Quantitative rainfall estimation method and system based on multi-source data correction radar echo

Abstract

The invention relates to the technical field of meteorology and hydrology, and discloses a quantitative rainfall estimation method and system based on multi-source data correction radar echoes, and the method comprises the steps: collecting the multi-source data, such as rainfall station actually measured rainfall data, elevation data, hydrometric station flow, radar reflectivity, soil humidity and environmental attributes, in a target region; a Z-R relation parameter of each radar sampling grid is dynamically calibrated by combining a collaborative Kriging interpolation and physical constraint hybrid neural network model, and a radar reflectivity correction parameter of each grid is calculated by using a difference between rainfall station observation data and radar data; in real-time application, based on the corrected radar reflectivity and the calibrated Z-R parameter, the rainfall intensity of each grid is inverted, and the precision of quantitative rainfall estimation is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of meteorological hydrology, and in particular to a quantitative rainfall estimation method and system based on multi-source data correction of radar echoes. BACKGROUND

[0002] In recent years, global climate change has led to an increase in extreme rainfall events, making flood disasters caused by extreme rainfall increasingly frequent and their destructive power growing stronger, and they have become one of the natural disasters that cause the most deaths. These flood disasters usually have the characteristics of great intensity, wide range, and strong destructiveness, posing a serious threat to social and economic development and people's life and property safety.

[0003] China is one of the countries most severely affected by flood disasters. Due to the influence of complex terrain conditions (such as mountainous areas and plateaus) and monsoon climate, regional rainstorms and extreme rainfall events are particularly common. However, traditional rainfall monitoring methods mainly rely on observation data from ground hydrological stations and rain gauges. Given the vast territory and complex terrain of China, the existing rain gauge network has a relatively low distribution density, especially in remote areas such as mountainous areas and plateaus, where the sites are sparse, which directly leads to insufficient spatial representativeness of rainfall data, making it difficult to meet the needs of fine-scale runoff calculation.

[0004] Although meteorological radar technology can provide large-scale, high-temporal and high-spatial resolution rainfall monitoring data, there are still significant systematic biases in quantitative rainfall estimation based on radar reflectivity factor inversion of rainfall rate. These problems mainly include the influence of radar calibration errors, beam blocking, and changes in the phase state of precipitation particles, resulting in a large difference between radar-estimated rainfall and actual ground measurements. These problems limit the precise application of radar technology in the fields of flood forecasting and mountain flood warning.

[0005] Therefore, there is an urgent need for a quantitative rainfall estimation method and system based on multi-source data correction of radar echoes to improve the accuracy of quantitative rainfall estimation. SUMMARY

[0006] To solve the above technical problems, the present application provides a quantitative rainfall estimation method and system based on multi-source data correction of radar echoes, which can make radar inversion rainfall closer to the actual rainfall distribution and improve the accuracy of quantitative rainfall estimation.

[0007] The present application provides a quantitative rainfall estimation method based on multi-source data correction of radar echoes, comprising the following steps: S1, collecting the measured rainfall data of rain gauges in the target prediction area, elevation data, cross-section measured flow time series of downstream hydrological stations, measured radar reflectivity factor, soil moisture data, and environmental attribute data; S2, determining the Z-R relationship parameter of each radar sampling grid dynamically based on the measured rainfall data of the rain gauge station, the elevation data, the measured cross-section flow time series of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data through a physically constrained hybrid neural network model; S3, calculating the radar reflectivity correction parameter of each radar sampling grid according to the measured rainfall data of the rain gauge station, the measured radar reflectivity factor and the Z-R relationship parameter of each radar sampling grid at the position of the rain gauge station; S4, monitoring the radar reflectivity of the target prediction area in real time, calculating the rainfall data of each radar sampling grid by using the Z-R relationship parameter of each radar sampling grid and the radar reflectivity correction parameter of each radar sampling grid, and completing quantitative rainfall estimation.

