KR102639424B1 - Technique to predict the frequency of typhoons landing in East Asia using a dynamic model and a statistical method - Google Patents

Abstract

The present invention relates to a method for predicting typhoon landing frequency, which is performed on a processor of a computing device, comprising: a) setting a forecast area and period, calculating the typhoon landing frequency during the set area and period, and b) observing. c) calculating a standard pattern by analyzing the statistical relationship between the predicted lower-level relative vorticity and the observed typhoon landing frequency; c) calculating a predicted pattern by analyzing the statistical relationship between the predicted lower-level relative vorticity and the observed typhoon landing frequency; Steps of selecting potential predictors, d) selecting optimal predictors that produce the best prediction results among potential predictors, e) constructing and verifying a prediction model using the optimal predictors, and , f) It may include the step of predicting future typhoon landing frequency by substituting future predictors into the prediction model.

Description

Method for predicting the frequency of typhoons landing in East Asia using a dynamic model and statistical method {Technique to predict the frequency of typhoons landing in East Asia using a dynamic model and a statistical method}

The present invention relates to a method for predicting the frequency of typhoon landings in East Asia, and more specifically, to a method for predicting by combining a dynamical model and statistical techniques.

A typhoon is a type of tropical cyclone that occurred in the tropical waters of the Northwestern Pacific Ocean. The Korea Meteorological Administration and the Japan Meteorological Administration classify tropical cyclones with a maximum wind speed of 17.2 m/s or more near the center as typhoons.

Recently, strong super typhoons have occurred frequently, such as Haiyan, which struck the Philippines in 2013, and Meranti, which hit Taiwan in 2016, causing damage to life and property in East Asian countries.

As society develops, the scale of damage increases exponentially, so the development of reliable typhoon landing prediction technology is necessary to minimize such damage.

However, because the process of typhoon occurrence and development is very complex, it is not easy to predict it. The first forecasting method attempted was to use statistical relationships between typhoon activity and atmospheric and ocean conditions prior to the period when typhoons occur frequently.

These statistical methods were developed under the premise that existing relationships will continue in the future. However, the climate continuously moves due to natural climate fluctuations such as interannual and decadal fluctuations and artificial climate changes caused by humans.

Therefore, it is difficult to trust the prediction results because statistical methods have errors from the beginning.

Afterwards, as the Coupled General Circulation Model (CGCM) was developed, typhoon activity was attempted to be predicted through a dynamic climate model that can reflect climate movements.

However, there were still limitations in predicting typhoons smaller than the spatial resolution of dynamical climate models.

Despite these limitations, CGCM can significantly predict large-scale environmental variables such as vertical shear and low-level vorticity related to typhoon activity. Therefore, various prediction methods have been developed using statistical relationships between large-scale environmental variables produced by dynamic climate models and typhoon activity.

In this regard, a hybrid seasonal prediction method for typhoon activity in the Northwest Pacific is disclosed in Korean Patent No. 10-1646587.

This invention relates to a method of predicting the ACE (Accumulated tropical cyclone kinetic energy) typhoon activity index required to calculate the location and movement path of a typhoon. The predicted ACE typhoon activity index can be used to predict the location of typhoon occurrence and movement path.

The landing of a typhoon is the result of a complex effect of the typhoon's location and movement path. Therefore, it is difficult to practically predict a typhoon landing only with the typhoon occurrence location and movement path information predicted by the method disclosed in No. 10-1646587.

In addition, because it uses dynamic modeling forecast data from the seasonal forecast system used by the Korea Meteorological Administration, it has the disadvantage that it may be useless if the forecast of the forecast system fails.

The purpose of the present invention, taking into account the above problems, is to provide a method for predicting the landing frequency of typhoons without additional processes using the correlation between lower-level relative vorticity and typhoon landing frequency.

In addition, since the present invention can be applied to other prediction data in the future, another purpose is to enable stable prediction by expanding and establishing various prediction systems rather than just one prediction system.

