A potential intelligent forecasting method for extreme rainfall intensity based on optimized physical factors

Abstract

The present invention discloses a method for intelligently forecasting extreme rainfall potential based on optimized physical factors. The method comprises the following steps: preparing numerical model forecast data for training the intelligent extreme rainfall potential forecast model; preparing precipitation observation data for training the intelligent extreme rainfall potential forecast model; calculating physical diagnostic quantities based on the numerical model; constructing forecast factors for the intelligent extreme rainfall forecast model; performing sensitivity analysis and preliminary screening on the forecast factors; performing dimensionality reduction on the forecast factors; constructing fuzzy matching observation labels for spatiotemporal neighborhoods; constructing an optimal extreme rainfall balance sample training set; data cleaning and standardization preprocessing; constructing the intelligent extreme rainfall forecast model and optimizing key parameters; and intelligently forecasting and evaluating the forecast results. The present invention addresses the problems of magnitude underestimation and area deviation in existing numerical model forecasts of extreme rainfall intensity for the next three days.

Description

Technical Field

[0001] The present invention relates to a forecasting method, in particular to an extreme rainfall intensity potential intelligent forecasting method based on optimized physical factors. Background Art

[0002] Extreme rainfall is characterized by localized, sudden, and highly catastrophic nature. Its causes are complex, involving the influence of multiple scale factors (synoptic-scale systems, small and medium-scale convective systems, and topographic effects). Against the backdrop of global warming, extreme rainfall events are becoming more frequent, and their impacts are becoming increasingly severe.

[0003] Currently, numerical weather prediction models (NWMs) are the primary means of conducting precipitation forecasts. However, due to numerous limitations in initial fields, physical processes, and dynamical frameworks, NWMs exhibit significant deficiencies in forecasting heavy precipitation, particularly extreme rainfall intensities. This is manifested in significant underestimation of extreme rainfall central values and significant deviations in the predicted rainfall areas. In particular, there is virtually no forecast skill for extreme rainfall intensities exceeding 24 hours.

[0004] Furthermore, traditional bias correction methods, such as "frequency matching," eliminate systematic biases in numerical model precipitation forecasts by correcting the frequency of numerical model precipitation forecasts to the observed precipitation frequency. However, this approach is limited in its ability to correct forecast biases only for small and medium-sized precipitation events, but is unable to correct forecast biases for heavy precipitation, especially extreme precipitation. Furthermore, while recent advances in artificial intelligence (AI) have made some progress in correcting nonlinear biases in numerical model precipitation forecasts, learning remains ineffective for small sample sizes of events, such as extreme precipitation. Summary of the Invention

[0005] In order to address the deficiencies of the above technologies, the present invention provides an intelligent forecasting method for extreme rainfall intensity potential based on optimal physical factors.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: an extreme rainfall intensity potential intelligent forecasting method based on optimal physical factors, comprising the following steps:

[0007] Step 1: Prepare numerical model forecast data for training the extreme rainfall intensity intelligent forecast model, covering both ground and high-altitude elements;

[0008] Step 2: Prepare precipitation observation data for training the extreme rainfall intensity intelligent forecast model. The precipitation observation data is hourly grid observation data.

[0009] Step 3: Calculate the physical diagnostic quantity based on the numerical model based on the numerical model forecast data collected in step 1;

[0010] Step 4: Based on the direct output of the numerical model and the physical diagnostic quantity calculated in step 3, the prediction factors of the extreme rainfall intensity intelligent forecast model are constructed to obtain the prediction factor data set;

[0011] Step 5: Conduct sensitivity analysis on the prediction factors and conduct preliminary screening, and select highly sensitive physical quantities as prediction factors for the extreme rainfall intensity intelligent forecast model;

[0012] Step 6: Perform dimensionality reduction processing on the prediction factors of the extreme rainfall intensity intelligent forecast model to optimize the screening;

[0013] Step 7: Construct fuzzy matching observation labels of the spatiotemporal neighborhood for precipitation observation data;

[0014] Step 8: Construct the optimal extreme rainfall intensity balanced sample training set;

[0015] Step 9: Perform data cleaning and standardization preprocessing on the optimal extreme rainfall intensity balance sample training set;

[0016] Step 10: construct an extreme rainfall intensity intelligent forecast model and optimize key parameters to obtain the optimal extreme rainfall intensity intelligent forecast model;

[0017] Step 11: Apply the optimal extreme rainfall intensity intelligent forecast model to make forecasts, output potential forecasts of extreme rainfall intensity for every 3 hours from 0 to 72 hours, and evaluate the forecast effect.

