Rainfall estimation method based on multi-source remote sensing data fusion

Abstract

The invention discloses a precipitation estimation method based on multi-source remote sensing data fusion, and the method comprises the following steps: S1, obtaining multi-source remote sensing data, and carrying out the preprocessing of the obtained data; s2, performing feature extraction based on the preprocessed data, and constructing a dimension reduction feature vector; s3, inputting the matrix of the dimensionality reduction feature vector as data into an artificial neural network model, designing a loss function and performing parameter updating; and S4, verifying the model and estimating the precipitation amount. According to the method, data of multiple observation stations are adopted, dynamic adjustment of parameters in a rainfall estimation model according to regional characteristics and climate conditions is achieved, local change differences can be more effectively represented, errors caused by adoption of fixed parameter values are reduced, and the accuracy of rainfall estimation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological precipitation estimation, and in particular to a precipitation estimation method based on multi-source remote sensing data fusion. Background Art

[0002] Accurate precipitation estimation is crucial for disaster prevention and mitigation and water resources management. Currently, it mainly relies on single data sources such as satellite remote sensing, meteorological radar and ground rain gauges, but these methods have problems such as low spatial resolution, poor real-time performance and limited accuracy. Simple multi-source data fusion methods do not fully consider the characteristics and uncertainties of each data source, resulting in limited accuracy. In addition, traditional algorithms have difficulty in handling the nonlinear relationship and high-dimensional characteristics of multi-source data.

[0003] The main shortcomings of existing technologies: 1) Limitations of single data source methods: Satellite remote sensing: difficult to capture small-scale precipitation, poor real-time performance, and easily affected by the environment. Meteorological radar: limited coverage, affected by terrain, and large estimation errors. Ground rain gauges: uneven distribution and poor real-time performance. 2) Shortcomings of simple fusion methods: data characteristics are not fully considered, and the fusion results have large deviations. Error accumulation reduces accuracy. The spatial and temporal resolutions of different data sources are inconsistent and difficult to unify. 3) Model and algorithm limitations: single-source models have large errors and inaccurate parameterization. Advanced technologies are not used to process nonlinear and high-dimensional data.

[0004] Therefore, a precipitation estimation method that considers the characteristics of multi-source data and utilizes advanced algorithms is needed to improve accuracy and real-time performance. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a precipitation estimation method based on multi-source remote sensing data fusion, comprising the following steps: S1. Acquire multi-source remote sensing data and pre-process the acquired data; S2. Perform feature extraction based on the pre-processed data and construct a reduced-dimensional feature vector; S3. Input the matrix of the reduced-dimensional feature vector as data into an artificial neural network model, design a loss function and perform parameter update; S4. Verify the model and estimate the precipitation; wherein the multi-source remote sensing data in step S1 specifically includes: microwave remote sensing data, infrared remote sensing data, and ground meteorological station observation data.

[0006] Furthermore, the method for preprocessing the acquired data in step S1 specifically includes: data denoising, radiation correction, and geographic registration.

[0007] Furthermore, the step S2 specifically includes the following sub-steps: S21. Extracting features from microwave remote sensing data and infrared remote sensing data respectively; S22. Combining the features extracted by microwave remote sensing and infrared remote sensing to form an initial eigenvector X for precipitation; S23. Calculating the covariance matrix C and performing eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors; S24. Calculating and obtaining the eigenvector after dimensionality reduction.

[0008] Furthermore, the features extracted from the microwave remote sensing data in step S21 are specifically: brightness temperature Tb, precipitation scattering index PSI;

[0009] The features extracted from infrared remote sensing data are as follows: cloud top temperature T cloud , cloud thickness τ, cloud type Ctype;

[0010] Among them, the calculation formula of precipitation scatter index PSI is: PSI = Tb low -Tb high ;

[0011] The calculation formula of cloud thickness τ is: τ=a×(T cloud -T surface )+b;

[0012] In the above formula, Tb low Indicates the brightness temperature of the low-frequency channel; Tb high represents the brightness temperature of the high-frequency channel; a and b represent empirical coefficients; T surface Represents the surface temperature.