[0008] Further, the S2 specifically comprises: S21, constructing a physically constrained hybrid neural network model and inputting the measured rainfall data of the rain gauge station, the elevation data, the measured cross-section flow time series of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data in the target prediction area; S22, performing collaborative Kriging interpolation processing on the measured rainfall data of the rain gauge station, and incorporating the elevation data as an auxiliary variable in the interpolation process to generate the corresponding rainfall estimation value of each radar sampling grid; S23, initializing and pre-training the Z-R relationship parameter corresponding to each radar sampling grid based on the measured rainfall data of the rain gauge station and the corresponding rainfall estimation value of each radar sampling grid, and predicting the Z-R relationship parameter; S24, using the predicted Z-R relationship parameter in S23 to perform quantitative precipitation retrieval on the radar reflectivity factor of each radar sampling grid, and obtaining the quantitative rainfall estimation value of each radar sampling grid; S25, using a distributed runoff yield model to perform watershed surface confluence calculation according to the quantitative rainfall estimation value of each radar sampling grid, and outputting the simulated flow of the downstream hydrological station cross-section; S26, constructing a loss function containing multiple physical constraint parameters according to the measured flow time series data of the downstream hydrological station cross-section, the quantitative precipitation estimation value of each radar sampling grid and the simulated flow of the downstream hydrological station cross-section; S27, dynamically determining the Z-R relationship parameter of each radar sampling grid through an iterative optimization process until the loss function containing multiple physical constraint parameters converges.

[0009] Further, the S23 specifically comprises: S231, extracting spatial features of the reflectivity factor time series data of the radar sampling grid through a convolutional layer; S232, extracting time series features of the reflectivity factor time series data of the radar sampling grid and the cross section flow time series of the downstream hydrological station through an LSTM layer; S233, calculating the concentration time based on the environmental attribute data of the radar sampling grid; S234, inputting the spatial features, the time series features, the soil moisture and the concentration time into a fully connected layer to predict the Z-R relationship parameters.

[0010] Further, in S22, the interpolation formula of the collaborative Kriging interpolation method is: ; wherein, wherein, is the rainfall estimation value of the radar sampling grid x0, is the rainfall data of the i-th rainfall station, n is the total number of rainfall stations; E(x0) is the elevation data of the radar sampling grid x0, and is the weight coefficient of the collaborative Kriging. Further, the distributed runoff yield model is a grid-based Xinanjiang model.

[0011] Further, the loss function containing multiple physical constraint parameters is: ; wherein, indicates the loss function containing multiple physical constraint parameters; Q obs indicates the measured flow of the cross section of the downstream hydrological station; Q sim indicates the simulated flow of the cross section of the downstream hydrological station; Z i indicates the reflectivity factor of the i-th radar sampling grid; a i , b i indicates the parameters of the Z-R relationship of the i-th radar sampling grid; R i indicates the rainfall data of the i-th radar sampling grid calculated based on the Z-R relationship; S i indicates the grid area of the i-th radar sampling grid; ɑ, β, γ are parameters of the loss function .

[0012] Further, the value range of ɑ, β, γ is: , used to control the flow error weight; , used to constrain the consistency of Z-R relationship; , used to force water balance.

[0013] Further, the S3 specifically comprises: S31, at each rain gauge position, according to the Z-R relationship parameters of each radar sampling grid and the measured rainfall data of each radar sampling grid corresponding to the rain gauge, the theoretical value of the radar reflectivity factor is calculated through the Z-R relationship; The Z-R relationship is: ; wherein Z is the theoretical value of the radar reflectivity factor, R is the rainfall data corresponding to the radar sampling grid; a, b are the Z-R relationship parameters of the radar sampling grid; S32, according to the theoretical value of the radar reflectivity factor and the measured radar reflectivity factor, the radar reflectivity correction parameter of each radar sampling grid is calculated.

[0014] Further, the S4 specifically comprises: S41, real-time monitoring of the radar reflectivity of the target prediction area; S42, using the Z-R relationship parameters of each radar sampling grid and the radar reflectivity correction parameters of the corresponding radar sampling grid, the radar reflectivity of each radar sampling grid after correction is generated; S43, according to the radar reflectivity of each radar sampling grid after correction and the Z-R relationship parameters of the corresponding radar sampling grid determined in S2, the rainfall data of each radar sampling grid is calculated; S44, according to the rainfall data of each radar sampling grid, the rainfall estimation in the entire prediction area is completed.