The method for predicting the East Asia landing frequency of typhoons of the present invention to solve the above technical problems is a method performed on a processor of a computing device, which includes a) setting a forecast area and period, and determining the typhoon landing frequency during the set area and period. a step of calculating, b) a step of calculating a standard pattern by analyzing the statistical relationship between the observed lower-level relative vorticity and the observed typhoon landing frequency, and c) a statistical relationship between the predicted lower-level relative vorticity and the observed typhoon landing frequency. Steps of analyzing prediction patterns and selecting potential predictors, d) selecting optimal predictors that produce the best prediction results among potential predictors, and e) predicting using the optimal predictors. It may include the step of building and verifying the model, and f) predicting the future typhoon landing frequency by substituting future predictors into the prediction model.

The present invention combines a dynamic model and statistical techniques, and has the effect of immediately predicting the landing frequency of typhoons without complicated processes.

In addition, since the present invention can be applied to other forecast data, it has the effect of minimizing uncertainty in a single forecast system and enabling stable typhoon landing frequency prediction.

Figure 1 is a flowchart of a method for predicting the frequency of typhoon landings in East Asia according to a preferred embodiment of the present invention.
Figure 2 is a relationship schematic diagram for supplementary explanation of Figure 1.
Figure 3 is an example of a predicted landing area classification in the East Asia region.
Figure 4 is a graph showing the average monthly landing frequency of typhoons observed in each region from 1980 to 2018.
Figure 5 is an example diagram of a standard pattern.
Figure 6 is an example of a prediction pattern.
Figure 7 is a table listing potential predictors and optimal predictors.
Figure 8 is a graph of the observed typhoon landing frequency, predicted typhoon landing frequency, and cross-validation results during the forecast period.

In order to fully understand the configuration and effects of the present invention, preferred embodiments of the present invention will be described with reference to the attached drawings. However, the present invention is not limited to the embodiments disclosed below, but can be implemented in various forms and various changes can be made. However, the description of this embodiment is provided to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the present invention of the scope of the invention. In the attached drawings, components are shown enlarged in size for convenience of explanation, and the proportions of each component may be exaggerated or reduced.

Terms such as 'first' and 'second' may be used to describe various components, but the components should not be limited by the above terms. The above terms may be used only for the purpose of distinguishing one component from another. For example, the 'first component' may be named 'the second component' without departing from the scope of the present invention, and similarly, the 'second component' may also be named 'the first component'. You can. Additionally, singular expressions include plural expressions, unless the context clearly dictates otherwise. Unless otherwise defined, terms used in the embodiments of the present invention may be interpreted as meanings commonly known to those skilled in the art.

In particular, the present invention relates to a method for predicting the frequency of typhoon landings in East Asia, and this method is performed on a computing device that includes at least a processor, a data storage means, and a display.

In other words, even if there is no special explanation, the person performing each step constituting the present invention may be a processor.

Figure 1 is a flow chart of a method for predicting the landing frequency of typhoons in East Asia according to a preferred embodiment of the present invention, and Figure 2 is a relationship schematic diagram for supplementary explanation of Figure 1.

Referring to Figures 1 and 2, the method for predicting the frequency of typhoon landings in East Asia according to the present invention includes the steps of setting the prediction area and period (S11), calculating the observed typhoon landing frequency (S12), and determining the landing frequency of the typhoon that landed in East Asia. A step (S13) of determining the statistical relationship between the lower-layer relative vorticity of the typhoon and the observed landing frequency of the typhoon, and a step (S14) of determining the statistical relationship between the lower-layer relative vorticity of the predicted typhoon and the observed landing frequency of the typhoon, A step of selecting a statistically significant region as a potential predictor (S15), a step of selecting the optimal predictor among the potential predictors (S16), and building and verifying a statistical model using the optimal predictors. It may include a step (S17) and a step (S18) of predicting the future typhoon landing frequency as a future predictor.

Hereinafter, the specific configuration and operation of the method for predicting the frequency of typhoon landings in East Asia according to the present invention, configured as described above, will be described.

First, the present invention combines a dynamical model and statistical techniques, and the combination of a dynamical model and statistical techniques is intended to complement the shortcomings that occur when a dynamical model or statistical technique is used alone in predicting typhoon activity.