[0013] Furthermore, the initial feature vector X in step S22 is specifically:

[0014] X=[Tb1,Tb2,…,Tb n ,PSI,T cloud ,τ,C type , P];

[0015] The calculation formula of the covariance matrix C in step S23 is:

[0016] C=(1 / (N-1))×Σ[(X_i-X_mean)×(X_i-X_mean) t ], CV = λV;

[0017] The calculation formula of the feature vector after dimension reduction in step S24 is: Z = v k t ×X;

[0018] In the above formula, N represents the number of samples; X_mean represents the mean of the eigenvector; λ represents the eigenvalue; V represents the eigenvector matrix; Z represents the eigenvector after dimensionality reduction; v k t represents the k most important features.

[0019] Furthermore, the S3 step specifically includes the following sub-steps: S31. Processing the input reduced-dimensional feature vector matrix in the hidden layer of the artificial neural network model to obtain the predicted rainfall Y; S32. Calculating the loss value through the loss function; S33. Updating the gradient W of the weight and the gradient b of the bias.

[0020] Furthermore, the calculation formula for processing the input reduced-dimensional feature vector matrix in step S31 is: h(l)=σ(W (l) h (l-1) +b (l) ), where W (l) represents the weight matrix of the Lth layer; σ represents the activation function: σ(z)=max(0,z); b (l) represents the bias term; the calculation formula of the predicted rainfall Y is: Among them, L is the number of layers in the network.

[0021] Furthermore, the calculation formula of the loss function in step S32 is: Calculate the derivative of the loss function with respect to the output: Calculate the error of each layer: δ (l) =(W (l+1) ) T δ (l+1) ·σ′(h (l) );In the above formula, N represents the number of samples; Represents the predicted value of the model; y i represents the true value of the i-th sample; Y represents the true value; represents the derivative of the activation function.

[0022] The present invention provides a precipitation estimation method based on multi-source remote sensing data fusion, which has the following beneficial effects:

[0023] (1) The present invention adopts a causal relationship analysis method to introduce key features that are neglected in multi-source remote sensing data into the precipitation estimation model, enriching the representation of the precipitation process and achieving a refined characterization of precipitation characteristics; the present invention introduces a deep learning model with spatial distribution and temporal dynamics, combines convolutional neural networks, long short-term memory networks and attention mechanisms, expands the precipitation estimation method from both spatial and temporal dimensions, and better demonstrates the spatial heterogeneity and temporal variation characteristics of precipitation.

[0024] (2) The present invention uses observation data from ground meteorological stations to calibrate the model, which has higher regional representativeness, reduces the uncertainty caused by regional differences, and improves the applicability of the model in different regions.

[0025] (3) The present invention uses data from multiple observation stations to achieve dynamic adjustment of parameters in the precipitation estimation model according to regional characteristics and climatic conditions, which can more effectively characterize local change differences, reduce the errors caused by using fixed parameter values, and improve the accuracy of precipitation estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0027] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0028] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0029] The following is a detailed description of the implementation method of the present invention in conjunction with the accompanying drawings. Only some embodiments are described, not all embodiments. For the purpose of clarity, representations and descriptions that are not related to the present invention are omitted in the drawings and descriptions.

[0030] In order to have a clearer understanding of the technical features, purposes and beneficial effects of the present invention, the technical solution of the present invention is now described in detail below. Obviously, the implementation cases described are part of the embodiments of the present invention, not all of the embodiments, and cannot be understood as limiting the scope of the implementation of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0031] like Figure 1 As shown, the present invention provides a precipitation estimation method based on multi-source remote sensing data fusion, comprising the following steps: S1. Acquire multi-source remote sensing data and preprocess the acquired data; S2. Perform feature extraction based on the preprocessed data and construct a reduced-dimensional feature vector; S3. Input the matrix of the reduced-dimensional feature vector as data into an artificial neural network model, design a loss function and perform parameter update; S4. Verify the model and estimate the precipitation.

[0032] The multi-source remote sensing data in step S1 specifically include: microwave remote sensing data, infrared remote sensing data, and ground meteorological station observation data. Microwave remote sensing data is obtained from GPM satellite data, infrared remote sensing data is obtained from FY-4 satellite data, and ground meteorological station observation data is precipitation data. De-noising, radiation correction and geo-registration are performed on each data source to obtain data preprocessing microwave remote sensing data, infrared remote sensing data, and ground meteorological station data.

