A method for correcting Qinghai-tibet plateau-tropical indian ocean zonal anomaly field enso response

Abstract

The application belongs to the technical field of data analysis, and provides an Qinghai-Tibet Plateau-tropical Indian Ocean longitude-latitude abnormal field ENSO response correction method, specifically: first, the strong and weak year average temperature field and the strong and weak year average field of multi-height layer meteorological elements in the Qinghai-Tibet Plateau-tropical Indian Ocean region are calculated respectively, the elements are subtracted from the long-term average to obtain the abnormal field and the significance test is carried out, then the spatial displacement and the shape direction characteristics of the abnormal field are extracted by using the longitude and latitude gradient operators, the significance statistical test of the abnormal direction distribution is carried out, and finally the direction coupling analysis of the multi-layer meteorological elements is carried out; based on the coupling analysis result, the ENSO response is corrected. Through the direction coupling of the multi-layer meteorological elements, the structural coupling analysis of the elements is carried out in the spatial displacement dimension and the shape principal axis direction dimension, so that the ENSO response recognition result of the Qinghai-Tibet Plateau-tropical Indian Ocean system as the research object has more robust and stable classification and quantitative output.

Description

Technical Field

[0001] This invention belongs to the field of data analysis technology, specifically relating to a method for correcting the ENSO response of the zonal anomaly field of the Tibetan Plateau-Tropical Indian Ocean. Background Technology

[0002] ENSO is a significant signal of global interannual climate change. The anomalous air-sea interactions it triggers can influence the temperature gradient and monsoon system of the Tibetan Plateau and surrounding areas through heat transport and circulation reorganization in the upper troposphere. The 500-200 hPa meridional temperature gradient between the Tibetan Plateau and the tropical Indian Ocean directly characterizes the strength of heat transport from the tropics to the mid-latitudes and is one of the important energy channels determining the ENSO teleconnection response. Existing studies mostly characterize the relationship between ENSO and upper-level temperature anomalies through the mean fields of meteorological elements, but have failed to reveal the dynamic coupling between the spatial morphology and orientation of the anomaly field and the circulation transport path.

[0003] Because anomalous signals in atmospheric circulation are not statically distributed but exhibit directional transport characteristics along the meridional and zonal axes, the plateau-tropical energy exchange driven by ENSO often manifests as a directionally consistent structure that evolves with altitude. However, existing technologies lack modeling methods for directional consistency and spatial offset in ENSO response diagnosis, resulting in the smoothing of the true morphology of the anomalous field, blurring the physical boundaries between strongly coupled and weakly coupled regions, and the conventional single-time-point static determination cannot characterize the evolutionary stability of the ENSO response on an interannual timescale, nor can it identify the synergistic or competitive relationships between different meteorological elements at the same anomalous center. This makes ENSO response correction susceptible to noise, time drift, and localized sporadic interference. Summary of the Invention

[0004] The purpose of this invention is to propose an ENSO response correction method for the zonal anomaly field of the Tibetan Plateau-Tropical Indian Ocean, in order to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create conditions.

[0005] To achieve the above objectives, according to one aspect of the present invention, a method for correcting the ENSO response of the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field is provided, the method comprising the following steps: S100, calculate the strong and weak annual average temperature fields and the strong and weak annual average fields of meteorological elements at multiple altitudes in the Tibetan Plateau-Tropical Indian Ocean region respectively. S200, subtract the long-term average value from each element to obtain the outlier field and perform a significance test; S300 uses meridional and zonal gradient operators to extract the spatial offset and morphological orientation features of the anomaly field; S400, perform a statistical test for the significance of abnormal directional distributions; S500, combined with multi-layer circulation element direction for coupled analysis; S600 performs corrections to the ENSO response based on the results of coupling analysis; The method of coupling analysis by combining the directions of multi-layer meteorological elements is to calculate the directional atomic contribution of each connected domain based on the spatial offset and the direction of the principal axis of morphology; and to integrate the directional atomic contributions obtained from each element into a grid coupling value DCI, and to determine whether the grid point is strongly coupled or weakly coupled based on the grid coupling value DCI.

[0006] Further, in step S100, the method for calculating the strong and weak annual average temperature fields and the strong and weak annual average fields of meteorological elements at multiple altitudes in the Tibetan Plateau-Tropical Indian Ocean region is as follows: the strong and weak annual average temperature data collection range is 200-500 hPa. Several years of 500–200 hPa temperature fields, wind fields, divergence, and geopotential heights at 200 hPa, 500 hPa, and 850 hPa, and the 500 hPa potential vortex field are obtained from ERA5 monthly reanalysis data. Summer (June-August) average temperatures are calculated for both the Tibetan Plateau (TP) and the Tropical Indian Ocean (TIO). Years are classified according to the strength of the meridional temperature gradient (MTG) index in the upper troposphere. Years with an MTG index greater than or equal to one standard deviation are classified as strong years; otherwise, they are classified as weak years. The middle and upper-level average temperature fields and the composite average fields of meteorological elements at multiple altitudes corresponding to strong and weak years are then calculated. The MTG index is calculated as follows: MTG = Formula (I).

[0007] The MTG index is the meridional temperature gradient index of the upper troposphere. TIO and TIO represent the Tibetan Plateau (75°E-103°E, 28°N-38°N) and the tropical Indian Ocean (60°E-100°E, 15°S-5°N), respectively. This represents the average temperature at 500-200 hPa.

[0008] Further, in step S200, the method for obtaining the anomalous field by subtracting the long-term average of each element and performing a significance test is as follows: the anomalous field is obtained by subtracting the long-term average of the corresponding elements from the strong and weak year average fields of the obtained temperature field and the average fields of meteorological elements at multiple altitude levels; and the anomalous field is subjected to one or more of the following methods: t-test, bootstrap resampling or Monte Carlo simulation, and anomalous grid points that are significant at a preset confidence level are selected.

[0009] The significance test here is used to identify statistically reliable spatial anomalies. In anomaly field calculations, the value of each grid point is the difference between the multi-year composite field and the long-term mean field. However, due to natural variations in climate itself, mainly interannual and intra-seasonal fluctuations, and the uncertainty of reanalysis data, outliers at grid points may be random fluctuations rather than genuine MTG anomaly responses. Therefore, the purpose of the significance test is to determine whether the outlier at each grid point exceeds the range of such natural fluctuations, i.e., to express statistical reliability. Ensuring spatially reliable anomaly signals is a crucial support for subsequent technical analysis and classification model training.

