Satellite data precipitation inversion method and system based on deep learning

Abstract

The invention discloses a satellite data rainfall inversion method and system based on deep learning, and the method comprises the steps: taking data obtained from satellite data in a preset region within a specified time as to-be-analyzed data; determining a first rainfall area through the determined data, performing uncertainty estimation on the uncertain data to obtain a second rainfall area, and performing rainfall scale prediction on the to-be-analyzed data in the first rainfall area and the second rainfall area to obtain a rainfall scale probability; performing precipitation scale feature extraction on the to-be-analyzed data to obtain precipitation scale features, and performing feature fusion on the precipitation scale probability and the precipitation scale features to obtain fusion data; and constructing a deep learning rainfall inversion model by using the fused data based on the attenuation rate, outputting the to-be-inverted data to the deep learning rainfall inversion model, and outputting an inversion result. The method not only can improve the precision of satellite data rainfall inversion, but also can be directly applied to a satellite data rainfall inversion system.

Description

Technical Field

[0001] The present invention relates to the field of precipitation retrieval, and particularly to a method and system for satellite data precipitation retrieval based on deep learning. Background Art

[0002] Precipitation is one of the core elements of the global water cycle and climate change. High-precision, high spatio-temporal resolution precipitation monitoring is of great significance for meteorological forecasting, disaster warning, agricultural planning, and water resource management. Traditional precipitation retrieval methods mainly rely on ground meteorological stations and weather radar observations, but their coverage is limited, and there are significant data gaps, especially in the ocean, plateau, and remote areas. Satellite remote sensing technology, with its advantages of wide-area coverage and continuous observation, has become an important means to supplement ground observations. However, satellite precipitation retrieval faces challenges such as insufficient multi-source data fusion, difficulty in quantifying uncertainty, and low modeling accuracy for complex precipitation processes.

[0003] Currently, the mainstream satellite precipitation retrieval methods can be divided into two categories: statistical retrieval and physical models. Statistical methods rely on empirical relationships of historical data but are difficult to adapt to dynamic atmospheric conditions; physical models, although having clear physical mechanisms, have high computational complexity and are sensitive to initial fields and auxiliary data. In recent years, deep learning technology, with its powerful non-linear modeling ability, has gradually been applied to the field of precipitation retrieval, but there are still the following limitations: the complementary nature of satellite multi-source data has not been fully exploited; there is a lack of systematic uncertainty estimation for uncertain data in satellite observations, resulting in insufficient robustness of the model in complex scenarios; existing deep learning models mostly adopt an end-to-end structure, ignoring the time decay characteristics of the precipitation process and the spatial correlation of multi-scale features, affecting the retrieval accuracy of extreme precipitation events.

[0004] In addition, existing studies usually simply splice deterministic data and uncertain data and input them into the model, without designing a differential feature fusion mechanism for their confidence differences, which may introduce noise interference. At the same time, the influence of environmental factors such as terrain undulation and surface type heterogeneity on the vertical structure of precipitation has not been effectively modeled, resulting in a significant increase in precipitation retrieval errors in complex terrain areas. Therefore, there is an urgent need to develop a deep learning retrieval method that integrates multi-source satellite data, incorporates an uncertainty estimation mechanism, and adapts to the spatio-temporal evolution law of precipitation to improve the reliability and refinement level of global-scale precipitation monitoring. Summary of the Invention

[0005] The object of the present invention is to provide a method for satellite data precipitation retrieval based on deep learning.

