Cold wave extension period forecasting method and system based on NAO index

Abstract

The invention discloses a cold wave extension period forecasting method and a cold wave extension period forecasting system based on an NAO (National Atlantic Oscillation) index, which are used for researching an indication effect of a low-frequency signal of North Atlantic Oscillation (NAO) in 30-60 days in a cold air process in China and a change of the low-frequency signal on a cold air event; the phase and the index change threshold value of the AO-NAO index 10-30 days before the cold air event occurs are established, and probability forecasting can be performed on national cold air events 10-30 days in advance. And further, for national cold waves with relatively high forecasting difficulty, independent modeling is carried out by utilizing a self-organizing neural network method and a single-particle Lagrange trajectory model method, an NAO index corresponding to the national cold waves is established, the forecasting hit rate of the national cold waves is increased, and information such as cold wave invasion positions and paths can be provided.

Description

Technical Field

[0001] The present invention relates to the technical field of weather forecasting, and in particular to a method and system for forecasting the extended period of a cold wave based on the NAO index. Background Art

[0002] Cold waves are one of the most important disastrous weather in China in winter and spring. The process in which cold air activities in the north reach a certain intensity and invade southward on a large scale is called a cold wave. Cold waves generally cause drastic cooling and strong winds, sometimes accompanied by a series of weather phenomena such as rain, snow, freezing rain, and frost. During the invasion of cold waves, they often bring great impacts and losses to various aspects such as transportation, agriculture, forestry, electricity, and communications. In addition, cold waves will also directly affect human and animal health. Strong cold air activities pose a great threat to different groups of people, especially patients with weather-sensitive diseases, and are very likely to cause respiratory diseases. They will also freeze livestock to death and cause great economic losses. Against the background of intensified global climate change in the 21st century, cold waves have produced new changing characteristics. From the perspective of climate change, although the fifth assessment report of the IPCC pointed out that the world has shown warming characteristics in the past 130 years, and many studies in recent years have found that the frequency of cold air activities in my country has shown a trend of interdecadal reduction. Although the overall frequency of cold air and cold wave activities affecting my country has shown a decreasing trend, the new situation of cold air in the context of global warming has shown more complex characteristics and mechanisms, whether from the perspective of frequent extreme events or responding to climate change. Therefore, it is of great practical and scientific significance to conduct research on the extended period forecast of cold waves with an extended period of 10-30 days and establish a forecast index based on the dominant factors of atmospheric circulation changes.

[0003] Numerous previous studies have shown that many long-life atmospheric phenomena have significant impacts on weather-scale systems, such as the intensity and position of planetary-scale Rossby waves, the North Atlantic Oscillation (NAO), the Arctic Oscillation (AO), etc. When some specific atmospheric circulation patterns appear, or when some abnormal circulation situations are obvious and persistent, extended-period weather processes tend to have better predictability. From the perspective of influencing factors on this time scale, the role of AO and NAO is very important. AO and NAO are significant modes of low-frequency variability in the extra-equatorial atmosphere with periodic signals on the intraseasonal time scale in the Northern Hemisphere, and are also the dominant modes of variability from intraseasonal to decadal winter half-year. The AO is characterized by opposite fluctuations in air pressure in the Arctic Circle and mid-latitudes, accompanied by opposite fluctuations in the intensity of subpolar and subtropical westerlies. NAO is a localized representation of AO in the North Atlantic region, and is essentially no different from AO. AO and NAO have a profound impact on winter weather in the United States, Europe, and Asia, and the impact increases with the altitude from the troposphere to the lower stratosphere. Their most significant contribution is in the cold season from October to March of the following year, especially in the eastern half of America, the North Atlantic region, Eurasia, and the Arctic cap region. Therefore, for the extended-range forecast of cold air or cold wave weather processes in the cold season, starting with the AO and NAO indices has clear scientific connotations and practical feasibility.

