Typhoon ventilation flow adaptive calculation method and system based on maximum probability feature and capable of overcoming radius sensitivity

Abstract

The invention relates to the technical field of tropical cyclone dynamics and numerical diagnosis, and discloses a typhoon ventilation flow adaptive calculation method and system based on maximum probability characteristics and capable of overcoming radius sensitivity. The method comprises the following steps: firstly, positioning a typhoon center and constructing a relative wind field based on reanalysis data or numerical forecasting data; then establishing a polar coordinate system and extracting a first-order harmonic component of the wind field; constructing radius-wind direction two-dimensional joint probability distribution P (r, alpha) on the basis, introducing a sample number threshold and direction consistency constraint, and adaptively identifying a connected region omega with the maximum probability and meeting statistical stability; and obtaining a ventilation flow vector and a quality index by adopting robust statistics and self-service re-sampling in the omega region, and realizing unified output of direction, strength and credibility. The system is designed in a modular manner, and has standardized input and output, automatic processing and business integration capabilities. The method eliminates the problem of radius sensitivity, remarkably improves the stability of ventilation flow estimation, and has important application value.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of tropical cyclone dynamics and numerical diagnosis, and relates to objective calculation of typhoon steering flow and path analysis, in particular to a typhoon steering flow self-adaptive calculation method and system based on maximum probability characteristics and overcoming radius sensitivity, which is used for robust extraction of steering flow vector in weak guiding, near-land and strong asymmetric scenes, correction of guiding air flow and support for path judgment and business integration. BACKGROUND

[0002] Tropical Cyclone (TC) as a highly destructive disastrous weather system, its path change and atmospheric circulation interaction mechanism has been a key problem in weather dynamics research and business forecasting. In the process of typhoon movement, the external guiding air flow is usually regarded as the main power to determine the large-scale translation direction. However, the actual observation shows that there is a systematic deviation between the typhoon moving direction and the steering flow vector, and the deviation is particularly prominent in the weak guiding air flow background, the β effect significant area and near the land boundary. Existing researches have pointed out that the deviation is related to the asymmetric circulation structure formed by the interaction of the typhoon vortex itself and the gradient of geostrophic potential vorticity (i.e. β plane effect), and its characteristic performance is that there is a steering flow channel approximately passing through the center of the cyclone between the typhoon center area and the β vortex pair, which not only affects the movement of the typhoon relative to the environment field, but also affects the precipitation distribution and intensity maintenance process of the typhoon.

[0003] With the development of reanalysis data, high spatiotemporal resolution satellite observations and large eddy simulation technology, the importance of steering flow in typhoon path prediction, uncertainty assessment and dynamic diagnosis has gradually been paid attention to. Through theoretical analysis, ideal numerical experiment and actual case analysis, it is found that the direction of steering flow has a strong correlation with the actual moving direction of typhoon, especially in the slow moving typhoon or before landing on the coast. Therefore, it is of great significance to construct an objective and standardized steering flow diagnostic quantity for improving the accuracy of path prediction, explaining the model bias, and identifying the mechanism between the large-scale guiding field and the vortex β drift.

[0004] However, the existing research still has the following main problems: (1) the ventilation flow has been theoretically defined, but there is a lack of unified calculation and diagnosis standard, and different literatures have great differences in the selection range of wind field, coordinate transformation method, signal extraction strategy and the like, resulting in that the calculation results are difficult to compare between different studies; (2) the existing ventilation flow diagnosis results are highly sensitive to the selected radial integral range, and the ventilation flow direction, intensity and even pointing type obtained under different radii can be significantly different, so that the ventilation flow is difficult to be used as a stable physical quantity for quantitative analysis; (3) under the condition of approaching land, gradient wind structure destruction or strong environmental shear, the typhoon wind field presents obvious non-axisymmetry, and the traditional radial average or simplified wind field processing method is easy to introduce significant error; (4) although the concept of ventilation flow has been gradually introduced into the path prediction model and path deviation analysis, there is a lack of a robust algorithm that can be automatically processed, and it is difficult to realize business promotion.

[0005] In summary, the existing ventilation flow calculation method has problems such as low standardization, strong sensitivity of results to radius, and subjectivity of parameter selection, which makes it difficult to reliably and quantitatively integrate this key physical factor into the typhoon path prediction model. Therefore, how to construct a ventilation flow adaptive calculation method with reliable physical basis, suitable for multiple types of typhoon environmental fields and capable of effectively weakening the radial dependence, has become a technical problem to be solved to improve the accuracy of typhoon path prediction. SUMMARY

[0006] (I) Invention purposes The purpose of the present application is to overcome the above-mentioned defects and shortcomings of the prior art, and to provide a typhoon ventilation flow adaptive calculation method and system based on maximum probability characteristics and overcoming radius sensitivity, which realizes the essential weakening of radius dependence by establishing a standardized ventilation flow diagnosis process and an adaptive optimization mechanism based on the maximization of two-dimensional probability distribution of radius and wind direction; on the basis of wind field preprocessing, coordinate transformation and first-order harmonic extraction, probability binning statistics are introduced to identify high-consistency flow direction intervals, and the zonal / meridional wind speed is robustly averaged in the interval to obtain the ventilation flow vector. Thereby, the stability and repeatability of the calculation results in key situations such as weak guidance and near-land asymmetric enhancement are improved, and an objective quantity that can be automatically integrated for path analysis and guiding air flow correction is provided.

[0007] (II) Technical solutions To achieve the purpose of the present application and solve its technical problems, the present application adopts the following technical solutions: The first purpose of the present application is to provide a typhoon ventilation flow adaptive calculation method based on maximum probability characteristics and overcoming radius sensitivity, which is used for robust extraction and quantitative estimation of typhoon ventilation flow in horizontal wind field of reanalysis data or numerical prediction data, and improves the calculation stability and physical representativeness under different cases and different data conditions, at least comprising the following steps: S100. Typhoon center positioning and relative wind field construction: Based on meteorological reanalysis data or numerical prediction data, determine the geographical position parameters of the typhoon center at the target time, and intercept the horizontal wind field data with the typhoon center as the center and covering the typhoon influence range; Determine and subtract the typhoon translation speed at the target time from the horizontal wind field to obtain the relative wind field in the stationary coordinate system relative to the typhoon movement; S200. Coordinate transformation and first-order harmonic decomposition: Establish a polar coordinate system with the typhoon center as the origin, convert the zonal wind component u and the meridional wind component v in the relative wind field data to the tangential wind speed Vt and the radial wind speed Vr relative to the typhoon center; At each radial radius r position, respectively, the first-order Fourier harmonic extraction is carried out on Vt and Vr along the azimuth angle λ, and the first-order harmonic component Vt1 and the first-order harmonic component Vr1 of the tangential wind speed and the radial wind speed are obtained, which are used to represent the asymmetric circulation characteristics caused by the β drift effect of the typhoon; Then convert the first-order harmonic components Vt1 and Vr1 to the rectangular coordinate system to obtain the corresponding first-order zonal wind component u1 and the meridional wind component v1, forming the horizontal wind field after asymmetric signal enhancement; S300. Probability binning modeling and radius-wind direction probability distribution construction: On the first-order horizontal wind field (u1, v1), group the radius r according to the preset radial binning interval Δr, and group the wind direction α according to the preset wind direction binning interval Δα, and discretize the analysis domain into (r, α) two-dimensional bins; Calculate the instantaneous wind direction α i in each two-dimensional bin i according to u1 and v1, and count its frequency of occurrence, combined with the speed threshold v min , eliminate low wind speed or abnormal samples, construct and normalize to obtain the radius-wind direction two-dimensional joint probability distribution P(r, α), which is used to quantify the consistency and concentration of wind direction at different radii; S400. Maximum probability feature recognition and adaptive optimization: In the two-dimensional joint probability distribution space P(r, α), search for a high-consistency connected region set that satisfies the sample number N ≥ N min and the direction consistency index C ≥ C min , N min is the lower limit of the number of valid samples in the connected region, and C min is the direction consistency threshold, determine the connected region Ω with the maximum probability as the dominant region of the ventilation flow, Ω is composed of several adjacent (r, α) bins and its determination process does not depend on the fixed radius value, thereby realizing the radius adaptive selection in a data-driven manner, and through the constraints of N min and C min , suppress the result sensitivity caused by empirical radius setting, low wind speed noise and / or insufficient sample number; S500. Vent flow vector robust estimation and quality control: robust statistical estimation is implemented on the first order zonal and meridional wind components within Ω to obtain the corresponding vent flow zonal and meridional components u v , v v , and the vent flow vector V v =(u v , v v ), its direction a v and modulus |V v | are formed accordingly, wherein the robust statistics adopts weighted average, trimmed average or median of outlier-resistant estimation to suppress the influence of residual outliers, and a quality index Q v is calculated to quantify the estimation reliability; S600. Result output and constraint judgment: the vent flow vector V v , direction a v , and modulus |V v | are output; when Q v is lower than a preset threshold or Ω does not satisfy N min and C min constraints, result rejection or uncertainty identification is triggered to ensure the stability and reproducibility of the results under weak guidance, near land and strong asymmetric conditions.