[0010] Furthermore, in step S300, the method for extracting the spatial offset and morphological direction features of the anomalous field using meridional and latitudinal gradient operators is as follows: the anomalous field is filled with boundaries, and the anomalous field is spatially differentiated using meridional and latitudinal gradient operators to generate gradient magnitude and direction fields. Invalid grid points in the anomalous field are filtered out and anomalous connected regions are identified. For each anomalous connected region, the gradient directions are aggregated and smoothed within a set sliding window, and the spatial offset features of the anomalous center and its dominant morphological direction are extracted.

[0011] If only ERA5 is used, interpolation is not required; the gradient magnitude refers to the magnitude of the calculated gradient vector at that grid point, i.e., the maximum spatial rate of change of the scalar field locally. The gradient vector is the maximum rate of increase in all spatial directions at any location in any scalar field, and its direction is the direction of the fastest change in the field. Its magnitude is the gradient magnitude, representing the rate of change in the direction of the fastest change in the field. It is used to calibrate significantly changing anomalous structures, thereby helping to identify important morphologies and migration characteristics of anomalous fields. The gradient magnitude is calculated as |▽f(x,y)|=sqrt((∂f / ∂x)). 2 +(∂f / ∂y) 2 In the formula, x and y are coordinate values, and ∂f / ∂x and ∂f / ∂y represent the first-order partial derivatives in the longitudinal (x-axis) and latitudinal (y-axis) directions, respectively.

[0012] Further, in step S400, the method for conducting a statistical test on the significance of abnormal directional distribution is as follows: the directional concentration is tested for each abnormal connected component using circular statistical methods, selecting either the Rayleigh test or the Watson test; robustness is then assessed using bootstrap resampling; when the confidence level reaches a preset threshold and passes multiple test correction, the directional feature is confirmed to be valid; otherwise, it is marked as invalid and the abnormal connected component is removed.

[0013] The purpose of this step is to ensure that the directional features are not caused by random fluctuations, eliminate false directions caused by random fluctuations, and ensure that the directional features originate from the true MTG-ENSO signal. False positives are controlled through significance and multiple corrections, reducing bias in downstream mechanism interpretation and prediction.

[0014] Furthermore, in step S500, the method for calculating the directional atom contribution for each connected domain based on spatial offset and morphological principal axis direction is as follows: select the temperature anomaly connected domain that overlaps with the grid point or is closest to the center, and calculate the morphological direction consistency and offset direction consistency respectively. Perform distance transformation on the grid points in the study domain to obtain their geographical distance to the boundary of the abnormal connected domain. Define the domain influence kernel of grid points within the abnormal connected domain as 1, and the domain influence kernel of grid points outside the abnormal connected domain as monotonically decreasing with geographical distance. Rayleigh aggregation is performed on samples of the same element in the principal axis direction at different height layers to obtain vertical consistency. If the vertical consistency is lower than a preset threshold, the corresponding grid point is masked so that its vertical gate value is 0; otherwise, it is 1. The directional atom contribution of each grid point is calculated using the vertical gate value vm, the domain influence kernel ws, the morphological direction consistency s1, and the offset direction consistency s2: atom=vm×ws×[α·s1+(1-α)·s2]; where α is the consistency bias weight, with a value range of [0.2,0.8], and a default of 0.5; the directional atom contribution is a grid-level parameter, and its smallest unit of affiliation includes features, year categories, abnormal connected components, and grid points. The directional atom contribution of each grid point in the abnormal connected component is obtained respectively.

[0015] Further, in step S500, the method for integrating the directional atomic contributions obtained from each element into a grid coupling value DCI is as follows: a quadruple consisting of year category, element, abnormal connected domain and grid point is used as the analysis unit, and the maximum value of the directional atomic contribution in different elements of an analysis unit is used as the representative contribution; quadruples with a vertical gating value of 1 and a domain influence kernel greater than 0 are selected from all analysis units to form an effective atom set; robust aggregation is performed on any element in the effective atom set based on year category and grid point to obtain the grid coupling value DCI, and the robust aggregation method is truncated mean.

[0016] Further, in step S500, the method for determining whether a grid point is strongly or weakly coupled based on the grid point coupling value (DCI) is as follows: An upper and lower threshold are preset for the grid point coupling value; the binary pair formed by the year category and the grid point is used as the principal analytical element (MAE). If the number of elements participating in the grid point coupling value calculation of the MAE is less than 2, it is judged as weakly coupled; otherwise, if the grid point coupling value of the MAE exceeds the upper threshold, it is judged as strongly coupled; if the grid point coupling value of the MAE is lower than the lower threshold, it is judged as weakly coupled; if the grid point coupling value of the MAE is between the upper and lower thresholds, the label is determined using the 3×3 neighborhood majority consensus rule. However, judging the degree of coupling directly from the perspective of DCI is essentially a single-point-in-time determination, which is difficult to adapt to interannual drift and local multimodal distribution. It is prone to sensitivity to short-term noise and salt-and-pepper noise at spatial boundaries. Therefore, it is impossible to enforce constraints based on temporal and spatial consistency, which leads to the risk of overfitting to accidental noise in step S600 and causes overcorrection.

[0017] Further, in step S500, the method for determining whether a grid point is strongly coupled or weakly coupled based on the grid point coupling value (DCI) is as follows: the binary pair consisting of the year category and the grid point is used as the principal analytical element; for any principal analytical element, the persistence factor within its backtesting window is obtained through the CUSUM change point detection operator, and the trend factor within its backtesting window is obtained through the EWMA exponential moving average operator; years in which both the persistence factor and the trend factor within the backtesting window are greater than their corresponding internal mean are judged as having strong temporal coupling, otherwise they are judged as having weak temporal coupling; the corresponding strong / weak temporal coupling labels and grid point coupling values ​​for each year in the backtesting window are input into the HMM model to obtain the posterior strong coupling probability for each year; if the posterior strong coupling probability of a weakly coupled year exceeds a threshold, it is relabeled as having strong temporal coupling; the labels for strong and weak temporal coupling are stored as temporal labels. Since DCI is a single-year indicator, it is susceptible to random disturbances and observation noise. Direct threshold determination leads to interannual label jitter and boundary misjudgment. Therefore, the high persistence transition of HMM is used to characterize interannual stability, and the Beta emission distribution is used to absorb the amplitude difference between different years. The output posterior strong coupling probability can form an interpretable probabilistic label, thereby achieving the ability to resist ENSO phase drift through adaptive threshold and time smoothing, reduce false alarms and false negatives, and provide conservative time dimension constraints for subsequent fusion with clustering channels.