[0006] To achieve the above object, the present invention is implemented according to the following technical solution:

[0007] The present invention includes the following steps:

[0008] Take the data obtained from satellite data in a preset area within a specified time as the data to be analyzed; the satellite data includes satellite observation data, auxiliary data, and label data; the data to be analyzed includes determined data and uncertain data; the determined data has a higher certainty than the uncertain data; the satellite observation data includes microwave radiation data, infrared data, multi-channel radiometer data, and radar joint observation data; the auxiliary data includes terrain elevation, surface type, and atmospheric reanalysis data;

[0009] Determine the first precipitation area through the determined data, perform uncertainty estimation on the uncertain data to obtain a second precipitation area, and predict the precipitation scale probability for the data to be analyzed within the first precipitation area and the second precipitation area;

[0010] Extract precipitation scale features from the data to be analyzed to obtain precipitation scale features, and perform feature fusion on the precipitation scale probability and the precipitation scale features to obtain fusion data;

[0011] Based on the attenuation rate, construct a deep learning precipitation inversion model using the fusion data, input the data to be inverted into the deep learning precipitation inversion model, and output the inversion result.

[0012] Further, the method for determining the first precipitation area through the determined data includes:

[0013] Adopt a neural network algorithm to learn the satellite images and their corresponding rainfall data in the determined data, automatically extract rainfall features and establish a mapping relationship, predict future rainfall conditions according to the mapping relationship, and output the rainy area as the first precipitation area.

[0014] Further, the method for performing uncertainty estimation on the uncertain data to obtain a second precipitation area includes:

[0015] Adopt a Bayesian neural network by taking the weights of each layer of the neural network as random variables, estimate the posterior distribution of the weights of each layer of the neural network, and predict the uncertainty estimation of the uncertain data during prediction according to the posterior distribution;

[0016] Perform spatial analysis based on the uncertain data, and combine the uncertain information to obtain an uncertainty evaluation value for the predicted rainfall area; output the rainfall area with an uncertainty evaluation value lower than 0.152 as the second precipitation area.

[0017] Further, the method for obtaining the precipitation scale probability includes:

[0018] The hybrid model with feature enhancement combines statistics and machine learning algorithms, uses advanced feature engineering to improve prediction accuracy, analyzes the data to be analyzed within the precipitation area, estimates the probability distributions of multiple rainfall scales, and outputs the probability distributions of multiple rainfall scales as precipitation scale probabilities.

[0019] Furthermore, the method for extracting precipitation scale features from the data to be analyzed includes:

[0020] Input the data to be analyzed in the first precipitation area and the second precipitation area into the precipitation scale feature extractor based on multi-head attention. The expression is:

[0021]

[0022] where the output feature of the i-th encoder layer is The intermediate feature after passing through the multi-head attention layer is The multi-layer attention layer is MSA, the multi-layer perceptron layer is MLP, and the output feature of the (i - 1)-th encoder layer is Layer normalization is LN(·), the position encoding function is PosEnc(·), and the feature after position encoding in the i-th layer is

[0023] Output the output feature as the rainfall scale feature.

[0024] Furthermore, the method for feature fusion of the precipitation scale probability and the precipitation scale feature to obtain fusion data includes:

[0025] Input the precipitation scale probability and the precipitation scale feature into the multi-scale feature fusion device, and divide the precipitation scale feature into a global token aggregating the entire area and an embedded representation of the image patch sequence;

[0026] Convert the global token into a form combined with the detailed feature K through the multi-layer perceptron layer, introduce the sampling feature, and the expression of feature fusion is:

[0027]

[0028] where the multi-scale feature data is The fusion feature is The feature embedding from the backbone network is K, and the feature after reorganizing the global token is z tk , the feature obtained after sampling through the factor is u, the multi-layer perceptron layer is MLP, the fusion module is Fusion(·), the sampling feature is u, and the preliminary fusion result is The precipitation scale probability is p, and the precipitation scale feature is c;

[0029] The classification result of the usage scale classification module constrains the precipitation feature fusion process for different scales, and the expression is:

[0030]

