KR20230028864A - System and method for predicting medium-range track of tropical cyclone - Google Patents

Abstract

A method for mid-term prediction of the track of a typhoon is disclosed. The method comprises the steps of: extracting data of previous typhoons that satisfy preset criteria including initial characteristics of the target typhoon whose tack is to be predicted from among data of previous typhoons that have historically occurred during a predetermined previous period; clustering tracks of previous typhoons into multiple track patterns using a clustering algorithm from among the extracted data of previous typhoons; calculating a daily average-synthesized observed background fields of predetermined predictors associated with each of the track patterns for 6 to 8 days following an analysis time; calculating daily average model background fields of the predictors from one or more dynamic model background fields for the 6 to 8 days in the future; calculating a pattern correlation between the observed background fields and the model background fields of the predictors to generate weight factors for each of the track patterns; and generating a relative frequency map using the tracks of previous typhoons and the weight factors for each of the track patterns and obtaining a final predicted track based on the relative frequency map.

Description

System and method for mid-term prediction of typhoon course {SYSTEM AND METHOD FOR PREDICTING MEDIUM-RANGE TRACK OF TROPICAL CYCLONE}

The present disclosure relates to a technique for predicting a path of a typhoon, and more specifically, to a technique for predicting a path of a typhoon in the mid-term (after 6 days to 8 days).

Among the various characteristics of typhoons, course prediction is very important. Without accurate information about future storm locations, preparedness and mitigation measures can be ineffective. Risk analysis of typhoons in Korea shows that damage caused by typhoons is more sensitive to the course of the typhoon than to the intensity and/or size of the typhoon. In recent years, the decrease in damage caused by typhoons may be the result of substantial improvement in course prediction. Because typhoons seriously threaten coastal and coastal areas with strong rains, strong winds, storms, and floods, predicting the course of typhoons is particularly important in the Northwest Pacific region, where large populations are distributed along long coastlines.

An average of about 30 typhoons occur annually in the Northwest Pacific region. Figure 1 is a distribution chart showing the lifespan of typhoons that occurred during the period from 1979 to 2019 in the Northwest Pacific region. Referring to Figure 1, about 37% of typhoons in the western North Pacific lasted more than 6 days. 60% of these sustained typhoons reached 30 degrees north latitude, and 20% headed for the South China Sea (south of 25 degrees north latitude, west of 120 degrees east longitude). Most of these long-lasting TCs eventually affect large populations, so accurate predictions of TC tracks over longer forecast periods are very helpful, especially in East Asia.

Currently, predictions of hurricane tracks for all preceding periods are highly dependent on epidemiological models. In the past, it was known that dynamic model predictions contained systematic errors that could lead to large errors in typhoon tracks, requiring additional statistical adjustment techniques or systematic approaches. However, due to advances in computing technology and substantial improvements (or reductions in error) in career prediction, the results of epidemiological models have improved, and the utilization of statistical models has decreased accordingly.

In addition, there are several previous studies to develop statistical models for predicting the course of typhoons, but these studies do not focus on predicting the course of the preceding period from 6 days to 8 days.

One object of the present disclosure is to extract and extract data of past typhoons that satisfy preset criteria including the initial characteristics of a target typhoon whose course is to be predicted from among data of past typhoons that have historically occurred during a predetermined past period. The paths of past typhoons that have been detected are clustered into a plurality of path patterns, and the pattern correlation between the daily observation background fields related to the path patterns and the background fields of the dynamics model is calculated, and a weighting factor for each path pattern is generated and used. A system for mid-term prediction of a typhoon course, which generates a relative frequency map and obtains a final predicted course based on the relative frequency map, thereby enabling course prediction for a preceding period 6 days to 8 days after the typhoon, and is to provide a way

An embodiment of the present disclosure is a method for predicting the course of a typhoon in the mid-term, which includes (1) among data of past typhoons that have historically occurred during a predetermined past period, the initial characteristics of a target typhoon whose course is to be predicted are included. Extracting data of past typhoons that satisfy preset criteria; (2) Clustering tracks of past typhoons into a plurality of course patterns from the extracted data of past typhoons using a clustering algorithm; (3) Analysis Calculating daily average synthesized observation background fields of predetermined predictors associated with each course pattern on the day 6 to 8 days after the time, (4) one or more days on the day 6 to 8 days after the time Calculating daily average model backgrounds of predictors from epidemiological models, (5) Calculating pattern correlation between predictors, observed backgrounds and model backgrounds to generate a weighting factor for each course pattern and (6) generating a relative frequency map using the paths of past typhoons and weighting factors for each path pattern, and obtaining a final predicted path based on the relative frequency map.

In one embodiment, in step (1), the predetermined past period includes the period from 1979 to 2019, and the data of past typhoons is among the best course data provided by the Regional Specialized Meteorological Center (RSMC) Tokyo Typhoon Center. It includes typhoon central location data and central air pressure data at 6-hour intervals, and the initial characteristics of the target typhoon may include the predicted location from the date of occurrence of the target typhoon provided by the Korea Meteorological Administration to the following 5 days.

In one embodiment, past typhoons satisfying the criteria set in advance in step (1) are centered on (i) the position of the target typhoon at the analysis time and the positions of the target typhoon on each forecast day for the following 5 days of the sequential array of six extraction sources that pass through each of the six extraction sources while entering the first extraction source and exiting the final extraction source, (ii) have a lifespan of 6 days or more, and (iii) generate a target typhoon It may be past typhoons that occurred within a total period of 3 months before and after, centering on work.

In one embodiment, in step (2), the clustering algorithm is a fuzzy c-means clustering algorithm, and the extracted data of past typhoons may include only data of past typhoons after the time of appearance in the final extraction source.

In one embodiment, in step (3), the predetermined predictors are a 500 hPa station altitude, a directing east-west wind component, and a directing north-south wind component, and the daily average synthesized observation background fields are provided by ECMWF's ERA-Interim. It can be calculated from data with horizontal resolution of 0.75°Х0.75° at 6-hour intervals.

In one embodiment, the directing east-west wind component and the direct-flow north-south wind component may be synthesized as daily averages on barometric altitudes having different directing layer depths according to the central air pressure among data of past typhoons.

In one embodiment, in step (4), the one or more dynamics models include CFSv2, GEFS and GFS dynamics models provided by NCEP, and the model background fields may be gridded to have the same horizontal grid as the observation backgrounds. .

In one embodiment, in step (5), the pattern correlation is calculated using only the values of the predictors in the region where there are significant differences between the observational backgrounds and the course patterns in the backgrounds, and One difference may be a statistically significant region at the 95% confidence level identified by one-way analysis of variance.