[0008] The second inventive objective of the present application is to provide a typhoon vent flow adaptive calculation system, which utilizes the above-mentioned typhoon vent flow adaptive calculation method based on maximum probability characteristics and overcoming radius sensitivity, and at least includes the following modules: A data acquisition and preprocessing module is used to receive meteorological reanalysis data or numerical prediction data, locate the typhoon center at the target time and intercept the horizontal wind field data covering the typhoon influence range, estimate the typhoon translation speed and subtract from the wind field to form a relative wind field; A coordinate conversion and harmonic decomposition module is used to convert the zonal and meridional wind components u, v in the relative wind field into the tangential wind speed Vt and radial wind speed Vr relative to the typhoon center, and extract the first order harmonic of Vt1 and Vr1 along the azimuth angle λ for each radial r, and then inverse transform to obtain the first order horizontal wind u1 and v1, forming the wind field after asymmetric signal enhancement; A probability distribution construction module is used to discretize the (u1, v1) field into two-dimensional bins according to the preset radial interval Ar and wind direction interval Aa, calculate the wind direction occurrence frequency of each bin, and combine with the speed threshold v min to remove low wind speed or abnormal samples, and normalize to obtain the radius-wind direction two-dimensional joint probability distribution P(r, a); An adaptive parameter identification module is configured to search for a high-consistency connected region in P(r, alpha) that satisfies a sample size and direction consistency constraint, determine a connected region Omega with the maximum probability and satisfying the constraint as a ventilation flow dominant region, and adaptively determine a representative radius and a dominant wind direction to avoid fixed radius dependence. A ventilation flow calculation and quality evaluation module is configured to perform robust aggregation on u1 and v1 in the ventilation flow dominant region Omega to obtain a ventilation flow vector V v =(u v ,v v ), a direction alpha v , and a modulus |V v |, and generate a quality index Q v . A result checking and output control module is configured to output V v , alpha v , and |V v |, and trigger a result rejection or an uncertainty warning flag when Q v is lower than a preset threshold or Omega does not satisfy N min and C min constraints.

[0009] (Three) Technical effects Compared with the prior art, the typhoon ventilation flow adaptive calculation method and system based on the maximum probability feature and overcoming the radius sensitivity of the present application have the following beneficial and significant technical effects: (1) The present application replaces the empirical fixed radius with the maximum probability connected region Omega of the radius-wind direction joint probability distribution P(r, alpha), and realizes adaptive positioning and aggregation of ventilation channels on the relative wind field after Fourier first harmonic enhancement, thereby fundamentally eliminating the structural sensitivity of the radius selection of the traditional method. Compared with the fixed radius average, this strategy significantly reduces the direction deviation and amplitude drift in the beta vortex pair dominant asymmetric scene, and improves the stability and repeatability of the results.

[0010] (2) The present application significantly improves the stability of the ventilation flow calculation result. Through multiple quality control mechanisms such as probability threshold constraint, connectivity test, and direction consistency check, the interference of the typhoon eye area minimum wind area and the beta vortex pair main structure area is effectively avoided, and the ventilation flow characteristics between the beta vortex pairs are accurately extracted. Taking the No. 2 typhoon in 2021 as an example, the ventilation flow calculated by the present application method presents high consistency with the actual moving track of the typhoon, and the direction deviation is less than 5°; while the direction deviation of the traditional method under different radii can reach 23°. The 30-year typhoon sample statistics show that the present application method reduces the average deviation of the ventilation flow and the actual moving direction by about 35%, and the calculation quality index generally reaches a high quality level of 0.8 or above.

[0011] (3) The traditional method cannot be applied to business forecasting due to the problems of calculation ambiguity and radius sensitivity. The present application converts the ventilation flow calculation from a qualitative concept to an operable quantitative method through a clear standardized calculation process, adaptive parameter selection and quantitative uncertainty evaluation. Especially in the scenario where the typhoon moves slowly, approaches the land and has strong asymmetry, the importance of ventilation flow increases significantly, which has significant application value. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 is a typhoon ventilation flow adaptive calculation method flowchart provided by the embodiment of the present application based on the maximum probability feature and overcoming the radius sensitivity.

[0013] Figure 2 is a typhoon ventilation flow adaptive calculation system architecture diagram provided by the embodiment of the present application.

[0014] Figure 3 The pre-process flowchart of the traditional ventilation flow calculation method of Typhoon No. 2 in 2021 is shown, wherein: (a) is an original wind field distribution diagram, the arrow represents the original wind field in the ERA5 data, the black solid line represents the typhoon path, and the black dot represents the position of the current typhoon center (April 17, 2021 00:00 UTC); (b) is a relative wind field distribution diagram after deducting the typhoon moving speed; (c) is a tangential wind (Vt) decomposition result diagram of the typhoon wind field relative to the typhoon center; (d) is a radial wind (Vr) decomposition result diagram of the typhoon wind field relative to the typhoon center.

[0015] Note: Figure 3 In each diagram, the horizontal coordinate Lon is the longitude (°E), and the vertical coordinate Lat is the latitude (°N).

[0016] Figure 4 The subsequent flowchart of the traditional ventilation flow calculation method of Typhoon No. 2 in 2021 is shown, wherein: (e) is a wave component distribution diagram obtained after Fourier decomposition of the tangential wind Vt; (f) is a wave component distribution diagram obtained after Fourier decomposition of the radial wind Vr; (g) is a Vt and Vr wave component combined wind field diagram, wherein the red circle and arrow represent the average wind within 3° of the typhoon center (i.e. the ventilation flow calculated by the traditional method), there is a cyclone (marked as C) on the left side of the typhoon path and an anticyclone (marked as A) on the right side, C and A form a beta vortex pair of the typhoon; (h) is a comparison diagram of ventilation flow results calculated under different radius conditions, showing the differences of ventilation flow when the radius R is 3°, 5°, 7°, 9° and 11°, which shows that the traditional method has significant sensitivity to the radius parameter.

[0017] Note: Figure 4In (g) and (h), the large colored arrows used to represent ventilation flow are enlarged 10 times relative to the black arrows in the background to improve visualization clarity; in each figure, the horizontal axis Lon represents longitude (°E) and the vertical axis Lat represents latitude (°N).

[0018] Figure 5 The diagram shows the improved adaptive ventilation flow calculation method of the present invention, wherein: (a) is a schematic diagram of the traditional fixed radius ventilation flow algorithm, showing the method of determining the ventilation flow by using the average wind field within the calculation range of a fixed radius. This method is significantly sensitive to the selection of the radius; (b) is a two-dimensional probability distribution map of radius and wind direction constructed based on the wind direction information in (a). The horizontal axis represents the radius (unit: degree), the vertical axis represents the wind direction (unit: degree), and the color depth represents the probability magnitude. The figure shows that the maximum probability occurs at the position with a radius of 2° and a wind direction of approximately 315°; (c) is a schematic diagram of the calculation result of the average value of the zonal wind component u and the meridional wind component v corresponding to the interval with the maximum probability as the ventilation flow. The red arrow represents the ventilation flow calculated by the method of the present invention. This ventilation flow is highly consistent with the actual ventilation flow between the β vortex pairs (C and A).