[0018] Preferably, in step 1, the prepared numerical model forecast data are the output variables of the numerical model with a forecast time of 0-72 hours in the past five years, covering ground element data and high-altitude element data, including total precipitation, convective precipitation, large-scale precipitation, sea level pressure, surface pressure, ground 2m air temperature, ground 2m dew point, atmospheric temperature field, potential height field, and three-dimensional wind field; the spatial resolution of the data is 0.125°, and the temporal resolution is 3 hours.

[0019] Preferably, in step 2, the precipitation observation data is hourly grid observation data with a spatial resolution of 1 km and a temporal resolution of 1 hour.

[0020] Preferably, in step 3, the physical diagnostic quantities calculated are the vorticity, divergence, water vapor flux divergence, pseudo-equivalent potential temperature of each pressure layer, and the K index and A index describing the stability of the atmospheric stratification.

[0021] Preferably, in step 4, the process of constructing the predictor data set is:

[0022] Based on the meteorological elements directly output by the numerical model for each forecast time collected in step 1 and the calculation results of the physical diagnostic quantities calculated in step 3, a forecast factor dataset for the extreme rainfall intensity intelligent forecast model for each forecast time is constructed. The forecast time is 3 hours, 6 hours, 9 hours, ..., 72 hours.

[0023] Preferably, in step 5, the sensitivity analysis and preliminary screening process of the predictor factors is as follows:

[0024] Based on the hourly precipitation grid observation data collected in step 2 and the zero-field forecast data of the numerical model physical quantities collected in step 3, the probability density distribution of the physical quantities corresponding to the occurrence of extreme rainfall intensity and non-extreme rainfall intensity events is calculated grid by grid; the physical quantities with a mean difference exceeding one standard deviation are used as forecast factors for the extreme rainfall intensity intelligent forecast model.

[0025] Preferably, in step 6, the dimensionality reduction process of the prediction factors is performed by creating a group of unrelated prediction factors through principal component analysis, and the number of principal components to be retained is selected by testing the impact of retaining different numbers of principal components on the model score.

[0026] Preferably, in step 7, the spatiotemporal neighborhood fuzzy matching method of precipitation observation data is:

[0027] First, based on hourly precipitation grid observation data, we construct the first-level raw observation labels for hourly extreme rainfall intensity. That is, we determine whether there is observed precipitation >50 mm / h at each grid point on an hourly basis. If so, the raw observation label is recorded as 1 at that grid point in that hour; otherwise, it is recorded as 0.

[0028] Then, a secondary observation label for extreme rainfall intensity is constructed. That is, under the condition that the primary original observation label is 1, based on the forecast timeliness of the numerical model, it is determined whether there is observed precipitation > 50 mm / h within the 3-hour period covered by each timeliness. If so, the secondary observation label is recorded as 1, otherwise it is recorded as 0.

[0029] Finally, a third-level observation label for extreme rainfall intensity is constructed. That is, under the condition that the second-level label is 1, it is determined whether there is observed precipitation >50 mm / h at the target grid point and the surrounding 0.25° range within the 3 hours covered by each time period. If so, the third-level observation label is recorded as 1, otherwise it is recorded as 0.

[0030] Preferably, in step 8, the method for constructing the optimal extreme rainfall intensity balanced sample training set is:

[0031] Using the random downsampling method, samples with 1, 2, and 3 times the number of extreme rainfall intensity samples are randomly extracted from the non-extreme rainfall intensity samples to construct a new non-extreme rainfall intensity sample set. The non-extreme samples are samples with the observation label 0.