In one embodiment, step (5) normalizes the predictors of the observed background fields and the model background fields at each day from 6 days to 8 days later (5-1) for each case of past typhoons. (5-2) Calculating a pattern correlation coefficient between normalized observation background fields and model background fields, (5-3) Using the membership coefficient of each past typhoon as a weight, 6 days to 8 days later Calculating the member average correlation coefficient for each career pattern on each day after work, and (5-4) 3-day average of the average member correlation coefficient for each career pattern on each day from 6 to 8 days later Calculating a weighting factor for each pattern may be included.

In one embodiment, step (6) includes: (6-1) calculating the course frequencies by gridating the tracks of past typhoons belonging to each course pattern; (6-2) the calculated course for each grid Multiplying the frequencies by the weighting factor for each career pattern to calculate a specific relative frequency map and a total relative frequency map, (6-3) determining a representative curve for each career pattern based on the specific relative frequency map, and ( 6-4) determining a final predicted pathway based on the representative curve for each pathway pattern and the total relative frequency map.

In one embodiment, in step (6-3), starting from the location of each of the past typhoons at the time of analysis on the specific relative frequency map for each course pattern, the longitude and longitude of the grid with the maximum value in each latitude grid box , Points corresponding to the latitude of the grid with the maximum value in each longitude grid box are sequentially detected until the maximum value in the latitude grid box or longitude grid box is less than 0.5, and then a specific relative frequency for each course pattern A representative curve for each course pattern can be determined by connecting the moving averaged points of five sequentially detected points on the map.

로 정의되는 대표 곡선의 점수(s(m,c))가 가장 높은 대표 곡선을 최종 예측 진로로 결정할 수 있다.In one embodiment, in step 6-4, for the grid point index (i) on the representative curve for each course pattern, m represents the dynamic model used, c represents the course pattern, and L(m, c ) represents the length of the representative curve associated with the epidemiological model m and the career pattern c, and f[m, x(i), y(i)] is passed by the representative curve of the career pattern c associated with the dynamic model m , where s(m,c) represents the score of the representative curve associated with dynamic model m and path pattern c, case,

The representative curve with the highest score (s(m,c)) of the representative curve defined as can be determined as the final predicted course.

A computing system according to an embodiment of the present disclosure is configured to implement a method for predicting a typhoon course in the mid-term.

According to the system and method for predicting the course of a typhoon in the mid-term in the western North Pacific region according to the present disclosure, within a short calculation time using reduced computing resources compared to a dynamic model, 6 to 8 days after a typhoon occurred in the Northwest Pacific region Future course can be predicted with improved accuracy.

Figure 1 is a distribution chart showing the lifespan of typhoons that occurred during the period from 1979 to 2019 in the Northwest Pacific region.
2 is a flowchart of a method for predicting a mid-term typhoon path in the western North Pacific region according to an embodiment of the present disclosure.
3 is a block diagram conceptually illustrating the flow of a method for predicting a mid-term typhoon path in the western North Pacific region according to an embodiment of the present disclosure.
Figure 4 is a predicted course from the date of occurrence of Typhoon Kongrei provided by the Korea Meteorological Administration to 5 days and past typhoons used in the method for mid-term prediction of the typhoon course in the Northwest Pacific region according to an embodiment of the present disclosure. It shows extraction circles for extracting a part.
5 shows two path patterns in which the extracted past paths can be clustered in the case of mid-term prediction of the path of Typhoon Kongrei in the method for mid-term predicting the path of a typhoon in the Northwest Pacific region according to an embodiment of the present disclosure. show
6 is an example of observation background fields and model background fields that can be used in the case of mid-term prediction of the typhoon course in the method for mid-term prediction of the typhoon course in the Northwest Pacific region according to an embodiment of the present disclosure. shown as
7 is a method for predicting the course of a typhoon in the Northwest Pacific region in the mid-term according to an embodiment of the present disclosure, calculating a pattern correlation between an observed background field and a model background field, and calculating a weighting factor for each course pattern based on the pattern correlation It is a flow chart showing the flow of steps.
8 is a method for predicting the course of a typhoon in the Northwest Pacific region in the mid-term according to an embodiment of the present disclosure, calculating a pattern correlation between an observed background field and a model background field, and calculating a weighting factor for each course pattern based on the pattern correlation It is a block diagram conceptually showing the flow of steps.
9 is a method for predicting a typhoon path in the mid-term in the western North Pacific region according to an embodiment of the present disclosure, a relative frequency map is generated using past paths and pattern correlation coefficients, and based on the generated relative frequency map It is a flowchart showing the flow of the step of obtaining the final predicted course.
10 is a diagram conceptually illustrating the flow of calculating a specific relative frequency map and a total relative frequency map in a method for predicting a typhoon path in the mid-term in the western North Pacific region according to an embodiment of the present disclosure.
11 is a maximum value shown on a specific relative frequency map for a first course pattern (C1) and a second course pattern (C2) when predicting the course of Typhoon Kongrei in the mid-term according to an embodiment of the present disclosure. Points and representative curves are shown.
12 shows GEFS dynamics when the method for predicting the typhoon path in the Northwest Pacific region in the mid-term according to an embodiment of the present disclosure is applied (TP-GFS) and when the vortex detection algorithm of the CFSv2 dynamics model is used (VT-CFSv2) In the case of using the vortex detection algorithm of the model (VT-GEFS) and the case of using the eddy detection algorithm of the GFS dynamics model (VT-GFS), the error of the predicted course for the period after 6 days to 8 days is shown.
13 is a method for predicting a typhoon course in the Northwest Pacific region in the mid-term according to an embodiment of the present disclosure, in which the predictors used are configured differently, and the error of the predicted course for the period after 6 to 8 days is shown.
14 shows the actual course of Typhoon Bolaven that occurred in 2012, the course of Typhoon Bolaven predicted according to an embodiment of the present disclosure, and the predicted course of Typhoon Bolaven calculated using the vortex tracking algorithm of the dynamics model do.

Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

Terms used throughout this disclosure may have nuanced meanings suggested or implied in context beyond those explicitly stated.

As used in this disclosure, phrases such as “in one embodiment” or “in an exemplary embodiment” do not necessarily refer to the same embodiment, and phrases such as “in another embodiment” or “in another exemplary embodiment” may or may not necessarily refer to another embodiment.

As used in this disclosure, terms such as “and”, “or” or “and/or” may include a variety of meanings that may depend at least in part on the context in which such terms are used.

As used in this disclosure, the term "one or more" may be used to describe any feature, structure, or characteristic in a singular sense, or a feature, structure, or combination thereof in a plural sense, depending at least in part on the context. can be used to describe

Further, the terms "based on," "in response to," and "in response to" are not intended to convey an exclusive set of factors, but, at least in part, depending on the context, additional additions not necessarily explicitly delineated. It is also possible to allow for the presence of factors.

