Nowcasting method based on rapid cycle assimilation numerical model coupled with radar

Abstract

The present invention discloses a nowcasting method based on rapid cycle assimilation numerical model and radar coupling, and relates to the technical field of precipitation nowcasting. It includes: obtaining meteorological radar observation data corresponding to the precipitation period; preprocessing the obtained meteorological radar observation data; hourly assimilation operation; numerical model rapid cycle assimilation forecast output simulation results; hourly forecast output of forecast results for a preset time period in the future; building a deep learning nowcasting model based on multi-channel input; training the deep learning nowcasting model; the deep learning nowcasting model outputs a 0-3 hour near-term precipitation forecast result. The present invention helps to achieve the beneficial effects of advancing the forecast time (0-1h→0-3h) and improving the accuracy of precipitation nowcasting.

Description

Technical Field

[0001] The invention relates to the technical field of precipitation nowcasting, and in particular to a nowcasting method based on a rapid cycle assimilation numerical model and radar coupling. Background Art

[0002] Research on the data-driven physical prediction factor and observation coupling nowcasting model. The essence of data-driven models is to mine and learn the motion characteristics based on observation data. Usually, the forecast accuracy based on radar echo extrapolation drops rapidly after one hour; while the numerical model is based on thermal, dynamic and microphysical information, using physical equations and empirical parameters to simulate weather processes, and there are error accumulation and amplification effects. The refined numerical model can effectively describe the precipitation formation process and its physical mechanism, but it is easily affected by the boundary and initial field. The ground-based radar can obtain the regional cloud conditions and motion characteristics with high accuracy, and integrate the respective advantages of radar extrapolation and numerical models. In theory, it can improve the forecasting ability and level of near-fall precipitation. At the same time, the current operational rapid cycle assimilation numerical forecast model is capable of providing high-frequency physical information.

[0003] Therefore, how to utilize the advantages of data-driven in computing efficiency, data feature fusion, and nonlinear computing to jointly construct an efficient forecasting model with multi-source, multi-type data and multi-channel data formed by physical mechanism processes is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0004] In view of this, the present invention provides a nowcasting method based on a rapid cycle assimilation numerical model and radar coupling, which can achieve the effect of advancing the forecast time and improving the accuracy of precipitation nowcasting.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] The nowcasting method based on the rapid cycle assimilation numerical model and radar coupling includes the following steps:

[0007] S1. Filter out precipitation events using ground-based precipitation data and obtain meteorological radar observation data corresponding to the precipitation period;

[0008] S2. preprocessing the acquired meteorological radar observation data to obtain preprocessed meteorological radar observation data;

[0009] S3, assimilate the pre-processed meteorological radar observation data hour by hour, and then quickly cycle assimilate and forecast the output simulation results through the numerical model;

[0010] S4, forecast the simulation results output by S3 hour by hour and output the forecast results for the future preset time period;

[0011] S5, inputting the simulated physical quantities and radar observation data in the forecast results of S4 into the trained deep learning nowcasting model at preset time intervals;

[0012] S6. The deep learning nowcasting model outputs the nowcasting results for a preset time period.

[0013] In the above method, optionally, the preprocessing means in S2 include but are not limited to outlier removal, noise processing, redundancy removal and data format normalization.

[0014] In the above method, optionally, the numerical model in S3 includes but is not limited to multi-level physical quantities and common convection physical quantities.

[0015] The above method is optional, based on a deep learning nowcasting model with multi-channel input, specifically including: the first layer is a convolutional layer C1, the second layer is a pooling layer P2, the third layer is a convolutional layer C3 and the fourth layer is a fully connected layer F4.

[0016] In the above method, optionally, the deep learning nowcasting model is trained in S5, specifically comprising the following steps:

[0017] S501, extracting the matrix gradient of the proposed training data set to obtain the gradient matrix in six directions of x, y, xx, xy, yx, and yy; inputting the gradient matrix, outputting the predicted value of each time period of the rainfall behavior matrix sequence, performing error analysis through the loss function according to the output predicted value and actual value, and changing each weight value of the model;

[0018] S502: Determine whether the error analysis reaches a preset threshold. If not, return to the step of selecting the proposed training data; if yes, end the 3D-CNN network training step and output the deep learning nowcasting model.

[0019] The above method, optionally, a rainfall behavior matrix sequence: the rainfall time series data is formed into a rainfall behavior matrix sequence according to time periods;

[0020] The rainfall behavior matrix sequence is divided into a pseudo-training data set and a test data set.

