Rainfall nowcasting method based on microwave rainfall station data fusion

Abstract

The invention discloses a rainfall nowcasting method based on microwave rainfall station data fusion. The rainfall nowcasting method comprises the following steps: acquiring data of a wireless microwave link and a rainfall station in a research time period in a research area; using the data of the rainfall station and adopting an inverse distance weighting method to interpolate and construct a rainfall field; carrying out rainfall field reconstruction based on an RM algorithm by adopting microwave link average rainfall data as a nonlinear constraint; according to the reconstructed rainfall field, calculating a motion vector of each point in the space by using an LK optical flow method; and performing extrapolation forecasting on the input rainfall distribution at the last moment in combination with the obtained motion vector field by adopting a semi-Lagrange advection method. The invention innovatively provides a rainfall nowcasting method based on microwave rainfall station data fusion, and rainfall nowcasting with high temporal-spatial resolution can be realized. As the potential of the microwave communication base station in rainfall monitoring and forecasting is excavated, a large amount of manpower, material resources and infrastructure cost can be saved, and the rainfall forecasting precision is greatly improved.

Description

Technical Field

[0001] The invention relates to a rainfall nowcasting method based on microwave rain gauge station data fusion, and belongs to the field of meteorological forecasting. Background Art

[0002] In recent years, affected by global climate change and human activities, extreme rainstorms have frequently caused frequent and severe floods. Nearly 1.5 billion people (about 19% of the world's population) are directly at risk of serious floods, causing global economic losses of about US$50 billion each year. In the process of improving urban flood prevention and disaster reduction capabilities, obtaining forward-looking, timely and accurate rainfall monitoring and forecasting information plays a pivotal role. Improving the accuracy and spatiotemporal resolution of rainfall forecasts and developing high-quality and refined weather forecast services have also become the focus of attention in the field of meteorology and hydrology.

[0003] Traditional rainfall nowcasting methods are mainly based on radar data. However, radar data has monitoring blind spots, and the temporal and spatial resolution is not enough to meet the needs of refined forecasting at this stage. In recent years, by building a near-ground microwave transmission link, real-time collection of the attenuation of microwave propagation signals caused by rainfall in the atmosphere, and inverting precipitation based on the relationship between signal attenuation and rainfall intensity, it is an emerging technology for rainfall monitoring. Therefore, the patent of this invention innovates the rainfall nowcasting method based on the fusion of microwave data with high temporal and spatial resolution and rain gauge data. Summary of the invention

[0004] Purpose of the invention: In order to expand the application of microwave rainfall measurement technology and improve the accuracy of rainfall nowcasting, the present invention provides a rainfall nowcasting method based on microwave and rain gauge data fusion, uses data fusion to improve the accuracy of input rainfall fields, and performs extrapolation forecasting through the LK optical flow method.

[0005] Technical solution: To achieve the above purpose, the present invention provides a rainfall nowcasting method based on microwave rainfall station data fusion, comprising the following steps:

[0006] S1. Obtain data from wireless microwave links and rainfall stations in the study area during the study period;

[0007] S2, using the data from the rain gauge station to construct the rainfall field using the inverse distance weighted interpolation method;

[0008] S3, using microwave link average rainfall data as nonlinear constraints, and reconstructing rainfall fields based on RM algorithm;

[0009] S4, according to the reconstructed rainfall field, use the LK optical flow method to calculate the motion vector of each point in the space;

[0010] S5. The rainfall distribution at the last moment of input is combined with the obtained motion vector field and the semi-Lagrangian advection method is used for extrapolation forecasting.

[0011] Preferably, the data to be acquired in step S1 include: link signal attenuation strength, base station longitude and latitude and elevation, link length and rainfall monitored by a rain gauge.

[0012] Preferably, in step S2, the inverse distance square weighting method is adopted to take the distance power ρ=2.

[0013] Preferably, in step S3, the rainfall field reconstruction step based on the RM algorithm includes: constructing an objective function and adding nonlinear constraints, constructing a Lagrangian function, and iteratively solving.

[0014] Preferably, in step S4, based on the optical flow equation, a local smoothness constraint Lucas-Kanade method is used to solve the optical flow constraint equation to obtain the motion vector field.

[0015] Preferably, in step S5, after the motion vector field (u, v) at adjacent moments is obtained by using the LK optical flow method, the semi-Lagrangian advection method can be further used to extrapolate the obtained vector field to obtain the rainfall field predicted for the future period.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0017] 1. Improve the accuracy of rainfall monitoring: By integrating the monitoring data of microwave and rain gauges, it makes up for the shortcomings of uneven distribution of traditional rain gauges, sparse station networks, and difficulty in supporting refined forecasting needs, reduces the error of quantitative inversion of precipitation, and improves the accuracy of rainfall monitoring.