[0004] Limited by the development of atmospheric science and the level of understanding of the atmosphere, the extended-term weather forecast theory and technology are still a global problem. At present, the short-term forecast and near-term forecast technology for cold waves in actual business mainly rely on numerical forecasts and ensemble forecast products. The medium- and short-term forecast technology is relatively mature and the forecast effect is good. The common extended-term forecast duration in meteorological business is 11-30 days. The forecast of 10-30 days extended period has many difficulties and challenges in both theoretical research and actual business practice, and is also the weakest link in the entire forecast system. In particular, the extended-term forecast for cold waves in China is still imperfect and immature, mainly manifested in the low forecast accuracy and the relatively single forecast information provided. It can only judge whether a cold wave occurs or is a cold wave day within the forecast period, and the spatial characteristics and intensity of the cold wave are not reflected. The AO and NAO, which have an important impact on the weather and climate in Asia and Europe in the northern hemisphere in winter, are not only closely related to my country's cold wave weather, but also have significant low-frequency signal characteristics. From the previous research review, it can be found that NAO has a significant impact on the cold wave process, but the research on using AO and NAO indexes to make extended-term forecasts for my country's cold waves is still relatively scarce. In view of the above problems, the present invention solves the following main problems:

[0005] 1. How to use the AO-NAO index to establish a cold wave extended period forecast model.

[0006] 2. For nationwide cold waves that are more difficult to forecast, how to improve the forecast model and increase the accuracy of the forecast based on the establishment of a cold wave extended period forecast model. Summary of the invention

[0007] In view of the problems existing in the prior art, the present invention aims to provide a method and system for forecasting the extended period of cold waves based on the NAO index.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] A method for forecasting an extended period of a cold wave based on the NAO index, the method comprising the following steps:

[0010] S1 automatically obtains daily ERA-5 historical data;

[0011] S2 obtains daily ERA-5 historical data according to step S1, counts the low-frequency characteristics of the cold air process for 30-60 days, and obtains the phase transition characteristics of AO before the cold wave event breaks out;

[0012] S3 statistically analyzes the characteristics of AO and NAO indices before the occurrence of cold air based on daily ERA-5 historical data, and quantitatively describes the evolution of AO-NAO indices before the occurrence of cold air events;

[0013] S4 Establish a model for extended-range forecast of cold air events using the AO-NAO index;

[0014] S5 improves the prediction model of step S4 using a self-organizing map neural network SOM method;

[0015] After improving the forecast model according to the SOM method, S6 uses the hybrid single-particle Lagrangian integrated trajectory model to track the backward trajectory of the cold air mass, and finally obtains the model's prediction results for the cold air invasion path.

[0016] It should be noted that, in step S2 of the present invention, the Butterworth bandpass filtering method is used to retain the 30-60 day low-frequency signal on the 500 hPa isobaric surface in the daily ERA-5 historical data, so as to obtain the phase transition characteristics of AO before the outbreak of the cold wave event.

[0017] It should be noted that the quantitative description of the evolution law of the AO-NAO index before the cold air event in step S3 of the present invention includes:

[0018] Using the sea level pressure data from the daily ERA-5 historical data, the AO and NAO index characteristics were calculated and statistically analyzed 60 days before each cold air event from 1951 to 2014. The evolution of the AO-NAO index before a cold air event was quantitatively described. It was found that 10-30 days before a cold air event, the AO-NAO phase would change from "+" to "-" or the index would decrease continuously.

[0019] It should be noted that, in establishing the model described in step S4 of the present invention, it includes determining the index drop threshold before the cold air event, determining the period of occurrence of cold air after the threshold signal appears, and determining the intensity of the cold air event.

[0020] It should be noted that step S4 of the present invention also includes:

[0021] Using the daily AO-NAO index and the national cold air event dataset from 1951 to 2014 (a total of 805 cases), the continuous decline day threshold of the AO-NAO phase transition and the index decline amplitude threshold 10-30 days before the outbreak of the cold air event were counted to determine the representativeness of the index signal; and the daily AO-NAO index and the national cold air event dataset from 2014 to 2019 (a total of 69 cases) were used for verification, and the hit rate, missed reporting rate and false alarm rate were quantitatively calculated.