[0021] In the above method, optionally, the first 80% of the rainfall behavior matrix sequence is used as a test data set, and the last 20% is used as a quasi-training data set.

[0022] In the above method, optionally, the gradient extraction step is specifically as follows: continuously taking directional gradients in the directions of x, y, xx, xy, yx, and yy for the sample matrix sequence; first, selecting the Prewitt operator as a gradient extraction tool, convolving the horizontal Prewitt operator and the vertical Prewitt operator with the matrix sequence respectively to obtain first-order gradient images in the directions of x and y; and then convolving the horizontal and vertical Prewitt operators with the first-order gradient images in the directions of x and y respectively to obtain second-order gradient images in the directions of xx, xy, yx, and yy.

[0023] Compared with the prior art, the present invention provides a nowcasting method based on the coupling of rapid cycle assimilation numerical model and radar. The beneficial effects achieved by the present invention are embodied in the following aspects: 1) The present invention uses various physical quantities in the forecast results of the rapid cycle assimilation of numerical models and radar observation data of the same time frequency to simultaneously input the multi-channel input deep learning nowcasting model, so that the model can learn more information and produce high-precision long-delay nowcasting results. 2) The present invention inputs more atmospheric physical information that cannot be observed by meteorological radars, and combines it with high-precision real-time radar observation information. By taking advantage of the multi-channel deep learning model that can input multi-source and multi-type information, the present invention achieves the beneficial effects of advancing the forecast time (0-1h→0-3h) and improving the accuracy of precipitation nowcasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0025] Figure 1 It is a flow chart of the nowcasting method based on the coupling of the rapid cycle assimilation numerical model and radar disclosed in the present invention;

[0026] Figure 2 It is a flow chart of a specific forecasting method disclosed in the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, the elements defined by the sentence "comprise one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0029] Reference Figure 1 As shown, the present invention discloses a nowcasting method based on rapid cycle assimilation numerical model and radar coupling, comprising the following steps:

[0030] S1. Filter out precipitation events using ground-based precipitation data and obtain meteorological radar observation data corresponding to the precipitation period;

[0031] S2. preprocessing the acquired meteorological radar observation data to obtain preprocessed meteorological radar observation data;

[0032] S3, assimilate the pre-processed meteorological radar observation data hour by hour, and then quickly cycle assimilate and forecast the output simulation results through the numerical model;

[0033] S4, forecast the simulation results output by S3 hour by hour and output the forecast results for the future preset time period;

[0034] S5, inputting the simulated physical quantities and radar observation data in the forecast results of S4 into the trained deep learning nowcasting model at preset time intervals;

[0035] S6. The deep learning nowcasting model outputs the nowcasting results for a preset time period.

[0036] Furthermore, the preprocessing means in S2 include but are not limited to outlier removal, noise processing, redundancy removal and data format normalization.

[0037] Furthermore, the numerical model in S3 includes, but is not limited to, multi-level physical quantities, common convection physical quantities, and other features.

[0038] As shown in Table 1 , the numerical model in S3 includes, but is not limited to, multi-level physical quantities, common convection physical quantities, and other features;

[0039] Table 1 Optional physical prediction factors of numerical models

[0040]

[0041] Furthermore, the deep learning nowcasting model based on multi-channel input specifically includes: the first layer is a convolutional layer C1, the second layer is a pooling layer P2, the third layer is a convolutional layer C3 and the fourth layer is a fully connected layer F4.

[0042] Furthermore, the deep learning nowcasting model in S5 is trained, specifically including the following steps:

[0043] S501, extracting the matrix gradient of the proposed training data set to obtain the gradient matrix in six directions of x, y, xx, xy, yx, and yy; inputting the gradient matrix, outputting the predicted value of each time period of the rainfall behavior matrix sequence, performing error analysis through the loss function according to the output predicted value and actual value, and changing each weight value of the model;

[0044] S502: Determine whether the error analysis reaches a preset threshold. If not, return to the step of selecting the proposed training data; if yes, end the 3D-CNN network training step and output the deep learning nowcasting model.

[0045] Further, rainfall behavior matrix sequence: the rainfall time series data is composed into a rainfall behavior matrix sequence according to the time period;

[0046] The rainfall behavior matrix sequence is divided into a pseudo-training data set and a test data set.

[0047] Furthermore, the first 80% of the rainfall behavior matrix sequence is used as a test data set, and the last 20% is used as a quasi-training data set.