[0018] Construct feature vectors for each year within the backtesting window. The feature vectors include the DCI mean, DCI quantiles, morphological consistency quantiles, and offset consistency quantiles. Perform unsupervised clustering on all grid points of the same year category to obtain several clusters, which are denoted as coupled clusters. Map the clusters to cluster labels according to the ordering mapping rules of the DCI mean within the cluster. The cluster labels include strong coupling and weak coupling. If the temporal label of the principal analyte is the same as the cluster label, its coupling label directly inherits the corresponding strong and weak coupling classification. If the temporal label conflicts with the cluster label, the neighborhood majority consensus judgment is adopted. Compared to previous grid-based determination methods using static threshold rules, this method expands from single-year stationary states to dynamic temporal coupling by introducing temporal features and implicit state inference mechanisms. Its mathematical core lies in treating the temporal evolution of grid-based coupling values ​​as a hidden state sequence, using CUSUM and EWMA to extract persistence and trend factors to characterize interannual inertia driven by ENSO, and then establishing a highly persistent transition structure between strong and weak states through a hidden Markov model, thereby maximizing the posterior consistency of the time series in a statistical sense. This process can be viewed as a temporal smoothing mapping of meridional heat transport and upper tropospheric circulation in the ENSO response, naturally suppressing occasional anomalies caused by noise.

[0019] The main purpose of S500 is to establish spatial and morphological dimensions to pinpoint the true heat transport path by extracting directional and offset features, based on the physical laws governing the latitudinal transport of heat and momentum in atmospheric circulation and the natural laws governing the directional consistency of anomalous structures across different altitudes. This is equivalent to establishing a spatial dimension plus a morphological dimension, rather than relying solely on amplitude statistics. The DCI obtained by S500 represents the directional synchronization intensity of multiple elements at that grid point, thereby identifying regions of strong and weak mechanical coupling.

[0020] Because the ENSO-driven meridional temperature gradient response in the upper Tibetan Plateau exhibits significant inertial characteristics, the accumulation and release of energy and momentum in the climate system show a smooth multi-year evolution pattern. Therefore, by using the high-persistence transition matrix of the Hidden Markov Model (HMM) and the Beta emission distribution, the algorithm mathematically fits this natural hysteresis, making the judgment results more consistent with the actual process of heat transport and circulation reconstruction. This significantly improves the temporal robustness and spatial coherence of ENSO response identification, achieving cross-scale correction from static distribution to dynamic coupling, and overcoming the shortcomings of existing technologies in temporal consistency modeling and anomaly identification stability.

[0021] Further, in step S600, the method for correcting the ENSO response based on the coupling analysis results is as follows: the binary pair consisting of the year category and the grid point is used as the principal analysis element; if the coupling analysis result of the principal analysis element is strong coupling, its coupling weight is 1, otherwise it is λ; for any element, the anomaly field and the coupling weight are weighted and synthesized to form a coupled anomaly field; λ∈[0.3,0.5]; When considering the correlation between the ENSO index and any element field, weighted correlation calculation and weighted least squares regression are performed on each grid point using coupled anomaly fields to obtain the corrected correlation coefficient field, regression coefficient field, and lag assessment. Update the ENSO event classification based on the revised statistical results; when classification criteria conflict, the statistics of the strongly coupled region shall be given priority.

[0022] The weighted synthesis of outliers with coupling weights refers to multiplying the outlier of each element at the grid point by the coupling weight of the corresponding principal analytical element to form a weighted outlier. Then, the weighted outliers of all grid points in the entire study domain or selected region are weighted averaged or weighted cumulatively to calculate the synthesized outlier. Furthermore, when calculating the correlation or regression analysis with the ENSO index, the weighted outlier is used to replace the original outlier, and the weight is used as the sample weight for each grid point. The weighted correlation or weighted least squares regression method is used to obtain the weighted correlation coefficient and regression coefficient between the ENSO index and the element field, and lag assessment is further performed based on this. Updating the ENSO event classification based on the revised statistical results means that for each year category, the revised correlation coefficient field, regression coefficient field, and lag evaluation value are used as feature inputs fx for classification analysis. The feature inputs fx include the weighted correlation coefficient, regression coefficient, and best lag for each grid point or region. The existing ENSO event type determination method is compared with or retrained with the above feature fx, with the default threshold method based on Niño3.4 or MEI index. The fx is used to redetermine whether the year is an Eastern, Central, or Mixed ENSO event, and the revised classification results are obtained accordingly.

[0023] Preferably, all undefined variables in this invention, if not explicitly defined, can be manually set thresholds.

[0024] The beneficial effects of this invention are as follows: This invention provides a method for correcting the ENSO response of the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field. This method performs structural coupling analysis on temperature anomalies, wind fields, divergence, geopotential height, and potential vorticity anomalies in both spatial offset and morphological principal axis dimensions through multi-layer meteorological element directional coupling. It also constructs strongly coupled and weakly coupled grid labels, thereby effectively classifying ENSO-driven temperature circulation anomaly patterns. In further research on the relationship between the meridional temperature gradient anomaly and ENSO type responses in the upper Tibetan Plateau-Tropical Indian Ocean, this classification weakens the weighting of cases with low consistency in coupling relationships, making the boundaries between strongly coupled and weakly coupled regions with clear transport paths and consistent directions clearer. The principle behind this is based on the large-scale transport patterns of heat along meridional and zonal tendencies and the directional stability of the upper tropospheric circulation organization, such as the invariance of heat-momentum transport along the principal axis offset. This significantly improves the accuracy of physical processes in anomaly field identification. By coupling labels to weaken noise interference and enhance the real mechanism signal, the ENSO response identification results of the Tibetan Plateau-Tropical Indian Ocean system have more robust and stable classification and quantitative output. Attached Figure Description

[0025] The above and other features of the present invention will become more apparent from the detailed description of the embodiments shown in conjunction with the accompanying drawings. In the accompanying drawings, the same reference numerals denote the same or similar elements. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort. In the drawings: Figure 1 The diagram shows a flowchart of a method for correcting the ENSO response of the zonal anomaly field in the Tibetan Plateau-Tropical Indian Ocean region. Detailed Implementation

[0026] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0027] like Figure 1 The diagram shows a flowchart of a method for correcting the ENSO response of the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field. The following section will combine... Figure 1 This paper describes a method for correcting the ENSO response of the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to an embodiment of the present invention. The method includes the following steps: S100, calculate the strong and weak annual average temperature fields and the strong and weak annual average fields of meteorological elements at multiple altitudes in the Tibetan Plateau-Tropical Indian Ocean region respectively. S200, subtract the long-term average value from each element to obtain the outlier field and perform a significance test; S300 uses meridional and zonal gradient operators to extract the spatial offset and morphological orientation features of the anomaly field; S400, perform a statistical test for the significance of abnormal directional distributions; S500, combined with multi-layer meteorological element direction coupling analysis; S600 performs corrections to the ENSO response based on the results of coupling analysis; The method of coupling analysis by combining the directions of multi-layer meteorological elements is to calculate the directional atomic contribution of each connected domain based on the spatial offset and the direction of the principal axis of morphology; and to integrate the directional atomic contributions obtained from each element into a grid coupling value DCI, and to determine whether the grid point is strongly coupled or weakly coupled based on the grid coupling value DCI.