In addition, expressions such as "first" and "second" used in the present disclosure may modify various elements regardless of order and/or importance, and are only used to distinguish one element from another. The components are not limited.

In the present disclosure, the term "position of a typhoon" refers to the center position of a typhoon. It usually refers to a location that has the lowest regional mean sea level pressure (MSLP) and is surrounded by closed isobars at a particular geographic elevation or land/sea level. As generally referred to in the art, the location data of the typhoon refers to the latitude and longitude at which the center of the typhoon is located.

In this disclosure, the term "intermediate-term forecast" refers to a forecast for a period from the time of analysis (day t) to day 6 (day t+6) to day 8 (day t+8). That is, the mid-term prediction for a specific analysis time (day t) refers to predictions for days t+6, t+7, and t+8.

The term “initial” in this disclosure refers to the period from the time of analysis (day t) to the fifth day (t+5 days). That is, the initial period from the specific analysis time (t day) refers to t + 1 day, t + 2 day, t + 3 day, t + 4 day, and t + 5 day, wherein the analysis time (t day ) may or may not be included together.

The term "predictor" in this disclosure refers to any of a variety of meteorological variables. For example, predictors include, but are not limited to, 500 hPa station elevation, direct current (including east-west and north-south wind components), and the like.

Unless defined otherwise, all technical and scientific terms used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art.

Hereinafter, in the description according to the embodiments of the present disclosure, the present inventors applied the case of Typhoon Kongrei, which occurred in 2018, as an example for predicting the course of a typhoon in the mid-term. In addition, it is assumed that mid-term prediction is performed according to the occurrence time (day t) of Typhoon Kongrei. However, those skilled in the art will understand that the same technique can be applied to any typhoon occurring in the Northwest Pacific.

2 is a flowchart of a method for predicting a mid-term typhoon path in the western North Pacific region according to an embodiment of the present disclosure. 3 is a block diagram conceptually illustrating the flow of a method for predicting a mid-term typhoon path in the western North Pacific region according to an embodiment of the present disclosure.

Referring to FIGS. 2 and 3 , a method for predicting a typhoon path in the western North Pacific region in the mid-term according to an embodiment of the present disclosure includes typhoons that have historically occurred during a predetermined past period (hereinafter referred to as 'past typhoons'). Among the data of the typhoon (hereinafter referred to as 'target typhoon'), extracting data of past typhoons satisfying preset criteria including the initial characteristics of the typhoon (hereinafter referred to as 'target typhoon') to predict the mid-term course (S10), clustering Using an algorithm, clustering the paths of past typhoons into a plurality of path patterns from the data of past typhoons extracted in step S10 (S20), 6 days after the analysis time (day t) (day t+6) to Calculating daily average synthesized observation background fields of predetermined predictors associated with each course pattern for the day after 8 days (t+8 days) (S30), 6 days after the analysis time (day t) ( Calculating daily average model background fields of the predictors from one or more epidemiological models for days from t + 6 days) to 8 days later (t + 8 days) (S40), of the predictors, in step S30 Calculating the pattern correlation between the calculated observation background fields and the model background fields calculated in step S40, generating a weighting factor for each career pattern (S50), and using past career paths and weighting factors for each career pattern and generating a relative frequency map, and obtaining a final predicted course based on the generated relative frequency map (S60).

1. Extraction of Past Careers

Referring to FIGS. 2 and 3 , in one embodiment, in step S10, data of past typhoons satisfying preset criteria including initial characteristics of the target typhoon are extracted from data of past typhoons during a predetermined past period. It can be.

The data of all past typhoons for a predetermined historical period is the best track for the period from 1979 to 2019 provided, for example, by the Regional Specialized Meteorological Center (RSMC) Tokyo Typhoon Center. It can be data or a part thereof. However, those skilled in the art will understand that the historical period from 1979 to 2019 can be changed in any number of ways. The best course data provided by the RSMC Tokyo Typhoon Center includes, for example, parameters (central location, intensity level, etc.) of 6-hour intervals for past typhoons that occurred during a specific past period.

The initial characteristics of the target typhoon may include, for example, the location of the current (or analysis time) (day t) of the target typhoon and the predicted location until the following 5 days (day t+5). Data on the initial characteristics of the target typhoon may be provided by forecasting organizations (and/or companies). For example, if the target typhoon is Typhoon Kongrei, which occurred on September 29, 2018, as the initial course of Typhoon Kongrei, the first from the date of occurrence provided by the Korea Meteorological Administration (day t = September 29, 2018) Predictive data up to 5 days (t+5 days) may be used.

4 is a predicted course from the date of occurrence of Typhoon Kongrei provided by the Korea Meteorological Administration (FIG. 4a) up to 5 days, and is used in a method for predicting a mid-term typhoon course in the Northwest Pacific region according to an embodiment of the present disclosure. It shows extraction circles (FIG. 4b) for extracting some of the past typhoons.

Referring to FIGS. 2 and 4 , the criteria set in advance in step S10 are as follows.

First, as shown in FIGS. 4A and 4B, the position of the typhoon at the analysis time (day t) and the position of the typhoon on each predicted day for the following 5 days (day t + 1 to day t + 5) An array of six extraction circles, centered on , can be considered. The radii of these six extraction circles are, for example, 500 km (for day t), 600 km (for day t+1), 700 km (for day t+2), and 800 km (for day t+3). ), 900 km (in the case of t + 4 days), and 1000 km (in the case of t + 5 days). However, those skilled in the art will understand that the size of the extraction circles centered on the predicted initial position of the typhoon can be arbitrarily determined. The present inventors set the radius of the extraction circle to increase from day t + 1 to day t + 5 in order to consider the uncertainty of the predicted course up to 5 days later (t + 5 days) for the position of the typhoon. However, it was confirmed that the change in the radius of the six extraction circles did not significantly affect the mid-term prediction of the final typhoon course.

Then, i) entering the first extraction source (radius of 500 km) and passing through each of the six extraction sources while leaving (or disappearing) from the final extraction source (radius of 1000 km), ii) having a lifetime of 6 days or more, iii) The course of past typhoons that occurred in a period similar to the target typhoon can be extracted.

In this case, the similar period in iii) may refer to, for example, a period of 3 months in total before and after the analysis time (t day), but the total length of the similar period is not limited thereto. For example, in the case of typhoon Kongrei, since it occurred on September 29, 2018, other past typhoons that occurred between August 15 and November 13 during the period from 1979 to 2019 may also be included. In this case, in one embodiment, the course of a total of 72 past typhoons can be extracted.