[0032] The extreme rainfall intensity samples and new non-extreme rainfall intensity samples are divided into multiple extreme rainfall intensity balanced sample sets in different proportions, including 1:1, 1:2, and 1:3. Model training experiments are carried out to determine the optimal ratio of the number of extreme rainfall intensity samples to non-extreme rainfall intensity samples, and finally the optimal extreme rainfall intensity balanced sample set is constructed.

[0033] Preferably, in step 11, based on the binary classification concept, by constructing a confusion matrix under a specific threshold of 50 mm / h, different precipitation forecast evaluation indicators are calculated, including the risk score TS, the hit rate POD, the false alarm rate FAR, and the missed alarm rate MAR. The calculation formulas are shown in Equations 7-10:

[0034] Risk score TS = number of positive hits / (number of positive hits + number of false alarms + number of missed alarms) (7)

[0035] Hit rate POD = number of positive hits / (number of positive hits + number of missed reports) (8)

[0036] FAR = false alarm number / (positive hit number + false alarm number) (9)

[0037] Missing alarm rate MAR = number of missed alarms / (number of positive hits + number of missed alarms) (10).

[0038] This paper discloses an intelligent forecasting method for extreme rainfall intensity potential based on optimized physical factors. This method addresses the significant bias in existing numerical models' extreme rainfall intensity forecasts for the next three days, particularly underestimation of precipitation magnitude and biased prediction of rainfall areas. By integrating optimized physical factor forecasts from numerical weather forecast models with high-temporal and spatial resolution precipitation observation data, combined with machine learning modeling, the method effectively improves the hit rate of three-hourly extreme rainfall intensity potential forecasts for the next three days (0-72 hours). BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic flow chart of the technical solution of the present invention.

[0040] Figure 2 This is a comparison chart of the precipitation observation from 20:00 to 23:00 on August 5, 2023 in an embodiment of the present invention, the 63-hour time-sensitive forecast of the EC model, and the 63-hour time-sensitive forecast results of the model of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] The present invention utilizes the meteorological element forecast data of the numerical weather forecast model and the high temporal and spatial resolution multi-source fusion grid precipitation observation data, and constructs an extreme rainfall intensity potential intelligent forecast model based on a machine learning model, realizing the intelligent forecast of extreme rainfall intensity potential every 3 hours from 0 to 72 hours.

[0043] Among them, extreme rainfall intensity refers to the precipitation intensity when the cumulative precipitation exceeds 50 mm within one hour; extreme rainfall potential refers to the possibility that the rainfall in a certain place will exceed 50 mm in one hour within a specific time period in the future.

[0044] like Figure 1 As shown, the technical solution process of the present invention is specifically as follows:

[0045] Step 1: Prepare numerical model forecast data for training the extreme rainfall intensity intelligent forecast model: the numerical model forecast data covers ground elements and high-altitude elements;

[0046] Specifically, the data includes: collecting output variables of numerical models (such as the EC model) with a 0-72 hour forecast timeframe over the past five years, covering both ground and upper air element data, including total precipitation, convective precipitation, large-scale precipitation, sea level pressure, surface pressure, air temperature at 2 meters above ground, dew point at 2 meters above ground, atmospheric temperature field, geopotential height field, and three-dimensional wind field, with a spatial resolution of 0.125° (approximately 13 km) and a temporal resolution of 3 hours;

[0047] Step 2: Prepare precipitation observation data for training the extreme rainfall intensity intelligent forecast model. The precipitation data is hourly grid observation data.

[0048] The hourly precipitation grid observation data of the National Meteorological Information Center's Multi-Source Fusion Live Analysis Product (CMPAS) for the past five years were collected, with a spatial resolution of 1 km and a temporal resolution of 1 hour.

[0049] Step 3: Calculate physical diagnostic quantities based on the numerical model: Calculate physical diagnostic quantities based on the numerical model forecast data collected in step 1;

[0050] Specifically, based on the ground and upper air element data predicted by the numerical model (such as the EC model) collected in step 1, the vorticity, divergence, water vapor flux divergence, pseudo-equivalent potential temperature of each pressure layer, as well as physical diagnostic quantities such as the K index and A index that describe the stability of atmospheric stratification are calculated. The physical meaning and calculation formulas of the above physical diagnostic quantities can be found in formulas (1-6):

[0051] 1) Vorticity is used to characterize the degree of atmospheric rotation, using relative vorticity express:

[0052] (1),

[0053] in, and are the latitudinal and longitudinal wind speed components, respectively; and Represent the longitudinal distance and latitudinal distance of the unit grid, Represents the partial differential operation.

