Spring rainfall prediction method based on atmospheric circulation abnormal signal

Abstract

The invention discloses a spring rainfall prediction method based on atmospheric circulation abnormal signals. The method comprises the following steps: S1, obtaining and preprocessing multi-source data; s2, key circulation system identification and signal extraction; s3, construction of a synergistic effect index: constructing a circulating current synergistic effect index for comprehensively representing the configuration states of the plurality of circulating current systems in the specific region based on the circulating current signals extracted in the S2; s4, establishing a rainfall prediction model; and S5, testing a prediction result and integrating services. According to the method, abnormal signals of a plurality of key circulation systems including arctic oscillation and remote correlation wave train thereof, a south branch groove, subtropical high pressure, a water vapor transfer channel and a cold air activity path are systematically identified and quantified from multi-source data, so that the problems of high dependence on a single factor and incomplete physical mechanism of a traditional method are solved; and a comprehensive and three-dimensional circulation background support is provided for prediction.

Description

Technical Field

[0001] This application belongs to the field of precipitation prediction technology, specifically relating to a spring precipitation prediction method based on atmospheric circulation anomaly signals. Background Technology

[0002] Spring precipitation forecasting is a key and challenging area in meteorological and climate prediction. Spring is a transitional season with frequent adjustments in atmospheric circulation, high variability within the climate system, and complex and diverse influencing factors, making the predictability of spring precipitation relatively low and the forecasting process quite difficult.

[0003] Currently, spring precipitation forecasting methods can be mainly divided into two categories: dynamic forecasting based on numerical models and empirical forecasting based on statistical models. While numerical models can depict detailed physical processes, their ability to capture precursory signals of spring circulation anomalies is limited, and they exhibit significant systematic biases, particularly in regional-scale precipitation forecasting. Statistical models (such as regression and similarity models), although computationally simple, mostly rely on single or a few predictive factors, making it difficult to comprehensively and synergistically characterize the interactions and configurations among multiple key circulation systems such as the Arctic Oscillation, the Southern Branch Trough, the Subtropical High, moisture transport, and cold air activity. This results in insufficient forecast stability and weak physical mechanism support. Furthermore, in existing forecasting workflows, data preprocessing, signal extraction, model building, and product output are often relatively independent, with low levels of automation and integration, hindering continuous optimization of forecast models and rapid business integration, thus limiting the timeliness and operational availability of forecast results.

[0004] Therefore, there is an urgent need to develop a spring precipitation forecasting method that can comprehensively consider the synergistic effects of multiple circulation systems, has clear physical meaning, and is easy to integrate into operations, so as to improve the accuracy, stability and operational applicability of spring precipitation forecasting. Summary of the Invention

[0005] This application provides a spring precipitation prediction method based on atmospheric circulation anomaly signals, aiming to solve the problems of existing technologies being highly dependent on single factors and having incomplete physical mechanisms.

[0006] Firstly, a spring precipitation prediction method based on atmospheric circulation anomaly signals, the method comprising:

[0007] S1: Multi-source data acquisition and preprocessing. The system collects and processes multi-source meteorological data for spring precipitation forecasting and constructs a standardized dataset.

[0008] S2: Identification and signal extraction of key circulation systems. Identify and quantify anomalous signals of key circulation systems that affect spring precipitation from historical data. The key circulation systems include at least the Arctic Oscillation and its teleconnection wave train, the southern branch trough system, the subtropical high pressure system, water vapor transport channels, and cold air activity paths.

[0009] S3: Construction of the synergistic effect index. Based on the circulation signal extracted in S2, a circulation synergistic effect index is constructed to comprehensively characterize the configuration state of multiple circulation systems in a specific region.

[0010] S4: Establish a precipitation prediction model. Using the previously slow-varying factor and the circulation synergy index constructed in S3 as the core input, establish a spring precipitation prediction model and output the spring precipitation trend, precipitation anomaly level and key influencing system evolution prediction.

[0011] S5: Validation and integration of forecast results. The predictive capability of the forecast model is evaluated using cross-validation and independent sample validation. Models that pass the validation are integrated into the operational forecasting platform to achieve automated and visualized forecast product output.

