Regional rainfall prediction method

Abstract

The invention relates to a regional rainfall prediction method. The method comprises the following steps: acquiring EC rainfall forecast data; constructing a sun and rain prediction model, inputting the EC rainfall forecast data into the sun and rain prediction model to obtain sun and rain forecast, the sun and rain prediction model being obtained through training of a rainfall forecast training set; obtaining EC numerical forecasting data, wherein the EC numerical forecasting data comprises data of a height layer wind field, a potential height field and a temperature and humidity field; constructing a terrain synergy heavy rainfall model, and obtaining terrain amplification forecast according to the EC numerical forecast data and the terrain synergy heavy rainfall model; and the weather forecast and the terrain amplification forecast are fused to obtain a rainfall forecast field, and final rainfall forecast is obtained through the rainfall forecast field. According to the method, the limitation of an existing forecasting model in the aspects of accuracy, real-time performance, resolution and the like is solved, and the accuracy of weather and rainstorm forecasting is remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of precipitation prediction, and in particular to a regional precipitation prediction method. Background Art

[0002] As the most common and important component of daily meteorological services, the accuracy of the weather forecast is directly related to public life and safety, and is also one of the core criteria for evaluating meteorological forecasting capabilities. However, the existing weather forecast methods have significant deficiencies, especially in the prediction of minor precipitation events, which are manifested in the frequent false alarms of minor precipitation. Taking Baise City as an example, the accuracy of urban forecasts in 2021 has dropped by 1.5% compared with the average level of the past three years. This downward trend is largely attributed to the uneven distribution of showery precipitation in warm areas during the flood season, which has caused a decrease in the accuracy of weather forecasts.

[0003] In the field of heavy rain forecasting, this forecasting task has always been extremely challenging, mainly due to the influence of multiple complex climate systems, such as local convective heavy rain on the edge of the subtropical high pressure, monsoon and typhoon-induced precipitation. These factors work together to make forecasting difficult.

[0004] In recent years, the rapid development of artificial intelligence technology in the field of meteorology, especially in the application of heavy rain research and forecasting, has shown great potential. The use of deep learning models, such as the U-net rain and shine forecast model, has improved the ability of feature extraction and data analysis, which provides a new technical approach to improve the accuracy of forecasts. However, although these models have made progress in some aspects, they are still insufficient in integrating terrain effects for precipitation forecasting. In particular, the terrain-enhanced heavy precipitation model has not yet achieved the expected forecast accuracy and real-time requirements when dealing with terrain complexity and its relationship with precipitation events. Therefore, a regional precipitation prediction method is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a regional precipitation prediction method, which aims to solve the limitations of existing forecast models in terms of accuracy, real-time performance and resolution, and significantly improve the accuracy of sunny, rainy and rainstorm forecasts.

[0006] To achieve the above object, the present invention provides a regional precipitation prediction method, comprising:

[0007] Obtain EC precipitation forecast data;

[0008] Constructing a sunny or rainy weather prediction model, inputting the EC precipitation forecast data into the sunny or rainy weather prediction model, and obtaining a sunny or rainy weather forecast, wherein the sunny or rainy weather prediction model is obtained by training with a precipitation forecast training set;

[0009] Acquiring EC numerical forecast data, wherein the EC numerical forecast data includes altitude layer wind field, potential height field, and temperature and humidity field data;

[0010] Constructing a terrain enhancement heavy precipitation model, and obtaining a terrain enhancement forecast based on the EC numerical forecast data and the terrain enhancement heavy precipitation model;

[0011] The clear and rainy forecast and the terrain increase forecast are integrated to obtain a precipitation forecast field, and a final precipitation forecast is obtained through the precipitation forecast field.

[0012] Optionally, the sunny or rainy forecast model is constructed by converting each convolution layer in a U-Net network into a Bayesian convolution layer.

[0013] Optionally, training the sunny and rainy prediction model using a precipitation forecast training set includes:

[0014] Collect historical EC precipitation forecast data in the ECMWF forecast mode, process the historical EC precipitation forecast data into grid data of a preset resolution, and obtain input data;

[0015] Acquire multi-source real-time precipitation data, process the multi-source real-time precipitation data into grid data of the preset resolution, and acquire label data;

[0016] The precipitation forecast training set is constructed by using the input data and the label data, and the sunny or rainy weather prediction model is trained by using the precipitation forecast training set.