[0019] Note: Figure 5 In the middle (a) and (c), the large colored arrows used to represent ventilation flow are enlarged by 3 times relative to the black arrows in the background to improve visual clarity. Detailed Implementation

[0020] This invention aims to provide an adaptive calculation method and system for typhoon ventilation flow based on maximum probability characteristics and overcoming radius sensitivity. To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. The described embodiments are some, but not all, embodiments of this invention, and are exemplary and should not be construed as limiting the invention.

[0021] Example 1: Adaptive Calculation Method for Typhoon Ventilation Flow like Figure 1 As shown in the embodiments of the present invention, an adaptive calculation method for typhoon ventilation flow based on maximum probability characteristics and overcoming radius sensitivity is used to robustly extract and quantitatively estimate typhoon ventilation flow in horizontal wind fields of reanalysis data or numerical weather prediction data, improving the calculation stability and physical representativeness under different case and data conditions. The calculation method mainly includes the following steps in its implementation: S100. Typhoon Center Location and Relative Wind Field Construction: Based on meteorological reanalysis data or numerical prediction data, the geographical position parameters of the typhoon center at the target time are determined, and the horizontal wind field data covering the influence range of the typhoon with the typhoon center as the center are intercepted; the translation speed of the typhoon at the target time is determined and subtracted from the horizontal wind field to obtain the relative wind field in the stationary coordinate system relative to the typhoon movement. As an optimization, the typhoon center positioning adopts multi-index comprehensive judgment: taking the minimum sea level pressure and the maximum relative vorticity at 850 hPa as the main criterion, and combining the consistency check of the low-level convergence center and the maximum axisymmetric wind speed radius, when the multi-index positioning results are consistent within 0.5°, the weighted centroid average is adopted, when the positioning deviation exceeds 0.5°, the center consistent with the low-level relative vorticity and divergence coupling index is preferentially selected. When the center positioning is inconsistent, the center consistent with the low-level relative vorticity and divergence coupling index is selected to ensure the stability and physical representativeness of the subsequent polar coordinate transformation; the translation speed vector of the typhoon is obtained by least square linear fitting on the typhoon center position at the previous and next time, and the fitting time window is 3-6 hours before and after, and the consistency check is performed with the potential vorticity centroid tracking result, when the included angle between the two exceeds the threshold (for example, 30°), the translation speed is adjusted in the ensemble average manner, and the weight coefficient is dynamically allocated according to the historical consistency of each estimation method.

[0022] After determining the typhoon center positioning, the interception space range of the horizontal wind field is adaptively set: the outer radius R win 1.2-1.8 times of the isobaric line closing radius, the specific value is determined according to the typhoon intensity grade and the latitude, the larger value is taken for strong typhoon or super strong typhoon, and the smaller value is taken for tropical storm, to ensure the complete spatial structure of the beta vortex pair and its transition zone; the inner shielding radius R in 0.5°-1.0° according to the radius characteristics of the weak wind and high gradient area in the eye of the typhoon, wherein the smaller value is taken when the eye is clear and the eye wall is obvious, and the larger value is taken when the eye is blurred or filled, to suppress the interference of the eye; the regular gridding of the horizontal wind field is completed by bilinear or higher order shape-preserving interpolation, the interpolation grid spacing is 0.1°-0.25°, the number of radial sampling points is not less than 30, and the number of azimuth angle sampling points is not less than 72, and the missing data mask and abnormal value threshold limit detection are performed, the abnormal value threshold limit is defined as 3 times the standard deviation of the local climate average value or the absolute value of the wind speed exceeding 80 m / s, and the non-physical abrupt samples caused by terrain or land surface are removed, for the time when the proportion of missing data is more than 20%, it is marked as unqualified data quality and triggers a warning, to ensure the continuity and consistency of the input wind field.

[0023] S200. Coordinate transformation and first-order harmonic decomposition: An orthogonal coordinate system with the center of the typhoon as the origin is established, and the zonal wind component u and the meridional wind component v in the relative wind field data are converted into the tangential wind speed Vt and the radial wind speed Vr relative to the center of the typhoon, and the conversion relationship is Vt=-u*sinλ+v*cosλ, Vr=u*cosλ+v*sinλ, wherein λ is the azimuth angle; the first-order Fourier harmonic extraction is respectively performed on Vt and Vr along the azimuth angle λ at each radial radius r position, the harmonic extraction adopts the fast Fourier transform algorithm, only the component with the wave number k=1 is retained, and the first-order harmonic component Vt1 of the tangential wind speed and the first-order harmonic component Vr1 of the radial wind speed are correspondingly obtained, which are used to represent the asymmetric circulation characteristics caused by the beta drift effect of the typhoon; then the first-order harmonic components Vt1 and Vr1 are inversely converted into the orthogonal coordinate system, and the inverse conversion relationship is u1=-Vt1*sinλ+Vr1*cosλ, v1=Vt1*cosλ+Vr1*sinλ, the first-order zonal wind component u1 and the first-order meridional wind component v1 are obtained, and the horizontal wind field after the asymmetric signal enhancement is formed.

[0024] As preferred, the polar coordinate transformation adopts the sub-grid interpolation and the angle consistency constraint in space: the relative wind field is first resampled to a uniform r-λ grid with the center of the typhoon as the pole by the bicubic interpolation with energy preservation, the radial resolution Δr and the azimuth resolution Δλ are respectively adaptively set according to the spatial resolution of the wind field and the typhoon wind circle scale, Δr takes 0.5-1.0 times of the horizontal resolution of the original data, and Δλ takes 5° to ensure that there are not less than 72 sampling points in the azimuth direction; when the first-order harmonic extraction is performed, the band-pass filtering is performed on Vr and Vt along the λ direction to suppress high-order noise, the band-pass filtering retains the components with the wave number k=0.5-2, the filter adopts the Hann window or the Butterworth filter, the harmonic fitting adopts the weighted least squares, the weight is inversely proportional to the local wind speed module value and the observation error, the observation error is set according to the data type, the reanalysis data takes 1-2 m / s, and the numerical prediction data takes 2-4 m / s, so that the first-order harmonic components Vt1 and Vr1 are stably obtained; when the first-order harmonic variance interpretation degree is lower than a preset threshold η min (for example, 0.3-0.5), the radial sample point is marked as a low-confidence sample, and does not participate in subsequent probability statistics.

[0025] Further, the inverse transformation of the first-order harmonic component satisfies the geometric consistency and the energy consistency double constraints: after Vt1 and Vr1 are recovered into u1 and v1 by the orthogonal transformation, it is required that the dominant rotation direction in the local ring band is the same as that of the original relative wind field, and the ratio of |u1|, |v1| to the corresponding harmonic energy of the original field is located in the interval [0.7, 1.3], and the energy ratio is defined as (u1 2 +v1 2 ) / (u 2 +v 2) the integral ratio over corresponding harmonic components; if not satisfied, trigger local re-fitting and end-point smoothing, the latter using 3-point moving average or spline smoothing with window width of 3-5 grid points to reduce the cumulative impact of inverse transform error on subsequent probability estimation.

[0026] In addition, for local anomalies caused by heavy precipitation or terrain, an anomaly rejection and missing data filling mechanism is set up: the Mahalanobis distance based on the two-dimensional distribution covariance matrix of wind speed and the robust z-score based on the median and median absolute deviation normalization are calculated for the angular sequence of Vr, Vt at fixed r, and the λ position exceeding the threshold (e.g. Mahalanobis distance > 3 or robust z-score > 3) is marked as an anomaly and removed from the harmonic fitting; when the effective angular sample proportion at a certain r is lower than the set lower limit (e.g. 60%), the harmonic coefficients of the adjacent radius are used for radial extrapolation, the extrapolation uses first-order recursion and limits the coefficient drift with a Bayesian regularization term, the regularization term parameter λr eg = 0.1, 0.3, to constrain the radial variation rate of harmonic coefficients in the L2 norm, ensuring the continuity and energy conservation of u1, v1 in inverse transformation.

[0027] S300. Probability binning modeling and radius-wind direction probability distribution construction: On the first-order horizontal wind field (u1, v1), the radius r is grouped according to the preset radial binning interval Δr, and the wind direction α is grouped according to the preset wind direction binning interval Δα, the analysis domain is discretized into (r, α) two-dimensional bins, the total number of bins is N bins = (R max -R min ) / Δr×360° / Δα; in each two-dimensional bin i, the instantaneous wind direction α i is calculated according to u1 and v1, and its occurrence frequency is counted, combined with the speed threshold v min , low wind speed or abnormal samples are removed, and the radius-wind direction two-dimensional joint probability distribution P(r, α) is constructed and normalized to quantify the consistency and concentration of wind direction at different radii.