[0018] 2. Improve the real-time performance of nowcasting: The method based on microwave rain gauge data fusion can quickly process and analyze a large amount of rainfall monitoring data, thereby shortening the time of rainfall forecasting. Compared with the existing radar extrapolation forecasting technology, it has a higher time resolution and is of great significance for rainfall forecasting services that require rapid response (such as urban flood control, disaster warning, etc.).

[0019] 3. Improve the accuracy and stability of nowcasting: Higher-precision rainfall field input data can obtain more accurate forecast rainfall field results, which provides more reliable data support for the application of rainfall forecasting in multiple fields. The data fusion method based on microwave rain gauges can comprehensively consider the information of two rainfall monitoring methods, which helps to enhance the stability and robustness of the rainfall forecasting system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of the method of the present invention; DETAILED DESCRIPTION

[0021] The present invention will be further described below in conjunction with the accompanying drawings.

[0022] like Figure 1 As shown, the present invention provides a rainfall nowcasting method based on microwave rain gauge station data fusion, comprising the following steps:

[0023] S1. Obtain the data of wireless microwave links and rain gauges in the study area during the study period, including link signal attenuation strength, base station latitude and longitude and elevation, link length and rainfall monitored by the rain gauge;

[0024] S2, using the data from the rain gauge station to construct the rainfall field using the inverse distance weighted interpolation method;

[0025] S3, using microwave link average rainfall data as nonlinear constraints, and reconstructing rainfall fields based on RM algorithm;

[0026] S4, according to the reconstructed rainfall field, use the LK optical flow method to calculate the motion vector of each point in the space;

[0027] S5. The rainfall distribution at the last moment of input is combined with the obtained motion vector field and the semi-Lagrangian advection method is used for extrapolation forecasting.

[0028] The step S2 is specifically as follows:

[0029] The grid size is selected to grid the study area. The calculation formula for the interpolation point R(x0, y0) on the spatial grid is:

[0030]

[0031] Among them, R(x0, y0) is the interpolation result of the point to be interpolated; R obs，i is the measured rainfall value of rainfall station i (i = 1, 2, ..., n) in the study area; n is the number of rainfall stations involved in the interpolation; d i is the distance between the interpolation point and each rainfall station; ρ is the power of the distance, and the inverse distance square weighted method is used to take ρ=2.

[0032] The step S3 is specifically as follows:

[0033] S3-1. Constructing the objective function and adding nonlinear constraints

[0034] The objective function is to minimize the difference between the observed values ​​at the rain gauge and the reconstructed rainfall field:

[0035]

[0036] Among them, R(x i ,y i ) is the rainfall at the corresponding position of the rain gauge in the reconstructed rainfall field.

[0037] Using microwave link data as nonlinear constraints:

[0038]

[0039] Among them, A j is the signal attenuation strength observed by the jth microwave link; R j is the average rainfall along the path reconstructed by the jth microwave link, which can be expressed as R(x i ,y i ) is obtained after interpolation; a j , b j is the ITU-R model parameter value corresponding to the jth microwave link, which is related to the link frequency, polarization mode and rainfall particle morphology and can be found in the ITU-R Recommendations; L j is the link length of the jth link.

[0040] S3-2. Constructing Lagrangian functions

[0041]

[0042] Among them, λ j is the Lagrange multiplier and m is the number of microwave links.

[0043] S3-3. Iterative solution

[0044] The iterative method is used to solve R(x i ,y i ) and λ j , until the convergence condition is met, the random approximation (RM) algorithm is used for numerical optimization, that is, in each iteration, according to the current estimate R(x i ,y i ) and λ j , calculate the gradient of the objective function and the Lagrangian function, and update R(x i ,y i ) and λ j The initial rainfall field is set to be the rainfall field R(x0, y0) interpolated from the rainfall station, and the initial λ j =0.

[0045] The iterative update formula is usually in the form of:

[0046] θ (k+1) =θ (k) -α k g k (5)

[0047] Among them, θ represents the parameter to be optimized, that is, R(x i ,y i ) and λj The set of α k is the step size or learning rate of the kth step, g k is the Lagrangian function at θ (k) The gradient or approximate gradient at .

[0048] Then the updated rainfall field and Lagrange multiplier are:

[0049]

[0050] Among them, R (k) is the rainfall field obtained in the previous iteration, λ (k) is the Lagrange multiplier obtained in the previous iteration; R (k+1) and λ (k+1) are the updated rainfall fields and Lagrange multipliers; η is the step size of iterative update.