[0033] The method for preprocessing the acquired data in step S1 specifically includes: data denoising, radiation correction, and geographic registration.

[0034] Step S2 specifically includes the following sub-steps: S21. Extract features from microwave remote sensing data and infrared remote sensing data respectively; S22. Combine the features extracted by microwave remote sensing and infrared remote sensing to form an initial eigenvector X for precipitation; S23. Calculate the covariance matrix C and perform eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors; S24. Calculate and obtain the eigenvector after dimensionality reduction.

[0035] The features extracted from the microwave remote sensing data in step S21 are specifically: brightness temperature Tb and precipitation scattering index PSI.

[0036] The features extracted from infrared remote sensing data are as follows: cloud top temperature T cloud , cloud thickness τ, cloud type Ctype.

[0037] Among them, the calculation formula of precipitation scatter index PSI is: PSI = Tb low -Tb high ;

[0038] The calculation formula of cloud thickness τ is: τ=a×(T cloud -T surface )+b;

[0039] In the above formula, Tb low Indicates the brightness temperature of the low-frequency channel; Tb high represents the brightness temperature of the high-frequency channel; a and b represent empirical coefficients; T surface Represents the surface temperature.

[0040] The initial feature vector X in step S22 is specifically:

[0041] X=[Tb1,Tb2,…,Tb n ,PSI,T cloud ,τ,C type , P];

[0042] The calculation formula of the covariance matrix C in step S23 is:

[0043] C=(1 / (N-1))×Σ[(X_i-X_mean)×(X_i-X_mean) t ], CV = λV;

[0044] The calculation formula of the feature vector after dimension reduction in step S24 is: Z = v k t ×X;

[0045] In the above formula, N represents the number of samples; X_mean represents the mean of the eigenvector; λ represents the eigenvalue; V represents the eigenvector matrix; Z represents the eigenvector after dimensionality reduction; v k t represents the k most important features.

[0046] Step S3 specifically includes the following sub-steps: S31. Processing the input reduced-dimensional feature vector matrix in the hidden layer of the artificial neural network model to obtain the predicted rainfall Y; S32. Calculating the loss value through the loss function; S33. Updating the gradient W of the weight and the gradient b of the bias.

[0047] The calculation formula for processing the input reduced-dimensional feature vector matrix in step S31 is: h(l)=σ(W (l) h (l-1) +b (l) ), where W (l) represents the weight matrix of the Lth layer; σ represents the activation function: σ(z)=max(0,z); b (l) represents the bias term; the calculation formula of the predicted rainfall Y is: Among them, L is the number of layers in the network.

[0048] The calculation formula of the loss function in step S32 is: Calculate the derivative of the loss function with respect to the output: Calculate the error of each layer: δ (l) =(W (l+1) ) T δ (l+1) ·σ′(h (l) );In the above formula, N represents the number of samples; represents the predicted value of the model; Y represents the true value; represents the derivative of the activation function.

[0049] Update the weight gradient W and the bias gradient b:

[0050]

[0051] Finally, we use gradient descent to update the parameters with the minimum loss value, set an early stopping strategy, and stop training when the model improves very little within a few rounds. Finally, we get the most ideal model.

[0052] The model is tested on the prepared test data set, and the final evaluation is performed on the test set to verify the generalization ability of the model. If the accuracy is not ideal, increase the selection of feature values ​​during dimensionality reduction and adjust the initial weight value of the neural network.

[0053] The present invention adopts a causal analysis method to introduce key features that are neglected in multi-source remote sensing data into the precipitation estimation model, enriches the representation of the precipitation process, and realizes a refined characterization of precipitation characteristics; the present invention introduces a deep learning model with spatial distribution and temporal dynamics, combines convolutional neural networks, long short-term memory networks and attention mechanisms, expands the precipitation estimation method from both spatial and temporal dimensions, and better demonstrates the spatial heterogeneity and temporal variation characteristics of precipitation.

[0054] The present invention utilizes observation data from ground meteorological stations to calibrate the model, which has higher regional representativeness, reduces uncertainty caused by regional differences, and improves the applicability of the model in different regions.

