Intensity prediction method and system considering tropical cyclone scale

Abstract

The present invention relates to a method for predicting tropical cyclone intensity that takes into account the scale of tropical cyclones, comprising: obtaining a tropical cyclone optimal path dataset, corresponding atmospheric and oceanic variable data, sea temperature data, and land-sea distribution topography data; calculating the sea-to-land ratio within a tropical cyclone's level 10 wind circle based on the obtained dataset and corresponding data; calculating environmental background field prediction variables, ocean-land condition prediction variables, and characteristic variables related to the future path based on the obtained dataset and corresponding data; integrating and dividing the dataset based on the calculated environmental background field prediction variables, ocean-land condition prediction variables, characteristic variables related to the future path, and the land-sea ratio within a tropical cyclone's level 10 wind circle, constructing a tropical cyclone intensity prediction model based on the LightGBM algorithm; and using the tropical cyclone intensity prediction model to predict the intensity change value of the tropical cyclone. The present invention also relates to an intensity prediction system that takes into account the scale of tropical cyclones. The present invention can effectively improve the accuracy of tropical cyclone intensity prediction.

Description

Technical Field

[0001] The present invention relates to a method and system for predicting the intensity of tropical cyclones taking into account the scale. Background Art

[0002] Tropical cyclones are among the most complex weather systems in the atmosphere, encompassing thermal and dynamic processes at multiple scales, as well as interactions between these scales. Advances in observational techniques, particularly the use of unconventional data such as satellite radar, have significantly improved the accuracy of tropical cyclone track forecasts, but progress in intensity forecasts remains slow. In coastal areas, the losses caused by tropical cyclones are increasing with economic development, making timely and accurate tropical cyclone intensity forecasts essential for typhoon prevention and disaster reduction efforts.

[0003] Existing tropical cyclone intensity forecasting methods fall into two main categories: those based on numerical models and those based on statistical dynamical methods. Physics-based numerical simulations and statistical dynamical models have proven to be effective tools for predicting changes in tropical cyclone intensity. However, tropical cyclone intensity is affected by many factors and is difficult to interpret. Numerical models require accurate initial conditions and model parameters, but this information can be imperfect, leading to inaccurate model forecasts. Furthermore, physics-based numerical simulations cannot adequately represent complex dynamic processes and, by increasing the number of variables or equations, lead to an exponential increase in computational requirements. Statistical dynamical methods primarily use historical tropical cyclone data to establish empirical models linking various meteorological parameters to cyclone intensity. These methods also use multiple regression models, which use multiple meteorological parameters as predictors and determine their relationship with cyclone intensity through regression analysis. Statistical dynamical models typically require a large amount of historical data to build, and place high demands on data accuracy and completeness.

[0004] With the rapid development of artificial intelligence, many researchers have begun using machine learning algorithms to explore new methods for improving tropical cyclone intensity forecasts using satellite, radar, and environmental data. Machine learning methods can effectively address the inability of traditional statistical regression methods to handle nonlinear relationships between characteristic variables. However, both traditional statistical regression and machine learning-based forecasting methods currently analyze and predict based on characteristic variables such as the environmental background field and ocean and land conditions. They fail to consider the impact of tropical cyclone size on intensity variations. Tropical cyclone size has a significant impact on its intensity variation. For example, a tropical cyclone's size determines the ocean surface area it can cover and interact with. Larger cyclones generally require deeper and more extensive warm water to provide sufficient energy. Furthermore, the size of a tropical cyclone describes the radial extent of its wind field. Smaller cyclones are more likely to be located in areas of less vertical wind shear, which helps maintain cyclone symmetry and structural integrity, promoting intensification. Smaller cyclones may also be more likely to maintain strong convective activity in their cores, which helps release large amounts of latent heat, providing additional energy to maintain cyclonic intensity. Furthermore, given the same external wind field, smaller cyclones can more effectively conserve angular momentum, which can lead to faster rotation and intensification. Overall, there is a complex interrelationship between tropical cyclone size and intensity, and there is no simple linear relationship to describe the relationship between tropical cyclone size and intensity. Furthermore, other environmental conditions and internal dynamic processes also have a significant impact on tropical cyclone intensity changes. Therefore, this study calculated the sea-to-land ratio within a tropical cyclone's Category 10 wind circle as a predictor. Combining environmental background variables, ocean and land conditions, and characteristic variables related to future paths, the LightGBM algorithm was used to automatically learn and process the complex spatiotemporal nonlinear characteristics of tropical cyclone intensity evolution, establishing a model to predict future tropical cyclone intensity changes. Summary of the Invention

[0005] In view of this, it is necessary to provide a tropical cyclone intensity prediction method and system that takes into account the scale of tropical cyclones, which can effectively improve the accuracy of tropical cyclone intensity prediction.