Using only statistical techniques is vulnerable to climate fluctuations, and the dynamical model used for global seasonal prediction has a spatial resolution of approximately 50km to 150km, which has limitations in simulating and predicting typhoons, which are relatively small-scale meteorological phenomena.

To overcome these limitations, the landing frequency of typhoons was predicted by combining statistical techniques based on the prediction results of the dynamical model.

As a term used in the present invention, the lower layer refers to 850hPa to 500hPa, and the atmospheric circulation in this layer is closely related to the typhoon path, so it is irrelevant which of these layers is selected.

In addition, observations are actual data on the frequency of typhoons or landfalls that have occurred in the past or present, and predictions can be understood as the results of predicting the frequency of typhoons or landfalls that have occurred in the past or present by combining meteorological phenomena.

In the present invention, the dynamical model proved the effectiveness of the present invention based on an example of predicting the typhoon landing frequency in 2019 based on 39 years of data from 1980 to 2018 of PNU CGCM, one of the dynamical climate models.

First, set the typhoon's landing prediction area and period as in step S11.

Figure 3 is an example of a predicted landing area classification in the East Asia region.

For convenience, the East Asian region can be divided into Southeast Asia (SEA), Middle East Asia (MEA), and Northeast Asia (NEA), and this regional classification is also used in the present invention.

Figure 4 is a graph showing the average monthly landing frequency of typhoons observed in each region from 1980 to 2018.

Generally, in the western North Pacific, typhoons land on or affect land in various ways.

In the present invention, lower-level relative vorticity, which is closely related to the typhoon path, is used as a predictor to predict the landing frequency. Therefore, the entire area of East Asia is subdivided into three areas according to the typhoon path.

In other words, typhoons passing through one segmented area are all affected by similar lower-level atmospheric circulation.

Specific latitude and longitude information for the subdivided areas is as follows.

The Northeast Asia (NEA) region, including the Korean Peninsula and Japan, is 30~40°N 124~143°E, and the Middle East Asia (MEA) region includes Taiwan, Fujian, Zhejiang, Jiangsu, and Shanghai. ) area is 22~40°N 117~124°E, and the Southeast Asia (SEA) area including South China, Vietnam, and the Philippines is 5.7~23.5°N 117~127°E.

And the area that includes all three areas above was defined as East Asia (EEA), as shown in Figure 4.

In addition, when a typhoon passed through a segmented area, it was considered to have landed in that area. This takes into account the fact that even if a typhoon passes through the nearby sea without actually landing on land, it may actually suffer damage from the effects of the typhoon.

Meanwhile, the forecast period can be set based on the monthly average typhoon landing frequency in Figure 4. Forecasts can be made more effectively if the forecast period includes the period when typhoons frequently land in each forecast area.

Therefore, the forecast period is July to September, when more than 70% of the total annual typhoons land in the NEA, MEA, and EEA, and July to November, when about 72% of the total annual typhoons land in the SEA.

In other words, the forecast period can be applied differentially depending on the area, or it can be set from July to September, which is where the majority belong.

Next, calculate the observed typhoon landing frequency as in step S12.

The observed values of the lower relative vorticity are averaged over the prediction period and stored in the lower relative vorticity storage unit 21 of FIG. 2.

For example, when the forecast area is NEA, 39 years of lower relative vorticity data averaged from July to September every year are stored in the lower relative vorticity storage unit 21.

Similarly, the prediction data of the prediction model to be used is also stored in the storage unit 22 with the lower layer counterpart in the same manner.

In addition, the typhoon landing frequency is calculated for each region according to the set forecast area and period and stored in the typhoon landing frequency storage unit 23. For example, NEA's typhoon landing frequency is the number of typhoons that passed through the NEA area from July to September every year.

Next, as in step S13, the statistical relationship between the lower layer relative vorticity of the landed typhoon and the observed landing frequency of the typhoon is identified.

In other words, it spatially represents the correlation coefficient between the lower-layer relative vorticity of the lower-layer relative vorticity storage unit 21 and the typhoon landing frequency of the typhoon landing frequency storage unit 23. The spatial pattern calculated here is referred to as a standard pattern and is stored in the standard pattern storage unit 24.