[0055] The present invention uses data from multiple observation stations to achieve dynamic adjustment of parameters in the precipitation estimation model according to regional characteristics and climatic conditions, which can more effectively characterize local change differences, reduce the errors caused by using fixed parameter values, and improve the accuracy of precipitation estimation.

[0056] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A precipitation estimation method based on multi-source remote sensing data fusion, characterized in that: The following steps are involved: S1. Acquire multi-source remote sensing data and pre-process the acquired data; S2. Extract features based on the preprocessed data and construct a reduced-dimensional feature vector; S3. Input the matrix of the reduced-dimensional feature vector as data into the artificial neural network model, design the loss function and update the parameters; S4. Validate the model and estimate precipitation; The multi-source remote sensing data in step S1 specifically include: microwave remote sensing data, infrared remote sensing data, and ground meteorological station observation data.

2. The precipitation estimation method based on multi-source remote sensing data fusion according to claim 1 is characterized in that: The method for preprocessing the acquired data in step S1 specifically includes: data denoising, radiation correction, and geographic registration.

3. The precipitation estimation method based on multi-source remote sensing data fusion according to claim 1 is characterized in that: The S2 step specifically includes the following sub-steps: S21. extracting features from microwave remote sensing data and infrared remote sensing data respectively; S22. Combining the features extracted by microwave remote sensing and infrared remote sensing to form an initial feature vector X representing precipitation; S23. Calculate the covariance matrix C and perform eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors; S24. Calculate and obtain the feature vector after dimensionality reduction.

4. The precipitation estimation method based on multi-source remote sensing data fusion according to claim 3 is characterized in that: The features extracted from the microwave remote sensing data in step S21 are specifically: brightness temperature Tb, precipitation scattering index PSI; The features extracted from infrared remote sensing data are as follows: cloud top temperature T cloud , cloud thickness τ, cloud type Ctype; Among them, the calculation formula of precipitation scatter index PSI is: PSI = Tb low -Tb high ; The calculation formula of cloud thickness τ is: τ=a×(T cloud -T surface )+b; In the above formula, Tb low Indicates the brightness temperature of the low-frequency channel; Tb high represents the brightness temperature of the high-frequency channel; a and b represent empirical coefficients; T surface Represents the surface temperature.

5. The precipitation estimation method based on multi-source remote sensing data fusion according to claim 3 is characterized in that: The initial feature vector X in step S22 is specifically: X=[Tb1,Tb2,…,Tb n ,PSI,T cloud ,τ,C type ,P]; The calculation formula of the covariance matrix C in step S23 is: C=(1 / (N-1))×Σ[(X_i-X_mean)×(X_i-X_mean) t ],CV=λV; The calculation formula of the feature vector after dimension reduction in step S24 is: Z = v k t ×X; In the above formula, N represents the number of samples; X_mean represents the mean of the eigenvector; λ represents the eigenvalue; V represents the eigenvector matrix; Z represents the eigenvector after dimensionality reduction; v k t represents the k most important features.

6. The precipitation estimation method based on multi-source remote sensing data fusion according to claim 1 is characterized in that: The S3 step specifically includes the following sub-steps: S31. Processing the input reduced-dimensional feature vector matrix in the hidden layer of the artificial neural network model to obtain the predicted rainfall Y; S32. Calculate the loss value by using the loss function; S33. Update the weight gradient W and the bias gradient b.

7. The precipitation estimation method based on multi-source remote sensing data fusion according to claim 6 is characterized in that: The calculation formula for processing the input reduced-dimensional feature vector matrix in step S31 is: h(l)=σ(W (l) h (l-1) +b (l) ), where W (l) represents the weight matrix of the Lth layer; σ represents the activation function: σ(z)=max(0,z); b (l) represents the bias term; the calculation formula of the predicted rainfall Y is: Among them, L is the number of layers in the network.

8. The precipitation estimation method based on multi-source remote sensing data fusion according to claim 6 is characterized in that: The calculation formula of the loss function in step S32 is: Calculate the derivative of the loss function with respect to the output: Calculate the error of each layer: δ (l) =(W (l+1) ) T δ (l+1) ·σ′(h (l) );In the above formula, N represents the number of samples; Represents the predicted value of the model; y i represents the true value of the i-th sample; Y represents the true value; represents the derivative of the activation function.

Images