[0017] Optional, obtaining the weather forecast includes:

[0018] The precipitation forecast data is input into the sunny or rainy forecast model, and the encoder of the sunny or rainy forecast model extracts the features of the precipitation forecast data to obtain a feature map;

[0019] The feature map is input into the decoder of the weather prediction model, and the Bayesian convolution layer in the decoder assigns weights to the feature map to obtain the weather forecast, wherein the weather forecast is a weather distribution grid field.

[0020] Optionally, the Bayesian convolution layer in the decoder assigns weights to the feature map, and obtaining the weather forecast includes:

[0021] The Bayesian convolution layer in the decoder assigns weights to the feature map, outputs some uncertainty information, averages the uncertainty information, and obtains an average probability of precipitation;

[0022] The average probability of precipitation is used as the value of each grid point in the sunny or rainy forecast as the probability of precipitation at the corresponding position to obtain the sunny or rainy forecast.

[0023] Optionally, obtaining terrain increase forecast includes:

[0024] Determining the generation of terrain precipitation according to the temperature field data, the altitude layer wind field data and the potential height field;

[0025] After determining that terrain precipitation occurs, the water vapor condensation in the atmosphere is calculated based on the humidity field data to obtain the terrain increase forecast, wherein the terrain increase forecast is a heavy precipitation field, and the value of each grid point in the heavy precipitation field is the expected precipitation of the terrain.

[0026] Optionally, the method for determining whether terrain precipitation occurs is: determining whether terrain precipitation occurs by calculating a wet Froude number;

[0027] The calculation method of the wet Froude number is:

[0028]

[0029] Where U is the average velocity of the airflow; N is the Brant-Vaisala frequency, g is the acceleration due to gravity, θ is the potential temperature, is the variation of potential temperature with height, and h is the characteristic height of the terrain.

[0030] Optionally, obtaining the heavy precipitation field includes:

[0031]

[0032] Among them, P is the condensation amount, that is, the precipitation amount, R is the condensation rate per unit height, z h is the height of the saturated wet layer, and z0 is the ground height.

[0033] Optionally, fusing the clear and rainy forecast with the terrain increase forecast to obtain a precipitation forecast field includes:

[0034] Superimposing the terrain increase forecast with the EC precipitation forecast data to obtain a final terrain increase forecast;

[0035] Multiplying the clear and rainy forecast and the final terrain increase forecast by grid points to obtain the precipitation forecast field;

[0036] The final precipitation forecast is obtained through the precipitation forecast field.

[0037] The beneficial effects of the present invention are as follows: by introducing the Bayesian U-Net model, the model introduces Bayesian reasoning on the basis of the classic U-Net architecture, so as to more effectively handle the uncertainty in the input data, and optimize and train with ECWMF fine grid data to improve the accuracy of the weather forecast. The model can more accurately estimate the possibility of precipitation under different meteorological conditions through the probabilistic weights of the Bayesian neural network, especially showing a stronger ability in capturing subtle meteorological changes. Compared with the traditional deterministic model, the Bayesian U-Net model can give an uncertainty assessment of the prediction and provide more information for decision makers. This quantification of uncertainty reduces the sensitivity of the model to noise and outliers, thereby effectively reducing the situation of empty reports and missed reports in the weather forecast, and greatly improving the forecast accuracy. In addition, the Bayesian method also allows the model to make reasonable inferences through prior knowledge when data is scarce or changing, further enhancing the adaptability and accuracy of the model under variable meteorological conditions.

[0038] The present invention adopts a physical parameterization method, including the blocking and guiding effect of terrain on the atmospheric flow field, the influence on the rise and sinking of airflow, and factors such as temperature. By accurately incorporating these physical processes into the model calculation, the influence of terrain on precipitation distribution and intensity can be more realistically reproduced, and the forecast accuracy of precipitation events above heavy rain is improved. The present invention combines the Bayesian U-Net model with actual terrain data to improve the accuracy of precipitation forecast. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] 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 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 these drawings without paying creative labor.