[0015] The application also provides a quantitative rainfall estimation system for correcting radar echo based on multi-source data, comprising: A data collection module is used to collect the measured rainfall data of the rain gauge in the target prediction area, the elevation data, the measured flow time series of the section of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data; A physical constraint hybrid model neural network training module is used to dynamically determine the Z-R relationship parameters of each radar sampling grid based on the measured rainfall data of the rain gauge, the elevation data, the measured flow time series of the section of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data through the physical constraint hybrid neural network model; A radar reflectivity correction parameter calculation module is used to calculate the radar reflectivity correction parameter of each radar sampling grid according to the measured rainfall data of the rain gauge at each rain gauge position, the measured radar reflectivity factor and the Z-R relationship parameters of each radar sampling grid; The quantitative rainfall estimation module is used for monitoring radar reflectivity of a target prediction area in real time, calculating rainfall data of each radar sampling grid by using Z-R relationship parameters of each radar sampling grid and radar reflectivity correction parameters of each radar sampling grid, and completing quantitative rainfall estimation. The present application has the following technical effects: By introducing the measured rainfall data of the rain gauge station in the target area to dynamically correct the radar echo reflectivity factor, the systematic deviation in radar observation is effectively reduced, the consistency between the radar retrieved rainfall and the ground measured value is improved, and the accuracy of radar quantitative rainfall estimation is significantly enhanced. By fusing multi-source heterogeneous observation data, including rain gauge measured data, downstream hydrological station section flow time series, soil moisture, elevation, slope and Manning coefficient and other environmental attribute parameters, a hybrid neural network model with physical constraint mechanism is constructed, the collaborative calibration of Z-R relationship parameters is realized, the adaptability of the model to complex precipitation process is enhanced, and the physical rationality and precision of quantitative rainfall estimation are improved. And by using the grid independent parameter calibration strategy, the Z-R relationship parameters are dynamically optimized for each radar sampling grid, and the radar reflectivity factor is finely corrected in combination with the grid rainfall estimation value generated by the collaborative Kriging interpolation, so that the spatial resolution and local detail expression ability of the rainfall retrieval result are significantly improved, and the spatial accuracy can reach the order of 1km*1km. It provides high spatio-temporal resolution and high precision rainfall input data support for fine monitoring and early warning of extreme rainfall events and simulation of production and confluence of distributed hydrological models. Finally, the parameter calibration and correction method constructed has good generalization ability and regional adaptability, and can be applied to regions with different climate zones, terrain conditions and underlying surface characteristics, and has strong universality and business application potential. It provides a reliable and robust technical solution for meteorological and hydrological departments in flood forecasting, mountain flood disaster warning, urban waterlogging prevention and control, and water resources scheduling. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0017] Figure 1 is a flowchart of the quantitative rainfall estimation method based on multi-source data correction of radar echo provided by the embodiment of the present application.

[0018] Figure 2 is a schematic diagram of the quantitative rainfall estimation system based on multi-source data correction of radar echo provided by the embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions of the present application will be clearly and completely described below. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0020] Figure 1 is a flowchart of the quantitative rainfall estimation method based on multi-source data correction of radar echo provided by the embodiment of the present application. Referring to Figure 1 , specifically comprising: S1, collecting the measured rainfall data of the rain gauge in the target prediction area, elevation data, cross-section measured flow time series of the downstream hydrological station, measured radar reflectivity factor, soil moisture data and environmental attribute data; In the embodiments of the present disclosure, first, the measured rainfall data recorded by the rain gauges distributed in the target prediction area is obtained, the measured rainfall data is the cumulative rainfall or rainfall intensity of each time period (such as 5 minutes), which is used for subsequent collaborative Kriging interpolation and radar reflectivity factor correction; second, the digital elevation model (DEM) data of the target area is obtained, from which the elevation information of each radar sampling grid and the position of the rain gauge is extracted as an auxiliary variable in the collaborative Kriging interpolation, to reflect the influence of terrain on the spatial distribution of rainfall. Third, the cross-section measured flow time series observed by the hydrological station located downstream of the target area is collected, and the flow data is continuous time period flow record, which is used for subsequent simulation of the runoff process and calibration of the Z-R relationship parameter in the physically constrained hybrid neural network model.

[0021] The elevation data is obtained by digital elevation model (DEM), which is used to extract slope and elevation; at the same time, the meteorological radar observation data corresponding to the target area is obtained, including the measured radar reflectivity factor after preliminary quality control (such as denoising, ground object echo elimination, beam shielding correction, etc.), the spatial resolution can be 1km×1km, the time resolution can be 5, which is consistent with the time resolution of the measured rainfall data, as the main remote sensing input of quantitative rainfall inversion. In addition, the soil moisture data corresponding to each radar sampling grid in the target area is obtained, which can be obtained through remote sensing inversion product (such as SMAP, ASCAT) or land data assimilation system (such as GLDAS), which is used to reflect the influence of the underlying surface below the ground on the rainfall response. Finally, the environmental attribute data related to the radar sampling grid is collected, including but not limited to the surface slope (obtained by DEM calculation) and the soil Manning coefficient, which is used to construct the input features of the physically constrained hybrid neural network, and enhance the physical characterization ability of the model to the regional hydrological process.

[0022] All the above data need to be unified in space-time registration, unified to the same projection coordinate system, spatial resolution (based on radar grid) and time step, to ensure the matching of time stamp and spatial grid.