[0054] 2) Divergence D, which indicates the degree of airflow convergence or divergence within a unit volume:

[0055] (2),

[0056] in, and are the latitudinal and longitudinal wind speed components, respectively; and Represent the longitudinal distance and latitudinal distance of the unit grid, Represents the partial differential operation.

[0057] 3) Water vapor flux divergence (MFD), which indicates the convergence or divergence of water vapor transport:

[0058] (3),

[0059] Where V is the horizontal wind vector and q is the specific humidity.

[0060] 4) False equivalent temperature , represents the potential temperature after considering the latent heat effect of water vapor:

[0061] (4),

[0062] in, Indicates the position temperature, is the mixing ratio, is the dry air gas constant, is the specific heat capacity at constant pressure.

[0063] 5) K index, which represents the stability of atmospheric stratification. The larger the K index, the more unstable the stratification.

[0064] (5),

[0065] in, 、 and are the temperatures at the 850hPa, 700hPa, and 500hPa pressure layers, and They are the dew point temperatures at the 700hPa and 500hPa pressure layers respectively.

[0066] 6) A index, which represents the stability of atmospheric stratification. The larger the A index, the more unstable the stratification.

[0067] (6),

[0068] in, and are the temperatures at the 850hPa and 500hPa pressure layers, It is the dew point temperature at the 850hPa pressure layer.

[0069] Step 4: Construct prediction factors for the extreme rainfall intensity intelligent forecast model;

[0070] Based on the ground and upper-air elements predicted by the numerical model (such as the EC model) collected in step 1, and the physical diagnostic quantities (such as vorticity, divergence, water vapor flux, pseudo-equivalent potential temperature, K index, A index, etc.) calculated in step 3, a prediction factor dataset for each forecast time (3 hours, 6 hours, 9 hours,..., 72 hours) of the extreme rainfall intensity intelligent forecast model is constructed.

[0071] Step 5: Sensitivity analysis and preliminary screening of predictors;

[0072] Based on the hourly precipitation grid observation data (CMPAS) collected in step 2 for the past five years and the numerical model zero field (0-hour forecast) collected in step 3, the physical quantities corresponding to extreme rainfall intensity and non-extreme rainfall intensity events (such as meteorological elements directly output by the EC model and vorticity, divergence, water vapor flux divergence, pseudo-equivalent potential temperature, K index, A index, etc.) are calculated grid by grid point, and the physical quantities with mean differences exceeding one standard deviation are used as prediction factors for the extreme rainfall intensity forecast model.

[0073] Step 6: Dimensionality reduction and optimization screening of model prediction factors;

[0074] Since the model has many predictive factors (about 70) and there is a strong correlation between the predictive factors, it is easy to cause overfitting of the prediction model. Therefore, it is necessary to carry out dimensionality reduction of the model predictive factors.

[0075] The present invention creates a group of unrelated predictors through principal component analysis, and selects the final number of principal components to be retained by testing the impact of retaining different numbers of principal components on the model score (Formulas 7-10).

[0076] Step 7: Construct fuzzy matching observation labels of the spatiotemporal neighborhood;

[0077] Since grid observation precipitation data has higher spatiotemporal resolution, in order to retain the observation information of extreme rainfall intensity as much as possible and better match the output results of numerical models with lower resolution, the construction of spatiotemporal neighborhood fuzzy matching observation labels for precipitation observations is carried out.

[0078] First, based on the hourly precipitation grid observation data, the first-level original observation label of extreme rainfall intensity is constructed. That is, it is judged hour by hour and grid by grid point whether there is observed precipitation >50 mm / h. If so, the original observation label is recorded as 1 at that grid point in that hour; otherwise, it is recorded as 0.