[0040] Figure 1 is an overall flow chart of a regional precipitation prediction method according to an embodiment of the present invention;

[0041] Figure 2 A Bayesian convolutional layer weight distribution diagram of an embodiment of the present invention;

[0042] Figure 3 20:00 on September 9, 2019 - 20:00 on September 10, 2019, a comparison diagram of the precipitation correction effect of an embodiment of the present invention, wherein (a) is the terrain correction forecast from 20:00 on September 9, 2019 to 20:00 on September 10, 2019, and (b) is the EC original forecast from 20:00 on September 9, 2019 to 20:00 on September 10, 2019;

[0043] Figure 4It is a comparison diagram of the precipitation correction effect from 08:00 on April 28, 2023 to 08:00 on April 29, 2023 according to an embodiment of the present invention, wherein (a) is the terrain correction forecast from 08:00 on April 28, 2023 to 08:00 on April 29, 2023, and (b) is the EC original forecast from 08:00 on April 28, 2023 to 08:00 on April 29, 2023;

[0044] Figure 5 Schematic diagram of the overall network framework of the Bayesian U-Net according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] 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.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] This embodiment provides a regional precipitation prediction method, such as Figure 1 As shown, including:

[0048] Obtain EC precipitation forecast data;

[0049] Construct a clear and rainy forecast model, input the EC precipitation forecast data into the clear and rainy forecast model, and obtain the clear and rainy forecast, where the clear and rainy forecast model is trained by the precipitation forecast training set;

[0050] Obtain EC numerical forecast data, including altitude wind field, geopotential height field, and temperature and humidity field data;

[0051] Construct a terrain enhancement heavy precipitation model, and obtain terrain enhancement forecast based on EC numerical forecast data and the terrain enhancement heavy precipitation model;

[0052] The clear and rainy forecast and the terrain enhancement forecast are integrated to obtain the precipitation forecast field, and the final precipitation forecast is obtained through the precipitation forecast field.

[0053] The specific steps of building a sunny and rainy forecast model are as follows:

[0054] In terms of structure, the Bayesian U-Net still retains the basic architecture of the traditional U-Net, which is composed of two main parts: encoding (downsampling) and decoding (upsampling). In the present invention, each convolutional layer is converted into a Bayesian convolutional layer based on the traditional U-Net. These Bayesian convolutional layers not only learn the features of the data, but also learn the uncertainty of these features, so that not only the predicted value but also the uncertainty assessment of the prediction can be given in the prediction of the model.

[0055] The encoder of the Bayesian U-Net model extracts the features of the input data, obtains feature maps, and fully utilizes the feature maps of the same resolution from the encoding process during the decoding process, and splices these feature maps with the feature maps in the decoding process through cross-layer connections. This not only enables the model to use shallow features more effectively, but also provides important information about data uncertainty during prediction, greatly improving the performance and reliability of the model when processing complex or uncertain data.

[0056] Therefore, the Bayesian U-Net model not only inherits the advantages of traditional U-Net in image segmentation, but also enhances the model's ability to handle uncertainty by introducing Bayesian reasoning, making it more robust and adaptable in various application scenarios, especially in those fields that require precise processing of uncertain information, such as medical image analysis, meteorological data processing, etc.

[0057] In the Bayesian U-Net model, the introduction of the Bayesian convolutional layer (Bayesian neural network) not only improves the model's ability to handle data uncertainty, but also enables the model to more effectively identify and focus on key features in the input data. The core advantage of the Bayesian convolutional layer is that it can evaluate the uncertainty of these features while learning data features, thereby providing important uncertainty information for the model's predictions.

[0058] In the decoding step, the Bayesian convolutional layer assigns weights to the feature maps from the encoding step based not only on the importance of the features, but also on the uncertainty of the features. This approach makes the model more intelligent and accurate when integrating features from the encoder, and prioritizes the most important and reliable features for the current task, rather than relying solely on traditional feature importance evaluation.