[0023] S2, based on the measured rainfall data of the rain gauge, the elevation data, the measured section measured flow time series of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data, dynamically determining the Z-R relationship parameters of each radar sampling grid through a physically constrained hybrid neural network model.

[0024] In some embodiments, the S2 specifically comprises: S21, constructing a physically constrained hybrid neural network model, and inputting the measured rainfall data of the rain gauge, the elevation data, the section measured flow time series of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data in the target prediction area; Specifically, the input environmental attribute data includes: surface slope, soil Manning coefficient.

[0025] S22, performing collaborative Kriging interpolation processing on the measured rainfall data of the rain gauge, and incorporating the elevation data as an auxiliary variable in the interpolation process to generate the corresponding rainfall estimation value of each radar sampling grid.

[0026] In the embodiments of the present disclosure, in order to improve the accuracy of rainfall spatial interpolation, especially in areas with significant topographic relief, a collaborative Kriging (Cokriging) interpolation method considering the influence of terrain is adopted to interpolate the measured rainfall data of the rain gauge to all radar sampling grid points to generate a continuous gridded rainfall estimation field.

[0027] Specifically, first, the target area is divided into regular grids (such as 1km x 1km) consistent with the spatial resolution of radar data, and each grid center point is taken as a target position to be interpolated. The measured rainfall P(x i ) of each rain gauge position (i=1, 2, …, n, n is the total number of rain gauges) and the elevation value E(x i ) of the corresponding position are obtained, and the elevation value E(x0) of each radar sampling grid center point x0 is extracted, which is derived from a digital elevation model (DEM). The interpolation formula of the collaborative Kriging interpolation method is: ; Wherein, is the rainfall estimation value of the radar sampling grid x0, is the rainfall data of the i-th rain gauge, n is the total number of rain gauges; E(x0) is the elevation data of the radar sampling grid x0, and The weight coefficient of the cokriging (the weight coefficient of the cokriging is determined by solving a linear equation set under the structure of the covariance function, so that the unbiasedness and optimality conditions of the minimum estimation variance are met, and the cokriging interpolation process can be realized through a standard statistical algorithm). Finally, the interpolation method is applied to all radar sampling grids to generate a spatio-temporally matched gridded rainfall estimation field covering the entire target area. The rainfall estimation value not only integrates the real rainfall information of the ground observation, but also effectively reflects the influence of terrain uplift on the spatial distribution of precipitation by introducing the elevation covariant, especially significantly improving the interpolation accuracy in mountainous or complex terrain areas, providing high-quality, spatially continuous reference rainfall data for subsequent radar reflectivity factor correction and Z-R relationship parameter calibration.

[0028] S23, based on the measured rainfall data of the rain gauge station and the rainfall estimation value corresponding to each radar sampling grid, the Z-R relationship parameter corresponding to each radar sampling grid is initialized and pre-trained to predict the Z-R relationship parameter.

[0029] In the embodiments of the present disclosure, step S23 aims to initialize and pre-train the Z-R relationship parameters (i.e., parameters a and b) in the physical constraint hybrid neural network using multi-source observation information to provide reasonable initial parameter estimation and lay a foundation for subsequent joint optimization. Specifically: S231, the spatial features of the reflectivity factor time series data of the radar sampling grid are extracted through a convolutional layer.

[0030] That is, the radar reflectivity factor time series data in each radar sampling grid and its neighborhood range are organized as a two-dimensional spatial matrix (for example, centered on a 3x3 grid window), and input to a convolutional neural network (CNN) layer. By setting a convolution kernel with a size of 3x3, a step of 1, and a channel number of 32, local feature extraction of the reflectivity spatial distribution pattern is performed to capture the spatial structure and gradient change of the precipitation system. After convolution and activation function (such as ReLU) processing, a high-dimensional spatial feature vector H k corresponding to each radar sampling grid is output, which is used to represent the spatial organization characteristics of the current radar echo.

[0031] S232, the time series features of the reflectivity factor time series data of the radar sampling grid and the time series of the cross-section flow of the downstream hydrological station are extracted through an LSTM layer.

[0032] The historical time series of the reflectivity factor of each radar sampling grid and the historical time series of the measured flow of the downstream hydrological station are spliced into a joint input sequence and sent to a long short-term memory network (LSTM) layer. The LSTM unit effectively captures the nonlinear time series dependence relationship between the precipitation input and the basin response through a gating mechanism, and outputs the hidden state h tThe state encodes the dynamic evolution characteristics of the rainfall process and the potential influence on the hydrological response.

