Southwest vortex identification method and system based on multi-source data fusion

Abstract

The application discloses a southwest vortex identification method and system based on multi-source data fusion, comprising collecting multi-source data of a preset area, and preprocessing the multi-source data; adopting an eight-quadrant wind field determination method to filter the multi-source data to obtain a southwest vortex candidate point, and adopting a cyclonic circulation determination method to determine a southwest vortex center in combination with a maximum value of a vertical component of vorticity; constructing a southwest vortex event library according to the southwest vortex center, synthesizing multiple vortices based on the southwest vortex center to obtain common characteristics, and performing dynamic diagnosis based on the common characteristics to obtain key dynamic factors; constructing a multi-source fusion southwest vortex identification model based on the key dynamic factors, training the multi-source fusion southwest vortex identification model by using the southwest vortex event library, inputting to-be-identified data into the multi-source fusion southwest vortex identification model, and outputting an identification result.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of atmospheric coupling, and in particular to a southwest vortex identification method and system based on multi-source data fusion. BACKGROUND

[0002] With the increase in the frequency and intensity of extreme precipitation events under the background of global warming, the association between the southwest vortex and the PHP event is becoming increasingly close. The southwest vortex, which is a mesoscale cyclonic vortex generated on the southeast edge of the Qinghai-Tibet Plateau and the Sichuan Basin, is a key system for triggering persistent heavy precipitation (PHP), rainstorms and other disastrous weather in the southwest region of China and the Yangtze River Basin. Its generation location is scattered (covering the Jiulong, Sichuan Basin and other regions), the underlying surface topography is complex (highlands and basins are interlaced), and the structure is significantly affected by the interaction of multiple atmospheric layers. Accurate identification of the center position, dynamic evolution and intensity change of the southwest vortex plays an irreplaceable role in improving the timeliness and accuracy of disaster weather warning.

[0003] Current southwest vortex identification techniques revolve around vortex identification-feature extraction-mechanism analysis, and have formed a technical framework based on single meteorological elements, traditional diagnostic methods and preliminary synthetic analysis. However, due to limitations in data coverage and method limitations, multi-source data deep fusion and dynamic accurate identification have not been achieved. The identification method relies on a single element, making it difficult to avoid interference systems. The center positioning lacks three-dimensional support, and the accuracy is significantly affected by the terrain. The data source is single, and the fusion degree is at a shallow stage. The synthetic analysis and dynamic mechanism are not combined, and lack of standardized support. Therefore, there is an urgent need to invent a new southwest vortex identification method to improve the accuracy and reliability of southwest vortex identification. SUMMARY

[0004] The purpose of the present application is to provide a southwest vortex identification method based on multi-source data fusion.

[0005] To achieve the above purpose, the present application is implemented according to the following technical solutions:

[0006] The present application comprises the following steps:

[0007] Collecting multi-source data of a predetermined area, and preprocessing the multi-source data; the multi-source data includes atmospheric reanalysis data, satellite remote sensing data, radar network observation data, encrypted sounding observation data, terrain elevation data and water vapor data;

[0008] Using an eight-quadrant wind field determination method to filter the multi-source data to obtain southwest vortex candidate points, and using a cyclonic circulation determination method combined with the maximum value of the vertical component of vorticity to determine the center of the southwest vortex;

[0009] According to the southwest vortex center, a southwest vortex event library is constructed, common characteristics are obtained by synthesizing multiple vortexes based on the southwest vortex center according to a dynamic mechanism, and key dynamic factors are obtained by dynamic diagnosis according to the common characteristics;

[0010] According to the key dynamic factors, a multi-source fusion southwest vortex identification model is constructed, the multi-source fusion southwest vortex identification model is trained by using the southwest vortex event library, and identification data is input into the multi-source fusion southwest vortex identification model to output an identification result.

[0011] Further, the method for screening by using the eight-quadrant wind field determination method through the multi-source data comprises:

[0012] Any grid point (i, j) in a target region is randomly selected as a target grid point, when the meridional and zonal wind fields in the four grid ranges of the target grid point satisfy v1>0, u2<0, u3<0, v4<0, v5<0, u6>0, u7>0, and v8>0, the target grid point is a cyclonic circulation; wherein u is the zonal wind speed, and v is the meridional wind speed;

[0013] When the wind directions of the upper grid point (i, j+1), the lower grid point (i, j-1), the left grid point (i-1, j), and the right grid point (i+1, j) of the cyclonic circulation satisfy u(i, j+1)<0, v(i-1, j)<0, u(i, j-1)>0, and v(i+1, j)>0, the vortex core region is a cyclonic circulation, and all grid points are traversed to output the target grid points meeting the requirements as southwest vortex candidate points.