[0059] The Bayesian U-Net model not only inherits the advantages of traditional U-Net in image segmentation, but also significantly improves the model's ability to identify and utilize key features through the introduction of the Bayesian convolutional layer. Especially when facing data with high uncertainty, the Bayesian convolutional layer can not only more effectively distinguish between small precipitation and non-precipitation areas, but also quantify the uncertainty of the prediction, thereby greatly reducing the number of false positives and false negatives. The structure diagram of the Bayesian U-Net model is shown in the figure below. Figure 2 shown.

[0060] In solving this binary rainy or sunny forecast problem, we chose the binary cross entropy loss and combined it with the advantages of the Bayesian method, so that the model can not only measure the difference between the prediction and the actual value in the classification problem, but also consider the uncertainty of the prediction. This method helps guide the model to make more accurate and reliable predictions. At the same time, in order to ensure that the model has good generalization ability.

[0061] In terms of data set construction, EC precipitation data from 2019 to 2022 were selected. The inverse distance weighted method was used to interpolate the station precipitation data, that is, the EC precipitation forecast was interpolated to 5km, and a high-resolution training data set matching the EC precipitation data was generated. In the training data set, precipitation greater than 0.1mm is regarded as rain (1), and precipitation less than 0.1mm is regarded as sunny (0). In the face of the imbalance of sunny and rainy samples in the data set, an appropriate balanced processing strategy was adopted. That is, when constructing the data set, a constraint condition was used that the number of grid points with precipitation greater than 0.1mm accounted for more than 20% of the grid points or there were heavy precipitation events (maximum value greater than 50mm), and sunny and rainy days were marked. This ensures that during the model training process, both sunny and rainy weather conditions can obtain sufficient learning samples, thereby significantly improving the prediction performance.

[0062] In this embodiment, the Bayesian U-Net model uses the 3-hour precipitation data provided by EC (European Centre for Mesoscale Weather Forecasts) as input and the multi-source real-time precipitation data with a resolution of 5 km as training labels. The following steps are performed to achieve accurate prediction of precipitation events. The Bayesian U-Net model framework is as follows: Figure 5 As shown:

[0063] Step 1: Data preprocessing

[0064] Input data: Obtain the 3-hour historical precipitation forecast data provided by EC and perform necessary preprocessing, such as data normalization and missing value processing;

[0065] Label data: Integrate real-time precipitation data from multiple sources and convert them into 5km resolution grid data that matches the model input as the true label in the training and validation process.

[0066] Step 2: Model construction

[0067] Construct a Bayesian U-Net model, which incorporates the Bayesian reasoning mechanism based on the traditional U-Net architecture.

[0068] Step 3: Model training and validation

[0069] The Bayesian U-Net model was trained using the preprocessed historical EC precipitation data and precipitation data (labeled data);

[0070] During the training process, cross-validation and other techniques are used to avoid overfitting and ensure that the model has good generalization ability;

[0071] The model is validated using an independent test set to ensure the accuracy and reliability of the model prediction.

[0072] Step 4: Uncertainty Assessment

[0073] In the forecasting process, the model not only provides the forecast value, but also gives the corresponding uncertainty range, for example, the uncertainty of the forecast result is expressed through the posterior probability distribution;

[0074] Uncertainty assessment is of great significance in guiding practical meteorological decision-making and helps to evaluate the risk and credibility of forecasts.

[0075] Step 5: Output and Application

[0076] The EC precipitation forecast data is input into the sunny or rainy forecast model to obtain the sunny or rainy forecast. The model outputs precipitation forecast results with a resolution of 5km, including the predicted precipitation amount and its uncertainty assessment. The output results can be directly used for meteorological forecast services to provide more accurate and reliable precipitation information to the public and relevant departments.

[0077] The EC numerical forecast data is obtained, wherein the EC numerical forecast data includes the high-level wind field, the potential height field, and the temperature and humidity field data; a terrain enhancement heavy precipitation model is constructed, and a terrain enhancement forecast is obtained according to the EC numerical forecast data and the terrain enhancement heavy precipitation model.