[0028] Further, in step S100, the method for calculating the strong and weak annual average temperature fields and the strong and weak annual average fields of meteorological elements at multiple altitudes in the Tibetan Plateau-Tropical Indian Ocean region is as follows: the range of atmospheric layers for collecting strong and weak annual average temperatures is 200-500 hPa. The 500-200 hPa temperature field, wind field, divergence, geopotential height, and 500 hPa potential vortex field at 200 hPa, 500 hPa, and 850 hPa are obtained from ERA5 reanalysis data. The summer average temperature is calculated for the Tibetan Plateau region and the tropical Indian Ocean region respectively. The years are divided according to the strength of the upper meridional temperature gradient MTG index, and the average temperature fields and the annual average fields of meteorological elements at multiple altitudes corresponding to strong and weak years are obtained respectively. The data time span is from 1950 to 2024 by default. Summer refers to June to August. The latitude and longitude range of the Qinghai-Tibet Plateau region is (60E–100E, 20N–40N), and the tropical Indian Ocean region is (60E–100E, 10S–10N). The process of calculating the summer mean temperature for the Tibetan Plateau region and the tropical Indian Ocean region is as follows: a vertical weighted average is performed over the 500–200 hPa air layer range to obtain the temperature field of the upper troposphere. The summer mean temperatures of the Tibetan Plateau region and the tropical Indian Ocean region are selected, and the difference between the two is calculated to construct the upper meridional temperature gradient (MTG) index.

[0029] The process of classifying years based on the strength of the upper meridional temperature gradient MTG index is as follows: according to the time series characteristics of the MTG index, the standard deviation threshold method or the threshold after Z-score normalization is used to classify the years into MTG strong years and MTG weak years. The process of obtaining the temperature fields corresponding to strong and weak years is as follows: for the identified strong or weak MTG years, the 500–200 hPa temperature fields are collected by year. In each year, the summer temperature of the selected region, the Tibetan Plateau region or the tropical Indian Ocean region is averaged by grid point over many years to obtain the average temperature fields of strong and weak years, which are used to reflect the spatial distribution characteristics of the MTG anomaly pattern. The process of obtaining the annual mean field of meteorological elements at multiple altitudes is as follows: For the identified strong or weak MTG years, the horizontal wind, divergence, geopotential height at the 200hPa, 500hPa, and 850hPa altitudes, as well as the 500hPa potential vortex field, are accumulated year by year according to the set of strong and weak years and averaged to obtain the multi-year mean field of each element in strong and weak years, thereby revealing the atmospheric circulation structure and its variation characteristics under different MTG anomaly states. Horizontal wind, divergence, geopotential height, vortex field, and temperature are all elements; grid points are spatial coordinates containing corresponding longitude, latitude, and pressure layer information, with each grid point possessing its own corresponding spatial element. Strong and weak years are abbreviations for strong MTG years and weak MTG years.

[0030] Further, in step S200, the method for obtaining the anomalous field by subtracting the long-term average of each element and performing a significance test is as follows: the anomalous field is obtained by subtracting the long-term average of the corresponding elements from the strong and weak year average fields of the obtained temperature field and the average fields of meteorological elements at multiple altitude levels; and the anomalous field is subjected to one or more of the following methods: t-test, bootstrap resampling or Monte Carlo simulation, and anomalous grid points that are significant at a preset confidence level are selected.

[0031] The process of calculating the long-term average value of each element, including the average temperature fields of strong and weak years and meteorological elements at multiple altitudes obtained from the S100 step, is as follows: for any element, the average value of each grid point is calculated point by point over the data time span, i.e., from 1950 to 2024. The set of average values ​​obtained from each grid point constitutes the corresponding long-term average field.

[0032] Subtracting the corresponding grid points of the strong or weak annual mean field from the long-term mean field yields the anomaly field reflecting the spatial distribution under MTG anomaly conditions. There is a one-to-one correspondence between the grid points of the strong / weak annual mean field and the long-term mean field.

[0033] The outlier field is subjected to significance tests, including t-tests, to determine whether the difference between the grid point mean and the long-term mean is significant. The preset confidence level is 95%, with a range of ±4%. Grid points exceeding the preset confidence level are considered outliers. The outlier field remaining after filtering out outliers that are significant at the preset confidence level is considered a significant outlier field.

[0034] Furthermore, in step S300, the method for extracting the spatial offset and morphological direction features of the anomalous field using meridional and latitudinal gradient operators is as follows: the anomalous field is meshed and its boundaries are filled; the anomalous field is spatially differentiated using meridional and latitudinal gradient operators to generate gradient magnitude and direction fields; invalid grid points in the anomalous field are filtered out and anomalous connected regions are identified; for each anomalous connected region, the gradient directions are aggregated and smoothed within a set sliding window, and the spatial offset features of the anomalous center and its dominant morphological direction are extracted.

[0035] The process of mesh unification and boundary filling for the anomalous field is as follows: the anomalous field is reprojected onto a uniform 0.25°×0.25° latitude and longitude grid, bilinear remapping is used, and the time axis and variable dimensions are aligned before resampling; then periodic boundary processing is used in the meridional direction, and mirror filling or nearest neighbor extrapolation is used in the latitudinal direction and for missing measurements, supplemented by IDW local void interpolation, in order to maintain the continuity of gradient and morphology without changing the climatology of the field.