2. Clustering of Past Careers

Referring back to FIGS. 2 and 3 , in one embodiment, in step S20, the paths of past typhoons can be clustered into a plurality of course patterns from the data of past typhoons extracted in step S10 using a clustering algorithm. . For example, extracted past career paths may be classified into several career patterns by a fuzzy c-means clustering algorithm. The fuzzy c-means clustering algorithm causes each course to belong to all clusters, with various membership coefficients indicating how strongly that course belongs to a particular cluster. In one embodiment, each typhoon course may be classified into a course pattern having the largest membership coefficient, and clusters may be generated. These membership coefficients can then be used to calculate pattern correlation coefficients in step S50.

Since the typhoon path prediction technology according to embodiments of the present disclosure is to perform forward prediction 6 to 8 days after the analysis time (and/or occurrence date) during the life of the typhoon, the initial stage (step) of the path of past typhoons Steps before appearing in the final extraction source of S10) may not be included in the analysis target. Therefore, in one embodiment, the clustering of course patterns may be performed by considering only the courses of past typhoons after the point in time appearing in the final extraction source used in step S10. This is not overly sensitive since the variance of historical tracks contained within the first five circles of extraction is much smaller than the variance of historical paths contained within or proceeding beyond the final circle of extraction.

In one embodiment, the number of career patterns to be clustered may be determined by an Xie-Beni index known in the art. This index, which can be used for fuzzy clustering, is the ratio of compressibility (within a cluster) to separation (between clusters). A small value of this index means that the career patterns are well separated from each other and the dispersion of career paths within each career pattern is small. This index value may be calculated according to "the number (n) of career patterns". Here, n is greater than or equal to 2 and is less than or equal to the total number of extracted past routes. For example, considering the number of extractable past career paths, the maximum number of career patterns may be set to 5. For example, when the target typhoon is Typhoon Kongrei, n may be set to 2.

5 shows two path patterns in which the extracted past paths can be clustered in the case of mid-term prediction of the path of Typhoon Kongrei in the method for mid-term predicting the path of a typhoon in the Northwest Pacific region according to an embodiment of the present disclosure. show The determined two path patterns may be shown as a first path pattern C1 and a second path pattern C2. For example, each of the first course pattern C1 and the second course pattern C2 may include the course of 36 past typhoons. In FIG. 5, the paths of individual past typhoons are shown in gray, and the average path within each path pattern is shown in bold black. In FIG. 5, the dotted line indicates a course from the point closest to the occurrence location of Typhoon Kong-Rei to the point first included in the final extraction circle (eg, a circle with a radius of 1000 km) used in step S10.

Paths included in each path pattern (eg, C1, C2, etc.) may have different main geographical locations. For example, the second route pattern C2 may include routes heading toward the South China Sea, while the first route pattern C1 may include routes that bend while heading toward the poles.

3. Synthesis of observational background fields for each career pattern

Referring back to FIGS. 2 and 3 , in one embodiment, in step S30, each day is 6 days (t+6 days) to 8 days (t+8 days) from the analysis time (t day). Observation backgrounds synthesized by daily averages of predetermined predictors associated with the course pattern of can be calculated. As observational backgrounds, for example, reanalysis data from ERA-Interim of the European Center for Medium-range Weather Forecasts (ECMWF) can be used. For example, as observation background fields, data provided at a horizontal resolution of 0.75°×0.75° at intervals of 6 hours can be used.

In step S30, as predetermined predictors, a 500 hPa station altitude (GPH500) and a direct flow (STEER) may be used. The GPH500 is often used in the art to identify subtropical highs that directly affect the movement of typhoons due to expansion or retreat. Also, STEER in a deep layer in the troposphere can be presented as pressure-weighted wind and vertical-weighted wind in a specific layer of the troposphere.

In the case of direct currents, it may be inappropriate to use constant (unchanging) directed layers to predict the movement of a typhoon. Thus, as shown in Table 1, barotropic altitudes can be used to approximate changes in directed layer depth and storm intensity. Table 1 shows tropospheric pressure altitudes used to calculate direct currents for each typhoon intensity category.

Typhoon intensity classification (hPa) Tropospheric pressure altitude (hPa) > 1000 700-850 990-999 500-850 980-989 400-850 970-979 400-850 960-969 300-850 950-959 300-850 940-949 250-850 < 940 200-700

By using these barometric altitudes, the error of the predicted course according to the strength of the typhoon can be reduced. In one embodiment, only data from each 50 hPa barometric altitude interval may be used to efficiently perform the calculations that yield the observed background fields.

For each of the tracks of all past typhoons extracted in step S10, daily observation background fields for 6 to 8 days after the analysis time (but within the lifetime of the past typhoon) may be calculated. For example, in one embodiment for the mid-term prediction of Typhoon Kongrei, for the cases of past typhoons of 72 extracted and clustered (36 each for C1 and C2), the predictors (GPH500, STEER) The reanalysis background fields can be synthesized as daily averages.

6 is an example of observation background fields and model background fields that can be used in the case of mid-term prediction of the typhoon course in the method for mid-term prediction of the typhoon course in the Northwest Pacific region according to an embodiment of the present disclosure. shown as

6A, 6B, 6D, and 6E are examples of past typhoons having the highest membership coefficients belonging to two course patterns (C1, C2) for mid-term prediction of the course of Typhoon Kongrei according to an embodiment. For , the background fields of the predictors 6 days after the analysis time (t + 6 days) are shown. 6A and 6D, in the case of past typhoon cases belonging to the first course pattern (C1), the western edge of the subtropical high pressure (thick solid line) at 25 degrees north latitude is close to the meridian of 130 degrees east longitude, but at higher latitudes It can be seen that it retreated to the east. At this time, the strongest current near the deep low pressure area was the southwesterly wind. On the other hand, referring to FIGS. 6B and 6E , in the case of past typhoon cases belonging to the second course pattern C2, it can be seen that the western edge of the subtropical high pressure at 25 degrees north latitude is near the meridian of 120 degrees east longitude. The magnitude of the direct current wind vector was very uniform over the entire domain.

4. Model Predicted Backgrounds

Referring back to FIGS. 2 and 3 , in one embodiment, in step S40, one for the day of 6 days (t+6 days) to 8 days (t+8 days) from the analysis time (t day). Daily average model background fields of the predictors can be calculated from the above epidemiological models. Specifically, for each day from 6 to 8 days after the analysis time of all past typhoons extracted in step S10, a 500 hPa position altitude (GPH500) background field and a direct current (STEER) background field are calculated from the dynamic model. can

For example, one or more epidemiological models include the Climate Forecast System version 2 (CFSv2), Global Ensemble Forecast System (GEFS) and Global Forecast System (GFS), etc. Details of these models are shown in Table 2.

division CFSv2 GEFS GFS 2011 2012-2014 2015-2019 2011-2014 2015-2019 model resolution
(km) 0-192h 100 70 55 34 28 13 192-384h 100 70 70 55 70 13 Model cycles per day (times) 4 4 4 4 4 4 Output time interval (h) 0-192h 6 6 6 6 3 3 192-384h 6 6 6 6 12 3

Among the dynamic models listed in Table 2, CFSv2 and GFS provide a single simulation, while GEFS provides 21 ensemble members. Therefore, when using GEFS as the epidemiological model, the average value of all available members can be used. The data of the dynamic models may be grid-adjusted with the same horizontal grid as the observation (reanalysis) data used in step S30.