[0012] Optionally, in S1, the multi-source data includes atmospheric reanalysis data, station observation data, sea surface temperature data, atmospheric circulation index, and external forcing factor data;

[0013] The preprocessing includes quality control and missing data processing, spatiotemporal scale matching and alignment, data standardization and anomaly correction, and finally constructs a time-series dataset with years as samples, previous and contemporaneous circulation and external forcing factors as features, and spring precipitation conditions as labels.

[0014] Optionally, in S2, identifying the Arctic Oscillation and its teleconnected wave train includes: obtaining the Arctic Oscillation index and determining, through synthetic analysis, the wave train characteristics of the geopotential height field anomaly distribution in the mid-high latitudes of Eurasia under its positive and negative anomalous phases and its transmission influence on the intensity of the southern branch trough;

[0015] Identifying the southern branch trough system includes: calculating a multi-index system for defining the intensity of the southern branch trough based on at least two of the geopotential height field, vorticity field, and wind field.

[0016] Identifying the subtropical high-pressure system includes: extracting the ridge position, westward extension point and intensity index of the western Pacific subtropical high, as well as the center position and intensity information of the North African subtropical high in the Arabian Sea region;

[0017] Identifying the water vapor transport channels includes: calculating the southwest water vapor flux originating from the Bay of Bengal and the southeast water vapor flux originating from the periphery of the western Pacific subtropical high, and analyzing their flux intensity, direction, and convergence characteristics over the predicted region;

[0018] Identifying the cold air activity path includes analyzing the meridional position and intensity index of the Northern Hemisphere polar vortex, as well as the intensity of the Siberian High and the southern boundary of the key isobars.

[0019] Optionally, in step S3, constructing the circulation synergy index specifically includes:

[0020] S3.1: Quantifying the configuration relationship between the phase of the Arctic Oscillation and the intensity of the southern branch trough;

[0021] S3.2: Calculate the superposition and enhancement effect of the position and morphological characteristics of the Western Pacific subtropical high and the North African subtropical high on the airflow ahead of the southern branch trough;

[0022] S3.3: Assess the convergence of moisture transport intensity from different source regions and southward cold air activity intensity in the forecast area;

[0023] S3.4: The quantitative relationship factors obtained in steps S3.1, S3.2 and S3.3 are integrated into a single circulation synergy index by weighted combination or principal component analysis.

[0024] Optionally, in step S4, the spring precipitation prediction model is established using a phased modeling strategy, which includes:

[0025] In the first stage, a regression model was established using the previous sea surface temperature, snow cover, and soil moisture as input factors to predict the trend of the circulation synergy index.

[0026] In the second stage, the predicted circulation synergy index trend from the first stage is combined with real-time circulation signals near spring as input features. Random forest or gradient boosting decision tree models are then used to establish the final prediction model for spring precipitation and drought / flood levels.

[0027] Optionally, the business integration in S5 is specifically implemented as follows:

[0028] Establish an automated data pipeline on the business forecasting platform to regularly acquire and preprocess real-time data;

[0029] The deployed prediction model is automatically invoked to perform calculations and generate forecast results including quantitative precipitation prediction, drought and flood level probability, and circulation factor contribution analysis.

[0030] The forecast results are automatically applied to a preset template to generate a comprehensive forecast product that combines graphics and text, and then pushed to an internal consultation platform or external release channels.

[0031] Optionally, it also includes a model continuous optimization step: based on new data accumulated in business operations, the precipitation prediction model is incrementally trained or retrained periodically or triggered, and the performance of the updated model is evaluated using the same verification method as in step five. When the performance improvement meets a preset threshold, the updated model version replaces the original model and is put into business operations.

[0032] Secondly, an operational forecasting system for implementing spring precipitation forecasting methods includes:

[0033] The data preprocessing module is used to perform S1 to acquire and standardize multi-source meteorological data;

[0034] The signal extraction and index construction module is used to execute S2 and S3 to identify the circulation signal and calculate the circulation synergy index;

[0035] The prediction model module is used to execute S4 and has a built-in trained precipitation prediction model.

[0036] The verification and product integration module is used to execute S5, verify and evaluate the model prediction results, and automatically integrate the model into the business process to generate a visual forecast product.

[0037] Compared with the prior art, this application has at least the following beneficial effects:

[0038] This application overcomes the problems of traditional methods, such as strong dependence on single factors and incomplete physical mechanisms, by systematically identifying and quantifying anomalous signals of multiple key circulation systems from multi-source data, including the Arctic Oscillation and its teleconnection wave train, the southern branch trough, the subtropical high, the water vapor transport channel, and the cold air activity path. It provides comprehensive and three-dimensional circulation background support for prediction.