[0022] It should be noted that step S4 of the present invention also includes:

[0023] The daily sea level pressure data is obtained every day, and the NAO index is calculated using the formula built into the model. The NAO index is monitored daily and a forecast signal is generated when any of the following conditions are met:

[0024] A has a downward trend for 5 consecutive days; or

[0025] B has been decreasing for 3 consecutive days and the cumulative decrease value exceeds 1.0; or

[0026] C has been negative for 10 consecutive days.

[0027] It should be noted that the step S5 of the present invention includes:

[0028] The SOM method is used to train and cluster the 500hPa height field situation in the period of 15-30 days before the occurrence of a nationwide cold wave event. First, the 500hPa height field data of the extended forecast period of 36 nationwide cold wave events from 1961 to 2009 are sorted out, converted into one-dimensional data and input into the SOM neural network. The minimum competition layer latitude, the initial radius of the winning neighborhood, the learning rate and the maximum number of iterations are obtained according to the data volume and the model empirical formula. After these parameters are input into the neural network, the height field data is iteratively trained, and the two-dimensional mapping distribution of the winning neuron is estimated through the U-Matrix weight matrix to preliminarily judge whether its training clustering results are stable and effective. After adjusting the parameters and training for many times, the winning neuron with stable distribution is obtained, and its training results are output, and its two-dimensional data structure is restored, so that the weather situation at 500hPa height after clustering can be obtained.

[0029] It should be noted that the step S6 includes:

[0030] Based on the five weather situation models at 500hPa obtained by the SOM clustering method, combined with the East Asian classic cold wave model, the situation distribution in the obtained modes was analyzed and two key areas were identified:

[0031] Key area I: 40°W-20°E, 45°N-65°N; Key area II: 60°E-130°E, 45°N-80°N;

[0032] Like the calculation method of AO and NAO index, the calculation method of key area index is defined:

[0033]

[0034] Among them, represents the latitudinal average of the normalized sea level pressure anomaly at the southern boundary of the key area, represents the latitudinal average of the normalized sea level pressure anomaly at the northern boundary of the key area, and the key area index I is the difference between the two;

[0035] Then construct index Index1 and index lndex2 corresponding to key area I and key area II respectively;

[0036] The index lndex1 or index lndex2 generates a forecast signal if any of the following conditions are met:

[0037] A has a downward trend for 5 consecutive days; or

[0038] B has been decreasing for 3 consecutive days and the cumulative decrease value exceeds 1.0; or

[0039] C has been negative for 10 consecutive days.

[0040] Based on the cold wave extended period forecasting method of the NAO index of the present invention, the present invention also provides a system using the cold wave extended period forecasting method based on the NAO index. Further, the system of the present invention comprises:

[0041] The data acquisition module is used to obtain daily historical data;

[0042] A verification index module is used to analyze the low-frequency characteristics of the process based on the daily historical data and verify the feasibility of using the AO index to perform extended cold air forecasting;

[0043] The forecast model building module uses the NAO index to preliminarily build a forecast model;

[0044] The forecast model improvement module uses the SOM method to independently model the forecast key area, define new indices, and improve the forecast model;

[0045] The cold wave extended period forecast module is used to make a daily extended period forecast of the cold wave based on the final forecast model and the real-time daily sea level pressure data.

[0046] The beneficial effects of the present invention are:

[0047] 1. A cold wave extended-term forecast system based on the AO and NAO indices has been established, which can make probabilistic forecasts of nationwide cold air events 10-30 days in advance.

[0048] Based on the 30-60 day low-frequency signals of the Arctic Oscillation (AO) and the North Atlantic Oscillation (NAO) during cold air processes in my country and the indicative effect of their changes on cold air events, the phase and index change threshold of the AO-NAO index 10-30 days before the occurrence of cold air events are established, and a probabilistic forecast of nationwide cold air events can be made 10-30 days in advance.

[0049] 2. In response to nationwide cold waves that are difficult to predict, the NAO index corresponding to the nationwide cold waves was established to provide information such as the location and path of the cold wave invasion.