6c and 6f show the daily average model background of GPH500 and STEER after 6 days (t + 6 days) calculated from the dynamics model CFSv2 in the case of mid-term prediction of the course of Typhoon Kongrei according to an embodiment. do. Referring to FIG. 6c, it can be seen that the western edge of the subtropical high was near the meridian of 130 degrees east longitude, and the ridge was close to the southern coast of Japan. Referring to FIG. 6F, it can be seen that the dominant direct current on the East Sea around the meridian of 130 degrees east longitude is the southerly wind. Referring to FIGS. 6A to 6F, it can be seen that the background fields calculated from the dynamic model are more similar to the case of the first path pattern C1 than to the case of the second path pattern C2 (FIGS. 6A and 6B). The numbers in parentheses following C1 and C2 shown in the title in , indicate the membership coefficient (front) and pattern similarity with the model background (back).

5. Pattern correlation between observational background and model background

Referring back to FIGS. 2 and 3 , in one embodiment, in step S50, pattern correlation between the observed (reanalyzed) background fields calculated in step S30 and the model background fields calculated in step S40 of the predictors. The relationship is calculated, and based on it, a weighting factor for each career pattern may be created.

When the background fields related to a specific course pattern are similar to the predicted background fields including the target typhoon, since the target typhoon is highly likely to follow a similar course pattern, in order to quantify this similarity, 6 days from the analysis time (day t) For each day from after (t+6 days) to 8 days after (t+8 days), pattern correlations between the synthesized observed background fields and the predicted background fields from the dynamic model can be calculated.

In one embodiment, the pattern correlation is a region in which there are significant differences between any two course patterns (eg, in the case of Typhoon Kongrei) in daily average background fields (hereinafter referred to as 'target region'). ) may be considered. For example, the target area may be an area surrounded by 10 degrees north latitude to 40 degrees north latitude and 105 degrees east longitude to 150 degrees east longitude. Such significant differences can be identified through, for example, one-way ANOVA. For example, a target region can be a statistically significant region with a 95% confidence level identified through a one-way analysis of variance. 6A to 6F, the regions where the significant differences exist are stippled.

In one embodiment, the step of calculating the pattern correlation in step S50 and calculating the weighting factor for each course pattern based on the pattern correlation is, for each case of past typhoons, the observation backgrounds and model backgrounds for each medium-term forecast date. Normalizing predictors (S51), calculating pattern correlation coefficients between background fields normalized by mid-term prediction date for each case of past typhoons (S52), using the membership coefficient of each past typhoon as a weight Calculating the average correlation coefficient for each member by career pattern for each mid-term prediction date (S53), and calculating the weighting factor for each career pattern by averaging the average member correlation coefficient for each career pattern for each mid-term forecast date over 3 days (S54) ) may be included.

7 is a method for predicting the course of a typhoon in the Northwest Pacific region in the mid-term according to an embodiment of the present disclosure, calculating a pattern correlation between an observed background field and a model background field, and calculating a weighting factor for each course pattern based on the pattern correlation It is a flow chart showing the flow of steps. 8 is a method for predicting the course of a typhoon in the Northwest Pacific region in the mid-term according to an embodiment of the present disclosure, calculating a pattern correlation between an observed background field and a model background field, and calculating a weighting factor for each course pattern based on the pattern correlation It is a block diagram conceptually showing the flow of steps.

Referring to FIGS. 7 and 8 , in step S51, for each case of past typhoons, predictors of observation background fields and model background fields may be normalized for each medium-term forecast date. Since the size and physical unit of the 500 hPa geographic altitude (GPH500) and direct current (STEER) east-west/north-south components are different, the predictors can be normalized to remove the effect due to their size and physical unit. For example, the wind speeds of the east-west wind and the north-south wind may be normalized to values ranging from -1 to 1, respectively. For example, a 500 hPa geographic altitude (GPH500) may be normalized to values ranging from 0 to 1. Then, the normalized east-west wind speed, the normalized north-south wind speed, and the normalized 500 hPa geographic altitude (hereinafter referred to as 'normalized predictors') are combined to form one array at each of the grids within the target area. It can be. Accordingly, arrays composed of three normalized predictors may be generated on each of the grids in the target area for each observation background field and model background field, on each day after 6 to 8 days.

In step S52, for each case of past typhoons, a pattern correlation coefficient between normalized background fields for each medium-term forecast date may be calculated. Depending on embodiments, when the correlation coefficient calculated in this step is a negative number, it may be set to 0. This may be performed to prevent a negative pattern correlation representing a difference between the two background fields from canceling out the path patterns in the process of calculating the relative frequency map in a subsequent step.

In step S53, the member average correlation coefficient for each course pattern for each medium-term forecast date may be calculated using the membership coefficient of each past typhoon as a weight. That is, by using each membership coefficient used when past typhoons were clustered in step S20, the pattern correlation coefficient between the observed background field and the model background field in each typhoon case is adjusted, so that each of 6 to 8 days later On the day of , a member-averaged correlation coefficient for each career pattern can be calculated. For example, for each case of typhoons, the pattern correlation coefficient (pcc) calculated in step S52 multiplied by its own membership coefficient (mc) and the product of the membership coefficient (pcc × mc), By averaging the total number of member typhoons belonging to the course pattern, the member average pattern correlation coefficient can be calculated for each medium-term forecast date.

For example, in the case of mid-term prediction of the course of Typhoon Kongrei according to an embodiment, on the day of 6 days after the analysis time (t + 6), each past typhoon belonging to the first course pattern (C1) and In this regard, by multiplying the pattern correlation coefficients calculated from the observation background field and the model background field by the membership coefficient of each past typhoon, and then averaging the past typhoons belonging to the first course pattern (C1), the member average pattern correlation A value of 0.71 can be calculated as a coefficient (see the lower left value in the rounded blue rectangle in FIG. 8). Similarly, on the day of 6 days (t+6) from the time of analysis, the member average pattern correlation coefficient for the second career pattern (C2) can be calculated as 0.58 (see the lower right value in the rounded blue rectangle in FIG. 8). . In this way, calculating the member average pattern correlation coefficient may be performed in order to reduce the effect of the past typhoon's course, which is too far from the course pattern to which it belongs, on the pattern correlation for each course pattern.