[0031] Among them, the precipitation scale predictor is Γ(·), and the precipitation scale feature extractor is The feature map after weighted sum is The specified number of precipitation scales is n, the normalization exponential function is softmax(·), and the predicted precipitation scale of the multi-scale feature data is The i-th precipitation scale feature of the multi-scale feature data is The second precipitation scale feature of the multi-scale feature data is The n-th precipitation scale feature of the multi-scale feature number is The precipitation scale probability distribution is The fused precipitation feature map of the precipitation scale feature and the precipitation scale probability distribution in the a-th row and j-th column is The index of the x-th precipitation scale of the precipitation scale feature in the a-th row and j-th column is The index of the x-th precipitation scale of the precipitation scale probability distribution in the a-th row and j-th column

[0033] Based on the temperature coefficient, a loss function is given:

[0034]

[0035] Among them, the loss function is The cross-entropy loss function is The temperature coefficient is The mean squared error loss function is The fused precipitation feature map is output as fused data.

[0036] Furthermore, a method for constructing a deep learning precipitation inversion model based on the attenuation rate using the fused data includes:

[0037] Calculate the attenuation coefficient:

[0038]

[0039] Among them, the attenuation cross section is σ e , the wavelength is ρ, the particle diameter is D, the mass-weighted average diameter is D m , the normalized intercept is N w , the particle number concentration is N, the particle number concentration of the particle diameter is N(D), the gamma distribution is Β(·), the temperature is T, the particle diameter D and the mass-weighted average diameter D m The particle size distribution function of is f(D,D m) where the shape parameter of the precipitation particle size distribution is γ;

[0040] The attenuation coefficient is corrected according to the field monitored precipitation data, and the expression is:

[0041]

[0042] where the corrected normalized intercept is The vertical height at the earth's curvature is h, and the correction factor for height h is The actual precipitation is y, and the predicted precipitation is The temperature difference at the rainfall ground when rainfall starts is ΔT, the humidity difference at the rainfall ground when rainfall starts is ΔS, the constants are r1, r2, r3 respectively, the adjustment factor is ω, and the corrected attenuation coefficient is

[0043] Calculate the rainfall:

[0044]

[0045] where the precipitation falling speed is V(D) and the precipitation amount is Q;

[0046] Based on the attenuation coefficient, mass weighted average diameter and rainfall, a deep learning precipitation inversion model is used to invert the adjustment factor and mass weighted average diameter for the corrected normalized intercept, and then invert the rainfall according to the inverted normalized intercept, and output the precipitation inversion result.

[0047] In a second aspect, a satellite data precipitation inversion system based on deep learning includes:

[0048] Data acquisition module: used to obtain the data in the satellite data in a preset area within a specified time as the data to be analyzed; the satellite data includes satellite observation data, auxiliary data and label data; the data to be analyzed includes determined data and uncertain data; the determined data has a higher certainty than the uncertain data;

[0049] Regional probability module: used to determine the first precipitation area through the determined data, estimate the uncertainty of the uncertain data to obtain the second precipitation area, and predict the rainfall scale of the data to be analyzed within the first precipitation area and the second precipitation area to obtain the precipitation scale probability;

[0050] Feature fusion module: used to extract the precipitation scale features from the data to be analyzed to obtain precipitation scale features, and fuse the precipitation scale probability and the precipitation scale features to obtain fusion data;

[0051] Modeling output module: It is used to construct a deep learning precipitation inversion model based on the attenuation rate using the fusion data, output the data to be inverted to the deep learning precipitation inversion model, and output the inversion result.

[0052] The beneficial effects of the present invention are:

[0053] The present invention is a satellite data precipitation inversion method and system based on deep learning. Compared with the prior art, the present invention has the following technical effects:

[0054] By determining the first precipitation area, obtaining the second precipitation area, obtaining the precipitation scale probability, obtaining the precipitation scale characteristics, obtaining the fusion data, and the model construction steps, the present invention can improve the accuracy of satellite data precipitation inversion, thereby improving the precision of satellite data precipitation inversion. Optimizing the satellite data precipitation inversion can greatly save resources, improve work efficiency, can realize intelligent precipitation inversion of satellite data, partition and fuse data for satellite data precipitation inversion in real time, which is of great significance for satellite data precipitation inversion, and can adapt to different standards of satellite data precipitation inversion and different satellite data precipitation inversion requirements, and has a certain universality. Brief Description of the Drawings

[0055] Figure 1 It is a flowchart of the steps of the satellite data precipitation inversion method based on deep learning of the present invention. Detailed Embodiments

[0056] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention, but do not limit the present invention.