[0079] Then, a secondary observation label for extreme rainfall intensity is constructed. That is, under the condition that the primary original observation label is 1, according to the forecast timeliness of the numerical model, it is judged whether there is observed precipitation >50 mm / h within the 3 hours covered by each timeliness (for example, the observations corresponding to the 6-hour timeliness on July 29, 2024 should be 04:00, 05:00, and 06:00 on July 29, 2024). If so, the secondary observation label is recorded as 1, otherwise it is recorded as 0.

[0080] Finally, a third-level observation label for extreme rainfall intensity is constructed. That is, under the condition that the second-level label is 1, it is determined whether there is observed precipitation >50 mm / h at the target grid point and the surrounding 0.25° range within the 3 hours covered by each time period. If so, the third-level observation label is recorded as 1, otherwise it is recorded as 0.

[0081] Step 8: Constructing the optimal extreme rainfall intensity balanced sample training set;

[0082] Since the number of extreme rainfall intensity samples is extremely small (i.e., samples with the observation label recorded as 1 only account for 1% of the total samples), there is a problem of extreme imbalance between extreme rainfall intensity samples and non-extreme rainfall intensity samples in model training, which is not conducive to the construction of extreme rainfall intensity forecast model.

[0083] Based on steps 5 and 6, a random downsampling method is used to randomly extract 1 times, 2 times, and 3 times the number of extreme rainfall intensity samples from non-extreme rainfall intensity samples (i.e., samples with observation label 0) to construct a new non-extreme rainfall intensity sample set.

[0084] By using multiple extreme rainfall intensity balanced sample sets consisting of different ratios (1:1, 1:2, 1:3) of extreme rainfall intensity samples to non-extreme rainfall intensity samples, model training experiments were carried out to determine the optimal ratio of extreme rainfall intensity samples to non-extreme rainfall intensity samples, and finally the optimal extreme rainfall intensity balanced sample set was constructed.

[0085] Step 9: Data cleaning and standardization preprocessing;

[0086] The optimal extreme rainfall intensity balance sample set determined in step 8 is subjected to data cleaning (including removal of missing values, etc.) and normalization preprocessing to improve data quality and enhance model performance.

[0087] Step 10: Construction of an extreme rainfall intensity intelligent forecast model and optimization of key parameters;

[0088] Using the optimal balanced extreme rainfall intensity sample set constructed in steps 8-9, divide the sample set into training, validation, and test sets in a ratio (e.g., 7:2:1). Develop an intelligent extreme rainfall forecast model based on machine learning methods such as random forests. Use the training and validation sets of the optimal balanced extreme rainfall intensity sample set to train and validate the model. Optimal values for the model parameters (such as the number of decision trees and maximum tree depth) are determined using a grid search method. Cross-validation is used to verify the simulation results of key parameters.

[0089] Step 11: Evaluate the forecast effect of the extreme rainfall intensity intelligent forecast model.

[0090] The constructed extreme rainfall intelligent forecast model was fed with the test set of the optimal extreme rainfall balance sample set. It outputs a three-hourly extreme rainfall potential forecast for the 0-72 hour period (a 0-1 scale, with 0 indicating no extreme rainfall and 1 indicating extreme rainfall). Based on the model output, the model's forecast performance evaluation metrics were calculated, as shown in Table 1. The model's overall forecast performance was evaluated by calculating the false alarm rate (FAR), missed alarm rate (MAR), hit probability (POD), and risk score (TS) using Formulas 7-10.

[0091] The specific evaluation indicators of the extreme rainfall intensity intelligent forecast model are as follows:

[0092] Unlike temperature, which is a continuous variable, precipitation is typically a discontinuous variable. Therefore, meteorological accuracy testing of precipitation forecasts is typically based on binary classification. This is done by constructing a confusion matrix at a specific threshold (50 mm / hour) to calculate various precipitation forecast evaluation metrics.