[0028] As a preferred, the radial binning interval Δr is taken as 0.1°-0.25°, corresponding to a physical distance of about 10-25 kilometers, the wind direction binning interval Δα is taken as 5°-15°, a smaller Δα is beneficial to fine description of the direction distribution, but will reduce the sample quantity of a single bin, and is adjusted adaptively according to data density in implementation; the speed threshold v min is determined by the quantile method of local annular velocity distribution, specifically, the empirical distribution of wind speed modulus |V| = (u1 2 +v1 2 ) 0.5 is calculated at each radius r, and the 20th-30th percentile is taken. minThe speed should be no less than 2 m / s to avoid noise dominance; the joint probability distribution P(r,α) is obtained by kernel density estimation from a normalized histogram or an equivalent von Mises distribution, with the bandwidth parameter h used in kernel density estimation. r and h α The radial and azimuth resolutions are taken as 1.5 to 2.5 times, respectively. Optimization is performed using adaptive bandwidth or cross-validation, and the effective sample weights are monotonically increased with wind speed and fitting explanatory power. The weighting function is defined as w. i =|V i |·η i , where η i The first harmonic variance explanation for this radial position is given; finally, a normalization process with a sum of 1 is applied to each (r, α) bin, and the normalization formula is P(r, α) = N. w (r,α) / ∑ {r,α} N w (r,α), where N w (r,α) represents the weighted frequency, which characterizes the consistency and concentration of wind direction at different radii.

[0029] Furthermore, the wind direction α is calculated as α = arctan(v1 / u1), and the quadrant of the wind direction is determined based on the signs of u1 and v1. The wind direction is then uniformly converted to the meteorological definition of wind direction from 0° to 360°. The conversion relationships are: when u1>0, v1≥0, α = arctan(v1 / u1); when u1≤0, v1>0, α = 90° + arctan(-u1 / v1); when u1<0, v1≤0, α = 180° + arctan(v1 / u1); when u1≥0, v1<0, α = 270° + arctan(-u1 / v1). The statistical method for the frequency of instantaneous wind direction in each two-dimensional bin is as follows: traverse all grid points in the first-order horizontal wind field (u1, v1), and assign each grid point to its corresponding two-dimensional bin based on its radius r and wind direction α. ​​The bin assignment is determined using a floor rule, i.e., r... bin =floor((rR min ) / Δr), α bin =floor(α / Δα), which accumulates the number of grid points in each bin and simultaneously adds the corresponding weight w. i This forms a frequency matrix N(r,α), which is then normalized to transform into a probability distribution P(r,α)=N(r,α) / ∑N(r,α), or a weighted form P(r,α)=N w (r,α) / ∑N w (r,α).

[0030] S400. Maximum Probability Feature Recognition and Adaptive Optimization: On the two-dimensional joint probability distribution space P(r,α), search for samples N ≥ N min Consistency index with direction C≥C min A set of highly consistent connected regions, N min This is the lower limit for the number of effective samples within a connected region, and a value ranging from 50 to 200 grid points is recommended. The specific value should be determined based on the data resolution and typhoon scale. min For the directional consistency threshold, a value range of 0.7 to 0.9 is recommended. The connected interval Ω with the highest probability is determined as the dominant ventilation flow region. Ω is composed of several adjacent (r, α) sub-bins, and its determination process does not depend on a fixed radius value. This achieves data-driven adaptive radius selection, and through N... min With C min The constraints suppress the sensitivity of the results caused by empirical radius setting, low wind speed noise and / or insufficient sample size.

[0031] In this embodiment of the invention, the determination of highly consistent connected regions combines topological connectivity with morphological constraints: first, a threshold segmentation is applied to P(r,α) to obtain candidate regions. The threshold is taken as the 50th to 70th percentile of the probability distribution or the average plus 0.5 times the standard deviation. Connected components are identified using the 4-neighborhood or 8-neighborhood criteria. When the binning mesh is coarse, the 8-neighborhood is used to enhance connectivity; when the binning mesh is dense, the 4-neighborhood is used to improve resolution. The circular variance (σ) of the directional distribution is calculated for each candidate component. circ 2 =1-R, where R is the average composite vector length and the radial span Δr span =r max -r min In terms of shape compactness, excessive radial span (Δr) should be eliminated. span >3°) or the directional variance exceeds the threshold (σ circ 2 False peaks >0.5); when there are multiple candidate components that meet the conditions, select the component with both a higher peak probability and a smaller directional variance, and use the Bayesian information criterion BIC=k·ln(n)-2·ln(L), where k is the number of model parameters (2 here, i.e., the center radius and the center wind direction), n is the sample size, and L is the likelihood function value. The final selection is completed by combining the sample size and model complexity, and the connected component with the smallest BIC is selected as the optimal solution.

[0032] Furthermore, the directional consistency index C is defined as the average composite vector length of a unit wind direction vector within the region, and is calculated using the formula C=(∑cos(α)). i )) 2 +(∑sin(α i )) 2 ) 0.5 / N, with a value range of [0,1], or equivalently defined as the percentage of samples falling within the prevailing wind direction tolerance zone, where the prevailing wind direction tolerance zone is 10°~20°, and the prevailing wind direction is defined as the weighted average wind direction α. mean =arctan(∑w i ·sin(α i ) / ∑w i ·cos(α i The tolerance band is [α] mean -Δα tol , α mean +Δα_ tol When there exist multiple conditions satisfying N≥N min C≥C min When considering a connected region, first select Ω, which has the largest sum of probability masses within the region. The sum of probability masses is defined as M. Ω =∑ {(r,α) ∈Ω}P(r,α), then compare the effective sample size with the consistency index C. When the probability quality and the difference are less than 10%, the region with the higher C value is selected first.

[0033] S500. Robust estimation and quality control of ventilation flow vectors: Robust statistical estimations were performed on the first-order zonal and meridional wind components within Ω to obtain the corresponding zonal and meridional components u of the ventilation flow. v v v Based on this, the ventilation flow vector V is formed. v =(u v ,v v ) and its direction α v With modulus |V v |,α v =arctan(v v / u v ), |V v |=(u v 2 +v v 2 ) 0.5 Robust statistics employs weighted average, truncated average, or median estimation to suppress the influence of residual outliers. The selection criteria for these three methods are as follows: weighted average is used when the sample distribution within Ω is approximately Gaussian and the proportion of outliers is <5%; a 10% truncated average is used when the proportion of outliers is between 5% and 15%; and median estimation is used when the proportion of outliers is >15% or the distribution is significantly skewed. Simultaneously, the quality index Q is calculated. v To quantify the confidence level, Q v The value range is [0,1], and the higher the value, the more reliable the estimate.

[0034] In this embodiment of the invention, when performing robust estimation of the first-order horizontal wind component within Ω, the weighting coefficient is jointly determined by the binning probability P(r,α) and the local wind speed amplitude, and the weighting function is w. i =P(r i ,α i )·|V i | β Where β is the wind speed weighting exponent, ranging from 0.5 to 1.0, and the median absolute deviation is used as the scaling parameter for adaptive estimation; ≥200 resamplings are performed using the bootstrap method, with each resampling randomly drawing samples with replacement from within Ω, resulting in the ventilation flow vector V. v The confidence intervals and bias correction terms are calculated, with the confidence intervals taken at the 2.5% and 97.5% quantiles of the Bootstrap distribution, and the bias correction term being the difference between the Bootstrap mean and the original estimate; the quality index Q... v From the number of valid samples N eff The normalized weighted combination of directional consistency C and the half-width of the confidence interval is used to quantify the confidence level.