[0057] The satellite data precipitation inversion method and system based on deep learning of the present invention include the following steps:

[0058] As Figure 1 shown, in this embodiment, it includes the following steps:

[0059] Take the data obtained from the satellite data in the preset area within the specified time as the data to be analyzed; the satellite data includes satellite observation data, auxiliary data, and label data; the data to be analyzed includes determined data and undetermined data; the determined data has a higher certainty than the undetermined data; the satellite observation data includes microwave radiation data, infrared data, multi-channel radiometer data, radar joint observation data; the auxiliary data includes terrain elevation, surface type, and atmospheric reanalysis data;

[0060] Determine the first precipitation area based on the determined data, perform uncertainty estimation on the undetermined data to obtain the second precipitation area, and predict the rainfall scale probability of the data to be analyzed within the first precipitation area and the second precipitation area;

[0061] Extract the precipitation scale characteristics of the data to be analyzed to obtain precipitation scale characteristics, and perform feature fusion on the precipitation scale probability and the precipitation scale characteristics to obtain fusion data;

[0062] Construct a deep learning precipitation inversion model using the fusion data based on the attenuation rate, input the data to be inverted into the deep learning precipitation inversion model, and output the inversion result.

[0063] In this embodiment, the method for determining the first precipitation area based on the determined data includes:

[0064] Use the neural network algorithm to learn the satellite images and their corresponding rainfall data in the determined data, automatically extract rainfall features and establish a mapping relationship, predict future rainfall conditions according to the mapping relationship, and output the rainy area as the first precipitation area.

[0065] In this embodiment, the method for performing uncertainty estimation on the undetermined data to obtain the second precipitation area includes:

[0066] Use the Bayesian neural network to regard the weights of each layer of the neural network as random variables, estimate the posterior distribution of the weights of each layer of the neural network, and predict the uncertainty estimation of the undetermined data during prediction according to the posterior distribution;

[0067] Perform spatial analysis based on the undetermined data, combine the undetermined information to obtain the uncertainty evaluation value of the predicted rainfall area; output the rainfall area with the uncertainty evaluation value lower than 0.152 as the second precipitation area.

[0068] In this embodiment, the method for obtaining the precipitation scale probability includes:

[0069] Use the hybrid model with feature enhancement to combine statistical and machine learning algorithms, use advanced feature engineering to improve the prediction accuracy, analyze the data to be analyzed within the precipitation area, estimate the probability distribution of multiple rainfall scales, and output the probability distribution of multiple rainfall scales as the precipitation scale probability.

[0070] In this embodiment, the method for extracting the precipitation scale characteristics of the data to be analyzed to obtain precipitation scale characteristics includes:

[0071] Input the data to be analyzed in the first precipitation area and the second precipitation area into the precipitation scale feature extractor based on multi-head attention, and the expression is:

[0072]

[0073] where the output feature of the $i$-th encoder layer is The intermediate feature passing through the multi-head attention layer is The multi-layer attention layer is MSA, the multi-layer perceptron layer is MLP, and the output feature of the $(i - 1)$-th encoder layer is The layer normalization is LN(·), the position encoding function is PosEnc(·), and the feature after position encoding of the $i$-th layer is

[0074] The output feature is output as the rainfall scale feature.