[0014] Further, the method for determining the southwest vortex center by using the cyclonic circulation determination method combined with the maximum value of the vertical component of vorticity comprises:

[0015] The vortex center position is identified based on the maximum value of the vertical component of vorticity, when multiple maximum values appear, the vortex centers are merged, and the geometric center is taken as the vortex center; the expression of the vertical component of vorticity is:

[0016]

[0017] wherein is the longitudinal gradient of the zonal horizontal wind speed, is the transverse gradient of the meridional horizontal wind speed, and δ is the vertical component of vorticity.

[0018] Further, the method for constructing a southwest vortex event library according to the southwest vortex center comprises:

[0019] The fifth generation global reanalysis data of the European Centre for Medium-Range Weather Forecasts is used to concatenate the time, central longitude and latitude coordinates, and central intensity of the southwest vortex event into an independent vortex event by using a spatiotemporal correlation algorithm; the vortex event is set to have a minimum duration (more than 6 hours) and a minimum horizontal scale (diameter ≥ 200 km), and the vortex event set is output as a southwest vortex event library.

[0020] Further, according to the method for synthesizing multiple vortexes with the southwest vortex center as a reference based on a dynamic mechanism, the method comprises the following steps:

[0021] The multiple southwest vortexes identified during the PHP event are diagnosed and analyzed by using a classical vorticity equation to obtain main factors affecting local changes and trends of the vorticity; wherein the atmospheric variables used for the vorticity equation diagnosis include a horizontal wind field, a vertical velocity, and specific humidity; and the expression is as follows:

[0022]

[0023] wherein u is a zonal wind speed, v is a meridional wind speed, ζ is vorticity, f is a Coriolis parameter, w is a vertical velocity, p is air pressure, is a vertical gradient of the vortex flow, is a vertical gradient of the zonal wind speed, is a horizontal gradient of the Coriolis parameter, is a vertical gradient of the Coriolis parameter, is a vertical gradient of the meridional wind speed, is a horizontal divergence, represents a local change rate of the vorticity, is a relative vorticity advection term RVA, is a vorticity vertical transport, is a horizontal convergence and divergence DIV, is a vorticity tilting term, is a geostrophic vorticity advection term GVA;

[0024] The relative vorticity advection term and the horizontal convergence and divergence are split in the zonal and meridional directions;

[0025]

[0026] wherein is a Hamiltonian operator, RVA is a relative vorticity advection term, and DIV is a horizontal convergence and divergence;

[0027] By diagnosing and calculating the vorticity budget of multiple southwest vortexes, a multi-year data set of the vorticity budget is constructed, which is complete in time sequence and complete in physical quantities; and a dynamic mechanism of the interannual variation of the southwest vortex intensity is analyzed according to the multi-year data set of the vorticity budget from the perspective of atmospheric dynamics.

[0028] A synthetic analysis method based on the vortex power core is adopted to unify the accurate center position of the southwest vortex as the coordinate origin, and the alignment and synthetic average of each vorticity budget term are performed to obtain the common characteristics of the vorticity budget.

[0029] Further, the method for power diagnosis according to the common characteristics to obtain key power factors comprises:

[0030] The relative vorticity advection term, horizontal convergence and divergence, vorticity tilting term, geostrophic vorticity advection term and vorticity vertical transport are taken as power factors, and triple diagnosis analysis is performed based on the synthetic field:

[0031] (1) Quantitative contribution diagnosis: based on the high signal-to-noise ratio three-dimensional physical field, the power factor is calculated and compared at the grid point level to obtain the local change contribution rate of the power factor, and the power factor with a local change contribution rate greater than a contribution rate threshold is taken as a primary diagnosis, which includes a dominant positive contribution factor, a dominant negative contribution factor and a vertical structure difference.

[0032] (2) Spatio-temporal evolution diagnosis: the synthetic physical field is diagnosed based on the vortex center as the benchmark in different directions and life history stages to construct an evolution sequence of key power parameters and obtain a secondary diagnosis, which includes the causal time sequence relationship of the power factor and the vorticity intensity, and the spatial structure evolution of the power factor.

[0033] (3) Long-term trend diagnosis: the southwest vortex in different periods is independently synthesized and diagnosed, and the differences in the contribution rates of the power factors in different periods are compared to obtain a tertiary diagnosis, which includes trend power sources and dominant processes.

[0034] The key power factors include primary key factors, secondary key factors and modulation factors.