[0078] The specific construction of terrain enhanced heavy precipitation model is as follows:

[0079] Traditional terrain precipitation estimation methods, such as the formula proposed by Smith (1979), are effective in many cases, but they show certain limitations when dealing with precipitation forecasts under complex terrain conditions. In particular, there are obvious deficiencies in the upper limit of integration, precipitation efficiency, and judgment criteria, which are particularly evident when dealing with heavy precipitation forecasts in specific areas.

[0080] The present invention proposes to use the saturated wet layer height as the integral upper limit to more reasonably reflect the actual precipitation process. This improvement is based on a deeper understanding of the condensation and vertical movement of water vapor in the precipitation process.

[0081] Precipitation efficiency: The region is divided at a resolution of 5 km, and the precipitation efficiency of terrains at different heights is re-evaluated and quantified. This method takes into account the effect of terrain height on the precipitation process and more accurately simulates the effect of terrain on precipitation.

[0082] a. Wet Froude number calculation:

[0083] The wet Froude number (Fr) is a dimensionless number used to evaluate the flow pattern of a fluid as it flows over an obstacle. In the context of orographic precipitation, it can be used to determine whether the airflow is sufficient to pass over a terrain obstacle to cause precipitation. The formula for the wet Froude number is:

[0084]

[0085] Where: U is the average speed of the airflow (which can be obtained from the altitude wind field data of the ECMWF model), N is the Brant-Vaisala frequency, which represents the atmospheric stability, and h is the characteristic height of the terrain.

[0086] b. Brant-Vaisala frequency (N) calculation:

[0087] The Brant-Vaisala frequency is a measure of atmospheric stability and is calculated as:

[0088]

[0089] Where: g is the acceleration due to gravity, θ is the potential temperature (a meteorological quantity that indicates temperature change), is the rate of change of potential temperature with height.

[0090] c. Terrain precipitation judgment:

[0091] When the calculated wet Froude number is greater than 1, it indicates that the airflow is strong enough to pass over the terrain obstacle and may produce terrain precipitation. Otherwise, it is unlikely to produce terrain precipitation.

[0092] d. Calculation of condensation amount:

[0093] If it is determined to be orographic precipitation, the calculation of condensation amount can be based on the condensation of water vapor in the atmosphere. The calculation formula of condensation amount may be:

[0094]

[0095] Where: P is the condensation amount, i.e., precipitation, R is the condensation rate per unit height, z h and z0 are the lower and upper limits of the integration respectively. The upper limit is the saturated wet layer height mentioned above, and z0 is the ground height.

[0096]

[0097] Among them, ρ w =r s ρ air , r s is the saturation mixing ratio, ρ airis the air density, the saturated water vapor density (ρ ws ) using the actual air water vapor density (ρ w ) is replaced by , α is the terrain slope, and U(z) is the horizontal wind.

[0098] Judgment standard: The dimensionless wet Froude number is introduced as a new judgment standard for orographic precipitation. When it is greater than 1, it indicates that orographic precipitation has occurred. This standard provides a new and more scientific method for orographic precipitation forecasting.

[0099] This embodiment uses the altitude layer wind field, potential height field and temperature field data predicted by the ECMWF model to calculate the wet Froude number, so as to determine whether terrain precipitation occurs. If it is determined to be terrain precipitation, the condensation amount is calculated according to the condensation amount (the condensation of water vapor in the atmosphere is obtained by processing the humidity field) and the integral upper limit, and then the terrain heavy precipitation amount, i.e., the terrain enhanced forecast, is obtained, wherein the terrain heavy precipitation amount obtained is a terrain heavy precipitation amount field with a resolution of 5 km.

[0100] Precipitation efficiency considerations: The precipitation efficiency of different terrains may be different, and the calculated precipitation can be adjusted according to the terrain characteristics.