[0036] The process of generating gradient magnitude and direction fields by spatial differentiation of the anomalous field using meridional and latitudinal gradient operators is as follows: The anomalous field is spatially differentiated at the grid level by applying discrete gradient operators in the latitudinal and meridional directions, preferentially using central difference or Sobel operators, with one-sided difference used at the boundaries to connect with the aforementioned filling results; subsequently, the obtained two-directional gradient components are Gaussian or median smoothed to suppress isolated noise, and angle normalization and phase unwrapping are performed to obtain a continuous direction angle field. Each grid point generates a gradient magnitude field and a direction field, and the direction field is mapped to discrete direction indicators of north, east, south, and west according to quadrants, while retaining continuous azimuth angles for subsequent sliding window aggregation and principal axis extraction; when the gradient magnitude is lower than a preset threshold or the direction variance is higher than a threshold, the grid point is marked as invalid to avoid noise interference.

[0037] The process of aggregating and smoothing gradient directions within a defined sliding window to extract the spatial offset features and dominant morphological orientation of anomaly centers involves setting a fixed-scale sliding window on a significant anomaly field. The default window size is 5×5 pixels, or approximately 1°×1°. Gradient directions within the window are then subjected to circular statistical weighted averaging based on their gradient magnitudes to obtain the resulting vector and its average orientation. Orientation concentration and variance are used as confidence levels. Low-confidence windows are marked as invalid, and orientation-preserving smoothing methods are employed to eliminate isolated noise and retain the continuity of dominant orientations. Orientation-preserving smoothing methods can use any of the following: orientation median, anisotropic diffusion, or bilateral filtering. Orientation concentration refers to the normalized composite vector length obtained by performing axial circular statistics on orientation angle samples that have passed the validity screening within the sliding window; its value ranges from [0,1]. Anomalous connected components are identified using general gradient and connected component methods for anomaly fields. The mass center of the connected components excluded by the low-confidence window is taken as the anomaly center of the anomalous connected components, and the corresponding center of the long-term mean field is taken as the reference center. The displacement vectors of the two in latitude and longitude are taken as the spatial offset. For anomalous connected components, the first principal axis azimuth angle is obtained using existing structural tensor or second-order moment methods, and this azimuth angle is defined as the dominant morphological direction. When the anisotropy is below a preset threshold, the dominant morphological direction is not output.

[0038] In this invention's orientation-preserving smoothing of the Tibetan Plateau-Tropical Indian Ocean upper-level anomaly field, the primary objective is to preserve the continuity of the ENSO-driven meridional transport ridges and troughs, as well as the jet stream axis, while also considering the noise characteristics of the reanalysis data and computational constraints. When the orientation concentration is high and there is a significant anisotropic structure, anisotropic diffusion is preferentially used to diffuse along the main direction and suppress cross-directional diffusion, avoiding the smoothing of the dominant phase. When grid-level random noise accounts for a high proportion, the structure is already discernible, and rapid and robust noise reduction is required, directional median is used to cyclically and statistically suppress outliers and maintain azimuth consistency. When a compromise between edge preservation and noise reduction is needed, and the contrast of the regional boundaries is moderate, bilateral filtering is used to simultaneously utilize spatial proximity and orientation differences. The above selections are adaptively switched or applied in partitions based on the gradient magnitude quantiles, orientation concentration thresholds, and dynamic sensitivity of the target elements calculated in this invention, ensuring that the dominant morphology and phase information of the MTG anomaly are not destroyed.

[0039] The confidence level is determined using directional concentration and directional variance as confidence indices. The process for setting a low-confidence window is as follows: directional concentration is defined as the normalized length of the weighted composite vector, ranging from [0,1], with larger values ​​indicating more consistent directions; directional variance is defined as 1 minus directional concentration, also ranging from [0,1], with larger values ​​indicating higher dispersion. The confidence threshold is preset as follows: a low-confidence window is identified and marked as invalid when directional concentration < 0.5 or directional variance > 0.5. Furthermore, in small samples (i.e., fewer than 9 effective pixels) or when the average gradient magnitude of the window is lower than the weak gradient at the P25 quantile of the entire field, the threshold is raised to directional concentration < 0.6 or directional variance > 0.4 to reduce the risk of misjudgment. These thresholds can be calibrated within the range of [0.4, 0.7] based on resolution and climatic characteristics.

[0040] The above definitions are all implemented using existing technical standards, directly producing orientation and displacement features that can be used for subsequent coupling and classification. This step outputs spatial offset features defined by the displacement of the center of mass and dominant morphological orientation defined by the first principal axis of the structural tensor for significant anomalous connected regions in strong and weak years; when anisotropy is below a threshold, the dominant morphological orientation is not output.

[0041] Further, in step S400, the method for conducting a statistical test on the significance of abnormal directional distribution is as follows: the directional concentration is tested for each abnormal connected component using circular statistical methods, selecting either the Rayleigh test or the Watson test; robustness is then assessed using bootstrap resampling; when the confidence level reaches a preset threshold and passes multiple test correction, the directional feature is confirmed to be valid; otherwise, it is marked as invalid and the abnormal connected component is removed.

[0042] Furthermore, in step S500, the method for calculating the directional atom contribution for each connected domain based on spatial offset and morphological principal axis direction is as follows: select the temperature anomaly connected domain that overlaps with the grid point or is closest to the center, and calculate the morphological direction consistency and offset direction consistency respectively. The morphological direction consistency is defined as: s1 = axis-sim(θ(E,Y,i),θ_T); the offset direction consistency is defined as: s2 = axis-sim(angle(Δ(E,Y,i)), angle(Δ_T)); where E is an element, referring to an element in the set of meteorological elements participating in the coupled analysis, used for directional characterization, including: Temperature Anomaly, Wind200 / 500 / 850 Horizontal Wind, Div200 / 500 / 850 Divergence, Z200 / 500 / 850 Geopotential Height, PV500 Potential Vortex; Y is the strength category of the year, i is the anomaly connected domain number or index; θ(E,Y,i) represents the principal stretching direction of the corresponding connected domain morphology, which is essentially an angle; θ_T is the first principal axis azimuth angle of the temperature anomaly connected domain that overlaps with or is closest to the target grid point under the year category Y of the temperature element; Δ(E,Y,i) is the angle from element E The reference center is the displacement vector pointing to the mass center of the connected domain; Δ_T is the spatial offset vector of the temperature anomaly connected domain that overlaps with or is closest to the target grid point under the year category Y of the temperature element; angle() is the vector axis angle operator, which obtains its axial azimuth angle by calling the two-dimensional displacement vector; axis-sim(, ) is the axial consistency measurement operator, which obtains the consistency score by calling the two axis angles, with values ​​between [0,1], where axis refers to the undirected geometric direction, that is, 0° and 180° are considered to be the same direction, and sim() is the similarity operator. In the direction coupling module of this invention, axis-sim() is fixedly used as the specific implementation of the similarity function to ensure axial equivalence and implementation determinism.