[0093] Table 1 Confusion matrix between extreme rainfall intensity observation and forecast

[0094]

[0095] Risk score (TS) = number of positive hits / (number of positive hits + number of false alarms + number of missed alarms) (7)

[0096] FAR = False Alarms / (Number of Hits + Number of False Alarms) (8)

[0097] Missing alarm rate (MAR) = number of missed alarms / (number of positive hits + number of missed alarms) (9)

[0098] Percentage of Odds (POD) = Number of hits / (Number of hits + Number of missed hits) (10)

[0099] Formulas 7-10 provide the specific calculation formulas for the risk score (TS), false alarm rate (FAR) score, missed alarm rate (MAR) score, and hit rate (POD) score.

[0100] Application examples:

[0101] In this example, the extreme rainfall events in the North China Huanghuai region from May to September 2023 were taken as the research object. The EC model (a mainstream numerical model) forecast with a 0-72 hour forecast time from May to September 2019 to 2022 (a time resolution of 3 hours and a spatial resolution of 0.125°) and the multi-source fusion grid precipitation observation data of the National Meteorological Information Center (a time resolution of 1 hour and a spatial resolution of 1 km) were used as the input data of the extreme rainfall intelligent forecast model.

[0102] Through sensitivity analysis of predictive factors, 72 predictive factors, including model-based forecasts such as total precipitation, horizontal wind field, and vertical velocity, as well as physical diagnostic quantities such as vorticity and divergence calculated from these model forecasts, were selected as model inputs (see Table 2). Key model parameters, training windows, and the number of principal components retained in principal component analysis were calibrated. At each grid point, an intelligent extreme rainfall intensity prediction model based on optimized physical factors (constructed by the present invention) was used to predict the potential for extreme rainfall intensity.

[0103] Table 3 presents the evaluation results of the proposed model's back-calculated 0-72-hour forecast for extreme rainfall intensity from May to September 2023, and compares them with the EC model's forecast for the same period. It can be seen that the EC model's 0-72-hour extreme rainfall intensity forecast has a hit rate of only 0.001 and a TS score of only 0.001. However, the proposed model significantly improves the extreme rainfall intensity hit rate (0.101) and reduces the false negative rate compared to the EC model, resulting in an overall TS score that is an order of magnitude higher (0.036) than the EC model's forecast.

[0104] Figure 2 Shown is the comparison between the precipitation observations from 20:00 to 23:00 on August 5, 2023, and the forecast results of the model of the present invention and the EC model forecast. Figure 2 (a) in the figure shows the grid points where extreme rainfall intensity >50 mm / h occurred between 20:00 and 23:00 on August 5, 2023 (dark purple grid points); Figure 2 (b) is the 3-hour precipitation forecast (unit: mm) from the EC model at 08:00 on August 3 with a validity period of 63 hours; Figure 2 (c) is the potential forecast of extreme rainfall intensity reported by the extreme rainfall intensity intelligent forecast model with a time limit of 63 hours starting from 08:00 on August 3. The dark purple grid value is 1, indicating that there is extreme rainfall intensity potential; the blue-green grid value is 0, indicating that there is no extreme rainfall intensity potential.

[0105] In this example, from 20:00 to 23:00 on August 5, 2023, extreme rainfall occurred in northern Henan and northern Shandong. The 3-hour precipitation forecast of the EC model with a time validity of 63 hours was significantly weaker than the actual situation, with the precipitation extreme center only about 20 mm. Therefore, the hit rate and TS score were both 0. However, the model of the present invention successfully predicted the extreme rainfall belt from northern Henan to northern Shandong, such as Figure 2 As shown in (c), the hit rate reaches 0.26 and the TS score reaches 0.11.

[0106] In this example, the extreme rainfall intensity intelligent forecast model is closer to actual observations than the EC model, both in terms of its overall performance from May to September and the forecast results of typical cases.

[0107] Table 2 Prediction factors input into the extreme rainfall intensity intelligent forecast model based on optimal physical factors

[0108]

[0109] Table 3 Comparison of overall scores between the EC model and the extreme rainfall intensity intelligent forecast model from May to September 2023

[0110]

[0111] Therefore, the disclosed method for intelligently forecasting extreme rainfall potential based on optimized physical factors uses numerical model output data and high-resolution gridded precipitation observation data to perform intelligent forecasting of extreme rainfall potential (e.g., steps 1-11). This forecasting method has the following key points:

[0112] (1) Using sensitivity analysis, physical diagnostic quantities directly related to extreme rainfall intensity were selected as key predictors;

[0113] (2) Using random downsampling method to alleviate the problem of extreme and non-extreme rainfall intensity samples being extremely unbalanced;

[0114] (3) Construct fuzzy matching observation labels based on spatiotemporal neighborhoods to alleviate the problem of insufficient description of the spatial distribution characteristics of extreme rainfall intensity caused by machine learning point-to-point modeling.