[0050] 3. Independent modeling was carried out using methods such as self-organizing neural networks and single-particle Lagrangian trajectory models, and the NAO index corresponding to the nationwide cold wave was established, which improved the forecast hit rate of the nationwide cold wave and provided information such as the location and path of the cold wave invasion. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The low-frequency signal characteristics of the typical cold wave event 30-60 days before the outbreak in early 2008 were obtained by the Butterworth bandpass filtering method in the present invention;

[0052] Figure 2The low-frequency signal characteristics of the typical cold wave event 30-60 days before the outbreak in early 2016 were obtained by the Butterworth bandpass filtering method in the present invention;

[0053] Figure 3 The present invention is based on the time series graph of AO and NAO indexes in the winter of 2007 and the winter of 2015;

[0054] Figure 4 A modeling flow chart of the present invention using a self-organizing neural network (SOM) to train the 500h Pa height field in the extended period forecast period of a nationwide cold wave;

[0055] Figure 5 is the U-Matrix weight matrix of the SOM model of the present invention;

[0056] Figure 6 These are the five main modes of 500hPa height field obtained by the present invention using SOM training;

[0057] Figure 7 For the present invention to test the actual case of cold wave;

[0058] Figure 8 The present invention traces the trajectory of the cold air mass in an actual case of a cold wave.

[0059] Fig. 9 This is a flow chart of the cold wave extended period forecasting method based on the NAO index of the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further described below in conjunction with the accompanying drawings. It should be noted that this embodiment is based on the technical solution and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to this embodiment.

[0061] like Figure 1 As shown, the present invention is a method for forecasting the extended period of cold waves based on the NAO index, and the method comprises the following steps:

[0062] S1 automatically obtains daily ERA-5 historical data;

[0063] S2 obtains daily ERA-5 historical data according to step S1, counts the low-frequency characteristics of the cold air process for 30-60 days, and obtains the phase transition characteristics of AO before the cold wave event breaks out;

[0064] S 3 Statistically analyze the characteristics of AO and NAO indices before the occurrence of cold air based on daily ERA-5 historical data, and quantitatively describe the evolution of AO-NAO indices before the occurrence of cold air events;

[0065] S4 Establish a model for extended-range forecast of cold air events using the AO-NAO index;

[0066] S5 improves the prediction model of step S4 using a self-organizing map neural network SOM method;

[0067] After improving the forecast model according to the SOM method, S6 used the hybrid single-particle Lagrangian integrated trajectory model to track the backward trajectory of the cold air mass, and finally obtained the model's prediction result for the cold air invasion path.

[0068] Example

[0069] First, if Figure 1 , Figure 2 As shown in the figure, taking two typical cold wave events in early 2008 and early 2016 as examples, the Butterworth bandpass filtering method is selected to retain the 30-60 days low-frequency signal on the 500hPa isobaric surface to study the characteristics of the low-frequency signal 30-60 days before the outbreak of the cold air event. The 500hPa geopotential height anomaly in Eurasia showed an AO positive phase in late December 2007 (2015), and an AO negative phase in early January 2008 (2016). Before the outbreak of the cold wave event, the AO phase experienced a transition from "+" to "-", and the 30-60 days low-frequency oscillation can well reflect the transition of the AO phase, indicating that the AO has a 30-60 days low-frequency signal during the occurrence of cold air, which further verifies the feasibility of using the AO index for extended cold air forecasting.

[0070] As Li et al. (2002) defined the AO and NAO indices, the AO formula is as follows:

[0071]

[0072] here and They are the daily average values ​​of the Shanghai surface pressure field at the 35°N and 65°N latitudes in the Northern Hemisphere respectively.

[0073] The NAO index is defined as follows:

[0074]

[0075] here and They are the daily average values ​​of the sea level pressure field in the longitude range 80°W-30°E on the 35°N and 65°N latitudes, respectively.

[0076] P is the normalized value of the daily sea level pressure field, and the formula is as follows:

[0077]

[0078] Here p′ is the anomaly from the 1981-2010 climate state, S p is the standard deviation of the daily data of sea level pressure field from 1981 to 2010.