[0043] Distance transformation is performed on the grid points in the study domain to obtain their geographic distance to the boundary of the abnormal connected domain. The domain influence kernel of the grid points within the abnormal connected domain is defined as 1, and the domain influence kernel of the grid points outside the abnormal connected domain decays monotonically with geographic distance. By default, a Gaussian kernel is used as the spatial scale parameter. The domain influence kernel of the grid points outside the domain is calculated by selecting the abnormal connected domain with the minimum geographic distance. The preferred geographic distance is either a great circle or a Vincenty distance; the distance between the grid points outside the domain decreases monotonically with respect to the boundary, and the decay kernel is selected from Gaussian kernel, exponential kernel or linear kernel, with a spatial scale parameter of r0∈[100,200]km; this process does not change the geometric position of the original abnormal connected domain, but only provides continuous and differentiable spatial weights for the subsequent robust projection of domain-level directional information to the grid level; Rayleigh aggregation is performed on samples of the same feature along the principal axis at different height layers to obtain vertical consistency. If the vertical consistency is lower than a preset threshold, the corresponding grid point is masked and its vertical gate value is set to 0; otherwise, it is set to 1. Rayleigh aggregation uses the length of the Rayleigh composite vector as the vertical consistency, and its preset threshold is 0.6 by default. The directional atom contribution of each grid point is calculated using the vertical gate value vm, the domain influence kernel ws, the morphological direction consistency s1, and the offset direction consistency s2: atom=vm×ws×[α·s1+(1-α)·s2]; where α is the consistency bias weight, with a value range of [0.2,0.8], and a default of 0.5; the directional atom contribution is a grid-level parameter, and its smallest unit of affiliation includes features, year categories, abnormal connected components, and grid points. The directional atom contribution of each grid point in the abnormal connected component is obtained respectively.

[0044] Further, in step S500, the method for integrating the directional atomic contributions obtained from each element into a grid coupling value DCI is as follows: a quadruple consisting of year category, element, abnormal connected domain and grid point is used as the analysis unit, and the maximum value of the directional atomic contribution in different elements of an analysis unit is used as the representative contribution; quadruples with a vertical gating value of 1 and a domain influence kernel greater than 0 are selected from all analysis units to form an effective atom set; robust aggregation is performed on any element in the effective atom set based on year category and grid point to obtain the grid coupling value DCI, and the robust aggregation method is truncated mean.

[0045] When selecting quadruples from all analysis units that have a vertical gate value of 1 and a domain influence kernel greater than 0, the year category and grid point are used as the smallest analysis unit. That is, any quadruples with the same year category and grid point must contain at least two quadruples with a vertical gate value of 1 and a domain influence kernel greater than 0. Robust aggregation uses a truncated mean, removing the top and bottom 20% of extreme values ​​by default before averaging.

[0046] Further, in step S500, the method for determining whether a grid point is strongly or weakly coupled based on the grid point coupling value (DCI) is as follows: An upper and lower threshold are preset for the grid point coupling value; the binary pair formed by the year category and the grid point is used as the principal analytical element (MAE). If the number of elements participating in the grid point coupling value calculation of the MAE is less than 2, it is judged as weakly coupled; otherwise, if the grid point coupling value of the MAE exceeds the upper threshold, it is judged as strongly coupled; if the grid point coupling value of the MAE is lower than the lower threshold, it is judged as weakly coupled; if the grid point coupling value of the MAE is between the upper and lower thresholds, the label is determined using the 3×3 neighborhood majority consensus rule. The upper and lower threshold values ​​for grid coupling are preset to 0.7 and 0.5 by default, respectively. In addition to the default values, the difference between the upper and lower threshold values ​​must be greater than 0.2. At the same time, the upper and lower threshold values ​​are calibrated within the ranges of [0.65, 0.80] and [0.45, 0.60], respectively. The 3×3 neighborhood majority consensus rule determines whether the majority in the 8 neighborhoods is strongly coupled or weakly coupled based on the pre-condition rule. If the number of strong couplings is greater than that of weak couplings, the principal analyte is classified as strongly coupled, and vice versa. If the number is the same, it is classified as weakly coupled.

[0047] When the temperature reference domain is missing, the grid point is directly judged as weakly coupled.

[0048] Further, in step S500, the method for determining whether a grid point is strongly coupled or weakly coupled based on the grid point coupling value (DCI) is as follows: the binary pair consisting of the year category and the grid point is used as the principal analytical element; for any principal analytical element, the persistence factor within its backtesting window is obtained through the CUSUM change point detection operator, and the trend factor within its backtesting window is obtained through the EWMA exponential moving average operator; years in which both the persistence factor and the trend factor within the backtesting window are greater than their corresponding internal mean are judged as having strong temporal coupling, otherwise they are judged as having weak temporal coupling; the corresponding strong / weak temporal coupling labels and grid point coupling values ​​for each year in the backtesting window are input into the HMM model to obtain the posterior strong coupling probability for each year; if the posterior strong coupling probability of a weakly coupled year exceeds a threshold, it is relabeled as having strong temporal coupling; the labels for strong and weak temporal coupling are stored as temporal labels. The year range of the backtest window is [5,10] years, and the analysis unit whose year category is consistent with the main analysis element is selected; the CUSUM variable point detection operator and the EWMA exponential moving average operator are applied to the coupling values ​​of each grid point in the backtest window, and the results are normalized to [0,1] in the backtest window; Because of the year category distinction within the backtesting window, there are data gaps in the corresponding DCI backtesting windows under the strong and weak year categories. Therefore, when there are missing DCI data, it is necessary to synthesize using a phase-weighted time window: the weight wf for the same phase year is set to 1, and the weight wf for the opposite phase year is set to 0.1-0.4. An exponential time decay weight wtd is established by applying exponential time decay according to the year distance. The missing DCI input of the CUSUM and EWMA functions is supplemented by the weighted quantile and the resulting weighted mean (the product of wf and wd). The same and opposite phase refers to whether it has the same label in the year category as the current principal analyte. When the weighted effective sample size is less than 5, the ±1 year event alignment synthesis with the ENSO event peak year as the anchor is used as an alternative input.