[0006] The present invention provides an intensity prediction method taking into account the scale of tropical cyclones, which includes the following steps: a. obtaining a tropical cyclone optimal path data set, corresponding atmospheric and ocean variable data, sea temperature data, and sea-land distribution topography data; b. calculating the sea-to-land ratio within a tropical cyclone level 10 wind circle based on the above-obtained data set and corresponding data; c. calculating environmental background field prediction variables, ocean and land condition prediction variables, and characteristic variables related to the future path based on the above-obtained data set and corresponding data; d. integrating and dividing the data set based on the calculated environmental background field prediction variables, ocean and land condition prediction variables, and characteristic variables related to the future path, as well as the sea-to-land ratio within the tropical cyclone level 10 wind circle, and constructing a tropical cyclone intensity prediction model based on the LightGBM algorithm; e. using the tropical cyclone intensity prediction model to predict the intensity change value of the tropical cyclone.

[0007] Preferably, step a comprises:

[0008] Step S11, downloading a tropical cyclone best path dataset for the Northwest Pacific region from the official website of the Joint Warning Center, set as Q1; wherein the tropical cyclone best path dataset is composed of elements, each of which is a tropical cyclone best path record;

[0009] The best track record of each tropical cyclone includes: the entire time from the formation to the extinction of the tropical cyclone, its latitude and longitude, its intensity, and the radius of its wind circles at levels 7, 10, and 12;

[0010] Step S12: Download the fifth-generation atmospheric reanalysis global climate dataset of the European Centre for Medium-Range Weather Forecasts from the official website of the European Centre for Medium-Range Weather Forecasts (ECMWF), set it as Q2, and then perform data analysis;

[0011] Step S13: Download daily mean sea surface temperature data from the official website of the National Oceanic and Atmospheric Administration of the United States, and set it as Q3.

[0012] Step S14: Based on the underlying surface distribution of the ocean and land in the Northern Hemisphere, a binary grayscale image is drawn and set as Q4.

[0013] Preferably, the step b comprises:

[0014] Step S21, extracting the occurrence time, center location, and 10-force wind circle radius in the northeast, southeast, northwest, and southwest quadrants of the tropical cyclone in Q1;

[0015] Step S22, according to the 10-force wind circle radii of the northeast, southeast, northwest, and southwest quadrants, select the largest value as the radius of the study area;

[0016] Step S23, based on Q4, calculate the sea-to-land ratio in the study area with the center of the tropical cyclone as the center as the prediction variable.

[0017] Preferably, the step c comprises:

[0018] Step S31, extracting the time, location, and intensity information of the tropical cyclone from Q1, calculating the tropical cyclone moving speed, Julian day, intensity change over the past 12 hours, etc., and obtaining the prediction factors of longitude, latitude, initial tropical cyclone intensity, tropical cyclone moving speed, Julian day, intensity change over the past 12 hours, and prediction variables of tropical cyclone intensity after 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours;

[0019] Step S32: Find the data file corresponding to Q2 based on the time information in Q1, and then set a circular area with a radius of 200 to 800 km centered on the longitude and latitude of the tropical cyclone center in Q1 as the study area; then calculate the mean of the corresponding variables in the area to obtain the environmental background field variables, including: temperature at 200 hPa, relative humidity at 850-700 hPa, relative humidity at 500-300 hPa, vertical vorticity at 850 hPa, vertical vorticity at 500 hPa, horizontal divergence at 200 hPa, water vapor flux at 850 hPa, and water vapor flux at 500 hPa;

[0020] Step S33: Find the data file corresponding to Q2 based on the time information in Q1, and then set the area with a radius of 0 to 500 km centered on the longitude and latitude of the tropical cyclone center in Q1 as the study area; then calculate the mean of the corresponding variables in the area to obtain the environmental background field variables, including: U wind at 200 hPa, V wind at 200 hPa, vertical wind shear and zonal wind shear at 200-850 hPa, vertical wind shear and zonal wind shear at 500-850 hPa, and the product of vertical wind shear and the sin value of the latitude of the tropical cyclone center;

[0021] Step S34: Find the daily mean sea surface temperature in Q3 based on the time and longitude and latitude information in Q1, and calculate the mean sea surface temperature in an area with a radius of 0 to 800 km centered on the longitude and latitude in Q1.