[0079] like Figure 3 As shown in the figure, the AO and NAO index characteristics 60 days before each cold air event from 1951 to 2014 were calculated and statistically analyzed using the reanalysis of daily sea level pressure data, and the evolution of the AO-NAO index before the cold air event was quantitatively described. It was found that 10-30 days before the cold air event, the AO-NAO phase would change from "+" to "-" or the index would continue to decline. Taking the low temperature rain, snow and freezing event in 2008 and the super cold wave process in January 2016 as examples, Figure 3 As shown, the AO-NAO index showed a positive value 20-30 days before the two cold wave events, and the value in 2015 was greater than that in 2007. About 10 days before the event, the AO-NAO index showed a negative value, that is, the AO-NAO phase changed from "+" to "-" 10-30 days before the outbreak of the event, which is a process of continuous decline in the AO-NAO index.

[0080] After verifying the feasibility of using the signal of the AO-NAO index before the outbreak of a cold air event for extended-term forecasting and conducting a preliminary statistical analysis, we began to establish a model for using the AO-NAO index for extended-term forecasting of cold air events. The main basic processes include determining the index decline threshold before a cold air event, determining the time period for the occurrence of cold air after the threshold signal appears, and determining the intensity of the cold air event.

[0081] (1) Determination of the threshold for the index signal to drop before a cold air event

[0082] Using the daily AO-NAO index and national cold air event data set from 1951 to 2014 (a total of 805 cases), the threshold of continuous decline days and index decline amplitude of AO-NAO phase transition 10-30 days before the outbreak of cold air events were studied. The analysis showed that from September 1951 to April of the following year, about 76% of the national cold air events 15-30 days before the outbreak, the AO index would show a downward trend for 5 consecutive days or a continuous decline for 3 consecutive days with a cumulative decline value exceeding 1.0. About 79% of the cold air events 10-30 days before the outbreak, the NAO index showed a negative phase for 10 consecutive days, a downward trend for 5 consecutive days, or a continuous decline for 3 consecutive days with a cumulative decline value exceeding 1.0.

[0083] In order to further verify whether the index signals appearing in the above events are representative, the daily AO-NAO index and the national cold air event data set from 2014 to 2019 (a total of 69 cases) were used for testing, and the hit rate, missed alarm rate and false alarm rate were quantitatively calculated.

[0084] The determination of the above thresholds was verified using the daily AO-NAO index and the national cold air event dataset from 2014 to 2019 (a total of 69 cases), and the hit rate, missed rate, false alarm rate and TS score were quantitatively calculated:

[0085] Hit rate:

[0086]

[0087] False negative rate:

[0088]

[0089] Empty report rate:

[0090]

[0091] TS Rating:

[0092]

[0093] In the formula, NA k To predict the correct number of times, NB k NC is the number of empty reports. k is the number of missed reports.

[0094] The test results show that 59 nationwide cold air events met the above conclusions, 10 were missed and 15 were falsely reported, the forecast hit rate reached 85.5%, the missed reporting rate was about 14.5%, the false reporting rate was about 20%, and the TS score was 0.702.

[0095] The above results show that through the test and evaluation of the model initially established using the threshold standard, it is shown that the threshold has a good indicative significance for the forecast of national cold air events. Therefore, the forecast process of the model is determined: daily sea level pressure data is obtained every day, and the NAO index is calculated by calling the built-in formula of the model, and monitored daily. When the NAO index produces one of the following situations, a forecast signal is generated:

[0096] A has a downward trend for 5 consecutive days; or

[0097] B has been decreasing for 3 consecutive days and the cumulative decrease value exceeds 1.0; or

[0098] C has been negative for 10 consecutive days.

[0099] It should be pointed out that the built-in formula of the above model is:

[0100]

[0101] in, and They are the daily average values ​​of the sea level pressure field in the longitude range 80°W-30°E on the 35°N and 65°N latitudes, respectively.

[0102] (2) Determination of the onset time of cold air events

[0103] After the forecast process was formed and the supporting software was produced according to the forecast process, the application test was carried out on the business platform. Since January 2020, the NAO index has been applied in the cold air extended period forecast business and services of the National Meteorological Center, and has been running in real time in the medium- and long-term weather forecast business integrated application system. During the application test, a total of 21 national cold air events occurred in the cold season of 2020-2022, of which 20 events could be predicted in advance by the NAO index model 10-30 days before the outbreak of the cold air event, with a hit rate of 95%. The advance forecast time of these cold air events was as short as 11 days and as long as 33 days. In subsequent longer application tests, it was found that most index signals appeared 15-30 days before the occurrence of cold air events. Therefore, it was determined that the advance forecast time of the model was within 15-30 days.