In step S54, the 3-day average of the member average correlation coefficients for each career pattern for each mid-term prediction day may be used to calculate a weighting factor for each career pattern. The member average pattern correlation coefficients for each career pattern (eg, C1 or C2) calculated for the three mid-term prediction days (ie, after 6 days, 7 days, and 8 days) range from 6 days to 8 days. Averaged over the 3-day period after 3 days, a single weighting factor per career pattern can be calculated for medium-term forecasting. Meanwhile, although FIG. 8 shows only the case where CFSv2 is used as the dynamics model for simplicity, it should be understood that the same process can be applied to other dynamics models as well.

6. Mid-term Career Prediction

Referring back to FIGS. 2 and 3 , in one embodiment, in step S60, a relative frequency map is generated using a weighting factor for each past route and each route pattern, and a final predicted route is determined based on the generated relative frequency map. can be obtained.

The weighting factors for each course pattern calculated in step S50 may indicate the possibility that the typhoon follows a specific course pattern. However, only a few scalar values may be insufficient to quantify the prediction that a typhoon is likely to occur at a specific point in two-dimensional space. Therefore, a two-dimensional distribution of typhoon occurrences can be generated by combining these weighting factors with past courses belonging to each course pattern.

In an embodiment, the step of generating a relative frequency map using past routes and pattern correlation coefficients and obtaining a final predicted route based on the generated relative frequency map may include gridding past routes belonging to each route pattern. calculating the path frequencies (S61), calculating the specific relative frequency map and the total relative frequency map by multiplying the calculated path frequencies by the weighting factor for each path pattern (S62), the specific relative frequency map Determining a representative curve for each career pattern based on (S63), and determining a final predicted career path based on the representative curve for each career pattern and the total relative frequency map (S64).

9 is a method for predicting a typhoon path in the mid-term in the western North Pacific region according to an embodiment of the present disclosure, a relative frequency map is generated using past paths and pattern correlation coefficients, and based on the generated relative frequency map It is a flowchart showing the flow of the step of obtaining the final predicted course. 10 is a diagram conceptually illustrating the flow of calculating a specific relative frequency map and a total relative frequency map in a method for predicting a typhoon path in the mid-term in the western North Pacific region according to an embodiment of the present disclosure.

Referring to FIGS. 9 and 10 , in step S61, past routes belonging to each route pattern may be gridded to calculate route frequencies. For example, the past routes belonging to each course pattern (eg, referring to the past routes shown in FIGS. 4A and 4B ) may be gridded 1°×1° to obtain a course frequency. In this case, each lattice value may be set to the number of past routes passing through an area with a radius of 250 km centered on a 1°×1° lattice box to which the corresponding lattice belongs. Travel frequencies thus calculated may correspond to grid values in a color map located above the center of FIG. 10 , for example.

In step S62, the calculated path frequencies for each grid are multiplied by the weighting factor for each path pattern, so that a specific relative frequency map and a total relative frequency map can be calculated. At this time, first, by multiplying the track frequencies calculated in step S61 by the weight factor for each track pattern calculated in step S50, a "weighted track frequency" can be calculated for each grid. By adding all of these weighted career frequencies regardless of career patterns and then dividing by the total number of past career paths belonging to the career patterns, a two-dimensionally distributed "overall relative frequency map" can be calculated. (far right map in Figure 10). On the other hand, "specific relative frequency map" is the weighted career frequencies in each grid for each career pattern (eg, C1 or C2, etc.), as the total number of past careers belonging to the career pattern It can be calculated by dividing (refer to the color shading of FIG. 11 described later).

In step S63, a representative curve for each career pattern may be determined based on the specific relative frequency map. In this step, starting from the location of the typhoon at the analysis time (day t) on the specific relative frequency map for each course pattern, the grid with the maximum value in each latitude grid box (ie, the grid box extended in the row direction) The points corresponding to the longitude and the latitude of the grid having the maximum value in each longitude grid box (ie, the grid box extended in the column direction) (hereinafter, the maximum value point), It can be detected sequentially until the value is, for example, less than 0.5.

Then, on a specific relative frequency map for each career pattern (e.g., in the north and west directions), for example, by connecting the moving averaged points for five sequential maximum value points, a representative curve for each career pattern is obtained. can be determined 11 is a diagram on a specific relative frequency map for each path pattern, that is, the first path pattern C1 and the second path pattern C2, in the case of mid-term prediction of the path of Typhoon Kongrei according to an embodiment. The maximum value points and representative curves shown are shown (FIG. 11A shows a representative curve for the first course pattern C1, and FIG. 11B shows a representative curve for the second course pattern C2).

In step S64, a final predicted path may be determined based on the representative curve for each path pattern and the total relative frequency map. At this time, the final prediction path may be determined as the representative curve having the highest score (s) according to Equation 1 below with respect to the lattice point index (i) on the representative curves calculated in step S63.

Here, m represents the dynamic model used, c represents the path pattern, and i represents grid point indices on the representative curve for the path pattern c. In addition, L(m, c) represents the length of a representative curve associated with the epidemiological model m and the career pattern c. f[m, x(i), y(i)] is the total relative frequency at grid point [x(i), y(i)] associated with epidemiological model m and passed by the representative curve of path pattern c value on the map. In addition, s(m,c) represents the score of the representative curve associated with the epidemiological model m and the career pattern c. In Equation 1 above, the latter part of the representative curve (the second term in the numerator) is highlighted.

As described above, in the method for predicting the mid-term typhoon path in the western North Pacific region according to the embodiments of the present disclosure, two or more typhoons coexist in the western North Pacific region compared to the mid-term prediction result of the typhoon track predicted from the dynamics model. This can be particularly useful when For example, coexisting typhoons centered west of 140 degrees east longitude at the analysis time (day t) can be considered. Here, a co-existing typhoon is defined as a case in which two or more typhoons exist at the same time, and the distance between the centers of the typhoons when they first coexist is less than 20 degrees (in latitude/longitude). When two typhoons are close together, the movement of one typhoon can be influenced by the circulation of the other, and therefore, large-scale direct currents cannot fully explain the movement of these typhoons. In contrast, the method for predicting the path of a typhoon in the mid-term in the western North Pacific region according to the embodiments of the present disclosure can reduce errors in predicting the path of coexisting typhoons.