[0033] S233, calculating the concentration time based on the environmental attribute data of the radar sampling grid.

[0034] In some embodiments, based on the environmental attribute data obtained in step S1, including the surface slope of the radar sampling grid and the soil Manning coefficient determined by land use and soil type data, combined with the runoff path length extracted from the digital elevation model (DEM) (i.e. the shortest surface concentration distance from the radar sampling grid to the control section of the downstream hydrological station), the concentration time of the radar sampling grid is calculated using the hydrodynamic empirical formula , the specific calculation formula is: ; Where L k is the runoff length from radar sampling grid k to the downstream hydrological station, n k is the Manning coefficient of radar sampling grid k.

[0035] S234, inputting the spatial features, the time sequence features, the soil moisture and the concentration time into the fully connected layer to predict the Z-R relationship parameters.

[0036] The spatial feature vector H k of the sampling radar grid k extracted in S231, the LSTM hidden state h t obtained in S232, the soil moisture W k,t of the sampling radar grid k at the current time, and the concentration time T of the sampling radar grid k calculated in S233 are concatenated to form a comprehensive feature vector. This vector is input into the fully connected neural network (Fully Connected Layer), and after multiple layers of nonlinear transformation, the Z-R relationship parameters a k and b k corresponding to the radar sampling grid are finally output, a k and b k represent the Z-R relationship parameters corresponding to the kth radar sampling grid. To ensure the physical reasonableness of the parameters, the output layer uses a nonlinear activation function such as Softplus to ensure that the prediction results a k > 0, b k > 0.

[0037] In the initialization pre-training stage, the measured rainfall data of the rain gauge station and the grid rainfall estimation value generated in S22 are taken as the target true value, an mean square error loss function is constructed, the above network parameters are supervised training, and the preliminary spatialization estimation of the Z-R relationship parameters is realized. After the initialization pre-training of the Z-R relationship parameters is completed, the application further optimizes the parameters by introducing the hydrological physical process constraint, so as to improve the physical consistency and overall precision of the quantitative rainfall estimation.

[0038] S24, using the Z-R relationship parameters predicted in S23, the radar reflectivity factor of each radar sampling grid is quantitatively precipitated to obtain the quantitative rainfall estimation value of each radar sampling grid.

[0039] Based on the Z-R relationship parameters a k and b k predicted in S23, combined with the measured radar reflectivity factor Z k after preliminary quality control, the Z-R relationship formula is used for quantitative precipitation inversion of each radar sampling grid to obtain the quantitative rainfall estimation value R k of each radar sampling grid, and the specific formula is: .

[0040] S25, according to the quantitative rainfall estimation value of each radar sampling grid, a distributed runoff yield model is used for watershed surface confluence calculation to output the simulated flow of the downstream hydrological station section.

[0041] In this embodiment, the distributed runoff yield model is a grid-based Xin'anjiang model.

[0042] S26, according to the measured flow time series data of the downstream hydrological station section, the quantitative rainfall estimation value of each radar sampling grid and the simulated flow of the downstream hydrological station section, a loss function containing multiple physical constraint parameters is constructed.

[0043] In order to realize the dynamic optimization of the Z-R relationship parameters, a composite loss function integrating multiple physical constraints is constructed, which is used to measure the deviation between the model output and the observation data under the current parameter configuration.

[0044] The loss function containing multiple physical constraint parameters is: ; Wherein, represents the loss function containing multiple physical constraint parameters; Q obs represents the measured flow of the downstream hydrological station section; Q sim represents the simulated flow of the downstream hydrological station section; Z i represents the reflectivity factor of the i-th radar sampling grid; a i , bi parameters of the Z-R relationship of the i-th radar sampling grid; R i rainfall data of the i-th radar sampling grid calculated based on the Z-R relationship, S i grid area of the i-th radar sampling grid; a, β, γ are parameters of the loss function .

[0045] Further, the value range of a, β, γ is: , used to control the flow error weight; , used to constrain the consistency of Z-R relationship; , used to force water balance.

[0046] S27, dynamically determine the Z-R relationship parameters of each radar sampling grid through an iterative optimization process until the loss function containing multiple physical constraint parameters converges.

[0047] Embed the above loss function containing multiple physical constraint parameters into the overall training framework of the physical constraint hybrid neural network, and use the gradient back propagation algorithm to train the network parameters end-to-end. In each forward propagation, update the Z-R relationship parameters a k , b k , re-execute the quantitative precipitation retrieval and runoff yield simulation of S24 and S25, calculate the new simulated flow and evaluate the loss value; in the back propagation, the loss gradient is passed back to each layer of the network through the automatic differentiation mechanism, and the weights of CNN, LSTM and fully connected layer are adjusted step by step, so as to realize the joint optimization of the Z-R relationship parameters of each radar sampling grid.