[0039] This application also constructs a "circulation synergy index," which comprehensively quantifies the configuration relationships and interactions between multiple circulation systems and condenses them into a predictive factor with clear physical meaning. This index, as a whole, characterizes whether the circulation system is in a synergistic state conducive to precipitation, enhances the model's ability to depict complex climate interactions, and improves the physical consistency and stability of the prediction.

[0040] This application also employs a phased modeling strategy. In the first phase, slow-changing factors such as sea surface temperature and snow cover are used to predict the trend of the circulation synergy index. In the second phase, real-time circulation signals near spring are combined to make the final precipitation prediction. This strategy not only utilizes the good predictive properties of slow-changing factors but also integrates the dynamic information of real-time circulation, effectively improving the model's predictive ability and real-time update flexibility. Attached Figure Description

[0041] Figure 1 A flowchart of a spring precipitation prediction method based on atmospheric circulation anomaly signals provided in this application. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0043] This application provides a spring precipitation prediction method based on atmospheric circulation anomaly signals, comprising the following steps:

[0044] S1: Multi-source data acquisition and preprocessing, collecting and processing multi-source meteorological data for spring precipitation forecasting, and constructing a standardized dataset;

[0045] Specifically, the core of S1 lies in the systematic collection and processing of various meteorological data for spring precipitation forecasting. Through a series of standardized operations, a high-quality, spatiotemporally consistent initial dataset is constructed, laying a reliable data foundation for subsequent accurate extraction of circulation signals and establishment of prediction models.

[0046] First, we obtain multi-source raw data from authoritative institutions, mainly including: global atmospheric reanalysis data from NCEP / NCAR or ERA5, which provides information on circulation fields, humidity fields, etc.; daily surface observation data from the China Meteorological Administration and climate centers of various provinces, which are used to obtain actual precipitation and temperature data; global sea surface temperature data from NOAA or the Hadley Centre in the UK; and large circulation index data such as the Arctic Oscillation and the Southern Oscillation from NOAA and other institutions. We also collect data on external forcing factors such as snow cover and soil moisture.

[0047] After acquiring the data, a rigorous preprocessing procedure was performed. The first step was to conduct quality control and handle missing values, using physical thresholds and climatological boundaries to remove obvious erroneous values ​​and verify the physical consistency within the data. For a small number of missing values ​​in the station data, a weighted average of spatial neighboring stations was used for interpolation; for gridded data, the complete field of the reanalysis data itself was used directly.

[0048] The second step is to achieve matching and alignment of spatiotemporal scales. In terms of time, all daily data are uniformly processed into monthly values. For example, daily precipitation is accumulated into monthly totals, and the daily circulation field is averaged into a monthly average field. In terms of space, the precipitation data observed at the stations are normalized to a regular grid consistent with the circulation field through methods such as Kriging interpolation. At the same time, based on the geographical boundaries of the prediction area, the grid data is averaged regionally to obtain the circulation index and sea surface temperature index that represent the overall condition of the area.

[0049] The third step is to standardize and anomalyize key predictive factors. The Z-Score method is used to standardize circulation index, sea surface temperature index, etc., to eliminate the dimensional differences between different factors. At the same time, the anomaly series of precipitation and circulation field are calculated, that is, their long-term climate average is subtracted to highlight the interannual anomaly signals required for prediction.

[0050] Finally, the processed data are integrated to construct a structured time series dataset. This dataset uses the year as a sample, and uses signals such as sea surface temperature, circulation, and snow cover from autumn to winter, as well as circulation coordination index in spring, as predictive features. The regional average precipitation anomaly or drought / flood level in spring of that year is used as the prediction target, thus forming a standardized data set that can be directly read and learned by machine learning models.

[0051] For example, in the practice of forecasting spring precipitation in Yunnan, daily precipitation data, corresponding reanalysis circulation data, and various circulation indices from over a hundred meteorological stations in the province were collected. By performing quality control interpolation on the station data, calculating the total spring precipitation, and obtaining the regional average anomaly as the forecast label, and simultaneously extracting and standardizing the sea surface temperature index and circulation index of key sea areas in the previous winter, a structured data file containing multiple years and multiple features was finally generated, completing all data preparation work before forecasting.