12 shows GEFS dynamics when the method for predicting the typhoon path in the Northwest Pacific region in the mid-term according to an embodiment of the present disclosure is applied (TP-GFS) and when the vortex detection algorithm of the CFSv2 dynamics model is used (VT-CFSv2) In the case of using the vortex detection algorithm of the model (VT-GEFS) and the case of using the eddy detection algorithm of the GFS dynamics model (VT-GFS), the error of the predicted course for the period after 6 days to 8 days is shown.

Referring to FIG. 12, when the background field of the GFS dynamics model is used for pattern correlation calculation in an embodiment of the present disclosure (TP-GFS), the error of the typhoon course predicted for the period from 6 days to 8 days later is, It can be seen that during the 3-day period, it is generally reduced compared to the errors of the typhoon path predicted using only the eddy current detection algorithm of the dynamic models.

On the other hand, in the method for predicting the course of a typhoon in the mid-term in the Northwest Pacific region according to embodiments of the present disclosure, a direct current (STEER) to which a barometric altitude is applied along with a 500 hPa position altitude (GPH500) can be used as a predictor. there is. In contrast, in the art, it is common to consider more thermodynamic meteorological variables known to be more related to changes in typhoon intensity as predictors for use in predicting the typhoon path. That is, in general, for forecasting the course of a typhoon, as additional meteorological variables considered to play a specific role in the movement of a typhoon due to the relationship between the typhoon intensity and the direction layer, Relative humidity needs to be considered.

However, in the method for predicting the mid-term course of a typhoon in the Northwest Pacific region according to embodiments of the present disclosure, these relative humidities are not included in the predictor, but are equal to or higher than the mid-term course prediction results when these relative humidities are included. It is possible to obtain an improved prediction course.

13 is a method for predicting a typhoon course in the Northwest Pacific region in the mid-term according to an embodiment of the present disclosure, in which the predictors used are configured differently, and the error of the predicted course for the period after 6 to 8 days. 13 shows the case of using the background field of the CFSv2 dynamics model for pattern correlation calculation (TP-CFSv2), the case of using the background field of the GEFS dynamics model (TP-GEFS), and the case of using the background field of the GFS dynamics model ( TP-GFS) the predicted path errors are shown. At this time, FIG. 13a shows the results using only the 500 hPa position altitude (GPH500) background field and the direct current (STEER) background field as predictors, and FIG. The results of using the relative humidity background fields (RH300, RH500, RH850) together are shown. Comparing FIGS. 13A and 13B , even if calculation of relative humidity background fields, which are meteorological variables generally considered for predicting the course of a typhoon in the present art, is not included, there is a large difference in the error of the predicted course as the final result. It can be seen that there is no Accordingly, in the embodiments of the present disclosure, the overall calculation efficiency and speed can be improved by significantly reducing the variables used in the calculation for predicting the path of the typhoon.

Furthermore, the method for predicting the mid-term typhoon course in the Northwest Pacific region according to embodiments of the present disclosure has the advantage of being able to continuously predict over a wide latitude and longitude range. Often, the path of a typhoon detected from the outputs of the dynamic model does not dissipate at a position similar to that of an actual typhoon. In this case, the path of the typhoon detected from the output of the dynamics model may be considerably shorter than the actual path of the typhoon. This may occur because the vortex tracking algorithm of the dynamics model fails to track the typhoon, or because of an error in the dynamics model simulating the typhoon when the typhoon moves to mid-latitudes. However, in the predicted course according to the embodiments of the present disclosure, this problem does not occur because the latitude and longitude ranges of the extracted past courses can be wide enough to cover the length of the actual typhoon course.

14 shows the actual course of Typhoon Bolaven that occurred in 2012, the course of Typhoon Bolaven predicted according to an embodiment of the present disclosure, and the predicted course of Typhoon Bolaven calculated using the vortex tracking algorithm of the dynamics model. show

In FIG. 14 , the star marker, triangle marker, and square marker indicate the (predicted) position of the typhoon 6 days later, 7 days later, and 8 days later, respectively. Referring to FIG. 13, it can be seen that the predicted course (red) of Typhoon Bolaven according to an embodiment of the present disclosure is more closely predicted to the actual course (black) than the predicted course (blue) from the CFSv2 dynamics model. can

On the other hand, the method for predicting the mid-term course of a typhoon in the Northwest Pacific region according to the embodiments of the present disclosure is faster in calculation time and more efficient than the case of predicting the mid-term course of a typhoon using a vortex tracking algorithm of a dynamics model. For example, when performing mid-term prediction of a typhoon course according to embodiments of the present disclosure using a computer including Intel's Xeon(R) Gold 5118 as a CPU and 8 Hynix's DDR4 16GB 2666MHz Х as RAM , the time taken to calculate may be in the range of, for example, 84 seconds (about 1.4 minutes) to 762 seconds (about 12.7 minutes).

As described above, the method for predicting the mid-term typhoon path in the western North Pacific region according to embodiments of the present disclosure may be configured to be implemented by a computing system. Computing systems may include, but are not limited to, servers, desktops, notebooks, workstations, personal digital assistants (PDAs), mainframes, one or more types of digital computing devices, and the like, for example.

A computing system may include one or more processor(s), one or more memory(s), and the like. Each of the processor, memory, etc. may be interconnected using various wired/wireless communication means and/or buses. Depending on embodiments, multiple processors and/or multiple memories may be used, or multiple computing devices may be connected. Each computing device may provide some of the necessary operations, for example as a server, group of servers, or multiprocessor system. Where the plurality of functions and/or operations are performed by processor(s), the plurality of functions and/or operations may be performed by any number of processor(s) included in any number of computing device(s). can Further, when a function and/or operation is performed by the processor(s), the function and/or operation may be performed on any number of computing devices included in any number of computing device(s), for example in a distributed computing system. may be performed by the processor(s).

The memory(s) may be non-volatile memory unit(s) that store information within the computing system. The memory(s) may be or include, for example, a floppy disk, hard disk, magnetic disk, optical disk, magnetic tape, flash memory, solid state memory, various forms of computer readable media storing computer readable instructions, and the like. can

One or more methods as described above may be configured to be implemented when instructions stored in memory(s) are executed by one or more processor(s). These methods may be implemented in digital electronic circuits, integrated circuits, application specific integrated circuits (ASICs), computer hardware, firmware, software, and/or combinations thereof. These various implementations may include implementation on one or more computer programs executable and/or interpretable on a programmable system comprising at least one programmable processor configured to execute instructions.

These computer programs (also referred to as programs, software, software applications, or code) may be implemented in high-level procedural and/or object-oriented programming languages and/or in assembly language, machine language, and/or the like.

Components of different embodiments described in this disclosure may be combined to form other embodiments not specifically described above. Components may be excluded from processes, computer programs, databases, etc. described in embodiments of the present disclosure without affecting their function and/or operation.