[0048] The training process continues for multiple iteration periods until the loss function converges to a preset threshold or reaches a maximum number of iterations, at which time the obtained Z-R relationship parameter set is the optimal estimation result under the dual constraints of multi-source observation and physical process. The parameter set has good spatiotemporal adaptability and physical consistency, and can significantly improve the accuracy and reliability of radar quantitative precipitation estimation in complex basins.

[0049] S3, at each rain gauge location, calculate the radar reflectivity correction parameter of each radar sampling grid according to the measured rainfall data of the rain gauge at that location, the measured radar reflectivity factor and the Z-R relationship parameters of each radar sampling grid.

[0050] This step aims to use the calibrated Z-R relationship parameters and ground rain gauge observation data to finely correct the radar reflectivity factor to eliminate systematic bias and improve the accuracy of quantitative precipitation estimation. Specifically, it includes: S31, at each rain gauge position, according to the Z-R relationship parameters of each radar sampling grid and the measured rainfall data of each radar sampling grid corresponding to the rain gauge, the theoretical value of the radar reflectivity factor is calculated through the Z-R relationship; The Z-R relationship is: ; wherein Z is the theoretical value of the radar reflectivity factor, R is the rainfall data corresponding to the radar sampling grid; a, b are the Z-R relationship parameters of the radar sampling grid; At the spatial position of each rain gauge, one or more adjacent radar sampling grids corresponding to the rain gauge are identified (usually matched in the nearest neighbor or bilinear interpolation manner). For each matched radar sampling grid k, the Z-R relationship parameters a k and b k of the grid at the corresponding moment are obtained obs , and the measured rainfall intensity R theo observed by the rain gauge at the same time is obtained. Based on the Z-R relationship, the radar reflectivity factor Z that should be theoretically observed under the current rainfall data is calculated.

[0051] S32, according to the theoretical value of the radar reflectivity factor and the measured radar reflectivity factor, the radar reflectivity correction parameter of each radar sampling grid is calculated.

[0052] The radar reflectivity factor theoretical value Z theo calculated above is compared with the radar reflectivity factor Z obs actually observed by the radar at the same space-time position (which has been pre-processed such as ground object echo elimination and attenuation correction), the difference between the two is calculated, and then the radar reflectivity correction parameter E1 is determined, that is .

[0053] S4, real-time monitoring of the radar reflectivity of the target prediction area, using the Z-R relationship parameters of each radar sampling grid and the radar reflectivity correction parameters of each radar sampling grid, calculating the rainfall data of each radar sampling grid, and completing quantitative rainfall estimation.

[0054] After the model training and parameter calibration are completed, the application enters the actual application stage, that is, based on the calibrated Z-R relationship parameters and radar reflectivity correction parameters, high-precision, near-real-time quantitative rainfall estimation of the target prediction area is realized.

[0055] In this embodiment, the S4 specifically includes: S41, real-time monitoring of the radar reflectivity of the target prediction area.

[0056] The real-time radar-based data of the target prediction area, including radar reflectivity factor, are continuously obtained through the meteorological radar data receiving system, with a time resolution of 5 minutes and a spatial resolution of 1 km x 1 km. The data are pre-processed, including ground echo filtering, non-meteorological echo identification, beam blocking compensation and attenuation correction, to ensure the reliability of the input data.

[0057] S42, using the Z-R relationship parameters of each radar sampling grid and the radar reflectivity correction parameters of the corresponding radar sampling grid, generating the radar reflectivity of each corrected radar sampling grid.

[0058] For each radar sampling grid k, the radar reflectivity correction parameter E1(k) corresponding to the radar sampling grid is determined by calling the physical constraint hybrid neural network model training stage (obtained by step S3). The radar reflectivity correction parameter reflects the comprehensive influence of radar system bias, local propagation attenuation and environmental interference and other factors.

[0059] The measured reflectivity Z obs (k) of the radar sampling grid k is corrected to generate the radar reflectivity Z coor (k) of the corrected radar sampling grid k, specifically: Z coor (k)=E1(k)×Z obs (k).

[0060] S43, according to the radar reflectivity of each corrected radar sampling grid and the Z-R relationship parameters of the corresponding radar sampling grid determined in S2, the rainfall data of each radar sampling grid is calculated, and the specific calculation formula is: Where R k is the rainfall data of the radar sampling grid k, a k and b k are the Z-R relationship parameters of the grid optimized by the physical constraint hybrid neural network.