[0022] Step S35, calculating the potential maximum intensity of the tropical cyclone based on the above-mentioned mean sea temperature;

[0023] Step S36, based on the longitude and latitude position information of the tropical cyclone in Q1 and the land-sea distribution in Q4, calculate the land-sea ratio within the area with a radius of 0 to 500 km centered at the longitude and latitude;

[0024] Step S37, based on the latitude and longitude position information of the tropical cyclone forecast time in Q1, calculate the land-sea ratio and sea temperature of the location at the current time, and calculate the vertical wind shear of the location at the future time.

[0025] Preferably, the vertical wind shear includes: vertical wind shear and latitudinal wind shear of 200-850 hPa, vertical wind shear and latitudinal wind shear of 500-850 hPa, and the product of vertical wind shear and the sin value of the latitude of the center of the tropical cyclone.

[0026] Preferably, the step d comprises:

[0027] Step S41, integrating the extracted and calculated tropical cyclone prediction variables into an atmospheric variable dataset R, predicting the tropical cyclone intensity changes in the next 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours, and dividing the combined atmospheric variable dataset R into atmospheric variable datasets R6, R12, R24, R36, R48, R60, R72, R96, and R120 for different prediction time periods;

[0028] Step S42: The atmospheric variable datasets of different prediction time periods are divided into four life history categories: full life cycle, post-landfall, nearshore, and offshore, according to the land-sea ratio within 500 km, totaling 36 datasets, to form the final training set X.

[0029] Step S43: constructing a tropical cyclone intensity forecast model based on the LightGBM algorithm.

[0030] Preferably, the step S43 includes:

[0031] Step S431: read the characteristic variables and tropical cyclone intensity change values ​​in the training set X, and divide the data into three groups for cross-validation, with six consecutive years as one group: one group is used as the test set, and the other two groups are used as the training set;

[0032] Step S432: input the training set into the LightGBM regression model to train the model and obtain a tropical cyclone intensity forecast model;

[0033] Step S433: Automatically adjust the model hyperparameters using the Bayesian optimization algorithm to obtain the optimal hyperparameter combination after Bayesian optimization, and save the prediction model of the optimal hyperparameter combination.

[0034] Preferably, the step e comprises:

[0035] The trained forecast model is used to predict the intensity of the test sets at different stages and times, and the intensity change values ​​of tropical cyclones in the future 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours for the entire life cycle, after landfall, nearshore, and offshore are obtained.

[0036] The present invention provides an intensity prediction system considering the scale of tropical cyclones, the system comprising an acquisition module, a calculation module, a construction module, and a prediction module, wherein:

[0037] The acquisition module is used to obtain the tropical cyclone optimal path data set, the corresponding atmospheric and ocean variable data, sea temperature data, and sea and land distribution topography data;

[0038] The calculation module is used to calculate the sea-to-land ratio within the tropical cyclone level 10 wind circle based on the above-obtained data set and corresponding data;

[0039] The calculation module is further used to calculate environmental background field prediction variables, ocean and land condition prediction variables, and characteristic variables related to the future path based on the above-obtained data set and corresponding data;

[0040] The construction module is used to integrate and divide the data set according to the calculated environmental background field prediction variables, ocean and land condition prediction variables and characteristic variables related to the future path, and the sea-to-land ratio within the tropical cyclone level 10 wind circle, and build a tropical cyclone intensity forecast model based on the LightGBM algorithm;

[0041] The prediction module is used to predict the intensity change value of the tropical cyclone using the tropical cyclone intensity forecast model.

[0042] Based on the current and future circulation background and ocean and land conditions as prediction variables, the present invention considers the land and sea distribution within a tropical cyclone's level 10 wind circle as a characteristic variable, constructs a tropical cyclone intensity forecast model based on the LightGBM algorithm, and predicts the intensity of tropical cyclones throughout their life cycle, after landfall, near the coast, and offshore. Furthermore, the present invention has the following beneficial effects:

[0043] First, compared to existing tropical cyclone intensity forecast models, whose prediction factors only include variables such as the tropical cyclone's structure, environmental background field, and ocean and land conditions, this invention considers the impact of the scale of the tropical cyclone (the size of the 10-level wind circle) on its intensity changes;

[0044] Secondly, compared with the tropical cyclone intensity forecast model in the existing technology, the present invention divides the different life cycles of tropical cyclones by calculating the sea-to-land ratio within a range of 500km, and constructs a tropical cyclone intensity forecast model based on the machine learning LightGBM regression algorithm, and uses the sea-to-land ratio within the 10-level wind circle as the feature input of the model, which significantly improves the accuracy of tropical cyclone intensity prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flowchart of the intensity prediction method considering the tropical cyclone scale according to the present invention;

[0046] Figure 2 This is a hardware architecture diagram of the tropical cyclone intensity prediction system of the present invention. DETAILED DESCRIPTION

[0047] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] See Figure 1 FIG. 1 is a flowchart of a preferred embodiment of the intensity prediction method considering the scale of tropical cyclones according to the present invention.