[0048] Furthermore, the gradient extraction step is specifically as follows: continuously take the directional gradients in the directions of x, y, xx, xy, yx, and yy for the sample matrix sequence. First, select the Prewitt operator as the gradient extraction tool, and convolve the horizontal Prewitt operator and the vertical Prewitt operator with the matrix sequence to obtain the first-order gradient images in the directions of x and y, and then convolve the horizontal and vertical Prewitt operators with the first-order gradient images in the directions of x and y to obtain the second-order gradient images in the directions of xx, xy, yx, and yy.

[0049] Reference Figure 2 As shown, the present invention discloses a specific forecasting method flow chart.

[0050] The present invention uses a numerical model to quickly cycle and assimilate various physical quantities in the forecast results and radar observation data of the same time frequency, and simultaneously inputs them into a multi-channel input deep learning nowcasting model, so that the model can learn more information and produce high-precision long-delay nowcasting results. This helps to achieve the beneficial effects of advancing the forecast time (0-1h→0-3h) and improving the accuracy of precipitation nowcasting.

[0051] The embodiments in this specification are described in a progressive manner. For the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For the relevant parts, please refer to the partial description of the method embodiment. The system and system embodiments described above are only schematic, in which the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0052] In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0053] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A nowcasting method based on rapid cycle assimilation numerical model and radar coupling, characterized in that: The following steps are involved: S1. Filter out precipitation events using ground-based precipitation data and obtain meteorological radar observation data corresponding to the precipitation period; S2. preprocessing the acquired meteorological radar observation data to obtain preprocessed meteorological radar observation data; S3, assimilate the pre-processed meteorological radar observation data hour by hour, and then quickly cycle assimilate and forecast the output simulation results through the numerical model; S4, forecast the simulation results output by S3 hour by hour and output the forecast results for the future preset time period; S5, inputting the simulated physical quantities and radar observation data in the forecast results of S4 into the trained deep learning nowcasting model at preset time intervals; S6. The deep learning nowcasting model outputs the nowcasting results for a preset time period.

2. The nowcasting method based on rapid cycle assimilation numerical model and radar coupling according to claim 1 is characterized in that: The preprocessing methods in S2 include but are not limited to outlier removal, noise processing, redundancy removal and data format normalization.

3. The nowcasting method based on rapid cycle assimilation numerical model and radar coupling according to claim 1 is characterized in that: The numerical models in S3 include but are not limited to multi-level physical quantities and common convection physical quantities.

4. The nowcasting method based on rapid cycle assimilation numerical model and radar coupling according to claim 1 is characterized in that: The deep learning nowcasting model based on multi-channel input in S5 specifically includes: the first layer is the convolution layer C1, the second layer is the pooling layer P2, the third layer is the convolution layer C3 and the fourth layer is the fully connected layer F4.

5. The nowcasting method based on rapid cycle assimilation numerical model and radar coupling according to claim 4 is characterized in that: The deep learning nowcasting model in S5 is trained, which specifically includes the following steps: S501, extracting the matrix gradient of the proposed training data set to obtain the gradient matrix in six directions of x, y, xx, xy, yx, and yy; inputting the gradient matrix, outputting the predicted value of each time period of the rainfall behavior matrix sequence, performing error analysis through the loss function according to the output predicted value and actual value, and changing each weight value of the model; S502: Determine whether the error analysis reaches a preset threshold. If not, return to the step of selecting the proposed training data; if yes, end the 3D-CNN network training step and output the deep learning nowcasting model.

6. The nowcasting method based on rapid cycle assimilation numerical model and radar coupling according to claim 5 is characterized in that: Rainfall behavior matrix sequence: Rainfall time series data are organized into rainfall behavior matrix sequences according to time periods; The rainfall behavior matrix sequence is divided into a pseudo-training data set and a test data set.

7. The nowcasting method based on rapid cycle assimilation numerical model and radar coupling according to claim 6 is characterized in that: The first 80% of the rainfall behavior matrix sequence is the test data set, and the last 20% is the quasi-training data set.

8. The nowcasting method based on rapid cycle assimilation numerical model and radar coupling according to claim 6 is characterized in that: The specific gradient extraction steps are as follows: continuously take the directional gradients in the directions of x, y, xx, xy, yx, and yy for the sample matrix sequence. First, select the Prewitt operator as the gradient extraction tool, and convolve the horizontal Prewitt operator and the vertical Prewitt operator with the matrix sequence to obtain the first-order gradient images in the x and y directions, and then convolve the horizontal and vertical Prewitt operators with the first-order gradient images in the x and y directions to obtain the second-order gradient images in the four directions of xx, xy, yx, and yy.

Images