[0035] As a preferred approach, to avoid instability caused by single-time fluctuations, time consistency and multi-time smoothing are implemented: for consecutive time periods V v A combined exponentially weighted moving average and Kalman filter are used for smoothing. The state equation models the ventilation flow evolution as a random walk and uses the environmentally guided airflow rate of change as a priori process noise. The state equation is X. t =A·X {t-1} +w t The observation equation is Z t =H·X t +v t , where the state vector X t =[u v ,v v ] T The state transition matrix A is the identity matrix, the observation matrix H is the identity matrix, and w t For process noise, v t To observe the noise, the process noise covariance matrix Q is initially set to (2 m / s). 2 The observation noise covariance matrix R is adaptively set to R = (5·(1-Qv) m / s) based on the quality index Qv. 2 • I; Based on the generalized likelihood ratio test, when a sudden change is detected, the process noise is adaptively increased to track it quickly, ensuring that a physically reasonable estimate can still be given during the rapid structural adjustment phase.

[0036] S600. Result Output and Constraint Determination: Output ventilation flow vector V v Direction αv Modulus | V v | and its 95% confidence interval; when the quality index Q v Below the preset threshold or Ω does not meet N min and C min When constraints are applied, a result rejection or uncertainty flag is triggered to ensure the stability and reproducibility of the results under weak guidance, near-shore, and strong asymmetric conditions. Preferably, the quality index Q... v By normalizing to 0-1 and setting a level threshold, it is divided into high quality (Q) v ≥0.8), medium quality (0.6≤Q) v <0.8), low quality (0.4≤Q) v <0.6) and unavailable (Q) v (<0.4) Four levels, when Q v In the middle range (0.6≤Q) v When Q < 0.8, output interval estimation and uncertainty strips, with the uncertainty strip width being 1.5 times the half-width of the confidence interval; when Q v Below the minimum threshold (Q) v When the value is less than 0.4, the result is marked as unusable and reverted to the backup strategy. The backup strategies are in the following order of priority: ① Appropriately increase Δr (increase by 20%~50%) or Δα (increase by 30%~100%) to increase the sample size per bin; ② Enable the suboptimal connected region, i.e., select the connected region with the second largest probability quality and satisfying the constraints; ③ Use the alternative estimate of the adjacent height layer (e.g., 700 hPa or 500 hPa) and correct for the height layer difference; ④ When all backup strategies fail, output the missing data flag and record the reason for failure, including increasing Δr or Δα to reduce spatial resolution in exchange for sample stability, enabling the suboptimal connected region, i.e., selecting the region with the second highest probability quality and satisfying the weakening constraint (N≥0.7·Nmin, C≥0.85·Cmin), or using the alternative estimate of the adjacent height layer, with the 700 hPa layer preferred and the 500 hPa layer used when necessary, and corrected according to the climatological differences of the steering airflow between height layers. The correction coefficient is obtained through regression of historical statistical samples. The final output includes: ventilation flow vector V v and its confidence interval, quality grade label, and dominant region radius [r] min ,r max ], Dominant azimuth range [α min ,α max ] , Number of valid samples N eff The direction consistency index C value, and the strategy type identifier when the backup strategy is adopted.

[0037] Example 2: Typhoon Ventilation Flow Adaptive Calculation System Based on Embodiment 1 above, Embodiment 2 further provides a typhoon ventilation flow adaptive calculation system based on the above calculation method. This system adopts a modular design, with each module interacting through standardized data interfaces, supporting deployment and operation in single-machine or distributed computing environments. For example... Figure 2 As shown, the computing system mainly includes the following modules: The data acquisition and preprocessing module receives meteorological reanalysis data or numerical weather prediction data, locates the typhoon center at the target time, and extracts horizontal wind field data covering the typhoon's influence area. The location methods include: ① reading typhoon path databases (such as the CMA optimal path dataset or JTWC path data) to obtain the initial position; ② if path data is unavailable, automatically identifying the typhoon center using the multi-index comprehensive judgment method in step S100; ③ performing a spatiotemporal continuity check on the location results and eliminating locations with anomalous jumps. The typhoon's translational velocity is estimated and subtracted from the wind field to form a relative wind field. The translational velocity is calculated using finite difference or least squares fitting of the center position time series, with a time window of 3-6 hours before and after the typhoon.

[0038] As a preferred approach, the data preprocessing workflow of this module includes the following steps: ① Data quality check: detecting missing values, outliers (outside the physically reasonable range), and data integrity; ② Spatial interpolation and gridding: interpolating non-uniform or sparse observation data to a regular latitude and longitude grid, using bilinear, bicubic, or Kriging interpolation methods; ③ Time alignment: when the time resolution of the input data is inconsistent with the required analysis time, performing time interpolation or selecting the most recent time; ④ Coordinate system transformation preparation: extracting the geographical range of the analysis area and calculating the grid spacing to establish a mapping relationship for subsequent polar coordinate transformation.

[0039] The coordinate transformation and harmonic decomposition module is used to convert the zonal and meridional wind components u and v in the relative wind field into tangential wind speed Vt and radial wind speed Vr relative to the typhoon center. The transformation process includes: ① establishing a polar coordinate grid with the typhoon center as the origin, and the radial range [R] in， R out], radial step size Δr, azimuth range [0°, 360°], azimuth step size Δα; ② For each polar coordinate grid point (r, λ), calculate the latitude and longitude (lon, lat) in the geographic coordinate system; ③ Interpolate the wind speed components of the polar coordinate points on the (u, v) field of the geographic coordinate grid, using bilinear or bicubic interpolation; ④ Convert (u, v) to (Vt, Vr), the conversion formula is Vt=-u·sin(λ)+v·cos(λ), Vr=u·cos(λ)+v·sin(λ), and extract the first harmonics of each radial r along the azimuth angle λ to obtain Vt1 and Vr1. The harmonic extraction uses Discrete Fourier Transform (DFT) or Fast Fourier Transform (FFT), and then inverse transform to obtain the first-order horizontal wind u1 and v1. After inverse transform, it needs to be interpolated back to the geographic coordinate grid for subsequent processing to form the wind field after asymmetric signal enhancement.

[0040] The probability distribution construction module is used to discretize the (u1,v1) field into two-dimensional bins according to a preset radial interval Δr and wind direction interval Δα. The discretization process is as follows: ① Determine the radius bin boundary r bins =[R in ,R in +Δr,R in +2Δr,..,R out Number of boxes N r =(R out -R in ) / Δr;② Determine the wind direction sub-box boundary α bins =[0°,Δα,2Δα,..,360°], Number of boxes N α =360° / Δα;③ Traverse all grid points in the (u1,v1) field, calculate the radius and wind direction of each grid point relative to the typhoon center α=arctan2(v1,u1), and assign the grid points to the corresponding two-dimensional bins (i,j), where i=(rR in ) / Δr, j=α / Δα, calculate the frequency of wind direction occurrence in each sub-box and combine it with the velocity threshold v min Low-wind-speed or anomalous samples are removed, specifically by calculating the wind speed modulus |V| for each grid point, only removing samples if |V| ≥ v. min When this grid point is included in the frequency statistics, v min The wind speed distribution within each radial ring (r ± Δr / 2) is calculated independently. The 20th to 30th percentile of the wind speed distribution in that ring is taken, and finally normalized to obtain the two-dimensional joint probability distribution of radius and wind direction, P(r,α), which is used to quantify the consistency and concentration of wind direction at different radial locations.

[0041] As a preferred approach, the probability distribution construction module implements adaptive adjustment of binning parameters: ① When the original data resolution is coarse (grid spacing > 0.5°), Δr is reduced to increase radial resolution, but not less than 0.5 times the original grid spacing; ② When the average sample size per bin is too small (e.g., < 10 grid points), Δr or Δα is moderately increased, specifically by 1.5 to 2 times the original value, until the sample size per bin is ≥ 10; ③ When the typhoon scale is small (radius of the 7-level wind circle < 200 km), Δr and Δα are reduced proportionally; ④ The rationality of binning granularity is evaluated through cross-validation or information entropy criteria to avoid excessive discretization or excessive smoothing.

[0042] The construction of the joint probability distribution P(r, α) includes two options: the first option is the histogram method, P(r, α) = N(r, α) / ∑N(r, α), where N(r, α) is the number of effective grid points or weighted counts within the bin (r, α); the second option is the kernel density estimation method, P(r, α) = ∑ {i} K h ((rr i ) / h r , (α-α i ) / h α ) / Z, where K h For a two-dimensional kernel function (such as a Gaussian kernel or a von Mises kernel), h r and h α Let Z be the bandwidth parameter and Z be the normalization constant. Kernel density estimation can smooth discrete noise, but it is computationally intensive and suitable for high-resolution data; histogram method is computationally simple and suitable for cases with sufficient sample size and relatively uniform distribution. Both methods require periodic processing of the boundary bins, i.e., α=0° and α=360° are considered to be in the same direction.