[0104] When the forecast signal appears, it is preliminarily determined that a nationwide cold air event is likely to occur 15 days after the signal conditions are met. After that, in order to forecast the cold air process more accurately, other cold air forecast products of the Central Meteorological Observatory can be combined when the cold air is approaching. For example, 7-10 days before the cold air outbreak, attention can be paid to the westerly jet intensity index, blocking high pressure index and Siberian high pressure intensity index. In this way, quantitative products are used for overall control, so as to gradually obtain more accurate mid-term forecast results for the cold air process.

[0105] In the business application test, after determining the cold air onset time, we further focused on the intensity of the predicted cold air events. In the test of the 2020-2022 cold season, we found that among the cold air events predicted by the index model, 67% were moderate cold air, 27% were strong cold air, and only 6% were nationwide cold wave events. On the one hand, it is proved that as the intensity of the cold air process increases, its probability of occurrence decreases and the difficulty of forecasting increases; on the other hand, it is found that the index forecast model's forecast effect on strong cold air processes needs to be improved, and lacks accurate identification of strong cold air processes. Therefore, in order to more accurately forecast strong nationwide cold waves, the model has been improved for the medium-term and extended period forecast of nationwide cold waves.

[0106] It is proposed to apply the artificial intelligence method - the Self-Organizing Map neural network (SOM) method to train and cluster the 500hPa height field situation in the period 15-30 days before the occurrence of a nationwide cold wave event, so as to obtain the forecast key area, and then define and calculate the new NAO index, thereby improving the forecast effect of the nationwide cold wave.

[0107] like Figure 4 As shown in the figure, the training and modeling process of using the self-organizing neural network (SOM) to train the 500h Pa height field in the extended period forecast period of the national cold wave is given.

[0108] First, the 500hPa height field data of the extended forecast period of 36 national cold wave events from 1961 to 2009 were sorted out, converted into one-dimensional data and input into the SOM neural network. According to the data volume and the empirical formula of the model, the minimum competition layer latitude, the initial radius of the winning neighborhood, the learning rate and the maximum number of iterations were obtained. After these parameters were input into the neural network, the height field data was iteratively trained. Figure 5 As shown, the two-dimensional mapping distribution of the winning neuron is estimated through the U-Matrix weight matrix, and a preliminary judgment is made as to whether its training clustering results are stable and effective.

[0109] like Figure 6 As shown in the figure, after adjusting the parameters and training multiple times, the winning neurons with stable distribution are obtained, and the training results are output and the two-dimensional data structure is restored, so that the weather situation at the height of 500h Pa after clustering can be obtained. All five models show the typical NAO abnormal phase characteristics in the northern hemisphere 15 days before the outbreak of the cold wave, which further verifies the feasibility of selecting the NAO index for the extended period forecast of the cold wave. Combined with the East Asian classic cold wave model, the situation distribution in the obtained mode is analyzed, and two key areas are determined: Key Area I: 40°W-20°E, 45°N-65°N; Key Area II: 60°E-130°E, 45°N-80°N.

[0110] Among them, key area I is located in the Atlantic Ocean and the east coast, but its location and range are different from the range specified by the traditional North Atlantic Oscillation. It mainly reflects the signal that the cold air in the high latitudes north of the Atlantic Ocean moves southward and eastward to invade my country; key area II is located in Siberia and the north of Siberia, which mainly reflects the signal that the cold air brought by the splitting of the Arctic polar vortex directly moves southward into my country.

[0111] After determining the two key areas, such as the calculation method of AO and NAO index, the calculation method of key area index is defined:

[0112]

[0113] in The normalized sea level pressure anomaly represents the latitudinal average of the southern boundary of the key area. The normalized sea level pressure anomaly represents the latitudinal average of the northern boundary of the key area, and the key area index I is the difference between the two. According to this definition method, indexes Index1 and Index2 are constructed corresponding to key area I and key area II respectively.