[0049] A year in which both the persistence factor and the trend factor within the backtesting window are greater than their corresponding internal mean is defined as a year in which the persistence factor is greater than the average of all persistence factors within the backtesting window, and the trend factor and the average of all trend factors are also greater. In this case, the year is considered to be strongly coupled in time series; otherwise, it is considered to be weakly coupled in time series. The strong and weak coupling labels for time series are the judgment results of weak and strong coupling for each year obtained within the backtesting window. The HMM model refers to the Hidden Markov Model. In the process of obtaining the posterior strong coupling probability, the grid coupling value DCI of each year within the backtesting window is input, and the hidden states are {strong coupling, weak coupling}. Let the initial state distribution and the high-persistence transition matrix A (pSS, pWW∈[0.8,0.95]) be set. The observed emission distribution is selected as the Beta distribution matching the value range [0,1], which corresponds to the parameters of the strong and weak states, respectively. The parameters are estimated by the DCI quantiles within the backtesting window or the previously obtained strong and weak coupling labels. The posterior strong coupling probability of each year is obtained through the forward-backward algorithm, and an adaptive threshold Tadapt is set with a default value of 0.65. Years that are initially judged as weak coupling but have a posterior strong coupling probability greater than Tadapt are re-labeled as strong coupling, thus obtaining the final strong and weak coupling labels for time series within the window.

[0050] Construct feature vectors for each year within the backtesting window. The feature vectors include the DCI mean, DCI quantiles, morphological consistency quantiles, and offset consistency quantiles. Perform unsupervised clustering on all grid points of the same year category to obtain several clusters, which are denoted as coupled clusters. Map the clusters to cluster labels according to the ordering mapping rules of the DCI mean within the cluster. The cluster labels include strong coupling and weak coupling. The DCI mean refers to the DCI mean of each element in the backtesting window. The DCI quantile is the percentile value of the DCI corresponding to the principal analyte in the backtesting window. The morphological consistency quantile is the percentile of its morphological consistency in the backtesting window. The offset consistency quantile is the percentile of its offset consistency in the backtesting window. The unsupervised clustering uses the GMM algorithm. The ordering mapping rule means that the coupled clusters are arranged in descending order according to the mean DCI of all elements in the cluster. If the number of coupled clusters is 2, the cluster with the maximum mean DCI is labeled as strongly coupled, and the other is labeled as weakly coupled. If the number of clusters is greater than 2, the Jenks natural break method is used to classify each coupled cluster into strongly coupled and weakly coupled clusters. If the temporal label of the principal analyte is the same as the cluster label, its coupling label directly inherits the corresponding strong and weak coupling classification. If the temporal label conflicts with the cluster label, the neighborhood majority consensus judgment is adopted. The same time series label and cluster label means that both are either strongly coupled or weakly coupled; neighborhood majority consensus means that, based on the preconditions, the majority of the 8 neighborhoods is determined to be strongly coupled or weakly coupled. If the number of strong couplings is greater than that of weak couplings, the principal analyte is determined to be strongly coupled, and vice versa. If the number is the same, it is determined to be weakly coupled.

[0051] Further, in step S600, the method for correcting the ENSO response based on the coupling analysis results is as follows: the binary pair consisting of the year category and the grid point is used as the principal analysis element; if the coupling analysis result of the principal analysis element is strong coupling, its coupling weight is 1, otherwise it is λ; for any element, the anomaly field and the coupling weight are weighted and synthesized to form a coupled anomaly field; λ∈[0.3,0.5]; The process of weighted summation of outliers and coupling weights uses year category and grid point as indexes. For any element at each grid point, its outlier is multiplied by the coupling weight corresponding to that grid point to obtain the weighted outlier. When performing spatiotemporal synthesis such as regional, seasonal, or strong / weak year synthesis, a weighted average with normalized weights is used. That is, the coupling weight of each grid point is used as the unique weight, and the weighted average of the weighted outliers is calculated and normalized by weight. When used for statistical methods such as correlation, the coupling weight is also used as the sample weight to participate in weighted correlation or weighted least squares calculations. Weakly coupled grid points only participate by reducing their weights without changing their spatial location and original outlier. The weight of missing grid points is regarded as zero and does not participate in the synthesis.

[0052] When calculating the correlation between the ENSO index and any element field, weighted correlation calculation and weighted least squares regression are performed on each grid point using coupled anomaly fields to obtain the corrected correlation coefficient field, regression coefficient field and lag assessment. Update the ENSO event classification based on the revised statistical results; when classification criteria conflict, the statistics of the strongly coupled region shall be given priority.

[0053] Strongly coupled grid points are fully preserved (coupling weight is 1), while weakly coupled grid points are suppressed by λ. This amplifies the response with consistent direction and suppresses the noise with inconsistent direction in the synthesis and statistical stages, thereby completing the quantitative correction of the ENSO response.

[0054] Although the invention has been described in considerable detail and particularly with regard to several of the described embodiments, it is not intended to limit itself to any of these details or embodiments or any particular embodiment, thereby effectively covering the intended scope of the invention. Furthermore, the invention has been described above with respect to embodiments foreseeable by the inventors in order to provide a useful description, and non-substantial modifications to the invention that have not yet been foreseen may still represent equivalent modifications.

Claims

1. A method for correcting the ENSO response of the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field, characterized in that, The method includes the following steps: S100, calculate the strong and weak annual average temperature fields and the strong and weak annual average fields of meteorological elements at multiple altitudes in the Tibetan Plateau-Tropical Indian Ocean region respectively. S200, subtract the long-term average value from each element to obtain the outlier field and perform a significance test; S300 uses meridional and zonal gradient operators to extract the spatial offset and morphological orientation features of the anomaly field; S400, perform a statistical test for the significance of abnormal directional distributions; S500, combined with multi-layer meteorological element direction coupling analysis; S600 performs corrections to the ENSO response based on the results of coupling analysis; The method of coupling analysis by combining the directions of multi-layer meteorological elements is to calculate the directional atomic contribution of each connected domain based on the spatial offset and the direction of the principal axis of morphology; and to integrate the directional atomic contributions obtained from each element into a grid coupling value DCI, and to determine whether the grid point is strongly coupled or weakly coupled based on the grid coupling value DCI. The method for determining whether a grid point is strongly or weakly coupled based on the grid point coupling value (DCI) is as follows: The binary pair consisting of the year category and the grid point is used as the principal analytical unit (PMU). For any PMU, the persistence factor within its backtesting window is obtained using the CUSUM change point detection operator, and the trend factor within its backtesting window is obtained using the EWMA exponential moving average operator. Years in which both the persistence factor and the trend factor within the backtesting window are greater than their corresponding internal mean are considered strongly coupled in time series; otherwise, they are considered weakly coupled in time series. The corresponding strong / weak coupling labels and grid point coupling values ​​for each year in the backtesting window are input into the HMM model to obtain the posterior strong coupling probability for each year. If the posterior strong coupling probability of a weakly coupled year exceeds a threshold, it is relabeled as strongly coupled in time series. The labels for strongly and weakly coupled time series are stored as time series labels. Construct feature vectors for each year within the backtesting window. The feature vectors include the DCI mean, DCI quantiles, morphological consistency quantiles, and offset consistency quantiles. Perform unsupervised clustering on all grid points of the same year category to obtain several clusters, which are denoted as coupled clusters. Map the clusters to cluster labels according to the ordering mapping rules of the DCI mean within the cluster. The cluster labels include strong coupling and weak coupling. If the temporal label of the principal analyte is the same as the cluster label, its coupling label directly inherits the corresponding strong or weak coupling classification. If the temporal label conflicts with the cluster label, the neighborhood majority consensus judgment is used.