[0043] In addition, the anomaly detection and removal mechanism of this module includes: ① Local outlier detection: Within each radial ring, the circular standard deviation of wind direction is calculated, and samples exceeding 3 times the standard deviation are marked as anomalies; ② Spatial isolated point detection: If the probability difference between a bin and its 8 neighboring bins is too large (relative difference > 200%), and the sample size of the bin is < 50% of the neighborhood average, it is marked as isolated noise and smoothed or removed; ③ Temporal continuity test: When processing continuous time data, if the probability of a bin changes abruptly between adjacent time periods (change rate > 100% / hour), the weight of the bin in the current time period is reduced or manual review is triggered.

[0044] The adaptive parameter identification module is used to search for highly consistent connected regions in P(r,α) that satisfy the constraints of sample size and orientation consistency. The search algorithm adopts a two-stage method based on threshold segmentation and connected component labeling: In the first stage, a threshold P is applied to the probability distribution P(r,α). th We obtain the binary mask M(r,α), Pth The value is taken as the 50th to 70th percentile of the probability distribution or the mean plus 0.5 times the standard deviation, such that M(r,α)=1 when P(r,α)≥P th Otherwise, M(r,α)=0; In the second stage, a connected component labeling algorithm (such as a flood fill algorithm based on breadth-first search (BFS) or depth-first search (DFS)) is applied to the binary mask M(r,α), using the 8-neighborhood connectivity criterion to identify all connected regions {Ω k}, k=1,2,..,K, the connected interval Ω with the highest probability and satisfying the constraints is determined as the dominant ventilation flow region. Ω is composed of several adjacent (r,α) sub-bins, and its determination process does not depend on a fixed radius value, thus realizing adaptive radius selection in a data-driven manner, and through N min With C min The constraints suppress the sensitivity of results caused by empirical radius setting, low wind speed noise, and / or insufficient sample size. Preferably, when multiple high-scoring regions exist (score difference <10%), physical rationality criteria are used for further selection: ① Regions where the prevailing wind direction is consistent with the direction of the environmental steering flow (angle <30°), which can be extracted from the large-scale wind field of reanalysis data or the environmental field of numerical forecasts; ② Regions whose radial center is located near the characteristic radius of the typhoon wind circle (e.g., the radius of the 34-knot or 50-knot wind circle); ③ Regions that are stable over time (center position drift <1° between time intervals). If a unique selection is still not possible, the top two candidate regions and their uncertainty intervals are output.

[0045] The ventilation flow calculation and quality assessment module is used to perform robust aggregation of u1 and v1 within the dominant ventilation flow region Ω to obtain the ventilation flow vector V. v =(u v ,v v ), direction α v With modulus |V v |and generate quality index Q v Q v The overall reliability of the estimate is reflected, and the value range is [0,1]. Preferably, the quality index Q... v It is composed of a weighted combination of three sub-indicators, and the calculation formula is: Q v =w N ·Q N +w C ·Q C +w CI ·Q CI Where: ① Sample quantum index QN=1-exp(-N) Ω / N0), where N0 is the reference sample size taken as 100, Q N Q reflects the sufficiency of the sample; when the sample size reaches N0, Q... N≈0.63, the larger the sample size, the lower the Q value. N The closer to 1; ② Consistency sub-index Q C= C Ω That is, the direction of the dominant region is consistent, and the C calculated by the previous module is directly adopted. Ω Value; ③ Confidence sub-index Q CI =1-min(ΔCI / ΔCI max ,1), where ΔCI=(|u v97.5% -u v2.5% |+|v v97.5% -v v2.5% |) / 2 is the average of the half-width of the confidence interval, ΔCI max The normalized reference value is taken as 10 m / s, Q CI The narrower the confidence interval, the more accurate the estimation. CI The higher the value; ④ It is recommended that the weighting coefficient be w. N =0.3, w C =0.4, w CI =0.3, which can be adjusted according to application requirements. Final Q v Normalized to the [0,1] interval, Q v ≥0.8 indicates high quality, 0.6≤Q v <0.8 indicates medium quality, 0.4≤Q v <0.6 indicates low quality, Q v <0.4 indicates unavailable.

[0046] The result verification and output control module is used to output V. v α v 、|V v | and related quality and uncertainty information, and perform final verification and formatted packaging of the results, when Q v Below the preset threshold or Ω does not meet N min and C min When constraints are applied, a result rejection or uncertainty warning is triggered, and the reason for failure and the history of degradation operations are recorded. Preferably, the output of this module includes the following complete information set: ① Core result: Ventilation flow vector V v =(u v v v ) and its 95% confidence interval [u vlow u vhigh ],[v vlow v vhigh Ventilation flow direction α v (Meteorological wind direction, in degrees) and its standard error σ αv Ventilation flow modulus | V v | (unit: m / s) and its relative error δ|V v ② Quality Indicators: Comprehensive Quality Indicator Qv Its level identifier (high / medium / low / unavailable), sub-index Q N Q C Q CI ③ Dominant region information: radius range r of the dominant region min r max The azimuth range of the dominant region α min α max Valid sample size N Ω , directional consistency C Ω probability mass and M Ω ④ Metadata: Analysis time (UTC timestamp), typhoon number and name, typhoon center location, typhoon movement speed, data source identifier, binning parameters (Δr, Δα), constraint parameters (N) min C min ⑤ Diagnostic information: calculation method identifier (such as robust estimation scheme number, whether time smoothing is applied), downgrade operation record (if any), warning or error code, calculation time.

[0047] In this embodiment of the invention, the system adopts a pipelined data processing architecture, with each module sequentially coupled in a data flow manner: the output of the data acquisition and preprocessing module is sequentially supplied to the coordinate transformation and harmonic decomposition module, the probability distribution construction module, the adaptive parameter identification module, the ventilation flow calculation and quality assessment module, and the result verification and output control module; the system stores each parameter and intermediate result in the memory, and the processor executes the instructions of each module to output typhoon ventilation flow diagnosis results that are insensitive to the selection of a fixed radius.

[0048] Example 3: Application Case Based on Examples 1 and 2 above, Example 3 provides a case study based on an actual typhoon (Typhoon No. 2 of 2021, see...). Figure 3 In the application scenario of this invention, using hourly 0.25°×0.25° reanalysis data from ERA5 as input, adaptive extraction and effect verification of ventilation flow are performed to verify the operability and physical representativeness of the method in real wind fields.

[0049] Regarding the identification of the original wind field and the typhoon center. For example... Figure 3 As shown in Figure (a), April 17, 2021, 00:00 UTC was selected as the analysis time, at which time Typhoon No. 2 of 2021 was located in the Northwest Pacific Ocean. The location of the typhoon center was determined by the minimum sea level pressure and the maximum vorticity at 850 hPa, with latitude and longitude of approximately (13.6°N, 129.8°E). Wind field data within a 10-15 latitude and longitude range around the typhoon center were extracted as the analysis area. The relative wind field was obtained by subtracting the typhoon's translational velocity from the original wind field, as shown below. Figure 3 As shown in Figure (b), the large-scale circulation structure of the β-vortex pair can be clearly identified.

[0050] Polar coordinate transformation and first-order harmonic structure identification. The horizontal wind components u and v are converted into tangential Vt and radial Vr, as follows: Figure 3 As shown in Figures (c) to (d), Fourier first-order harmonic decomposition is applied to each radius to extract the main signal of the asymmetric circulation, as follows: Figure 4 As shown in Figures (e) to (f), the cyclonic (C) and anticyclonic (A) structures of the β-vortex pair are clearly visible in the figures, indicating that the wind field possesses the dynamic basis for the formation of ventilation flow. The first-order horizontal wind field obtained after the inverse transformation is shown below. Figure 4 As shown in Figure (g).

