0081] (5.2) Neural network training: Divide the training set into a training set and a test set. Use the training set sample data to train the neural network model, learn the spatial and temporal characteristics of rainfall inversion factors and measured rainfall, continuously adjust the model parameters, use the validation set data to evaluate the accuracy of the model-predicted rainfall data, determine the optimal model parameters, and establish a rainfall inversion model that integrates GNSS water vapor tomography and hydraulic rainfall radar to perform rainfall inversion.

[0082] (5.3) Model accuracy assessment: statistically calculate the accuracy, error rate, root mean square error (RMSE) and mean absolute error (MAE) of rainfall inversion.

[0083] The above-mentioned embodiments are only partial embodiments of the present invention and cannot cover the entire present invention. Based on the above-mentioned examples and drawings, those skilled in the art can obtain more implementation methods without paying any creative work. Therefore, these implementation methods obtained without paying any creative work should be included in the scope of protection of the present invention.

Claims

1. A rainfall inversion method integrating GNSS water vapor tomography and hydraulic rainfall radar, characterized in that: The following steps are involved: Step 1: Data collection: including GNSS data, water conservancy rainfall radar data, meteorological data and geographic information; Step 2: Estimate the atmospheric water vapor content SWV on the GNSS slant signal path; Step 3: Obtaining grid water vapor density values ​​using three-dimensional water vapor tomography: The tomography region is discretized and gridded according to the resolution of the hydrological rainfall radar data to obtain n three-dimensional voxel blocks. A GNSS water vapor tomography model is then constructed and solved based on the three-dimensional tomography algorithm to obtain the water vapor density value of each grid. The expression of the GNSS water vapor tomography model is: Where A is the coefficient matrix, representing the intercept information of the GNSS signal in each voxel block; B is the SWV observation value vector; X is the water vapor density vector; H is the coefficient matrix of horizontal constraints, V is the coefficient matrix of vertical constraints, I is the coefficient matrix of prior constraints, and C is the initial water vapor density value of the tomographic area; Step 4: Identify inversion factors: Based on the water vapor density value of each grid, integrate vertically to obtain the atmospheric precipitable water (PWV) for each grid. Use PWV and its related data, as well as water conservancy rainfall radar data, as the main inversion factors. Meteorological data and elevation data are used as auxiliary inversion factors to participate in rainfall inversion. Use linear regression to analyze the correlation between each factor and the measured rainfall data to determine the actual rainfall inversion factor. Step 5: Train the neural network and perform rainfall inversion: Use the rainfall inversion factor and actual rainfall data as input layer data, and the predicted rainfall as output layer data to construct a neural network. Train the neural network, learn the spatiotemporal characteristics between the rainfall inversion factor and the measured rainfall, obtain the optimal model parameters, construct a multi-factor rainfall inversion model, and perform rainfall inversion. In step 1, the GNSS data includes GNSS raw observation data and IGS precise orbit clock data; the water conservancy rain radar data includes water conservancy rain radar base data Level 2 and product data Level 3. The water conservancy rain radar base data Level 2 includes reflectivity factor Z, differential reflectivity factor Z DR , differential propagation phase Differential propagation phase shift rate K DP The product data Level 3 includes surface rainfall inversion products, 0-2h surface rainfall forecast products, rainfall-related particle phase classification, and liquid water content data; the meteorological data includes measured rainfall, air pressure, temperature, and relative humidity; the geographic information includes elevation and longitude and latitude; In step 4, PWV-related data include: PWV change and PWV change rate; Water conservancy rainfall radar data includes basic data Level 2 and product data Level 3 of water conservancy rainfall radar.

2. The rainfall inversion method integrating GNSS water vapor tomography and hydraulic rainfall radar according to claim 1 is characterized in that: In step 2, estimating the atmospheric water vapor content SWV on the GNSS slant signal path includes the following steps: Step 2-1: Solve the GNSS data based on the precise point positioning technology, use the VMF1 model as the zenith mapping function model, and obtain the zenith tropospheric total delay (ZTD); Step 2-2: Calculate the zenith static delay ZHD using the Saastamoinen model. The expression is: Among them, P s is the air pressure value at the underlying surface of the measuring station, hPa; θ is the latitude of the measuring station; H is the height of the measuring station, m; Step 2-3: Calculate the zenith wet delay ZWD in mm using the expression: ZWD=ZTD-ZHD Step 2-4: Calculate the wet delay (SWD) of the oblique path in mm using the following expression: SWD=M w (e)×ZWD+M g (e)×[G NS cos(α)+G EW sin(a)]+R Where ε and α represent the satellite cutoff elevation angle and azimuth angle respectively; M w and M g Represent the wet mapping function and gradient mapping function respectively; G NS and G EW denote the moist horizontal gradient terms in the north-south and east-west directions respectively; R denotes the residual term that is not modeled; Step 2-5: Calculate the atmospheric water vapor content SWV on the oblique signal path in mm using the expression: SWV=Π×SWD Among them, П is the atmospheric water vapor conversion coefficient, which is a dimensionless value and its calculation formula is: Where ρ is the density of liquid water in kg / m 3 ; R v is the water vapor gas constant, which is 461.495 J / (kg·K); k'2 and k3 are atmospheric refractive index constants, which are 16.48 K / hPa and 3.776×10 5 K 2 / hPa; T m is the weighted average temperature of the atmosphere, in K.

3. The rainfall inversion method integrating GNSS water vapor tomography and hydraulic rainfall radar according to claim 1 is characterized by: In step three, the tomography area is discretized and gridded based on the resolution of the water conservancy rainfall radar data: including determining the layer height, horizontal resolution and vertical resolution of the tomography area, discretizing the geometric space and regional water vapor elements of the tomography area; setting the heights of the bottom and top of the grid to 0km and 11km respectively; the horizontal range tomography grid division needs to ensure that the horizontal resolution of the GNSS water vapor tomography is consistent with the horizontal resolution of the water conservancy rainfall radar; the vertical direction tomography area grid division adopts an uneven division method of dense at the bottom and sparse at the top based on the distribution characteristics of water vapor in the vertical direction, and the grid division of the tomography area below 2km is encrypted to ensure that the vertical resolution of the GNSS water vapor tomography is consistent with the vertical resolution of the water conservancy rainfall radar.

4. The rainfall inversion method integrating GNSS water vapor tomography and hydraulic rainfall radar according to claim 1 is characterized by: In step three, the horizontal constraint assumes that the water vapor value of a voxel block in the horizontal direction is the weighted average of the water vapor parameters of several adjacent voxel blocks in the same layer based on the Gaussian weighted function; the vertical constraint uses the exponentially decreasing change characteristics of water vapor in the vertical direction to construct the vertical constraint relationship between two adjacent layers; the prior constraint obtains the initial atmospheric water vapor field as a constraint condition.

5. The rainfall inversion method integrating GNSS water vapor tomography and hydraulic rainfall radar according to claim 1 is characterized by: In step 5, the neural network: combines the convolutional neural network (CNN) and the long short-term memory (LSTM) neural network model structure, extracts the spatial features between each rainfall inversion factor and the actual rainfall based on CNN, and combines the LSTM model to capture the temporal features between each rainfall inversion factor and the actual rainfall, thereby enhancing the ability of rainfall spatiotemporal inversion.

Images