[0061] S44, according to the rainfall data of each radar sampling grid, the rainfall estimation in the entire prediction area is completed.

[0062] The rainfall data R k of all radar sampling grids is organized into a spatiotemporally consistent two-dimensional grid data set to form a high-resolution rainfall field covering the entire target prediction area. The data set can be further visualized, accumulated (such as hourly and daily cumulative rainfall), threshold alarmed (such as short-term heavy rainfall identification), or used as input to drive distributed hydrological models for flood forecasting, mountain flood warning and other applications.

[0063] ​This invention also provides a quantitative rainfall estimation system based on multi-source data-corrected radar echoes, such as... Figure 2 As shown, it includes: The data collection module is used to collect measured rainfall data, elevation data, measured flow time series of downstream hydrological stations, measured radar reflectivity factor, soil moisture data, and environmental attribute data from rain gauge stations within the target prediction area. Physically constrained hybrid model neural network training module: used to dynamically determine the ZR relationship parameters of each radar sampling grid based on the measured rainfall data of the rain gauge, the elevation data, the measured flow time series of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data, through a physically constrained hybrid neural network model; Radar reflectivity correction parameter calculation module: used to calculate the radar reflectivity correction parameter for each radar sampling grid at each rain gauge location based on the measured rainfall data, measured radar reflectivity factor, and ZR relationship parameters of each radar sampling grid at that location. Quantitative rainfall estimation module: Used to monitor the radar reflectivity of the target prediction area in real time, and calculate the rainfall data of each radar sampling grid using the ZR relationship parameters and radar reflectivity correction parameters of each radar sampling grid to complete the quantitative rainfall estimation. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for correcting quantitative precipitation estimates from radar echoes based on multi-source data, characterized in that, The method comprises the following steps: S1, collecting the measured rainfall data of the rain gauge stations in the target prediction area, elevation data, cross-section measured flow time series of the downstream hydrological stations, measured radar reflectivity factors, soil moisture data and environmental attribute data; S2, dynamically determining the Z-R relationship parameters of each radar sampling grid based on the measured rainfall data of the rain gauge stations, the elevation data, the cross-section measured flow time series of the downstream hydrological stations, the measured radar reflectivity factors, the soil moisture data and the environmental attribute data through a physically constrained hybrid neural network model; S3, at each rain gauge station position, calculating the radar reflectivity correction parameters of each radar sampling grid according to the measured rainfall data of the rain gauge station at the position, the measured radar reflectivity factor and the Z-R relationship parameters of each radar sampling grid; S4, monitoring the radar reflectivity of the target prediction area in real time, calculating the rainfall data of each radar sampling grid by using the Z-R relationship parameters of each radar sampling grid and the radar reflectivity correction parameters of each radar sampling grid, and completing quantitative rainfall estimation.

2. The quantitative rain estimation method based on multi-source data correction of radar echo according to claim 1, characterized in that, The S2 specifically comprises: S21, constructing a physically constrained hybrid neural network model and inputting the measured rainfall data of the rain gauge stations in the target prediction area, elevation data, cross-section measured flow time series of the downstream hydrological stations, measured radar reflectivity factors, soil moisture data and environmental attribute data; S22, performing collaborative Kriging interpolation processing on the measured rainfall data of the rain gauge stations, and incorporating the elevation data as auxiliary variables in the interpolation process to generate corresponding rainfall estimation values of each radar sampling grid; S23, initializing and pre-training the Z-R relationship parameters corresponding to each radar sampling grid based on the measured rainfall data of the rain gauge stations and the corresponding rainfall estimation values of each radar sampling grid, and predicting the Z-R relationship parameters; S24, using the predicted Z-R relationship parameters in S23 to quantitatively retrieve the radar reflectivity factors of each radar sampling grid, and obtaining quantitative rainfall estimation values of each radar sampling grid; S25, using a distributed runoff yield model to perform watershed surface confluence calculation according to the quantitative rainfall estimation values of each radar sampling grid, and outputting the simulated flow of the downstream hydrological station section; S26, constructing a loss function containing multiple physical constraint parameters according to the measured flow time series data of the downstream hydrological station section, the quantitative precipitation estimation values of each radar sampling grid and the simulated flow of the downstream hydrological station section; S27, through an iterative optimization process, until the loss function containing multiple physical constraint parameters converges, the Z-R relationship parameters of each radar sampling grid are dynamically determined.