[0115] Compared with the prior art, the present invention has the following technical advantages:

[0116] By innovatively constructing spatiotemporal fuzzy matching observation labels and sample balancing schemes, the imbalance problem of extreme precipitation data is effectively solved. Through sensitivity analysis, a large number of physical diagnostic quantities directly related to extreme rainfall intensity are incorporated as forecasting factors, thereby improving the numerical model's weak forecast of extreme rainfall intensity and large deviation in forecast of rainfall area, and enhancing the forecast capability of extreme rainfall intensity in the next three days.

[0117] The above embodiments are not limitations of the present invention, and the present invention is not limited to the above examples. Any changes, modifications, additions or substitutions made by technicians in this technical field within the scope of the technical solution of the present invention also fall within the scope of protection of the present invention.

Claims

1. A method for intelligent forecasting of extreme rainfall intensity potential based on optimized physical factors, characterized by: The following steps are included: Step 1: Prepare numerical model forecast data for training the extreme rainfall intensity intelligent forecast model, covering both ground and high-altitude elements; Step 2: Prepare precipitation observation data for training the extreme rainfall intensity intelligent forecast model. The precipitation observation data is hourly grid observation data. Step 3: Calculate the physical diagnostic quantity based on the numerical model based on the numerical model forecast data collected in step 1; Step 4: Based on the direct output of the numerical model collected in step 1 and the physical diagnostic quantity calculated in step 3, the prediction factors of the extreme rainfall intensity intelligent forecast model are constructed to obtain a prediction factor dataset; Step 5: Conduct sensitivity analysis on the prediction factors and conduct preliminary screening, and select highly sensitive physical quantities as prediction factors for the extreme rainfall intensity intelligent forecast model; The process of sensitivity analysis and preliminary screening of predictors is as follows: Based on the hourly precipitation grid observation data collected in step 2 and the numerical model physical quantity zero-field forecast data collected in step 3, the probability density distribution of the physical quantities corresponding to the occurrence of extreme rainfall intensity and non-extreme rainfall intensity events is calculated grid by grid point; The physical quantity with a mean difference exceeding one standard deviation is used as a prediction factor in the extreme rainfall intensity intelligent forecast model; Step 6: Perform dimensionality reduction processing on the prediction factors of the extreme rainfall intensity intelligent forecast model to optimize the screening; Step 7: Construct fuzzy matching observation labels of the spatiotemporal neighborhood for precipitation observation data; Step 8: Construct the optimal extreme rainfall intensity balanced sample training set; Step 9: Perform data cleaning and standardization preprocessing on the optimal extreme rainfall intensity balance sample training set; Step 10: construct an extreme rainfall intensity intelligent forecast model and optimize key parameters to obtain the optimal extreme rainfall intensity intelligent forecast model; Step 11: Apply the optimal extreme rainfall intensity intelligent forecast model to make forecasts, output potential forecasts of extreme rainfall intensity for every 3 hours from 0 to 72 hours, and evaluate the forecast effect.

2. The method for intelligent forecasting of extreme rainfall intensity potential based on optimal physical factors according to claim 1 is characterized by: In step 1, the prepared numerical model forecast data are the output variables of the numerical weather forecast model with a 0-72 hour forecast time limit over the past five years, covering ground element data and high-altitude element data, including total precipitation, convective precipitation, large-scale precipitation, sea level pressure, surface pressure, ground 2m air temperature, ground 2m dew point, atmospheric temperature field, geopotential height field, and three-dimensional wind field; the spatial resolution of the data is 0.125°, and the temporal resolution is 3 hours.