[0114] If one of the above two indices meets any of the following conditions, a forecast signal will be generated:

[0115] A has a downward trend for 5 consecutive days; or

[0116] B has been decreasing for 3 consecutive days and the cumulative decrease value exceeds 1.0; or

[0117] C has been negative for 10 consecutive days.

[0118] Similarly, according to the same test method and formula, five national cold wave events from 2010 to 2020 were used for testing. Among them, four national cold air events could be accurately predicted 15-30 days in advance, one was missed, and the number of false reports was relatively large, reaching 18 times. The forecast hit rate reached 80%, the missed rate was 20%, the false report rate was about 81.8%, and the TS score was 17.39%. From the results, it can be seen that the improved scheme timely predicted four national cold waves, and the hit rate was greatly improved compared with the previous model. However, it can also be seen that the false report rate of the index is high, resulting in a low TS score. The reason may be that the 500h Pa height field situation of the strong cold air and the national cold wave is not significantly different in the extended period, and the process of significantly increasing its intensity mainly occurs in the time period within two weeks. Therefore, the improved model incorrectly predicts the signal of general or strong cold air, resulting in a large false report rate. To address this shortcoming, it is necessary to reduce the false alarm rate of the improved model after the signal is generated in the extended period, that is, to further determine whether the intensity of the cold air has reached the intensity of a nationwide cold wave in the nearby period, and further select to reduce the false alarm rate.

[0119] Therefore, the exponential model can introduce an objective forecasting method 5-10 days after the forecast signal appears in the extended period and 5-15 days before the cold wave event occurs. Combined with quantitative products such as the westerly jet intensity index and the Siberian high pressure index, it can make a more accurate forecast of the intensity of cold air and the specific time of occurrence, thereby reducing the false alarm rate.

[0120] like Figure 7 As shown in Figure 2, the present invention is tested in actual cold wave events. The NAO index for the cold season in 2020-2022 is as follows: Figure 7As shown, there were 21 nationwide cold air events in the cold season of 2020-2022, and the NAO index experienced a continuous decline before the outbreak of the cold air events.

[0121] like Figure 8 As shown, the present invention traces the trajectory of the cold air mass in an actual cold wave event, proving that in this cold wave event, the cold air did gather in the Atlantic Ocean, and under the guidance of the high-altitude situation, it invaded my country from the northwest region eastward along the western route through the Caspian Sea and the Black Sea, verifying the cold wave invasion path information provided by the improved index.

[0122] For those skilled in the art, various corresponding changes and modifications can be made according to the above technical solutions and concepts, and all of these changes and modifications should be included in the protection scope of the claims of the present invention.

Claims

1. A cold wave extended period forecasting method based on NAO index, characterized in that: The method comprises the following steps: S1 automatically obtains daily ERA-5 historical data; S2 obtains daily ERA-5 historical data according to step S1, counts the low-frequency characteristics of the cold air process for 30-60 days, and obtains the phase transition characteristics of AO before the cold wave event breaks out; S 3 Statistically analyze the characteristics of AO and NAO indices before the occurrence of cold air based on daily ERA-5 historical data, and quantitatively describe the evolution of AO-NAO indices before the occurrence of cold air events; S4 Establish a model for extended-range forecast of cold air events using the AO-NAO index; S5 improves the prediction model of step S4 using a self-organizing map neural network SOM method; After improving the forecast model according to the SOM method, S6 uses the hybrid single-particle Lagrangian integrated trajectory model to track the backward trajectory of the cold air mass, and finally obtains the model's prediction results for the cold air invasion path.

2. The cold wave extended period forecast method based on NAO index according to claim 1 is characterized in that: In step S2, a Butterworth bandpass filtering method is used to retain the 30-60 day low-frequency signal on the 500 hPa isobaric surface in the daily ERA-5 historical data, and the phase transition characteristics of the AO before the outbreak of the cold wave event are obtained.