[0052] S2: Identification and signal extraction of key circulation systems. Identify and quantify anomalous signals of key circulation systems that affect spring precipitation from historical data. The key circulation systems include at least the Arctic Oscillation and its teleconnection wave train, the southern branch trough system, the subtropical high pressure system, water vapor transport channels, and cold air activity paths.

[0053] Specifically, S2 aims to accurately identify and quantify circulation systems and their anomalous states that have a dominant influence on spring precipitation from preprocessed historical climate data.

[0054] First, regarding the Arctic Oscillation (AO) and its downstream impacts, a standard monthly AO index sequence was obtained as the basic signal. More importantly, the teleconnection wave trains excited by AO anomalies were analyzed. Specifically, on a 500 hPa geopotential height field, the composite anomaly distribution of the height field in the mid-to-high latitudes of Eurasia was calculated when the AO was in a significant positive or negative phase. Through this composite analysis, it is possible to intuitively identify how the AO transmits energy downstream through wave trains in the form of "ridge-trough-ridge" or "trough-ridge-trough", ultimately affecting the strength of the southern branch trough in the Bay of Bengal region.

[0055] Secondly, the southern branch trough system is quantified in multiple dimensions. The southern branch trough is a direct weather system affecting spring precipitation. Its intensity is defined by a combination of multiple indicators to avoid the limitations of a single indicator. First, based on the 500 hPa geopotential height field, the average geopotential height anomaly of key areas in the Bay of Bengal (e.g., 80°E-100°E, 15°N-20°N) is calculated. The larger the negative anomaly, the deeper the trough usually is. Second, based on the 700 hPa vorticity field, the average relative vorticity of the same area is calculated. The larger the positive vorticity value, the stronger the cyclonic shear of the trough. Third, wind field characteristics are analyzed. For example, the vertical shear of the meridional wind in the range of 85°E-95°E or the north-south wind components at a specific longitude are calculated to determine the tilt structure and intensity of the trough. These signals extracted from the geopotential height field, vorticity field, and wind field together constitute a three-dimensional description of the activity of the southern branch trough.

[0056] Third, to accurately characterize the subtropical high-pressure system configuration, it is crucial to focus on the western Pacific subtropical high and the North African subtropical high (often manifested as the Arabian Sea high). For the western Pacific subtropical high, its ridge position, longitude of the westernmost ridge point, and intensity index should be extracted from the monthly mean geopotential height field. For the North African subtropical high, its central position and intensity in the northern Arabian Sea should be analyzed. The key purpose of extracting these signals is to determine whether the shape and position of the subtropical high are conducive to the superposition of its peripheral airflow with the southwest airflow ahead of the southern branch trough. For example, when the western Pacific subtropical high is stronger and more westerly, the southerly airflow on its southwest side is more likely to merge with the airflow ahead of the trough, enhancing the moisture transport to the predicted area; while when the North African subtropical high is more easterly and northerly, the peripheral airflow of its anticyclonic circulation can combine with the airflow ahead of the mid-latitude trough, enhancing the bypass flow on the west side of the plateau, thus affecting the depth of the southern branch trough.

[0057] Fourth, track anomalies in water vapor transport channels. Water vapor is the material basis for precipitation formation. Based on the wind field and specific humidity field of reanalysis data, calculate the water vapor flux vector of the entire layer integral. Focus on analyzing two main channels: one is a water vapor transport belt originating from the Bay of Bengal and carried by the southern branch westerly winds; the other is a water vapor transport belt originating from the western Pacific Ocean and the South China Sea and moving northward along the periphery of the subtropical high. The extracted signals include the water vapor flux intensity and direction of these two channels in key areas (such as the northern part of the Indochina Peninsula to the target area in my country), as well as their convergence (quantified by calculating water vapor flux divergence). In drought years, both water vapor flows are usually weak and their convergence is not obvious.

[0058] Fifth, diagnose the path and intensity of cold air activity. The southward movement of cold air is a crucial dynamic condition for triggering precipitation, primarily based on signals from two levels: first, in the upper troposphere, analyzing the meridional position and intensity index of the Northern Hemisphere polar vortex, whose positional shift indicates the approximate direction of the main body of cold air; second, at the surface level, analyzing the central intensity of the Siberian High and the positional changes of key isobars, such as the southern latitude of the 1025 hPa or 1030 hPa isobars on the sea level pressure field, which directly reflects the southward extent of cold air influence. Combining these two aspects allows for a comprehensive assessment of the activity level and main path of cold air in the predicted area.