Here, the correlation coefficient is the anomaly correlation coefficient (ACC), which is defined by Equation 1 below.

과

는 각각의 평균값이고, N은 예측년도 개수이다.ACC is calculated for each grid point, and V is the average value (i.e., the value of the lower relative vorticity storage unit 21) during the prediction period (July to September or July to November) of the lower relative vorticity observed in the corresponding year. , T is the total typhoon landing frequency observed during the forecast period of the corresponding year (i.e., the value of the typhoon landing frequency storage unit 23),

class

is each average value, and N is the number of forecast years.

The standard pattern calculated in this way shows the correlation between time-dependent deviations from the average value of the two observation data from 1980 to 2018. A value close to 1 indicates a positive correlation between the two variables, a value close to -1 indicates a negative correlation, and a value close to 0 indicates a low correlation between the two variables.

The center of the vorticity that shows a statistically significant negative (positive) correlation in the standard pattern is indicated as N(P). When negative and positive vorticity occurs at the points where N and P are located, typhoons occur more frequently or the forward current becomes stronger and typhoon landings become more frequent.

In other words, they represent relative vorticity, which plays an important role in predicting the landing frequency of typhoons, and are therefore used as indicators to select future predictors.

Figure 5 is an example of a standard pattern in which step S13 is performed for each configurable prediction region using 700hPa relative vorticity among the NCEP-DOEⅡ reanalysis data provided by the National Oceanic and Atmospheric Administration (NOAA) as observation data. In other words, it is a spatial representation of the correlation coefficient between the (a) NEA, (b) MEA, (c) SEA, and (d) EEA landing frequencies of typhoons observed during 1980-2018 and the observed 700hPa relative vorticity. The white solid line shows statistical significance at the 95%, 98%, and 99% confidence intervals, and N and P are indicated at the center of the opponent, which plays an important role.

The standard pattern of NEA consists of N in East China and the Korean Peninsula and P in South and East China and the tropical Pacific (Figure 5a). The standard pattern of MEA consists of the N of East China-Korean Peninsula-Japan, which extends in the east-west direction, and the P of the Indochina Peninsula-South China-East China Sea (Figure 5b), and the standard pattern of SEA is the N of South and East China, and the P of Indochina. It consists of the peninsula-Philippine Sea and the P of the Korean Peninsula (Figure 5c). Finally, the standard pattern of the EEA consists of N over the East China-Korean Peninsula and near the equator, and P over the Indochina Peninsula and the Philippine Sea (Figure 5d). This arrangement of N and P shows the unique environment associated with typhoon landfall in each region, and appears similarly even when using other observational data. At this time, the judgment of similarity can be limited to cases where there is a match of 90% or more.

Next, as in step S14, the statistical relationship between the predicted lower-level relative vorticity of the typhoon and the observed typhoon landing frequency is identified.

The spatial pattern of ACC is obtained for the prediction value stored in the lower layer counterpart storage unit 22, is referred to as a prediction pattern, and is stored in the prediction pattern storage unit 25.

Here, ACC uses the predicted value of the epidemiological model rather than the observed value for V in the correlation coefficient calculation formula in step S13. An example of a prediction pattern using PNU CGCM data is shown in Figure 6.

When comparing the arrangement of N and P in the standard pattern (Figure 5) and the predicted pattern (Figure 6), if they are similar and one or more of N and P in the predicted pattern (Figure 6) are statistically significant, the currently used epidemiological model This means that the typhoon landing frequency can be predicted by combining statistical techniques.

Next, in step S15, statistically significant regions are selected as potential predictors.

Here, the standard for statistical significance is the 95% confidence interval.

The relative vorticity of the N or P region, which is statistically significant in the prediction pattern (FIG. 6), is selected as a potential predictor, and the average value of the region is stored in the potential predictor storage unit 26. The location of statistically significant major vorticity centers may vary slightly depending on the prediction model used, lead time, and layer of relative vorticity.