2. The ENSO response correction method for the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to claim 1, characterized in that, In step S100, the method for calculating the strong and weak annual average temperature fields and the strong and weak annual average fields of meteorological elements at multiple altitudes in the Tibetan Plateau-Tropical Indian Ocean region is as follows: the range of atmospheric layers for collecting strong and weak annual average temperatures is 200-500 hPa. The 500-200 hPa temperature field, wind field, divergence, geopotential height, and 500 hPa potential vortex field at 200 hPa, 500 hPa, and 850 hPa are obtained from ERA5 reanalysis data. The summer average temperature is calculated for the Tibetan Plateau region and the tropical Indian Ocean region respectively. The years are divided according to the strength of the upper meridional temperature gradient MTG index, and the average temperature fields and the annual average fields of meteorological elements at multiple altitudes corresponding to strong and weak years are obtained respectively.

3. The ENSO response correction method for the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to claim 1, characterized in that, In step S200, the method for obtaining the anomalous field by subtracting the long-term average of each element and performing a significance test is as follows: the anomalous field is obtained by subtracting the long-term average of the corresponding elements from the strong and weak year average fields of the obtained temperature field and the average fields of meteorological elements at multiple altitude levels; and the anomalous field is subjected to one or more of the following methods: t-test, bootstrap resampling or Monte Carlo simulation, and anomalous grid points that are significant at a preset confidence level are selected.

4. The ENSO response correction method for the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to claim 1, characterized in that, In step S300, the method for extracting the spatial offset and morphological orientation features of the anomalous field using meridional and latitudinal gradient operators is as follows: the anomalous field is meshed and its boundaries are filled; the anomalous field is spatially differentiated using meridional and latitudinal gradient operators to generate gradient magnitude and orientation fields; invalid grid points in the anomalous field are filtered out and anomalous connected regions are identified; for each anomalous connected region, the gradient directions are aggregated and smoothed within a set sliding window, and the spatial offset features of the anomalous center and its dominant morphological orientation are extracted.

5. The ENSO response correction method for the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to claim 1, characterized in that, In step S400, the method for statistically testing the significance of abnormal directional distributions is as follows: the directional concentration is tested for each abnormal connected component using circular statistical methods, selecting either the Rayleigh test or the Watson test; robustness is then assessed using bootstrap resampling; when the confidence level reaches a preset threshold and passes multiple test correction, the directional feature is confirmed to be valid; otherwise, it is marked as invalid and the abnormal connected component is removed.

6. The ENSO response correction method for the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to claim 1, characterized in that, In step S500, the method for calculating the directional atom contribution for each connected domain based on spatial offset and morphological principal axis direction is as follows: select the temperature anomaly connected domain that overlaps with the grid point or is closest to the center, and calculate the morphological direction consistency and offset direction consistency respectively. Perform distance transformation on the grid points in the study domain to obtain their geographical distance to the boundary of the abnormal connected domain. Define the domain influence kernel of grid points inside the abnormal connected domain as 1, and the domain influence kernel of grid points outside the abnormal connected domain decreases monotonically with geographical distance. Rayleigh aggregation is performed on samples of the same element in the principal axis direction at different height layers to obtain vertical consistency. If the vertical consistency is lower than a preset threshold, the corresponding grid point is masked so that its vertical gating value is 0; otherwise, it is 1. The directional atom contribution of each grid point is calculated using the vertical gating value vm, the domain influence kernel ws, the morphological orientation consistency s1, and the offset orientation consistency s2: atom = vm × ws × [α·s1 + (1-α)·s2]; where α is the consistency bias weight.

7. The ENSO response correction method for the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to claim 1, characterized in that, In step S500, the method for integrating the directional atomic contributions obtained from each element into a grid coupling value DCI is as follows: the quadruple consisting of year category, element, abnormal connected domain and grid is used as the analysis unit, and the maximum value of the directional atomic contribution in different elements of an analysis unit is used as the representative contribution. The effective atom set is formed by selecting quadruples with a vertical gating value of 1 and a domain influence kernel greater than 0 from all analysis units; the grid coupling value DCI is obtained by robustly aggregating any element in the effective atom set based on the year category and grid point, and the robust aggregation method is the truncated mean.

8. The ENSO response correction method for the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to claim 1, characterized in that, In step S500, the method for determining whether a grid point is strongly coupled or weakly coupled based on the grid point coupling value DCI is as follows: an upper threshold and a lower threshold are preset for the grid point coupling value; the binary tuple formed by the year category and the grid point is used as the principal analysis element; if the number of elements participating in the calculation of the grid point coupling value of the principal analysis element is less than 2, it is judged as weakly coupled. Otherwise, if the grid coupling value of the principal analyte exceeds the upper threshold, it is judged as strong coupling; if the grid coupling value of the principal analyte is lower than the lower threshold, it is judged as weak coupling; if the grid coupling value of the principal analyte is between the upper and lower thresholds, the 3×3 neighborhood majority consensus rule is used to determine its label.

9. The ENSO response correction method for the Tibetan Plateau-Tropical Indian Ocean zonal anomaly field according to claim 1, characterized in that, In step S500, the method for correcting the ENSO response based on the coupling analysis results is as follows: the binary pair consisting of the year category and the grid point is used as the principal analysis element; if the coupling analysis result of the principal analysis element is strong coupling, its coupling weight is 1, otherwise it is λ; for any element, the anomaly field and the coupling weight are weighted and synthesized to form a coupled anomaly field; λ∈[0.3,0.5]; When calculating the correlation between the ENSO index and any element field, weighted correlation calculation and weighted least squares regression are performed on each grid point using coupled anomaly fields to obtain the corrected correlation coefficient field, regression coefficient field and lag assessment. The ENSO event classification is updated based on the corrected statistical results.

Images