[0075] In this embodiment, the method for obtaining the fusion data by fusing the precipitation scale probability and the precipitation scale feature includes:

[0076] Input the precipitation scale probability and the precipitation scale feature into the multi-scale feature fusion device, and divide the precipitation scale feature into the global token aggregating the entire region and the embedded representation of the image patch sequence;

[0077] Convert the global token into a form combined with the detailed feature K through the multi-layer perceptron layer, introduce the sampling feature, and the expression of feature fusion is:

[0078]

[0079] where the multi-scale feature data is The fusion feature is The feature embedding from the backbone network is K, and the feature after reorganizing the global token is z tk , the feature obtained after sampling through the factor is u, the multi-layer perceptron layer is MLP, the fusion module is Fusion(·), the sampling feature is u, and the preliminary fusion result is The precipitation scale probability is p, and the precipitation scale feature is c;

[0080] Use the classification result of the scale classification module to constrain the precipitation feature fusion process of different scales, and the expression is:

[0081]

[0082]

[0083] where the precipitation scale predictor is Γ(·), and the precipitation scale feature extractor is The feature map after weighted sum is The specified number of precipitation scales is n, the normalization exponential function is softmax(·), and the predicted precipitation scale of the multi-scale feature data is The $i$-th precipitation scale feature of the multi-scale feature data is The second precipitation scale feature of the multi-scale feature data is The nth precipitation scale feature of the multi-scale feature number is The precipitation scale probability distribution is The fused precipitation feature mapping of the precipitation scale feature and the precipitation scale probability distribution in the aj-th row is The index of the xth precipitation scale of the precipitation scale feature in the aj-th row is The index of the xth precipitation scale of the precipitation scale probability distribution in the aj-th row

[0085] Given a loss function based on the temperature coefficient:

[0086]

[0087] where the loss function is The cross-entropy loss function is The temperature coefficient is The mean square error loss function is Output the fused precipitation feature mapping as fused data.

[0088] In this embodiment, a method for constructing a deep learning precipitation retrieval model based on the attenuation rate using the fused data includes:

[0089] Calculate the attenuation coefficient:

[0090]

[0091] where the attenuation cross-section is σ e , the wavelength is ρ, the particle diameter is D, the mass-weighted average diameter is D m , the normalized intercept is N w , the particle number concentration is N, the particle number concentration of the particle diameter is N(D), the gamma distribution is Β(·), the temperature is T, the particle diameter D and the mass-weighted average diameter D m The particle size distribution function of is f(D,D m ), and the shape parameter of the precipitation particle size distribution is γ;

[0092] Revise the attenuation coefficient according to the field monitoring precipitation data, and the expression is:

[0093]

[0094] where the revised normalized intercept is The vertical height at the earth's curvature is h, and the correction factor of height h is The actual precipitation is y, and the predicted precipitation is When it starts to rain, the temperature difference between the rainfall ground is ΔT, the humidity difference between the rainfall ground is ΔS, the constants are r1, r2, r3 respectively, the adjustment factor is ω, and the corrected attenuation coefficient is

[0095] Calculate the rainfall:

[0096]

[0097] Among them, the precipitation falling speed is V(D), and the precipitation amount is Q;

[0098] Based on the attenuation coefficient, mass-weighted average diameter and rainfall deep learning precipitation inversion model, obtain the adjustment factor and mass-weighted average diameter to invert the corrected normalized intercept, and invert the rainfall according to the inverted normalized intercept, and output the precipitation inversion result.

[0099] In the second aspect, a satellite data precipitation inversion system based on deep learning includes:

[0100] Data acquisition module: used to obtain the data in the satellite data in the preset area within the specified time as the data to be analyzed; the satellite data includes satellite observation data, auxiliary data and label data; the data to be analyzed includes determined data and uncertain data; the determined data has higher certainty than the uncertain data;

[0101] Regional probability module: used to determine the first precipitation area through the determined data, estimate the uncertainty of the uncertain data to obtain the second precipitation area, and predict the rainfall scale of the data to be analyzed within the first precipitation area and the second precipitation area to obtain the precipitation scale probability;

[0102] Feature fusion module: used to extract the precipitation scale features of the data to be analyzed to obtain precipitation scale features, and fuse the precipitation scale probability and the precipitation scale features to obtain fusion data;

[0103] Modeling output module: used to construct a deep learning precipitation inversion model based on the attenuation rate using the fusion data, output the data to be inverted to the deep learning precipitation inversion model, and output the inversion result.