[0035] Further, the method for constructing a multi-source fusion southwest vortex identification model according to the key power factors comprises:

[0036] A multi-source fusion southwest vortex identification model based on a three-dimensional convolutional neural network is constructed, and atmospheric reanalysis data sets, radar networking data, encrypted sounding data and satellite data are taken as input data.

[0037] A hierarchical fusion strategy is adopted for feature fusion: a cross-modal attention gate is adopted to extract the spatial features of the input data in the area to be identified, and a primary fusion is performed according to the spatial features; a graph attention network is adopted to model the nonlinear coupling between the key power factors, and a secondary fusion is performed according to the nonlinear coupling; in the graph attention network, a feature vector composed of [low-level convergence intensity, middle-level vorticity vertical transport, relative vorticity advection, vorticity tilting] is taken as a node, and the edges between the nodes are constructed by calculating the physical interaction strength between the features.

[0038] According to the key dynamic factor, a factor feature vector is constructed, the low-layer convergence threshold is dynamically adjusted according to the ground elevation data, and the contributions of the horizontal wind shear and the buoyancy effect to the vorticity conversion are quantified based on the full-type vertical vorticity equation; wherein the factor feature vector = [low-layer convergence intensity, middle-layer vorticity vertical transport, relative vorticity advection, vorticity tilting]; the contribution degrees of the key dynamic factors are obtained, and the fusion weight is dynamically adjusted according to the contribution degrees;

[0039] The terrain-dependent threshold and the terrain complexity coefficient are obtained, and a decision output discrimination result is made according to the dynamic threshold self-adaptive mechanism; wherein the dynamic threshold self-adaptive mechanism is: when all the discrimination conditions of the southwest vortex are met, the discrimination target is output as the southwest vortex; the discrimination conditions of the southwest vortex are: the low-layer horizontal convergence intensity is greater than the terrain-dependent threshold, the middle-layer vorticity vertical transport is greater than 1.2*10 -9 hPa -1 s -2 , the absolute value of the relative vorticity advection is greater than 0.8*10 -9 s -2 , and the vorticity tilting is greater than the terrain complexity coefficient.

[0040] Further, the method for training the multi-source fusion southwest vortex discrimination model by using the southwest vortex event library comprises:

[0041] The binary cross-entropy loss is used to train the multi-source fusion southwest vortex discrimination model, and the difference between the probability distribution predicted by the multi-source fusion southwest vortex discrimination model and the true label is calculated;

[0042] The multi-source fusion southwest vortex discrimination model outputs a probability value belonging to the southwest vortex for each southwest vortex event, and the loss function calculates the cross-entropy of the probability value and the true label; through the back propagation algorithm, the parameters of the multi-source fusion southwest vortex discrimination model are updated in the direction of minimizing the average cross-entropy of all training samples, and the southwest vortex samples are judged as positive class and the non-southwest vortex samples are judged as negative class;

[0043] The training is continuously performed until the difference between the probability distribution predicted by the multi-source fusion southwest vortex discrimination model and the true label converges.

[0044] In the second aspect, a southwest vortex discrimination system based on multi-source data fusion comprises:

[0045] A data acquisition module is configured to acquire multi-source data of a preset area, and to pre-process the multi-source data; the multi-source data comprises atmospheric reanalysis data, satellite remote sensing data, radar network observation data, encrypted sounding observation data, terrain elevation data, and water vapor data;

[0046] Center determination module: for adopting eight-quadrant wind field judgment method to obtain southwest vortex candidate point through screening of the multi-source data, and adopting cyclonic circulation judgment method to determine the southwest vortex center in combination with the maximum of vertical component of vorticity;

[0047] Power diagnosis module: for constructing a southwest vortex event library according to the southwest vortex center, performing multi-vortex synthesis to obtain common characteristics according to the dynamic mechanism and taking the southwest vortex center as a reference, and performing power diagnosis to obtain key dynamic factors according to the common characteristics;

[0048] Modeling output module: for constructing a multi-source fusion southwest vortex identification model according to the key dynamic factors, training the multi-source fusion southwest vortex identification model by using the southwest vortex event library, inputting to-be-identified data into the multi-source fusion southwest vortex identification model, and outputting identification results.

[0049] The present application has the following beneficial effects:

[0050] Compared with the prior art, the present application has the following technical effects:

[0051] The present application has the following beneficial effects: BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The present application is a method and system for identifying southwest vortexes based on multi-source data fusion. DETAILED DESCRIPTION

[0053] The present application is a method and system for identifying southwest vortexes based on multi-source data fusion.