[0051] Set the convergence threshold, calculate the value of the objective function in each iteration, and compare it with the preset convergence threshold. If the error is less than the threshold, stop the iteration and output the final rainfall field and Lagrange multiplier. Otherwise, set k = k + 1 and continue the iteration.

[0052] The step S4 is specifically as follows:

[0053] For the rainfall R(x, y, t) at a point in space, there is an optical flow constraint equation:

[0054] I x u+I y v+I t =0 (8)

[0055] In the formula, Where u and v are the components commonly used in meteorology to represent motion vectors, namely optical flow; I x , I y and I t are the partial derivatives of the grid point rainfall R at x, y, and t respectively. The motion vector field is obtained by solving the optical flow constraint equation using the local smoothness constraint Lucas-Kanade method.

[0056] The image is divided into multiple small spaces. Assuming that in the small space neighborhood Ω centered on P, the optical flows of N points are the same, different weights are given to different points in the area, and equation (8) can be rewritten as:

[0057]

[0058] Where: Ω is W 2 (x) is the window weight coefficient, which can be defined as a Gaussian function, expressed as σ 2is the variance value of each point in the small space; V=(u,v) T is the optical flow of point P. Order:

[0059]

[0060] W=diag(W (x1) , W (x2) , ..., W (xN) ) (11)

[0061] b=-(I t(x1) , I t(x2) , ..., I t(xN) ) T (12)

[0062] Then the solution of the equation is:

[0063] V=(A T W 2 A) -1 A T W 2 b (13)

[0064] And A T W 2 A is actually a 2×2 matrix:

[0065]

[0066] During the calculation, it is assumed that the reflectivity changes between adjacent frames are relatively weak, and the first-order difference is used to replace the derivative of the reflectivity in each direction:

[0067]

[0068] According to formula (15), the optical flow field, that is, the motion vector field, can be calculated.

[0069] The step S5 is specifically as follows:

[0070] The LK optical flow method is used to obtain the motion vector field V = (u, v) at adjacent moments. T After that, the semi-Lagrangian advection method can be used to extrapolate the obtained vector field to obtain the rainfall field predicted for the future period. For each grid point (x, y), calculate its position (x) back to time t. b ,y b ), assuming that the velocity field is approximately constant within Δt:

[0071] x b =xu(x, y, t)Δt (16)

[0072] y b=yv(x, y, t)Δt (17)

[0073] When the backtracking point (x b ,y b ) does not fall on the grid point, it is necessary to obtain the rainfall value of the point through interpolation (such as bilinear interpolation). b ,y b ) rainfall value R(x b ,y b , t) is assigned to the grid point (x, y) at the prediction time point t1, t1 = t + Δt, that is:

[0074] R(x, y, t1) = R(x b ,y b , t) (18)

[0075] By repeating the above process, the rainfall fields at different time steps and different prediction moments can be obtained.

Claims

1. A precipitation nowcasting method based on microwave rain gauge station data fusion, characterized in that: The following steps are involved: S1. Obtain the data of wireless microwave links and rain gauges in the study area during the study period, including link signal attenuation strength, base station latitude and longitude and elevation, link length and rainfall monitored by the rain gauge; S2, using the data from the rain gauge station to construct the rainfall field using the inverse distance weighted interpolation method; S3, using microwave link average rainfall data as nonlinear constraints, and reconstructing rainfall fields based on RM algorithm; S4, according to the reconstructed rainfall field, use the LK optical flow method to calculate the motion vector of each point in the space; S5. The rainfall distribution at the last moment of input is combined with the obtained motion vector field and the semi-Lagrangian advection method is used for extrapolation forecasting.

2. The precipitation nowcasting method based on microwave rain gauge station data fusion according to claim 1 is characterized in that: The step S2 is specifically as follows: The grid size is selected to grid the study area. The calculation formula for the interpolation point R(x0, y0) on the spatial grid is: Among them, R(x0, y0) is the interpolation result of the point to be interpolated; R obs，i is the measured rainfall value of rainfall station i in the study area, i = 1, 2, ..., n, n is the number of rainfall stations involved in the interpolation; d i is the distance between the interpolation point and each rainfall station; ρ is the power of the distance, and the inverse distance square weighted method is used to take ρ=2.