[0051] Construct a joint probability distribution P(r,α) and identify the dominant region Ω. Based on step S300 of Example 1, on the wind field (u1,v1) after harmonic inverse transformation, the radius is grouped at intervals of Δr=1° latitude and longitude (1°~12°), and the wind direction is grouped at intervals of Δα=10° (0°~360°), constructing a two-dimensional statistical bin grid. For each radial ring, the wind speed threshold is calculated, and the distribution of grid points satisfying the conditions in each bin is statistically analyzed. After normalization, the probability distribution P(r,α) is obtained. Figure 5 As shown in Figure (b), the probability distribution clearly shows that the high-probability area is concentrated in a specific radius-wind direction combination. In this case, the probability peak is located at a radius r ≈ 2° and a wind direction α ≈ 315°, which corresponds to the maximum probability. This region satisfies the condition that the sample size N ≥ N0. min =50, wind direction consistency C≥0.65, forming a continuous high-probability region Ω. This spatial distribution characteristic accurately indicates the dominant region of ventilation flow between β-vortex pairs, without relying on any fixed radius setting, overcoming the limitations of traditional methods such as... Figure 4 The results in the middle (h) figure are sensitive to the radius value, which effectively avoids the interference of the minimum wind area in the eye of the typhoon and the β vortex on the main structure area.

[0052] Ventilation flow calculation and physical rationality verification. Within the dominant region Ω, a robust aggregation method with probabilistic weighting is used to calculate the ventilation flow: the zonal wind u and meridional wind v components corresponding to the interval with the highest probability are extracted, and their average value is calculated as the ventilation flow vector. For example... Figure 5 As shown in Figure (c), the calculated ventilation flow (red arrow) is highly consistent with the actual ventilation flow between the β-vortex pairs (C and A). Bootstrap resampling is performed to assess uncertainty, and the results show that the ventilation flow estimate has good stability. The computational quality index shows a high quality level, indicating that the estimate is robust and reliable. Physical plausibility checks show that, as... Figure 4 As shown in Figure (g), the calculated ventilation flow direction is highly consistent with the actual typhoon trajectory (solid black line). The angle between this direction and the actual typhoon movement direction is approximately 7°, which is significantly better than... Figure 4The deviations shown in Figure (h) when using fixed radii of 3° and 5° verify the physical rationality of the method. This consistency is particularly evident in scenarios where the typhoon's asymmetric structure is significant and the background steering airflow is weakened.

[0053] To verify the advantages of the method of this invention, comparative calculations were performed using the traditional fixed radius method (R=3°, 5°, 7°, 9°, 11°). Figure 4 As shown in Figure (h), the results of the fixed radius method reveal significant differences in the calculated ventilation flow under different radius settings. When the radius is 3°, the ventilation flow direction is relatively consistent with the actual movement of the typhoon; however, as the radius increases to 5°, 7°, 9°, and 11°, the ventilation flow direction gradually deviates, and the deviation from the actual movement of the typhoon becomes increasingly larger. This fully exposes the high sensitivity of the traditional method to radius selection. This sensitivity stems from the fact that when the radius is too large, the calculation region includes part or all of the circulation components of the β-vortex pair, causing the synthesized wind field to deviate from the true ventilation flow characteristics; while when the radius is too small, the extremely small wind area in the typhoon eye may introduce turbulence without physical meaning. In contrast, the adaptive method of this invention... Figure 5 As shown in Figure (c), the optimal radius (2° in this example) and wind direction (approximately 315°) are automatically identified based on probability distribution characteristics, without the need for manual parameter presets. The calculation results show that the ventilation flow is highly consistent with the actual movement of the typhoon, and the stability of the results is significantly better than the fixed radius method.

[0054] To verify the universality of the method of this invention, it was applied to the calculation of ventilation flow in 30 years of typhoons. Statistical results show that the method exhibits good applicability and stability in individual typhoon cases of different intensities, regions, and moving speeds. Compared with the fixed radius method, the adaptive method reduces the average deviation between the ventilation flow calculation results and the actual typhoon movement direction by about 35%, and eliminates the need for manual radius selection, achieving adaptive automation of ventilation flow calculation. Especially when the typhoon moves slowly, is close to land, and exhibits strong asymmetry, the importance of ventilation flow increases significantly. The method of this invention can more accurately capture ventilation flow characteristics, providing a reliable reference for typhoon track forecasting.

[0055] The objectives of this invention have been fully and effectively achieved through the above embodiments. Those skilled in the art will understand that this invention includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments described above. Although the invention has been described with reference to what is currently considered the most practical and preferred embodiments, it should be understood that the invention is not limited to the disclosed embodiments, and any modifications that do not depart from the functional and structural principles of the invention will be included within the scope of the claims.

Claims

1. An adaptive calculation method for typhoon ventilation flow based on maximum probability characteristics and overcoming radius sensitivity, characterized in that, It should include at least the following steps: S100. Determine the typhoon center at the target time and capture the horizontal wind field covering the typhoon's influence area with the typhoon center as the center. Determine and subtract the typhoon's translational velocity from the horizontal wind field to obtain the relative wind field. S200. Establish a polar coordinate system with the typhoon center as the origin, and convert the zonal and meridional wind components u and v in the relative wind field into tangential wind speed Vt and radial wind speed Vr relative to the typhoon center; at each radial radius r, perform first-order Fourier harmonic extraction on Vt and Vr along the azimuth angle λ to obtain the corresponding first-order harmonic components Vt1 and Vr1. Then, convert Vt1 and Vr1 to a rectangular coordinate system to obtain the first-order zonal and meridional wind components u1 and v1. S300. In a first-order horizontal wind field, the radius r is grouped according to the preset radial box interval Δr, and the wind direction α is grouped according to the preset wind direction box interval Δα, thus discretizing the analysis domain into (r, α) two-dimensional boxes; In each sub-box i, the instantaneous wind direction α is calculated based on u1 and v1. i And its frequency of occurrence is statistically analyzed, combined with the speed threshold v. min Remove low wind speed or abnormal samples and construct a radius-wind direction joint probability distribution P(r,α); S400. Search on P(r,α) for samples where the number of samples N ≥ N min Consistency index with direction C≥C min A set of highly consistent connected regions, N min C is the lower limit of the number of valid samples within a connected region. min As a threshold for directional consistency, the connected interval Ω with the highest probability is determined as the dominant ventilation flow region; S500. Robust statistics are performed on the first-order zonal and meridional wind components within Ω to obtain the corresponding zonal and meridional components u of the ventilation flow. v v v Based on this, the ventilation flow vector V is formed. v =(u v ,v v ) and its direction α v With modulus |V v |, and simultaneously calculate the quality index Q v Quantitatively estimate credibility; S600. Output ventilation flow vector V v Direction α v Modulus | V v |;When the quality index Q v Below the preset threshold or Ω does not meet N min and C min When constraints are applied, a result rejection or uncertainty flag is triggered.

2. The method according to claim 1, characterized in that, In step S100, the typhoon center location is determined by a combination of multiple indicators: the minimum sea level pressure and the maximum relative vorticity at 850 hPa are used as the main criteria, combined with the consistency check of the low-level convergence center and the radius of the maximum axisymmetric wind speed. When the center location is inconsistent, the center with consistent low-level relative vorticity and divergence coupling index is selected to ensure the stability and physical representativeness of the subsequent polar coordinate transformation. The typhoon translational velocity vector is obtained by least-squares linear fitting of the typhoon center position in adjacent time intervals, and the consistency check is performed with the potential vortex centroid tracking result. When the angle between the two exceeds the threshold, the translational velocity is adjusted by ensemble averaging.

3. The method according to claim 1 or 2, characterized in that, In step S100, the interception space range of the horizontal wind field is adaptively set: outer radius R win The radius of the shielding is taken from latitude and longitude of 6° to 12° or from the isobaric closure radius of 1.2 to 1.8 times to ensure a complete spatial structure covering the β-vortex pair and its transition zone; the inner shielding radius R in Based on the radius characteristics of weak winds and high gradient areas in the typhoon eye, a range of 0.5° to 1.0° is selected to suppress eye interference. Bilinear or higher-order conformal interpolation is used to complete the regular gridding of the horizontal wind field. Non-physical abrupt changes caused by topography or land surface are eliminated through missing measurement mask and outlier threshold detection.