[0101] The specific integration of the clear and rainy forecast and the heavy precipitation field is as follows:

[0102] In a complete terrain and AI fusion precipitation model strategy, combining the advantages of the Bayesian U-Net sunny and rainy forecast model and the terrain enhanced heavy precipitation model is a crucial step. This integration strategy aims to ensure that the model can provide comprehensive and accurate predictions under various precipitation conditions. In order to achieve the fusion of the two models, the terrain increase forecast and the sunny and rainy forecast are grid-multiplied to present a continuous and physically meaningful precipitation forecast field. Through this strategy, not only the accuracy of the prediction is improved, but also the continuity of the prediction results in space is ensured. Among them, the terrain increase forecast is superimposed with the EC precipitation forecast data before the terrain increase forecast of the terrain precipitation field is grid-multiplied with the sunny and rainy forecast of the sunny and rainy forecast field.

[0103] (1) Output of Bayesian U-Net weather forecast model:

[0104] The Bayesian U-Net model outputs a sunny or rainy forecast, which is a sunny or rainy distribution grid field, presented in the form of grid points, and the value of each grid point represents the probability of precipitation at that location. The decoder of the Bayesian U-Net model will calculate 30 uncertain values, and will average them as output for the purpose of deterministic forecasting. The output is the average probability of precipitation, and the output probability value is between 0 and 1. Finally, greater than 0.5 is 1 and less than 0.5 is 0, where 0 represents no precipitation and 1 represents precipitation.

[0105] (2) Output of terrain enhanced heavy precipitation model:

[0106] The terrain enhanced heavy precipitation model outputs a heavy precipitation amount field, which is also represented in grid form.

[0107] The value of each grid point indicates the expected precipitation under the terrain conditions. The final terrain-enhanced forecast is obtained by superimposing the original EC 3-hour precipitation forecast.

[0108] (3) Grid multiplication technique:

[0109] The output of the Bayesian U-Net sunny and rainy forecast model is grid-multiplied with the superposition results of the terrain enhancement heavy precipitation model and EC precipitation forecast data, and the fusion output is obtained. That is, (terrain enhancement forecast + EC precipitation forecast data) x sunny and rainy forecast = final rainfall forecast field.

[0110] (4) Final fusion output:

[0111] The fused output provides a comprehensive and accurate precipitation forecast field that not only improves the accuracy of the forecast but also maintains continuity in spatial distribution, allowing the model to effectively cope with a variety of precipitation and terrain conditions.

[0112] The method of the present invention can significantly improve the forecast performance of the model for heavy precipitation events, especially the forecast accuracy when the precipitation system interacts strongly with the terrain. Table 1 shows the TS score of the 24-hour precipitation forecast of the Bayesian U-Net model from April to September 2023.

[0113] Table 1

[0114] Precipitation threshold (mm) TS score of this model The original TS score of the control group Remark More than 0.1 mm 0.7343 0.6021 The accuracy of minor precipitation events has been significantly improved More than 10 mm 0.3645 0.3630 Moderate precipitation event forecast performance is comparable More than 25 mm 0.2630 0.2533 Moderate to heavy precipitation events increased slightly More than 50 mm 0.1757 0.1372 The accuracy of heavy precipitation events has been significantly improved More than 100 mm 0.0744 0.0543 Improved forecast accuracy of extreme heavy precipitation events

[0115] The data in Table 1 show the comparison of TS scores of the Bayesian U-Net model and the terrain enhanced heavy precipitation model of the present invention at different precipitation thresholds and the original EC numerical forecast of 24-hour rainfall. The data show that under all tested precipitation thresholds, the present model exhibits forecasting capabilities equal to or better than existing methods, especially in the forecast of light precipitation and heavy precipitation events, showing significant advantages.

[0116] Table 2 is a comparison of the precipitation correction effects from 20:00 on September 9, 2019 to 20:00 on September 10, 2019. Table 2 corresponds to Figure 3 (a)-(b), improved TS score of heavy rain.

[0117] Table 2

[0118] Precipitation threshold (mm) TS score of this model The original TS score of the control group More than 10 mm 0.58 0.57 More than 25 mm 0.38 0.31 More than 50 mm 0.19 0.16 More than 100 mm 0.04 0.03

[0119] Table 3 shows the precipitation comparison and Figure 4(a)-(b) match, and the data show that the present invention can improve the TS score of heavy rain.