[0059] S3: Construction of the synergistic effect index. Based on the circulation signal extracted in S2, a circulation synergistic effect index is constructed to comprehensively characterize the configuration state of multiple circulation systems in a specific region.

[0060] Specifically, S3 aims to integrate multiple independent circulation signals extracted from S2 into a circulation synergy index that can characterize the overall configuration of atmospheric circulation, based on their physical correlations.

[0061] First, the correlation between the Arctic Oscillation (AO) and the activity of the southern branch trough is analyzed. The time-delay correlation coefficient between the AO index and the southern branch trough intensity index (synthesized from multiple indicators such as geopotential height and vorticity) is calculated. For example, the correlation between the AO index in the previous winter (December-February) and the intensity of the southern branch trough in the spring (March-April) of the same year is analyzed. Through synthetic analysis, it is determined whether the propagation characteristics of the Eurasian teleconnection wave train (EU wave train) and the South Eurasian wave train (SEA wave train) are clear in typical years with positive and negative AO phases, and whether their final impact on the intensity of the southern branch trough in the Bay of Bengal is enhanced or weakened. An "AO-trough configuration coefficient" is constructed. When the phase and wave train propagation characteristics of the AO are conducive to the deepening of the southern branch trough, the coefficient is positive, and vice versa.

[0062] Secondly, the interaction between the subtropical high and the southern branch trough system is assessed, which involves two key subtropical highs: the western Pacific subtropical high and the North African subtropical high (often manifested as the Arabian Sea high).

[0063] For the Western Pacific subtropical high, calculate the position index of its western extension ridge and the intensity of the southerly wind component in its periphery (e.g., within the range of 110°E-120°E). When the subtropical high is westward and stronger, the southerly or southwesterly airflow in its periphery and the southwesterly airflow in front of the southern branch trough will overlap in the airspace from the Indochina Peninsula to the target area, which is likely to form a low-level jet stream. Define a "Western Pacific subtropical high superposition factor", which is jointly determined by the degree of westward extension of the subtropical high and the wind speed anomaly in a specific area in its periphery.

[0064] For the North African subtropical high, the focus is on analyzing the location (longitude and latitude) of its anticyclone center in the Arabian Sea and the westerly wind speed on its northern periphery. When its position is more easterly and northerly, the westerly winds on its periphery overlap with the southwesterly winds ahead of the Mediterranean-North African trough on the west side of the plateau, enhancing the flow around the plateau and facilitating the establishment of a "north ridge and south trough" pattern. Based on this, a "North African subtropical high flow enhancement factor" is constructed.

[0065] Third, precipitation requires abundant water vapor and sufficient lifting force (often provided by cold air activity), therefore, the convergence of the two needs to be comprehensively assessed;

[0066] The quantitative calculation of water vapor conditions comes from the southwest water vapor flux from the Bay of Bengal and the southeast water vapor flux from the periphery of the western Pacific subtropical high, as well as the magnitude of the vector summation and water vapor flux divergence (convergence intensity) over the target region (e.g., 95°E-105°E, 20°N-30°N).

[0067] Cold air conditions: Based on the surface Siberian High intensity index and the Northern Hemisphere polar vortex meridional position index, assess the potential intensity of southward cold air movement;

[0068] A "water vapor-cold air convergence index" is constructed, which is characterized by the product or weighted sum of the water vapor flux convergence intensity and the cold air activity intensity (after standardization). When both are enhanced simultaneously, the index is high, indicating that precipitation is likely to occur; if one is strong and the other is extremely weak, the index is low.

[0069] Finally, the three quantitative factors representing the coordination relationship between the systems are integrated. The integration method employs statistical methods such as weighted linear combination or principal component analysis (PCA), and the weights can be determined based on the correlation between each factor and historical spring precipitation or its contribution to precipitation variance.

[0070] The resulting "Circulation Synergy Index (CSI)" is a single time series (one value each spring). High values ​​indicate that the AO phase, through wave train transmission, favors the development of the southern branch trough; the position and shape of the subtropical high facilitates the superposition and enhancement of its peripheral airflow and the airflow ahead of the trough; and abundant moisture transport effectively converges with moderately cold air in the target area. Conversely, low values ​​indicate scattered configurations and mutual inhibition among the systems, hindering precipitation formation. This index, from a holistic perspective, encapsulates the core physical picture of the synergistic influence of multiple systems on precipitation, serving as the most comprehensive key input factor in the prediction model.