4. The method according to claim 1, characterized in that, In step S200, the polar coordinate transformation employs subgrid interpolation and angle consistency constraints in space: firstly, the relative wind field is resampled to a uniform r-λ grid with the typhoon center as the pole using bicubic interpolation to preserve energy; the radial resolution Δr and azimuth resolution Δλ are adaptively set according to the wind field spatial resolution and the typhoon wind circle scale, respectively; during first-order harmonic extraction, bandpass filtering is applied to Vr and Vt along the λ direction to suppress higher-order noise; harmonic fitting uses weighted least squares, with weights inversely proportional to the local wind speed modulus and observation error, thus robustly obtaining the first-order harmonic components Vt1 and Vr1; when the first-order harmonic variance interpretation is lower than the preset threshold η... min The radial sample points marked at time are low-confidence samples and are not included in subsequent probability statistics.

5. The method according to claim 1 or 4, characterized in that, In step S200, for local anomalies caused by heavy precipitation or topography, an anomaly removal and missing measurement filling mechanism is set up: the Mahalanobis distance and robust z-score are calculated for the angular sequences of Vr and Vt on a fixed r, and the λ positions that exceed the threshold are marked as anomalies and removed from the harmonic fitting; when the proportion of effective angular samples on a certain r is lower than the set lower limit, the harmonic coefficients of the neighboring radius are used for radial extrapolation, and the extrapolation adopts first-order recursion and limits the coefficient drift with Bayesian regularization.

6. The method according to claim 1 or 4, characterized in that, In step S200, the inverse transformation of the first harmonic components satisfies the dual constraints of geometric consistency and energy consistency: after Vt1 and Vr1 are orthogonally transformed back to u1 and v1, it is required that the dominant rotation direction with the same sign as the original relative wind field is maintained in the local ring zone, and the ratio of the harmonic energy of |u1| and |v1| relative to the original field is in the interval [0.7, 1.3]. If these conditions are not met, local refitting and endpoint smoothing are triggered.

7. The method according to claim 1, characterized in that, In step S300, the radial compartment spacing Δr is 0.1°~0.25°, and the wind direction compartment spacing Δα is 5°~15°; the velocity threshold v min The quantile method of local ring velocity distribution was used to determine the value, and the value was taken from the 20th to 30th percentile. The joint probability distribution P(r,α) was obtained by estimating the kernel density of the normalized histogram or the equivalent von Mises distribution, and was weighted with effective sample weights that monotonically increase with wind speed and fitting explanatory power. Finally, the sum of the values ​​of each (r,α) sub-box was normalized to 1.

8. The method according to claim 1 or 7, characterized in that, In step S300, the wind direction α is calculated as α=arctan(v1 / u1), and the quadrant of the wind direction is determined according to the signs of u1 and v1, and the wind direction is uniformly converted to the meteorological wind direction definition of 0°~360°; the statistical method of the occurrence frequency is as follows: traverse all grid points in the first-order horizontal wind field (u1,v1), and classify them into the corresponding two-dimensional bins according to the radius r and wind direction α of each grid point, accumulate the number of grid points in each bin and form a frequency matrix N(r,α), and the frequency matrix is ​​normalized and transformed into a probability distribution P(r,α)=N(r,α) / ∑N(r,α).

9. The method according to claim 1, characterized in that, In step S400, the determination of highly consistent connected regions adopts a combination of topological connectivity and morphological constraints: first, a threshold segmentation is applied to P(r,α) to obtain candidate regions, and the 4-neighborhood or 8-neighborhood criteria are used to identify connected components; for each candidate component, the circular variance, radial span, and shape compactness of the directional distribution are calculated, and spurious peaks with excessive radial span or directional variance exceeding the threshold are eliminated; when there are multiple candidate components that meet the conditions, the component with both higher peak probability and smaller directional variance is selected, and the Bayesian information criterion is used to combine sample size and model complexity to complete the final selection.

10. The method according to claim 1 or 9, characterized in that, In step S400, the directional consistency index C is defined as the average composite vector length of a unit wind direction vector within the region, or equivalently as the percentage of samples falling within the prevailing wind direction tolerance zone, where the prevailing wind direction tolerance zone is 10°~20°; when there are multiple conditions satisfying N≥N min C≥C min When considering connected regions, first select Ω, which has the highest probability quality within the region, and then compare the number of effective samples with the consistency index C.

11. The method according to claim 1, characterized in that, In step S500, when robustly estimating the first-order horizontal wind component within Ω, the weighting coefficients are jointly determined by the binning probability P(r,α) and the local wind speed amplitude, and the median absolute deviation is used as the scaling parameter for adaptive estimation; the ventilation flow vector V is obtained by performing ≥200 resamplings using the bootstrap method. v Confidence intervals and bias correction terms; quality index Q v From the number of valid samples N eff The normalized weighted combination of directional consistency C and the half-width of the confidence interval is used to quantify the confidence level.

12. The method according to claim 1 or 11, characterized in that, In step S500, to avoid instability caused by single-time fluctuations, time consistency and multi-time smoothing are implemented: for consecutive time periods V v A joint smoothing method using exponentially weighted moving average and Kalman filtering is employed, where the state equation models the evolution of ventilation flow as a random walk and uses the environmentally guided airflow change rate as a priori process noise. Based on the generalized likelihood ratio test, when a sudden change is detected, the process noise is adaptively increased to track it quickly.

13. The method according to claim 1, characterized in that, In step S600, the quality index Q v By normalizing to 0-1 and setting a level threshold, when Q... v When Q is in the middle interval, output interval estimation and uncertainty strips; when Q v When the result falls below the minimum threshold, it is marked as unavailable and falls back to an alternative strategy, including expanding Δr or Δα, enabling suboptimal connected regions, or using alternative estimates from adjacent height layers.

14. A typhoon ventilation flow adaptive calculation system, based on the typhoon ventilation flow adaptive calculation method based on maximum probability characteristics and overcoming radius sensitivity as described in any one of claims 1 to 13, characterized in that, It should include at least the following modules: The data acquisition and preprocessing module is used to receive meteorological reanalysis data or numerical forecast data, locate the typhoon center at the target time and extract horizontal wind field data covering the typhoon's influence range, estimate the typhoon's translation speed and subtract it from the wind field to form a relative wind field. The coordinate transformation and harmonic decomposition module is used to convert the zonal and meridional wind components u and v in the relative wind field into tangential wind speed Vt and radial wind speed Vr relative to the typhoon center, and extract the first harmonic of each radial r along the azimuth angle λ to obtain Vt1 and Vr1. Then, the inverse transformation is used to obtain the first horizontal wind u1 and v1, forming a wind field after asymmetric signal enhancement. The probability distribution construction module is used to discretize the (u1,v1) field into two-dimensional bins according to preset radial intervals Δr and wind direction intervals Δα, calculate the frequency of wind direction occurrence in each bin, and combine it with the velocity threshold v. min After removing low-wind-speed or abnormal samples, the radius-wind-direction two-dimensional joint probability distribution P(r,α) is obtained by normalization. The adaptive parameter identification module is used to search for highly consistent connected regions in P(r,α) that satisfy the constraints of sample size and direction consistency. The connected interval Ω with the highest probability and satisfying the constraints is determined as the dominant ventilation flow region, which is used to adaptively determine the representative radial direction and the dominant wind direction to avoid fixed radius dependence. The ventilation flow calculation and quality assessment module is used to perform robust aggregation of u1 and v1 within the dominant ventilation flow region Ω to obtain the ventilation flow vector V. v =(u v ,v v ), direction α v With modulus |V v |and generate quality index Q v ; The result verification and output control module is used to output V. v α v 、|V v |, when Q v Below the preset threshold or Ω does not meet N min and C min When constraints are applied, a result rejection or uncertainty warning flag is triggered.

15. The system according to claim 14, characterized in that, The modules are sequentially coupled in a data flow manner: the output of the data acquisition and preprocessing module is supplied in sequence to the coordinate transformation and harmonic decomposition module, the probability distribution construction module, the adaptive parameter identification module, the ventilation flow calculation and quality assessment module, and the result verification and output control module; the system stores each parameter and intermediate result in memory, and the processor executes the instructions of each module to output typhoon ventilation flow diagnosis results that are insensitive to the selection of a fixed radius.

Images