[0049] Step S1: Obtain the optimal tropical cyclone path data set, the corresponding atmospheric and oceanic variable data, sea temperature data, and land and sea distribution topography data. Specifically:

[0050] In this embodiment:

[0051] Step S11: Download the best track data set of tropical cyclones in the Northwest Pacific region from 2005 to 2022 from the official website of the Joint Warning Center (the elements are the best track records of tropical cyclones), and set it as Q1.

[0052] The best path record for each tropical cyclone includes: the total time (year, month, day, hour) from the formation to the extinction of the tropical cyclone, its latitude and longitude, its intensity, and the radius of its wind circles at levels 7, 10, and 12.

[0053] Step S12: Download the European Centre for Medium-Range Weather Forecasts' fifth-generation atmospheric reanalysis global climate dataset (ERA5) from the official website of the European Centre for Medium-Range Weather Forecasts (ECMWF) from 2005 to 2022, and reanalyze the ERA5 data. The temporal and spatial resolution of the ERA5 data is 0.25° × 0.25° grid and 1 hour time interval, respectively, and is set to Q2.

[0054] Step S13: Download daily mean sea surface temperature data from the official website of the National Oceanic and Atmospheric Administration of the United States, and set it as Q3.

[0055] Step S14: Based on the topographic distribution of the underlying surface of the ocean and land in the Northern Hemisphere, a binary grayscale image is drawn and set as Q4.

[0056] Step S2, based on the above-obtained data set and corresponding data, calculate the sea-to-land ratio within the tropical cyclone level 10 wind circle. Specifically:

[0057] Step S21: extract the occurrence time, center location, and force 10 wind circle radius of the four quadrants of northeast, southeast, northwest, and southwest of the tropical cyclone in Q1.

[0058] Step S22: According to the radii of the level 10 wind circles in the northeast, southeast, northwest, and southwest quadrants, the largest value is selected as the radius of the calculated study area.

[0059] Step S23, based on Q4, calculate the sea-to-land ratio in the study area with the center of the tropical cyclone as the center as the prediction variable.

[0060] Step S3: Calculate the environmental background field prediction variables, ocean and land condition prediction variables, and characteristic variables related to the future path based on the above-obtained data set and corresponding data. Specifically:

[0061] Step S31 extracts information such as the time, location, and intensity of the tropical cyclone from Q1 and calculates the tropical cyclone's movement speed, Julian day, and intensity change over the past 12 hours. The resulting prediction factors are longitude, latitude, initial tropical cyclone intensity, tropical cyclone movement speed, Julian day, intensity change over the past 12 hours, and the prediction variables for tropical cyclone intensity 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours later.

[0062] In step S32, the data file corresponding to Q2 is found based on the time information in Q1. Based on the longitude and latitude of the tropical cyclone center in Q1, a circular region with a radius of 200 to 800 km, centered at those longitude and latitude, is set as the study area. The mean values ​​of the corresponding variables within this region are then calculated to obtain some environmental background field variables, including: temperature at 200 hPa, relative humidity between 850 and 700 hPa, relative humidity between 500 and 300 hPa, vertical vorticity at 850 hPa, vertical vorticity at 500 hPa, horizontal divergence at 200 hPa, water vapor flux at 850 hPa, and water vapor flux at 500 hPa.

[0063] In step S33, the data file corresponding to Q2 is found based on the time information in Q1. Based on the longitude and latitude of the tropical cyclone center in Q1, a study area with a radius of 0 to 500 km is set, centered at those longitude and latitude. The mean of the corresponding variables within this area is then calculated to obtain the environmental background field variables, including: U wind at 200 hPa, V wind at 200 hPa, vertical wind shear and zonal wind shear from 200 to 850 hPa, vertical wind shear and zonal wind shear from 500 to 850 hPa, and the product of vertical wind shear and the sin value of the tropical cyclone center latitude.

[0064] Step S34, find the daily mean sea surface temperature in Q3 according to the time and longitude and latitude information in Q1, and calculate the mean sea surface temperature in the area with a radius of 0 to 800 km centered on the longitude and latitude in Q1.

[0065] Step S35: Calculate the potential maximum intensity of the tropical cyclone based on the above-mentioned mean sea temperature.