3. The precipitation nowcasting method based on microwave rain gauge station data fusion according to claim 1 is characterized in that: The step S3 is specifically as follows: S3-1. Constructing the objective function and adding nonlinear constraints The objective function is to minimize the difference between the observed values ​​at the rain gauge and the reconstructed rainfall field: Among them, R(x i ,y i ) is the rainfall corresponding to the i-th rainfall station in the reconstructed rainfall field; Using microwave link data as nonlinear constraints: Among them, A j is the signal attenuation strength observed on the jth microwave link; R j is the average rainfall along the path reconstructed by the jth microwave link, which can be expressed as R(x i ,y i ) interpolation; a j 、b j is the ITU-R model parameter value corresponding to the jth microwave link, which is related to the link frequency, polarization mode and rainfall particle morphology and can be found in the ITU-R Recommendations; L i is the link length of the jth link; S3-2. Constructing Lagrangian functions Among them, λ j is the Lagrange multiplier, m is the number of microwave links; S3-3. Iterative solution The iterative method is used to solve R(x i ,y i ) and λ j , until the convergence condition is met, the random approximate RM algorithm is used for numerical optimization, that is, in each iteration, according to the current estimate R(x i ,y i ) and λ j , calculate the gradient of the objective function and the Lagrangian function, and update R(x i ,y i ) and λ j , set the initial rainfall field to the rainfall field R(x0, y0) interpolated from the rainfall station, and the initial λ j =0; The iterative update formula is usually in the form of: i (k+1) =θ (k) -a k g k (5) Among them, θ represents the parameter to be optimized, that is, R(x i ,y i ) and λ j The set of α k is the step size or learning rate of the kth step, g k is the Lagrangian function at θ (k) The gradient or approximate gradient at ; Then the updated rainfall field and Lagrange multiplier are: Among them, R (k) is the rainfall field obtained in the previous iteration, λ (k) is the Lagrange multiplier obtained in the previous iteration; R (k+1) and λ (k+1) are the updated rainfall fields and Lagrange multipliers; η is the step size of iterative update; Set the convergence threshold, calculate the value of the objective function in each iteration, and compare it with the preset convergence threshold. If the error is less than the threshold, stop the iteration and output the final rainfall field and Lagrange multiplier. Otherwise, set k = k + 1 and continue the iteration.

4. The precipitation nowcasting method based on microwave rain gauge station data fusion according to claim 1 is characterized in that: The step S4 is specifically as follows: For the rainfall R(x, y, t) at a point in space, there is an optical flow constraint equation: I x u+I y v+I t =0 (8) In the formula, Where u and v are the components commonly used in meteorology to represent motion vectors, namely optical flow; I x , I y and I t are the partial derivatives of the obtained grid point rainfall R(x, y, t) at x, y, t, respectively. The motion vector field is obtained by solving the optical flow constraint equation using the local smoothness constraint Lucas-Kanade method; The image is divided into multiple small spaces. Assuming that in the small space neighborhood Ω centered on P, the optical flows of N points are the same, different weights are given to different points in the area, and equation (8) can be rewritten as: Where: Ω is W 2 The window weight coefficient of (x) can be defined as a Gaussian function, expressed as σ 2 is the variance value of each point in the small space; V=(u,v) T is the optical flow of point P, let: In=diag(In (x1) ,IN (x2) ,...,IN (xN) ) (11) b=-(I t(x1) ,I t(x2) ,...,I t(xN) ) T (12) Then the solution of the equation is: V=(A T W 2 A) -1 A T W 2 b (13) And A T W 2 A is actually a 2× 2 matrix: During the calculation, it is assumed that the reflectivity changes between adjacent frames are relatively weak, and the first-order difference is used to replace the derivative of the reflectivity in each direction: According to formula (15), the optical flow field, that is, the motion vector field, can be calculated.

5. The precipitation nowcasting method based on microwave rain gauge station data fusion according to claim 4 is characterized in that: The step S5 is specifically as follows: The LK optical flow method is used to obtain the motion vector field V = (u, v) at adjacent moments. T After that, the semi-Lagrangian advection method can be used to extrapolate the obtained vector field to obtain the rainfall field predicted for the future period; for each grid point (x, y), its position (x) back to time t is calculated. b ,y b ), assuming that the velocity field is approximately constant within Δt: x b =x-u(x,y,t)Δt (16) y b =y-v(x,y,t)Δt (17) When the backtracking point (x b ,y b ) does not fall on the grid point, it is necessary to obtain the rainfall value of the point through interpolation. b ,y b ) rainfall value R(x b ,y b , t) is assigned to the grid point (x, y) at the prediction time point t1, t1 = t + Δt, that is: R(x,y,t1)=R(x b ,y b ,t) (18) By repeating the above process, the rainfall fields at different time steps and different prediction moments can be obtained.

Images