Figure 7 is a table listing potential predictors and optimal predictors of the present invention using PNU CGCM.

As shown in Figure 7, for NEA, x1 (27-32˚N, 109-115˚E), x2 (10-15˚N, 138-145˚E), and for MEA, x1 (21-32˚N) , 110-115˚E), x2(49-54˚N, 140-150˚E), for SEA x1(38-43˚N, 110-118˚E), x2(13-18˚N, 96 -104˚E), in the case of EEA, the relative vorticity in the x1 (10-18˚N, 93-98˚E) and x2 (2-6˚N, 102-109˚E) areas fits the above conditions, making potential predictions was selected as a factor.

The corresponding area is indicated by a black box in Figure 6, and the black dotted line indicates statistical significance. Additionally, a relative vorticity of 500 hPa was used for NEA, SEA, and EEA, and a relative vorticity of 700 hPa was used for MEA.

Next, in step S16, the optimal predictor is selected among the potential predictors.

In this way, by finding the optimal predictor configuration, uncertainty can be removed and predictability can be effectively increased.

As a specific example of selecting the optimal predictor, the regression equation is repeatedly obtained through regression analysis while changing the predictor composition, and the results are compared.

At this time, the optimal regression equation is when the root mean squared errors (RMSE) is minimized and the coefficient of determination and F value are maximized.

Root mean square deviation (RMSE) is a measure used when dealing with the difference between the model predicted value and the observed value and expresses precision, and its definition is as in Equation 2 below.

In Equation 2 above, n means the total year, yi is the actual observed landing frequency of typhoons in the ith year, means the predicted typhoon landing frequency when substituting the ith predictor (xi) in the regression equation.

In addition, the F value is a value for comparing whether y (observed landing frequency of typhoons) is further explained by x (predictor, i.e. relative vorticity) and error, and is expressed in Equation 3 below.

In Equation 3 above, total sum of squares (SST), regression sum of squares (SSR), and error sum of squares (SSE) are each defined as Equation 4 below.

In equation 4 above, means the average landing frequency of observed typhoons over n years.

In addition, the coefficient of determination is the amount explained by x (predictor, i.e. relative vorticity) out of the total variation in y (observed typhoon landing frequency), which means the explanatory power of the regression model for the sample. The larger the value of the coefficient of determination (closer to 1), the more appropriate the model can be evaluated.

The coefficient of determination is defined by Equation 5 below.

The predictors finally selected through this process are stored in the predictor storage unit 27.

In this example, two potential predictors were selected for each prediction region, so it was performed for two cases in which x1 and x2 were used alone as predictors and one case in which they were used together.

The comparison results are shown in Figure 7, and the final selected cases are indicated in bold. The optimal regression equation was found when x1 and x2 were used together for NEA and SEA, when x2 was used alone for MEA, and when x1 was used alone for EEA.

Next, as in step S17, a statistical model is built and verified using the optimal predictors.

The optimal regression equation for each prediction region is stored in the statistical model prediction unit 28. However, since this is an optimal regression equation obtained from fixed data, it cannot be free from the overfitting problem.

The overfitting problem refers to a phenomenon in machine learning where training data is over-trained, resulting in good results on the training data but large errors on the actual data.

In other words, this means that although the typhoon landing frequency from 1980 to 2018, which was learned in this example, is well predicted, good prediction may not be shown for 2019, which was not learned.

Therefore, it is necessary to evaluate the predictability of untrained data, and if the predictability of the regression equation is high even though it has not been learned, the reliability of the predicted value increases.

For this purpose, leave-one-out cross-validation, a type of cross-validation, is used.

In this method, when calculating the typhoon landing frequency for each year, a regression equation is obtained using the predictors of the remaining years excluding the corresponding year, and then the landing frequency is predicted by substituting the predictors of the corresponding year.

Figure 8 shows examples of observed typhoon landing frequency (blue) and predicted typhoon landing frequency (yellow) and cross-validation results (orange) in the forecast period from 1980 to 2018. SCOR in the upper left is the correlation coefficient between observed values and predicted values, and FCOR in the upper right is the correlation coefficient between observed values and cross-validation results. And the * and ** marks on the correlation coefficient mean that it is significant at the 95% and 99% confidence levels, respectively.