[0066] Step S36, based on the longitude and latitude position information of the tropical cyclone in Q1 and the land and sea distribution in Q4, calculate the land and sea ratio of the area with a radius of 0 to 500 km centered at the longitude and latitude.

[0067] Step S37, based on the latitude and longitude position information at the time of the tropical cyclone forecast in Q1, calculate the land-sea ratio and sea temperature at the current time, and calculate the vertical wind shear at the future time. The vertical wind shear includes: the vertical wind shear and zonal wind shear at 200-850 hPa, the vertical wind shear and zonal wind shear at 500-850 hPa, and the product of the vertical wind shear and the sin value of the tropical cyclone center latitude.

[0068] In step S4, based on the calculated environmental background field prediction variables, ocean and land condition prediction variables, characteristic variables related to the future path, and the sea-to-land ratio within the tropical cyclone level 10 wind circle, the data set is integrated and divided, and a tropical cyclone intensity forecast model is constructed based on the LightGBM algorithm.

[0069] Specifically:

[0070] Step S41: Integrate the tropical cyclone prediction variables extracted and calculated over the past two decades into an atmospheric variable dataset R, predict the tropical cyclone intensity changes in the next 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours, and divide the combined atmospheric variable dataset R into atmospheric variable datasets R6, R12, R24, R36, R48, R60, R72, R96, and R120 for different prediction time periods.

[0071] In step S42, the atmospheric variable datasets of different prediction time periods are divided into four life history categories: full life cycle, post-landing, near sea, and offshore, according to the land-sea ratio within 500 km, totaling 36 datasets, to form the final training set X.

[0072] Step S43: constructing a tropical cyclone intensity forecast model based on the LightGBM algorithm. This specifically includes:

[0073] Step S431: Read the characteristic variables and tropical cyclone intensity change values ​​in the training set X. The data from 2005 to 2022 are divided into three groups (for example, the data from 2005 to 2010 are one group) with six consecutive years as one group for cross-validation: one group is used as the test set, and the other two groups are used as the training sets.

[0074] Step S432: Input the training set into the LightGBM regression model to train the model and obtain a tropical cyclone intensity forecast model.

[0075] Step S433: Automatically adjust the model hyperparameters using the Bayesian optimization algorithm to obtain the optimal hyperparameter combination after Bayesian optimization, and save the prediction model of the optimal hyperparameter combination.

[0076] Step S5: Use the tropical cyclone intensity forecast model to predict the intensity change value of the tropical cyclone. Specifically:

[0077] The trained forecast model is used to predict the intensity of the test sets at different stages and times, and the intensity change values ​​of tropical cyclones in the future 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours for the entire life cycle, after landfall, nearshore, and offshore are obtained.

[0078] See Figure 2 FIG. 1 is a hardware architecture diagram of a tropical cyclone intensity prediction system 10 according to the present invention. The system includes an acquisition module 101, a calculation module 102, a construction module 103, and a prediction module 104.

[0079] The acquisition module 101 is used to obtain the tropical cyclone optimal path data set, the corresponding atmospheric and ocean variable data, sea temperature data, and sea and land distribution topography data. Specifically:

[0080] In this embodiment:

[0081] The acquisition module 101 downloads the tropical cyclone best path dataset (elements are tropical cyclone best path records) in the northwest Pacific region from 2005 to 2022 from the official website of the Joint Warning Center, and sets it as Q1.

[0082] The best path record for each tropical cyclone includes: the total time (year, month, day, hour) from the formation to the extinction of the tropical cyclone, its latitude and longitude, its intensity, and the radius of its wind circles at levels 7, 10, and 12.

[0083] The acquisition module 101 downloads the European Centre for Medium-Range Weather Forecasts' fifth-generation atmospheric reanalysis global climate dataset (ERA5) from the official website of the European Centre for Medium-Range Weather Forecasts (ECMWF) from 2005 to 2022 and reanalyzes the ERA5 data. The temporal and spatial resolution of the ERA5 data is 0.25°×0.25° grid and 1 hour time interval, respectively, and is set to Q2.

[0084] The acquisition module 101 downloads daily mean sea surface temperature data from the official website of the National Oceanic and Atmospheric Administration of the United States, which is set as Q3.

[0085] The acquisition module 101 draws and obtains a binary grayscale image based on the underlying topography distribution of the ocean and land in the northern hemisphere, which is set as Q4.

[0086] The calculation module 102 is used to calculate the sea-to-land ratio within the tropical cyclone level 10 wind circle based on the above-obtained data set and corresponding data. Specifically:

[0087] The calculation module 102 extracts the occurrence time, center location, and radius of the force 10 wind circle in the northeast, southeast, northwest, and southwest quadrants of the tropical cyclone in Q1.