3. The cold wave extended period forecast method based on NAO index according to claim 1, characterized in that: The quantitative description of the evolution law of the AO-NAO index before the cold air event in step S3 includes: Using the sea level pressure data from the daily ERA-5 historical data, the AO and NAO index characteristics were calculated and statistically analyzed 60 days before each cold air event from 1951 to 2014. The evolution of the AO-NAO index before a cold air event was quantitatively described. It was found that 10-30 days before a cold air event, the AO-NAO phase would change from "+" to "-" or the index would decrease continuously.

4. The cold wave extended period forecast method based on NAO index according to claim 1, characterized in that: Establishing the model described in step S4 includes determining the index drop threshold before the cold air event, determining the cold air occurrence period after the threshold signal appears, and determining the intensity of the cold air event.

5. The cold wave extended period forecast method based on NAO index according to claim 4 is characterized in that: Also includes: Using the daily AO-NAO index and the national cold air event dataset from 1951 to 2014, the continuous decline day threshold of the AO-NAO phase transition and the index decline amplitude threshold 10-30 days before the outbreak of the cold air event were counted to determine the representativeness of the index signal; and the daily AO-NAO index and the national cold air event dataset from 2014 to 2019 were used for verification, and the hit rate, missed reporting rate and false alarm rate were quantitatively calculated.

6. The cold wave extended period forecast method based on NAO index according to claim 4, characterized in that: Also includes: The daily sea level pressure data is obtained every day, and the NAO index is calculated using the formula built into the model. The NAO index is monitored daily and a forecast signal is generated when any of the following conditions are met: A has a downward trend for 5 consecutive days; or B has been decreasing for 3 consecutive days and the cumulative decrease value exceeds 1.0; or C has been negative for 10 consecutive days.

7. The cold wave extended period forecast method based on NAO index according to claim 1, characterized in that: The step S5 includes: The SOM method is used to train and cluster the 500hPa height field situation in the period of 15-30 days before the occurrence of a nationwide cold wave event. First, the 500hPa height field data of the extended forecast period of 36 nationwide cold wave events from 1961 to 2009 are sorted out, converted into one-dimensional data and input into the SOM neural network. According to the data volume and the empirical formula of the model, the minimum competition layer latitude, the initial radius of the winning neighborhood, the learning rate and the maximum number of iterations are obtained. After these parameters are input into the neural network, the height field data is iteratively trained, and the two-dimensional mapping distribution of the winning neuron is estimated through the U-Matrix weight matrix to preliminarily judge whether its training clustering results are stable and effective. After adjusting the parameters and training for many times, the winning neuron with stable distribution is obtained, and its training results are output. After restoring its two-dimensional data structure, the weather situation at 500hPa height after clustering can be obtained.

8. The cold wave extended period forecast method based on NAO index according to claim 1, characterized in that: The step S6 includes: Based on the five weather situation models at 500hPa obtained by the SOM clustering method, combined with the East Asian classic cold wave model, the situation distribution in the obtained modes was analyzed and two key areas were identified: Key area I: 40°W-20°E, 45°N-65°N; Key area II: 60°E-130°E, 45°N-80°N; Like the calculation method of AO and NAO index, the calculation method of key area index is defined: Among them, represents the latitudinal average of the normalized sea level pressure anomaly at the southern boundary of the key area, represents the latitudinal average of the normalized sea level pressure anomaly at the northern boundary of the key area, and the key area index I is the difference between the two; Then construct index Index1 and index Index2 corresponding to key area I and key area II respectively; Index1 or Index2 generates a forecast signal if any of the following conditions are met: A has a downward trend for 5 consecutive days; or B has been decreasing for 3 consecutive days and the cumulative decrease value exceeds 1.0; or C has been negative for 10 consecutive days.

9. A system for implementing the cold wave extended period forecasting method based on NAO index as claimed in claim 1, characterized in that: The system comprises: The data acquisition module is used to obtain daily historical data; A verification index module is used to analyze the low-frequency characteristics of the process based on the daily historical data and verify the feasibility of using the AO index to perform extended cold air forecasting; The forecast model building module uses the NAO index to preliminarily build a forecast model; The forecast model improvement module uses the SOM method to independently model the forecast key area, define new indices, and improve the forecast model; The cold wave extended period forecast module is used to make a daily extended period forecast of the cold wave based on the final forecast model and the real-time daily sea level pressure data.

Images