The prediction results of the model developed in this example are statistically significant at the 99% confidence level for all prediction regions. The cross-validation results also show that for NEA and SEA, the correlation coefficient with observed values is significant at the 99% confidence level, and for MEA and EEA, it is also significant at the 95% confidence level. This suggests that the prediction value of the model presented in the present invention is stable.

The criterion for passing the verification of the present invention is when both SCOR and FCOR are statistically significant at a confidence level of 95% or higher.

Next, as in step S18, predict the future typhoon landing frequency using future predictors.

In other words, the future typhoon landing frequency is predicted by substituting the future predictors calculated from the future prediction data into the typhoon landing frequency prediction unit 28.

Specifically, this is done by calculating the optimal forecasting factors from the 2019 forecast data of PNU CGCM and substituting them into the typhoon landing frequency prediction unit (28).

Although embodiments according to the present invention have been described above, they are merely illustrative, and those skilled in the art will understand that various modifications and equivalent scope of embodiments are possible therefrom. Therefore, the true technical protection scope of the present invention should be determined by the following claims.

21: Lower level relative storage unit (observation)
22: Low-level counterpart and storage unit (prediction)
23: Typhoon landing frequency storage unit
24: Standard pattern storage unit
25: Prediction pattern storage unit
26: Potential predictor storage unit
27: Predictor storage unit
28: Typhoon landing frequency prediction department

Claims (7)

1. A method performed on a processor of a computing device, comprising:
a) Setting the forecast area and period and calculating the typhoon landing frequency during the set area and period;
b) calculating a standard pattern by analyzing the statistical relationship between the observed lower-level relative vorticity and the observed typhoon landing frequency;
c) calculating a prediction pattern and selecting potential predictors by analyzing the statistical relationship between the predicted lower-level relative vorticity and the observed typhoon landing frequency;
d) selecting the optimal predictor that minimizes the root mean squared errors (RMSE) and maximizes the coefficient of determination and F value when the regression equation is repeatedly obtained while changing the configuration of the predictors;
e) building and validating a prediction model using optimal predictors; and
f) Including the step of predicting future typhoon landing frequency by substituting future predictors into a prediction model,
The conditions for selecting potential predictors in step c) above are:
When the arrangement of negative relative vorticity and positive relative vorticity in the predicted pattern is more than 90% similar to the arrangement of negative relative vorticity and positive relative vorticity in the standard pattern, the negative relative vorticity and positive relative vorticity in the predicted pattern are A method for predicting the landing frequency of typhoons, characterized by selecting areas with high correlation for the 95% confidence interval among vorticity. According to paragraph 1,
The typhoon landing frequency in stage a) is,
A method for predicting the landing frequency of typhoons, characterized in that the total frequency of passage of the center of the typhoon during the forecast period each year for the set forecast area range. According to paragraph 1,
The correlation of step b) is,
A method for predicting the landing frequency of typhoons, characterized in that it is calculated using the anomalous correlation coefficient method expressed in Equation 1 below.
[Equation 1]

ACC is the correlation calculated for each grid point, V is the average value during the prediction period of the lower-level relative vorticity observed in the corresponding year, T is the total typhoon landing frequency observed during the prediction period in the corresponding year, class is the average value of each, N is the number of forecast years delete delete According to paragraph 1,
The statistical model construction and verification in step e) above is,
Each is performed through regression analysis and leave-one-out cross-validation, and the leave-one-out cross-validation uses predictors for years other than the relevant year when calculating the typhoon landing frequency for each year. A method for predicting the landing frequency of typhoons, which is characterized by obtaining a regression equation and then predicting the landing frequency by substituting the predictors of the corresponding year. According to clause 6,
The passing criteria for the above cross-verification are,
A method for predicting the landing frequency of typhoons, characterized in that when the correlation coefficient between the predicted value and the cross-validation result value is calculated for the observed value, both are statistically significant for the 95% confidence interval.

Images