[0088] The calculation module 102 selects the largest value among the 10-force wind circle radii in the northeast, southeast, northwest, and southwest quadrants as the radius of the study area.

[0089] The calculation module 102 calculates the sea-to-land ratio in the study area with the center of the tropical cyclone as the center as a prediction variable based on Q4.

[0090] The calculation module 102 is also used to calculate environmental background field prediction variables, ocean and land condition prediction variables, and characteristic variables related to the future path based on the above-obtained data set and corresponding data. Specifically:

[0091] The calculation module 102 extracts information such as the time, location, and intensity of the tropical cyclone from Q1 and calculates the tropical cyclone's movement speed, Julian day, and intensity change over the past 12 hours. The resulting prediction factors are longitude, latitude, initial tropical cyclone intensity, tropical cyclone movement speed, Julian day, intensity change over the past 12 hours, and prediction variables for tropical cyclone intensity after 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours.

[0092] The calculation module 102 finds the data file corresponding to Q2 based on the time information in Q1. Based on the longitude and latitude of the tropical cyclone center in Q1, it sets a circular area with a radius of 200 to 800 km centered on these longitude and latitude as the study area. It then calculates the mean of the corresponding variables within this area to obtain some environmental background field variables, including: temperature at 200 hPa, relative humidity between 850 and 700 hPa, relative humidity between 500 and 300 hPa, vertical vorticity at 850 hPa, vertical vorticity at 500 hPa, horizontal divergence at 200 hPa, water vapor flux at 850 hPa, and water vapor flux at 500 hPa.

[0093] The calculation module 102 locates the data file corresponding to Q2 based on the time information in Q1. Based on the longitude and latitude of the tropical cyclone center in Q1, it sets a study area with a radius of 0 to 500 km centered on the longitude and latitude. It then calculates the mean of the corresponding variables within the area to obtain environmental background field variables, including: U wind at 200 hPa, V wind at 200 hPa, vertical wind shear and zonal wind shear from 200 to 850 hPa, vertical wind shear and zonal wind shear from 500 to 850 hPa, and the product of vertical wind shear and the sin value of the latitude of the tropical cyclone center.

[0094] The calculation module 102 finds the daily mean sea surface temperature in Q3 according to the time and longitude and latitude information in Q1, and calculates the mean sea surface temperature in an area with a radius of 0 to 800 km centered on the longitude and latitude in Q1.

[0095] The calculation module 102 calculates the potential maximum intensity of the tropical cyclone based on the mean sea temperature.

[0096] The calculation module 102 calculates the land-sea ratio within a region with a radius of 0 to 500 km centered at the longitude and latitude of the tropical cyclone in Q1 and the land-sea distribution in Q4.

[0097] The calculation module 102 calculates the land-sea ratio and sea temperature at the current time, and the vertical wind shear at the future time, based on the latitude and longitude position information of the tropical cyclone at the time of the forecast in Q1. The vertical wind shear includes: the vertical wind shear and zonal wind shear at 200-850 hPa, the vertical wind shear and zonal wind shear at 500-850 hPa, and the product of the vertical wind shear and the sin value of the latitude of the tropical cyclone center.

[0098] The construction module 103 is used to integrate and divide the data set based on the calculated environmental background field prediction variables, ocean and land condition prediction variables, characteristic variables related to the future path, and the sea-to-land ratio within the tropical cyclone level 10 wind circle, and construct a tropical cyclone intensity forecast model based on the LightGBM algorithm. Specifically:

[0099] The construction module 103 integrates the tropical cyclone prediction variables extracted and calculated in the past twenty years into an atmospheric variable dataset R, predicts the changes in tropical cyclone intensity in the next 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours, and divides the combined atmospheric variable dataset R into atmospheric variable datasets R6, R12, R24, R36, R48, R60, R72, R96, and R120 for different prediction time periods.

[0100] The construction module 103 divides the atmospheric variable datasets of different prediction time periods into four life history datasets of full life cycle, post-landing, near sea and offshore according to the land-sea ratio within 500 km, totaling 36 datasets, to form the final training set X.

[0101] The construction module 103 constructs a tropical cyclone intensity forecast model based on the LightGBM algorithm.

[0102] Specifically include:

[0103] Read the characteristic variables and tropical cyclone intensity change values ​​in the training set X. Divide the data from 2005 to 2022 into three groups (for example, the data from 2005 to 2010 is one group) with six consecutive years as one group for cross-validation: use one group as the test set and the other two groups as the training sets.