3. The method for intelligent forecasting of extreme rainfall intensity potential based on optimized physical factors according to claim 2, characterized in that: In step 2, the precipitation observation data are hourly grid observation data with a spatial resolution of 1 km and a temporal resolution of 1 hour.

4. The method for intelligent forecasting of extreme rainfall intensity potential based on optimal physical factors according to claim 3 is characterized by: In step 3, the physical diagnostic quantities calculated are the vorticity, divergence, water vapor flux divergence, pseudo-equivalent potential temperature of each pressure layer, and the K index and A index that describe the stability of the atmospheric stratification.

5. The method for intelligent forecasting of extreme rainfall intensity potential based on optimal physical factors according to claim 4 is characterized by: In step 4, the process of constructing the predictor data set is: Based on the meteorological elements directly output by the numerical model for each forecast time collected in step 1 and the calculation results of the physical diagnostic quantities calculated in step 3, a forecast factor dataset for the extreme rainfall intensity intelligent forecast model for each forecast time is constructed. The forecast time is 3 hours, 6 hours, 9 hours, ..., 72 hours.

6. The method for intelligent forecasting of extreme rainfall intensity potential based on optimized physical factors according to claim 5, characterized in that: In step 6, the dimensionality reduction of the predictors is performed by principal component analysis to create a set of uncorrelated predictors. The number of principal components to be retained is selected by testing the impact of retaining different numbers of principal components on the model score.

7. The method for intelligent forecasting of extreme rainfall intensity potential based on optimized physical factors according to claim 6, characterized in that: In step 7, the spatiotemporal neighborhood fuzzy matching method for precipitation observation data is: First, based on hourly precipitation grid observation data, we construct the first-level raw observation labels for hourly extreme rainfall intensity. That is, we determine whether there is observed precipitation >50 mm / h at each grid point on an hourly basis. If so, the raw observation label is recorded as 1 at that grid point in that hour; otherwise, it is recorded as 0. Then, a secondary observation label for extreme rainfall intensity is constructed. That is, under the condition that the primary original observation label is 1, based on the forecast timeliness of the numerical model, it is determined whether there is observed precipitation > 50 mm / h within the 3-hour period covered by each timeliness. If so, the secondary observation label is recorded as 1, otherwise it is recorded as 0. Finally, a third-level observation label for extreme rainfall intensity is constructed. That is, under the condition that the second-level label is 1, it is determined whether there is observed precipitation >50 mm / h at the target grid point and the surrounding 0.25° range within the 3 hours covered by each time period. If so, the third-level observation label is recorded as 1, otherwise it is recorded as 0.

8. The method for intelligent forecasting of extreme rainfall intensity potential based on optimal physical factors according to claim 7, characterized in that: In step 8, the method for constructing the optimal extreme rainfall intensity balanced sample training set is: Using the random downsampling method, samples with 1, 2, and 3 times the number of extreme rainfall intensity samples are randomly extracted from the non-extreme rainfall intensity samples to construct a new non-extreme rainfall intensity sample set. The non-extreme rainfall intensity samples are samples with the observation label 0. The extreme rainfall intensity samples and new non-extreme rainfall intensity samples are divided into multiple extreme rainfall intensity balanced sample sets in different proportions, including 1:1, 1:2, and 1:

3. Model training experiments are carried out to determine the optimal ratio of the number of extreme rainfall intensity samples to non-extreme rainfall intensity samples, and finally the optimal extreme rainfall intensity balanced sample set is constructed.

9. The method for intelligent forecasting of extreme rainfall intensity potential based on optimized physical factors according to claim 8, characterized in that: In step 11, based on the binary classification concept, by constructing a confusion matrix at a specific threshold of 50 mm / h, different precipitation forecast evaluation indicators are calculated, including risk score TS, hit rate POD, false alarm rate FAR, and missed alarm rate MAR. The calculation formulas are shown in Equations 7-10: Risk score TS = number of positive hits / (number of positive hits + number of false alarms + number of missed alarms) (7) Hit rate POD = number of positive hits / (number of positive hits + number of missed reports) (8) FAR = false alarm number / (positive hit number + false alarm number) (9) Missing alarm rate MAR = number of missed alarms / (number of positive hits + number of missed alarms) (10).

Images