[0104] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A satellite data precipitation retrieval method based on deep learning, characterized in that Including the following steps: Taking the data obtained from satellite data in a preset area within a specified time as the data to be analyzed; the satellite data includes satellite observation data, auxiliary data, and label data; the data to be analyzed includes determined data and undetermined data; the certainty of the determined data is higher than that of the undetermined data; the satellite observation data includes microwave radiation data, infrared data, multi-channel radiometer data, and radar joint observation data; the auxiliary data includes terrain elevation, surface type, and atmospheric reanalysis data; Determining a first precipitation area through the determined data, performing uncertainty estimation on the undetermined data to obtain a second precipitation area, and predicting the precipitation scale probability for the data to be analyzed within the first precipitation area and the second precipitation area; Extracting precipitation scale features from the data to be analyzed to obtain precipitation scale features, and performing feature fusion on the precipitation scale probability and the precipitation scale features to obtain fusion data; Constructing a deep learning precipitation inversion model based on the attenuation rate using the fusion data, inputting the data to be inverted into the deep learning precipitation inversion model, and outputting the inversion result.

2. The satellite data precipitation retrieval method based on deep learning according to claim 1, wherein The method for determining the first precipitation area through the determined data includes: Using a neural network algorithm to learn the satellite images and their corresponding rainfall data in the determined data, automatically extracting rainfall features and establishing a mapping relationship, predicting future rainfall conditions according to the mapping relationship, and outputting the rainy area as the first precipitation area.

3. The precipitation retrieval method from satellite data based on deep learning according to claim 1, wherein The method for performing uncertainty estimation on the undetermined data to obtain a second precipitation area includes: Using a Bayesian neural network to regard the weights of each layer of the neural network as random variables, estimating the posterior distribution of the weights of each layer of the neural network, and predicting the uncertainty estimation of the undetermined data during prediction according to the posterior distribution; Performing spatial analysis on the undetermined data, and combining the undetermined information to obtain an uncertainty evaluation value for the predicted rainfall area; outputting the rainfall area with an uncertainty evaluation value lower than 0.152 as the second precipitation area.

4. The precipitation retrieval method from satellite data based on deep learning according to claim 1, wherein The method for obtaining the precipitation scale probability includes: Using a feature-enhanced hybrid model to combine statistical and machine learning algorithms, using advanced feature engineering to improve the prediction accuracy, analyzing the data to be analyzed within the precipitation area, estimating the probability distribution of multiple rainfall scales, and outputting the probability distribution of multiple rainfall scales as the precipitation scale probability.

5. The deep learning-based satellite data precipitation retrieval method according to claim 1, wherein The method for extracting precipitation scale features from the data to be analyzed to obtain precipitation scale features includes: Inputting the data to be analyzed in the first precipitation area and the second precipitation area into a precipitation scale feature extractor based on multi-head attention, and the expression is: The output feature of the i-th encoder layer is The intermediate feature after passing through the multi-head attention layer is The multi-layer attention layer is MSA, the multi-layer perceptron layer is MLP, and the output feature of the (i-1)-th encoder layer is The layer normalization is LN(·), the position encoding function is PosEnc(·), and the feature after position encoding in the i-th layer is Taking the output features as the rainfall scale features and outputting them.