3. The quantitative precipitation estimation method of correcting radar echoes based on multi-source data according to claim 2, characterized in that, The S23 specifically comprises: S231, extracting the spatial features of the reflectivity factor time series data of the radar sampling grid through a convolution layer; S232, extracting the time sequence features of the reflectivity factor time series data of the radar sampling grid and the flow time series data of the downstream hydrological station section through an LSTM layer; S233, calculating the confluence time based on the environmental attribute data of the radar sampling grid; S234, input the spatial features, the timing features, the soil moisture and the confluence time into a full connection layer to predict a Z-R relationship parameter.

4. The quantitative precipitation estimation method of correcting radar echoes based on multi-source data according to claim 2, characterized in that, In the S22, the interpolation formula of the collaborative Kriging interpolation method is: ； where, is the rainfall estimate for radar sampling grid x0, is the rainfall data of the ith rain gauge, n is the total number of rain gauges; E(x0) is the elevation data of radar sampling grid x0, and is the weight coefficient of the cokriging.

5. The quantitative precipitation estimation method of correcting radar echoes based on multi-source data according to claim 2, characterized in that, The distributed runoff generation model is a gridded Xinanjiang model.

6. The quantitative precipitation estimation method of correcting radar echoes based on multi-source data according to claim 2, characterized in that, The loss function comprising the plurality of physical constraint parameters is: ； wherein, represents a loss function containing multiple physical constraint parameters; Q obs represents the measured flow at the downstream hydrological station section; Q sim represents the simulated flow at the downstream hydrological station section; Z i represents the reflectivity factor of the i-th radar sampling grid; a i , b i represents the parameter of the Z-R relationship of the i-th radar sampling grid; R i represents the rainfall data of the i-th radar sampling grid calculated based on the Z-R relationship; S i represents the grid area of the i-th radar sampling grid; ɑ, β, γ are parameters of the loss function .

7. The quantitative rain estimation method based on multi-source data correction of radar echo according to claim 6, characterized in that, The value range of the ɑ, β and γ is: , for controlling the flow error weight; , for constraining Z-R relationship consistency; , for enforcing water balance.

8. The method of claim 1, wherein, The S3 specifically comprises: S31, at each rain gauge position, according to the Z-R relationship parameter of each radar sampling grid and the measured rainfall data of the rain gauge corresponding to each radar sampling grid, the theoretical value of the radar reflectivity factor is calculated through the Z-R relationship formula; The Z-R relationship is: ; wherein Z is a theoretical value of a radar reflectivity factor, R is rainfall data corresponding to a radar sampling grid; a and b are Z-R relationship parameters of the radar sampling grid. S32, according to the theoretical value of the radar reflectivity factor and the measured radar reflectivity factor, the radar reflectivity correction parameter of each radar sampling grid is calculated.

9. The method of claim 1, wherein, The S4 specifically comprises: S41, real-time monitoring of the radar reflectivity of the target prediction area; S42, using the Z-R relationship parameter of each radar sampling grid and the radar reflectivity correction parameter of the corresponding radar sampling grid, the radar reflectivity of each radar sampling grid after correction is generated; S43, according to the radar reflectivity of each radar sampling grid after correction and the Z-R relationship parameter of the corresponding radar sampling grid determined in S2, the rainfall data of each radar sampling grid is calculated; S44, according to the rainfall data of each radar sampling grid, the rainfall estimation in the entire prediction area is completed.

10. A quantitative precipitation estimation system for correcting radar echoes based on multi-source data, characterized in that, The system comprises: A data collection module is configured to collect the measured rainfall data of the rain gauge in the target prediction area, the elevation data, the measured cross-section flow time series of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data; A physical constraint hybrid model neural network training module is configured to dynamically determine the Z-R relationship parameter of each radar sampling grid through a physical constraint hybrid neural network model based on the measured rainfall data of the rain gauge, the elevation data, the measured cross-section flow time series of the downstream hydrological station, the measured radar reflectivity factor, the soil moisture data and the environmental attribute data; A radar reflectivity correction parameter calculation module is configured to calculate the radar reflectivity correction parameter of each radar sampling grid according to the measured rainfall data of the rain gauge at each rain gauge position, the measured radar reflectivity factor and the Z-R relationship parameter of each radar sampling grid; A quantitative rainfall estimation module is configured to real-time monitor the radar reflectivity of the target prediction area, calculate the rainfall data of each radar sampling grid by using the Z-R relationship parameter of each radar sampling grid and the radar reflectivity correction parameter of each radar sampling grid, and complete the quantitative rainfall estimation.

Images