[0071] S4: Precipitation prediction model establishment. Using the previously slow-varying factor and the circulation synergy index constructed in S3 as the core input, a spring precipitation prediction model is established, and the outputs spring precipitation trend, precipitation anomaly level and key influencing system evolution prediction are provided.

[0072] Based on the data preparation and core forecasting factor extraction completed in the previous steps, a spring precipitation prediction model that can effectively integrate this information is established. The aim is to use the patterns in historical data to make quantitative and qualitative predictions of future spring precipitation.

[0073] The predictive model is constructed by selecting and comparing different modeling paths, such as multiple regression, random forest, or neural networks. In most cases, a "phased" ensemble strategy is adopted.

[0074] First, slow-changing factors in the early stage (such as sea surface temperature in key sea areas before winter, snow depth on the Tibetan Plateau, and soil moisture in Eurasia) are used as inputs to the first-layer model. These factors change slowly, have long memory and good predictive power. Multiple linear regression or ridge regression models are usually used to predict the preliminary trend of the "circulation synergy index (CSI)" in spring. This step is equivalent to using background environmental conditions to make a preliminary forecast of the overall activity and configuration tendency of atmospheric circulation in spring.

[0075] Then, the CSI trend value predicted in the previous step is combined with the actual observed real-time circulation signals near spring (such as the position of the subtropical high in January and February of the current year, the preliminary shape of the southern branch trough, etc.) and used together as the input features of the second-layer model to finally predict spring precipitation. This layer model tends to use random forest or gradient boosting decision tree models with stronger nonlinearity. These models can better capture the complex interactions between multiple circulation factors and the nonlinear relationship between them and precipitation. For example, it can learn complex rules such as "only when water vapor transport is strong and cold air activity is moderate, the deepening of the southern branch trough has a significant positive contribution to precipitation".

[0076] The trained model can provide multi-layered predictive products to meet different business needs, including:

[0077] Spring precipitation trend forecast: The model directly outputs a quantitative forecast (in millimeters) of the regional average precipitation for the coming spring (March-April). By comparing this with long-term climate averages, it can be easily converted into a qualitative trend conclusion of "more," "near normal," or "less."

[0078] Precipitation anomaly level prediction: Historical spring precipitation is divided into multiple levels (for example, using percentile method to divide it into five levels: severe drought, moderate drought, normal, moderate flooding, and severe flooding); the model can be run in "classification" mode to directly predict the level that future spring precipitation is most likely to fall into, providing more specific risk warnings for disaster prevention and mitigation;

[0079] Predicting the evolution of key impact systems: As an extended application of the model, by analyzing the ranking of "feature importance" provided by models such as random forest, the relative contribution of each circulation factor (such as CSI, subtropical high index, water vapor flux, etc.) in the current year's forecast can be quantitatively assessed. This helps forecasters understand the main physical basis of the current forecast, determine whether water vapor anomalies or dynamic lifting conditions are dominant, and thus gain a deeper understanding of the certainty of the forecast conclusions and the possible uncertainties.

[0080] The model will be trained using data from a sufficiently long historical period (e.g., the previous 40 years) and rolled prediction tests will be conducted using data from the most recent 10-15 years as independent samples to evaluate its stability and actual forecasting skills. Finally, the model configuration that performs best in independent tests and has the clearest physical meaning will be selected as the fixed prediction model for operational use.

[0081] S5: Validation and integration of forecast results. The predictive ability of the forecast model is evaluated using cross-validation and independent sample validation. Models that pass the validation are integrated into the operational forecasting platform to achieve automated and visualized forecast product output.

[0082] Specifically, before a model is officially put into business use, its predictive accuracy and stability must be comprehensively evaluated, mainly using the following two complementary verification methods:

[0083] Cross-validation: Within the historical data range, cross-validation is performed using either the "leave-one-out" method or a "rolling window" approach. For example, assuming data from 1961 to 2020 is available, the model is first trained using data from 1961 to 1990 to predict spring precipitation in 1991; then, it is trained again using data from 1961 to 1991 to predict 1992, and so on, rolling over the years until 2020. This method maximizes the use of historical data to evaluate the model's overall fit and stability under simulated historical scenarios.