[0104] The training set is input into the LightGBM regression model, the model is trained, and a tropical cyclone intensity forecast model is obtained.

[0105] The Bayesian optimization algorithm is used to automatically adjust the model hyperparameters, obtain the optimal hyperparameter combination after Bayesian optimization, and save the prediction model of the optimal hyperparameter combination.

[0106] The prediction module 104 is used to predict the intensity change value of the tropical cyclone using the tropical cyclone intensity forecast model. Specifically:

[0107] The prediction module 104 uses the trained forecast model to predict the intensity of the test set at different stages and times, and obtains the tropical cyclone intensity change values ​​for the next 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours in the entire life cycle, after landfall, near sea, and offshore.

[0108] The present invention solves the problem that the characteristic variables of the existing tropical cyclone intensity forecast model do not contain prediction factors related to its size and the prediction accuracy is not high. The present invention constructs an intensity forecast model that takes the size of tropical cyclones into consideration based on the machine learning LightGBM regression model, thereby improving the accuracy of tropical cyclone intensity forecasting.

[0109] Although the present invention has been described with reference to the current preferred embodiments, those skilled in the art should understand that the above-mentioned preferred embodiments are only used to illustrate the present invention and are not used to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting tropical cyclone intensity taking into account the scale, characterized in that: The method comprises the following steps: a. Obtain the optimal tropical cyclone track dataset, corresponding atmospheric and oceanic variable data, sea temperature data, and land and sea distribution topography data; b. Calculate the sea-to-land ratio within the tropical cyclone's force 10 wind circle based on the dataset and corresponding data obtained above; c. Based on the above-obtained datasets and corresponding data, calculate environmental background field prediction variables, ocean and land condition prediction variables, and characteristic variables related to the future path; d. Based on the calculated environmental background field prediction variables, ocean and land condition prediction variables, characteristic variables related to the future path, and the sea-to-land ratio within the tropical cyclone's level 10 wind circle, integrate and divide the dataset and construct a tropical cyclone intensity forecast model based on the LightGBM algorithm; e. Use the tropical cyclone intensity forecast model to predict the intensity change of tropical cyclones; where: Step d includes: Step S41, integrating the extracted and calculated tropical cyclone prediction variables into an atmospheric variable dataset R, predicting the tropical cyclone intensity changes in the next 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours, and dividing the combined atmospheric variable dataset R into atmospheric variable datasets R6, R12, R24, R36, R48, R60, R72, R96, and R120 for different prediction time periods; Step S42: The atmospheric variable datasets of different prediction time periods are divided into four life history categories: full life cycle, post-landfall, nearshore, and offshore, according to the land-sea ratio within 500 km, totaling 36 datasets, to form the final training set X. Step S43: constructing a tropical cyclone intensity forecast model based on the LightGBM algorithm.

2. The method for tropical cyclone intensity prediction taking into account the scale of claim 1, wherein: Step a includes: Step S11, downloading a tropical cyclone best path dataset for the Northwest Pacific region from the official website of the Joint Warning Center, set as Q1; wherein the tropical cyclone best path dataset is composed of elements, each of which is a tropical cyclone best path record; The best track record of each tropical cyclone includes: the entire time from the formation to the extinction of the tropical cyclone, its latitude and longitude, its intensity, and the radius of its wind circles at levels 7, 10, and 12; Step S12: Download the fifth-generation atmospheric reanalysis global climate dataset of the European Centre for Medium-Range Weather Forecasts from the official website of the European Centre for Medium-Range Weather Forecasts (ECMWF), set it as Q2, and then perform data analysis; Step S13, downloading daily mean sea surface temperature data from the official website of the National Oceanic and Atmospheric Administration of the United States, set as Q3; Step S14: Based on the underlying surface distribution topography data of the ocean and land in the Northern Hemisphere, a binary grayscale image is drawn and set as Q4.

3. The method for tropical cyclone intensity prediction taking into account the scale of claim 2, wherein: Step b includes: Step S21, extracting the occurrence time, center location, and 10-force wind circle radius in the northeast, southeast, northwest, and southwest quadrants of the tropical cyclone in Q1; Step S22, according to the 10-force wind circle radii of the northeast, southeast, northwest, and southwest quadrants, select the largest value as the radius of the study area; Step S23, based on Q4, calculate the sea-to-land ratio in the study area with the center of the tropical cyclone as the center as the prediction variable.