Further, the logical flows depicted in the figures do not require any particular order shown or sequential order to achieve desirable results. It should be understood that the order of steps or order for performing a particular action is immaterial so long as the embodiments described in this disclosure remain operable. Also, two or more steps or actions may be performed concurrently.

Although the embodiments described in this disclosure have been shown and described in connection with specific embodiments, those skilled in the art can take many forms and details without departing from the spirit and scope of the claims as defined by the appended claims. You need to understand that change can happen.

Claims (13)

As a method for predicting the course of a typhoon in the mid-term,
(1) extracting data of past typhoons that satisfy a preset criterion including initial characteristics of a target typhoon whose course is to be predicted from among data of past typhoons that have historically occurred during a predetermined past period;
(2) clustering the paths of past typhoons into a plurality of path patterns from the extracted data of past typhoons by using a clustering algorithm;
(3) calculating daily average synthesized observation background fields of predetermined predictors associated with each of the course patterns for days 6 to 8 days after the analysis time;
(4) calculating daily average model background fields of the predictors from one or more epidemiological models for days after the 6th to 8th days;
(5) generating a weighting factor for each course pattern by calculating a pattern correlation between the observation background fields and the model background fields of the predictors; and
(6) Generating a relative frequency map using the paths of past typhoons and a weighting factor for each path pattern, and obtaining a final predicted path based on the relative frequency map, mid-term prediction of the path of typhoons way to do it. The method of claim 1, wherein in step (1),
The data of the past typhoons includes data of the central position of typhoons and central air pressure data of 6-hour intervals among the course data of the past typhoons,
The initial feature of the target typhoon includes a predicted location from the date of occurrence of the target typhoon to the following 5 days provided by the Korea Meteorological Administration. The method of claim 1, wherein past typhoons that satisfy the preset criteria in step (1),
(i) of the sequential array of six extraction circles centered on the position of the target typhoon at the analysis time and the positions of the target typhoon on each forecast day for the following 5 days, entering the first extraction source; passing through each of the six extraction sources while leaving the final extraction source;
(ii) has a lifespan of 6 days or more;
(iii) A method for predicting the course of typhoons in the mid-term, which are past typhoons that occurred within a period of three months before and after the target typhoon's occurrence date. The method of claim 3, wherein in step (2),
The clustering algorithm is a fuzzy c-means clustering algorithm,
The method for predicting the course of a typhoon in the middle term, wherein the extracted data of past typhoons includes only data of the past typhoons after the time when they appear in the final extraction source. The method of claim 1, wherein in step (3),
The predetermined predictors are a 500 hPa position altitude, a directing east-west wind component, and a directing north-south wind component,
The daily average synthesized observation background is calculated from data of 0.75 ° Х0.75 ° horizontal resolution at 6-hour intervals provided from ERA-Interim of ECMWF, a method for predicting the typhoon course in the medium term. According to claim 5,
Method for predicting the course of a typhoon in the medium term, wherein the directing east west wind component and the directing north south wind component are synthesized as daily averages on barometric altitudes having different directing layer depths according to the central air pressure among the data of the past typhoons. The method of claim 1, wherein in step (4),
The one or more dynamic models include CFSv2, GEFS and GFS dynamic models provided by NCEP;
The model background fields are grid-adjusted to have the same horizontal grid as the observation background fields. The method of claim 1, wherein in step (5),
The pattern correlation is calculated using only the values of the predictors within a region in which significant differences exist between the observation background fields and the course patterns in the background fields,
The above significant differences are statistically significant areas at the 95% confidence level identified through one-way ANOVA, a method for predicting the typhoon course in the medium term. The method of claim 1, wherein the step (5),
(5-1) normalizing the predictors of the observation background fields and the model background fields on each day after the 6 days to the 8 days for each case of the past typhoons;
(5-2) calculating a pattern correlation coefficient between the normalized observation background fields and the model background fields;
(5-3) calculating an average correlation coefficient for each member of each course pattern on each day from the 6th to the 8th day by using the membership coefficient of each of the past typhoons as a weight; and
(5-4) Calculating a weighting factor for each course pattern by 3-day averages of member average correlation coefficients for each course pattern on each day after the 6th to 8th days, mid-term typhoon course How to predict. The method of claim 1, wherein the step (6),
(6-1) calculating path frequencies by latticeizing the paths of the past typhoons belonging to each of the path patterns;
(6-2) calculating a specific relative frequency map and a total relative frequency map by multiplying the track frequencies calculated for each grid by a weighting factor for each track pattern;
(6-3) determining a representative curve for each course pattern based on the specific relative frequency map; and
(6-4) a method for predicting a typhoon course in the middle term, including a step of determining a final predicted course based on the representative curve for each course pattern and the total relative frequency map. The method of claim 10, wherein in the step (6-3),
Starting from the location of each of the past typhoons at the analysis time on the specific relative frequency map for each course pattern, the longitude of the grid with the maximum value in each latitude grid box, and the maximum value in each longitude grid box Points corresponding to the latitude of the grid with are sequentially detected until the maximum value in the latitude grid box or the longitude grid box is less than 0.5,
By connecting the moving averaged points for the five sequentially detected points on the specific relative frequency map for each path pattern, a representative curve for each path pattern is determined. The method of claim 10, wherein in the step (6-4),
Regarding the lattice point index (i) on the representative curve for each course pattern, m represents the dynamic model used, c represents the pathway pattern, and L (m, c) is related to the dynamic model m and the pathway pattern c Represents the length of the representative curve, f[m, x(i), y(i)] is the lattice point [x(i), y(i) ] represents the value in the total relative frequency map, and s (m, c) represents the score of the representative curve associated with the epidemiological model m and the career pattern c,

A method for predicting the typhoon course in the middle term, which determines the representative curve having the highest score (s (m, c)) of the representative curve defined as the final predicted course. A computing system comprising a processor and memory,
When the memory is executed by the processor, it causes the processor to:
(1) Among the data of past typhoons that have historically occurred during a predetermined past period, extract data of past typhoons that satisfy preset criteria including the initial characteristics of the target typhoon to predict the course,
(2) clustering the paths of past typhoons into a plurality of path patterns from the extracted data of past typhoons using a clustering algorithm;
(3) Calculate daily average synthesized observation backgrounds of predetermined predictors associated with each of the course patterns for days 6 to 8 days after the analysis time,
(4) Calculate daily average model background fields of the predictors from one or more epidemiological models for days after the 6th to 8th days,
(5) Calculate a pattern correlation between the observation background fields and the model background fields of the predictors to generate a weighting factor for each course pattern;
(6) To generate a relative frequency map using the paths of past typhoons and a weighting factor for each path pattern, and to obtain a final predicted path based on the relative frequency map, configured to store computer readable instructions. being, a computing system.

Images