[0084] Independent sample testing (post-report testing): In order to more realistically simulate business forecasting scenarios, a completely independent testing period will be set. For example, the data from the most recent ten years from 2011 to 2020 will be used as "future" data and will not be used in the training and adjustment of any model parameters. Only the data from 2010 and earlier will be used to build the final model and make continuous predictions for these ten years. This can most realistically reflect the model's "future" extrapolation forecasting ability and is the gold standard for measuring its business usability.

[0085] The evaluation will use several statistically recognized indicators in the meteorological field, such as: the correlation coefficient between predicted and actual values ​​to measure the correctness of the forecast trend; the root mean square error to measure the accuracy of the forecast; and the trend anomaly score (PS score) for drought and flood level forecasts to assess the ability to capture extreme weather events. Only when a model's key scoring indicators (such as the PS score) are systematically and consistently higher than the average level of existing operational forecasting methods in these tests will it be considered to have operational application value.

[0086] Business integration and automation processes include:

[0087] An automated data pipeline is established on the business platform. This automated task will periodically (e.g., on the 1st of each month) automatically retrieve the latest real-time data such as sea surface temperature, snow cover, and circulation index from internal databases and external authoritative data sources. It will also call the pre-packaged data preprocessing program to automatically complete the quality control, interpolation, and standardization processes consistent with historical data, and generate a real-time forecast factor dataset that meets the model input format requirements.

[0088] Model calling and calculation: The business system automatically loads the deployed prediction model, inputs the processed real-time data into the model, and after the model runs, it automatically outputs the quantitative precipitation forecast value, drought and flood level probability, and contribution analysis of various circulation factors for the future spring.

[0089] Product visualization and publishing: Calculation results are automatically transmitted to the product creation module. Based on preset templates, the system generates comprehensive forecast products with one click, including "Spring Precipitation Trend Forecast Distribution Map," "Contribution Ratio Map of Each Influencing Factor," and "Textual Explanation of Forecast Conclusions." These graphic and textual products are pushed to forecasters through the internal consultation platform as an important reference for their official forecast production; they can also be automatically published to decision-making service websites or new media platforms after review.

[0090] Furthermore, the model has continuous optimization capabilities. Based on new data accumulated during business operations, the precipitation prediction model is incrementally trained or retrained periodically or triggered. The performance of the updated model is evaluated using the same verification method as in step five. When the performance improvement meets the preset threshold, the updated model version replaces the original model and is put into business operation.

[0091] In one embodiment, an operational forecasting system for spring precipitation forecasting includes:

[0092] The data preprocessing module is used to perform S1 to acquire and standardize multi-source meteorological data;

[0093] The signal extraction and index construction module is used to execute S2 and S3 to identify the circulation signal and calculate the circulation synergy index;

[0094] The prediction model module is used to execute S4 and has a built-in trained precipitation prediction model.

[0095] The verification and product integration module is used to execute S5, verify and evaluate the model prediction results, and automatically integrate the model into the business process to generate a visual forecast product.