6. The precipitation retrieval method from satellite data based on deep learning according to claim 1, wherein, The method for performing feature fusion on the precipitation scale probability and the precipitation scale features to obtain fusion data includes: Inputting the precipitation scale probability and the precipitation scale features into a multi-scale feature fusion device, and dividing the precipitation scale features into a global mark aggregating the entire area and an embedded representation of an image patch sequence; Converting the global mark into a form combined with the detailed feature K through a multi-layer perceptron layer, introducing sampling features, and the expression for feature fusion is: Among them, the multi-scale feature data is The fused feature is The feature embedding from the backbone network is K, and the feature after reorganizing the global token is z tk , the feature obtained after sampling through a factor is u, the multi-layer perceptron layer is MLP, the fusion module is Fusion(·), the sampled feature is u, and the preliminary fusion result is The precipitation scale probability is p, and the precipitation scale feature is c; The classification results of the usage scale classification module are used to constrain the precipitation feature fusion process for different scales, and the expression is: where the precipitation scale predictor is Γ(·), and the precipitation scale feature extractor is The feature map after weighted sum is v, the specified number of precipitation scales is n, the normalization exponential function is softmax(·), and the predicted precipitation scale of the multi-scale feature data is The i-th precipitation scale feature of the multi-scale feature data is The 2nd precipitation scale feature of the multi-scale feature data is The n-th precipitation scale feature of the multi-scale feature number is The precipitation scale probability distribution is The fused precipitation feature map of the precipitation scale feature and the precipitation scale probability distribution at the j-th column of the a-th row is The index of the x-th precipitation scale of the precipitation scale feature at the j-th column of the a-th row is The index of the x-th precipitation scale of the precipitation scale probability distribution at the j-th column of the a-th row Given a loss function based on the temperature coefficient: where the loss function is The cross-entropy loss function is The temperature coefficient is The mean squared error loss function is Map the fused precipitation features to output the fused data.

7. The precipitation retrieval method from satellite data based on deep learning according to claim 1, characterized in that, A method for constructing a deep learning precipitation inversion model using the fusion data based on the attenuation rate, including: Calculating the attenuation coefficient: where the attenuation cross-section is σ e , the wavelength is ρ, the particle diameter is D, and the mass-weighted mean diameter is D m , the normalized intercept is N w , the number concentration of particles is N, the number concentration of particles with particle diameter is N(D), the gamma distribution is Β(·), the temperature is T, the particle diameter D and the mass-weighted mean diameter D m of the particle size distribution function is f(D, D m ), and the shape parameter of the precipitation particle size distribution is γ; Revising the attenuation coefficient according to the field monitored precipitation data, and the expression is: where the corrected normalized intercept is the vertical height at the earth's curvature is h, and the correction factor for height h is the actual precipitation is y, and the predicted precipitation is the temperature difference at the rainfall ground when rainfall starts is ΔT, the humidity difference at the rainfall ground when rainfall starts is ΔS, the constants are r1, r2, r3 respectively, the adjustment factor is ω, and the corrected attenuation coefficient is Calculating the rainfall: Where the precipitation falling speed is V(D) and the precipitation is Q; Based on the deep learning precipitation inversion model of the attenuation coefficient, mass weighted average diameter and rainfall, the adjustment factor and the mass weighted average diameter are used to invert the revised normalized intercept, and the rainfall is inverted according to the inverted normalized intercept, and the precipitation inversion result is output.

8. A satellite data precipitation retrieval system based on deep learning for performing the method according to any one of claims 1-7, characterized in that Including: Data acquisition module: used to obtain the data in the satellite data in the preset area within the specified time as the data to be analyzed; the satellite data includes satellite observation data, auxiliary data and label data; the data to be analyzed includes determined data and uncertain data; the determined data has higher certainty than the uncertain data; Regional probability module: used to determine the first precipitation area through the determined data, estimate the uncertainty of the uncertain data to obtain the second precipitation area, and predict the precipitation scale of the data to be analyzed within the first precipitation area and the second precipitation area to obtain the precipitation scale probability; Feature fusion module: used to extract the precipitation scale features of the data to be analyzed to obtain precipitation scale features, and fuse the precipitation scale probability and the precipitation scale features to obtain fusion data; Modeling output module: used to construct a deep learning precipitation inversion model using the fusion data based on the attenuation rate, output the data to be inverted to the deep learning precipitation inversion model, and output the inversion result.

Images