[0120] Table 3

[0121]

[0122]

[0123] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A regional precipitation prediction method, characterized in that: include: Obtain EC precipitation forecast data; Constructing a sunny or rainy weather prediction model, inputting the EC precipitation forecast data into the sunny or rainy weather prediction model, and obtaining a sunny or rainy weather forecast, wherein the sunny or rainy weather prediction model is obtained by training with a precipitation forecast training set; Acquiring EC numerical forecast data, wherein the EC numerical forecast data includes altitude layer wind field, potential height field, and temperature and humidity field data; Constructing a terrain enhancement heavy precipitation model, and obtaining a terrain enhancement forecast based on the EC numerical forecast data and the terrain enhancement heavy precipitation model; The clear and rainy forecast and the terrain increase forecast are integrated to obtain a precipitation forecast field, and a final precipitation forecast is obtained through the precipitation forecast field.

2. The regional precipitation prediction method according to claim 1, characterized in that: The sunny and rainy prediction model is constructed by converting each convolutional layer in the U-Net network into a Bayesian convolutional layer.

3. The regional precipitation prediction method according to claim 1, characterized in that: Training the sunny and rainy weather prediction model using the precipitation forecast training set includes: Collect historical EC precipitation forecast data in the ECMWF forecast mode, process the historical EC precipitation forecast data into grid data of a preset resolution, and obtain input data; Acquire multi-source real-time precipitation data, process the multi-source real-time precipitation data into grid data of the preset resolution, and acquire label data; The precipitation forecast training set is constructed by using the input data and the label data, and the sunny or rainy weather prediction model is trained by using the precipitation forecast training set.

4. The regional precipitation prediction method according to claim 1, characterized in that: Obtaining the weather forecast includes: The precipitation forecast data is input into the sunny or rainy forecast model, and the encoder of the sunny or rainy forecast model extracts the features of the precipitation forecast data to obtain a feature map; The feature map is input into the decoder of the weather prediction model, and the Bayesian convolution layer in the decoder assigns weights to the feature map to obtain the weather forecast, wherein the weather forecast is a weather distribution grid field.

5. The regional precipitation prediction method according to claim 4, characterized in that: The Bayesian convolution layer in the decoder assigns weights to the feature map, and obtaining the weather forecast includes: The Bayesian convolution layer in the decoder assigns weights to the feature map, outputs some uncertainty information, averages the uncertainty information, and obtains an average probability of precipitation; The average probability of precipitation is used as the value of each grid point in the sunny or rainy forecast as the probability of precipitation at the corresponding position to obtain the sunny or rainy forecast.

6. The regional precipitation prediction method according to claim 1, characterized in that: Obtaining terrain increase forecasts includes: Determining the generation of terrain precipitation according to the temperature field data, the altitude layer wind field data and the potential height field; After determining that terrain precipitation occurs, the water vapor condensation in the atmosphere is calculated based on the humidity field data to obtain the terrain increase forecast, wherein the terrain increase forecast is a heavy precipitation field, and the value of each grid point in the heavy precipitation field is the expected precipitation of the terrain.

7. The regional precipitation prediction method according to claim 6, characterized in that: The method for determining the occurrence of orographic precipitation is: judging whether orographic precipitation occurs by calculating the wet Froude number; The calculation method of the wet Froude number is: Where U is the average velocity of the airflow; N is the Brant-Vaisala frequency, g is the acceleration due to gravity, θ is the potential temperature, is the variation of potential temperature with height, and h is the characteristic height of the terrain.

8. The regional precipitation prediction method according to claim 6, characterized in that: Acquiring the heavy precipitation field includes: Among them, P is the condensation amount, that is, the precipitation amount, R is the condensation rate per unit height, z h is the height of the saturated wet layer, and z0 is the ground height.

9. The regional precipitation prediction method according to claim 1, characterized in that: The rainy and sunny forecast and the terrain increase forecast are integrated to obtain the precipitation forecast field, including: Superimposing the terrain increase forecast with the EC precipitation forecast data to obtain a final terrain increase forecast; Multiplying the clear and rainy forecast and the final terrain increase forecast by grid points to obtain the precipitation forecast field; The final precipitation forecast is obtained through the precipitation forecast field.

Images