[0096] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A spring precipitation prediction method based on atmospheric circulation anomaly signals, characterized in that, The method comprises: S1: multi-source data acquisition and preprocessing, the system collects and processes multi-source meteorological data for spring precipitation prediction, and constructs a standardized data set; S2: key circulation system identification and signal extraction, identify and quantify the abnormal signals of the key circulation system affecting spring precipitation from historical data, the key circulation system at least includes the Arctic Oscillation and its teleconnection wave train, the south branch trough system, the subtropical high pressure system, the water vapor transport channel and the cold air activity path; S3: construction of synergy index, based on the circulation signals extracted in S2, a circulation synergy index is constructed to comprehensively represent the configuration state of multiple circulation systems in a specific region; S4: establish a precipitation prediction model, use the slow-changing factors in the early stage and the circulation synergy index constructed in S3 as the core input to establish a spring precipitation prediction model, output the spring precipitation trend, precipitation anomaly grade and key influence system evolution prediction; S5: prediction result test and business integration, use cross-validation and independent sample test to evaluate the prediction ability of the prediction model, and integrate the tested model into the business prediction platform to realize automatic and visual prediction product output. 2.The spring precipitation prediction method based on the atmospheric circulation anomaly signal according to claim 1, characterized in that, In S1, the multi-source data includes atmospheric reanalysis data, station observation data, sea surface temperature data, atmospheric circulation index and external forcing factor data; The preprocessing includes quality control and missing value processing, time and space scale matching and alignment, data standardization and abnormalization, finally constructing a time series data set with year as sample, early and current circulation and external forcing factor as characteristics, and spring precipitation condition as label. 3.The spring precipitation prediction method based on the atmospheric circulation anomaly signal of claim 1, wherein, In S2, identifying the Arctic Oscillation and its teleconnection wave train includes: obtaining the Arctic Oscillation index, and determining the wave train characteristics of the geopotential height field anomaly distribution at middle and high latitudes of Eurasia and its conduction effect on the south branch trough intensity in positive and negative anomaly phase through synthesis analysis; Identifying the south branch trough system includes: calculating a multi-index index for defining the south branch trough intensity based on at least two of the geopotential height field, the vorticity field and the wind field; Identifying the subtropical high pressure system includes: extracting the ridge line position, westward ridge point and intensity index of the western Pacific subtropical high, and the center position and intensity information of the North African subtropical high in the Arabian Sea region; Identifying the water vapor transport channel includes: calculating the southwest water vapor flux from the Bay of Bengal and the southeast water vapor flux from the periphery of the western Pacific subtropical high, and analyzing the flux intensity, direction and convergence characteristics over the prediction area; Identifying the cold air activity path includes: analyzing the meridional position and intensity index of the North Polar Vortex, and the intensity and key isobaric line south boundary position of the Siberian High. 4.The spring precipitation prediction method based on the atmospheric circulation anomaly signal of claim 1, wherein, In S3, the circulation synergy index is constructed specifically as follows: S3.1: Quantify the configuration relationship between the Arctic Oscillation phase and the south branch trough intensity; S3.2: Calculate the superposition enhancement effect of the position and shape characteristics of the western Pacific subtropical high and the North African subtropical high on the south branch trough front airflow; S3.3: Evaluate the intersection of water vapor transport intensity from different sources and cold air southward activity intensity in the prediction area; S3.4: The quantitative relationship factors obtained in steps S3.1, S3.2 and S3.3 are integrated into a single circulation synergy index by weighted combination or principal component analysis. 5.The spring precipitation prediction method based on the atmospheric circulation anomaly signal according to claim 1, characterized in that, In the S4, the spring precipitation prediction model is established by using a phased modeling strategy, which includes: In the first phase, a regression model is established for predicting the trend of the circulation synergy index, using the previous sea surface temperature, snow cover and soil moisture as input factors; In the second phase, the trend of the circulation synergy index predicted in the first phase is combined with the real-time circulation signals in the approaching spring as input features, and a random forest or gradient boosting decision tree model is used to establish the final spring precipitation prediction model and the drought-flood grade prediction model. 6.The spring precipitation prediction method based on the atmospheric circulation anomaly signal according to claim 1, characterized in that, In the S5, the business integration is specifically implemented as: An automatic data pipeline is established on the business forecasting platform to regularly obtain and preprocess real-time data; The deployed prediction model is automatically called for calculation to generate prediction results including quantitative precipitation prediction, drought-flood grade probability and circulation factor contribution analysis; The prediction results are automatically applied to a preset template to generate comprehensive prediction products combining graphics and text, and are pushed to an internal consultation platform or an external release channel. 7.The spring precipitation prediction method based on the atmospheric circulation anomaly signal according to claim 1, characterized in that, The model continues to optimize the step: based on the new data accumulated in the business operation, the precipitation prediction model is incrementally trained or retrained regularly or triggered, and the performance of the updated model is evaluated using the same test method as step five. When the performance improvement meets the preset threshold, the updated model version replaces the original model and is put into business operation.

8. A business forecasting system for implementing the spring precipitation prediction method of any one of claims 1 to 7, characterized in that, It includes: A data preprocessing module for performing S1 to obtain and standardize multi-source meteorological data; A signal extraction and index construction module for performing S2 and S3 to identify circulation signals and calculate the circulation synergy index; A prediction model module for performing S4, which contains the trained precipitation prediction model; A test and product integration module for performing S5 to test and evaluate the model prediction results, and automatically integrate the model into the business process to generate visualized prediction products.

Images