4. The method for tropical cyclone intensity prediction taking into account the scale of claim 3, wherein: Step c includes: Step S31, extracting the time, location, and intensity information of the tropical cyclone from Q1, calculating the tropical cyclone moving speed, Julian day, and intensity change over the past 12 hours, and obtaining the prediction factors of longitude, latitude, initial tropical cyclone intensity, tropical cyclone moving speed, Julian day, intensity change over the past 12 hours, and prediction variables of tropical cyclone intensity after 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours; Step S32: Find the data file corresponding to Q2 based on the time information in Q1, and then set a circular area with a radius of 200 to 800 km centered on the longitude and latitude of the tropical cyclone center in Q1 as the study area; then calculate the mean of the corresponding variables in the area to obtain the environmental background field variables, including: temperature at 200 hPa, relative humidity at 850-700 hPa, relative humidity at 500-300 hPa, vertical vorticity at 850 hPa, vertical vorticity at 500 hPa, horizontal divergence at 200 hPa, water vapor flux at 850 hPa, and water vapor flux at 500 hPa; Step S33: Find the data file corresponding to Q2 based on the time information in Q1, and then set the area with a radius of 0 to 500 km centered on the longitude and latitude of the tropical cyclone center in Q1 as the study area; then calculate the mean of the corresponding variables in the area to obtain the environmental background field variables, including: U wind at 200 hPa, V wind at 200 hPa, vertical wind shear and zonal wind shear at 200-850 hPa, vertical wind shear and zonal wind shear at 500-850 hPa, and the product of vertical wind shear and the sin value of the latitude of the tropical cyclone center; Step S34: Find the daily mean sea surface temperature in Q3 based on the time and longitude and latitude information in Q1, and calculate the mean sea surface temperature in an area with a radius of 0 to 800 km centered on the longitude and latitude in Q1. Step S35, calculating the potential maximum intensity of the tropical cyclone based on the above-mentioned mean sea temperature; Step S36, based on the longitude and latitude position information of the tropical cyclone in Q1 and the land-sea distribution in Q4, calculate the land-sea ratio within the area with a radius of 0 to 500 km centered at the longitude and latitude; Step S37, based on the latitude and longitude position information of the tropical cyclone forecast time in Q1, calculate the land-sea ratio and sea temperature of the location at the current time, and calculate the vertical wind shear of the location at the future time.

5. The method for tropical cyclone intensity prediction taking into account the scale of claim 4, wherein: The vertical wind shear includes: vertical wind shear and zonal wind shear of 200-850hPa, vertical wind shear and zonal wind shear of 500-850hPa, and the product of vertical wind shear and the sin value of the latitude of the center of the tropical cyclone.

6. The method for tropical cyclone intensity prediction taking into account the scale of claim 5, wherein: Step S43 includes: Step S431: read the characteristic variables and tropical cyclone intensity change values ​​in the training set X, and divide the data into three groups for cross-validation, with six consecutive years as one group: one group is used as the test set, and the other two groups are used as the training set; Step S432: input the training set into the LightGBM regression model to train the model and obtain a tropical cyclone intensity forecast model; Step S433: Automatically adjust the model hyperparameters using the Bayesian optimization algorithm to obtain the optimal hyperparameter combination after Bayesian optimization, and save the prediction model of the optimal hyperparameter combination.

7. The method for tropical cyclone intensity prediction taking into account the scale of claim 6, wherein: Step e includes: The trained forecast model is used to predict the intensity of the test sets at different stages and times, and the intensity change values ​​of tropical cyclones in the future 6, 12, 24, 36, 48, 60, 72, 96, and 120 hours for the entire life cycle, after landfall, nearshore, and offshore are obtained.

8. A tropical cyclone scale intensity prediction system according to the tropical cyclone scale intensity prediction method of claim 1, characterized in that: The system includes an acquisition module, a calculation module, a construction module, and a prediction module, wherein: The acquisition module is used to obtain the tropical cyclone optimal path data set, the corresponding atmospheric and ocean variable data, sea temperature data, and sea and land distribution topography data; The calculation module is used to calculate the sea-to-land ratio within the tropical cyclone level 10 wind circle based on the above-obtained data set and corresponding data; The calculation module is further used to calculate environmental background field prediction variables, ocean and land condition prediction variables, and characteristic variables related to the future path based on the above-obtained data set and corresponding data; The construction module is used to integrate and divide the data set according to the calculated environmental background field prediction variables, ocean and land condition prediction variables and characteristic variables related to the future path, and the sea-to-land ratio within the tropical cyclone level 10 wind circle, and build a tropical cyclone intensity forecast model based on the LightGBM algorithm; The prediction module is used to predict the intensity change value of the tropical cyclone using the tropical cyclone intensity forecast model.

Images