[0054] The present application is a method and system for identifying southwest vortexes based on multi-source data fusion.

[0055] As shown in the figure, in the present embodiment, the following steps are included: Figure 1

[0056] Collecting multi-source data of a predetermined area, and preprocessing the multi-source data; the multi-source data includes atmospheric reanalysis data, satellite remote sensing data, radar network observation data, encrypted sounding observation data, terrain elevation data and water vapor data;

[0057] ​In the actual evaluation, the southwest vortex process in the southeast edge of A1 plateau and C1 basin (26°N-33°N, 100°E-108°E) from July 15 to July 20, 2020 is taken as the research object;

[0058] The preprocessing includes: converting radar polar coordinate data into Cartesian grid data, re-sampling satellite data and water vapor data to the same spatial grid through bilinear interpolation; converting the time dimension into UTC time to ensure that all data timestamps are synchronized;

[0059] The radar data adopts terrain shielding correction algorithm; the sounding data is removed by quality control formula; the ERA5 data is corrected by comparing the observation value of the sounding station, and the 700 hPa wind speed is corrected;

[0060] The eight-quadrant wind field determination method is used to obtain the southwest vortex candidate point through the multi-source data, and the cyclonic circulation determination method is used to determine the southwest vortex center combined with the maximum value of the vertical component of vorticity;

[0061] In the actual evaluation, the grid point (i, j) in the target area (for example, 29°N, 103°E) is selected, and based on the preprocessed ERA5 700 hPa wind speed data, the eight-quadrant wind field threshold condition is verified:

[0062] The determination range is: taking the target grid point as the center, the area with a radius of 4 grid distances (about 111 km, 0.25°×4) is divided into 8 quadrants, and the wind speed threshold of each quadrant needs to meet: v1>0 (the north quadrant meridional wind is south wind), u2<0 (the northeast quadrant zonal wind is west wind), u3<0 (the east quadrant zonal wind is west wind), v4<0 (the southeast quadrant meridional wind is north wind), v5<0 (the south quadrant meridional wind is north wind), u6>0 (the southwest quadrant zonal wind is east wind), u7>0 (the west quadrant zonal wind is east wind), and v8>0 (the northwest quadrant meridional wind is south wind);

[0063] Core area verification: for the target grid point meeting the above conditions, the wind direction of its upper, lower, left and right four adjacent points (i, j+1), (i, j-1), (i-1, j) and (i+1, j) is further verified: u(i, j+1)=-2.3 m / s<0 (west wind of upper grid point), v(i-1, j)=-1.8 m / s<0 (north wind of left grid point), u(i, j-1)=2.1 m / s>0 (east wind of lower grid point), and v(i+1, j)=2.5 m / s>0 (south wind of right grid point), which completely meets the core area determination condition of cyclonic circulation, so the grid point (29°N, 103°E) is marked as a southwest vortex candidate point;

[0064] The vertical component of vorticity distribution of the target area 700 hPa (the main activity layer of the southwest vortex) is calculated, and the results show that:

[0065] There are three vorticity maximum points around the candidate point, which are (28.9 °N, 102.9 °E, δ = 1.8 × 10 -5 s -1 ), (29.1 °N, 103.0 °E, δ = 2.1 × 10 -5 s -1 ), and (29.0 °N, 103.1 °E, δ = 1.9 × 10 -5 s -1 ).

[0066] According to the method, multiple maximum values are merged, and the geometric center coordinates are calculated: latitude = (28.9 + 29.1 + 29.0) / 3 = 29.0 °N, longitude = (102.9 + 103.0 + 103.1) / 3 = 103.0 °E, and finally the southwest vortex center is determined as (29.0 °N, 103.0 °E), which is completely coincided with the center of the Tbb low value area (Tbb = -50 °C, corresponding to the strong convective cloud system) of the FY-4A satellite, and the radar reflectivity shows that there is a spiral-shaped precipitation band ≥ 30 dBz in this area, verifying the accuracy of the center positioning;

[0067] According to the southwest vortex center, a southwest vortex event library is constructed, and common features are obtained by synthesizing multiple vortices based on the southwest vortex center according to the dynamic mechanism, and key dynamic factors are obtained by dynamic diagnosis according to the common features;

[0068] In the actual evaluation, the target area from July 15, 2020 to July 20 is identified by southwest vortex hour by hour, and the independent vortex events are connected in series by combining the spatiotemporal correlation algorithm (temporal continuity: center distance < 100 km between adjacent time points; spatial correlation: center intensity change < 50%), and finally 3 southwest vortex events meeting the minimum duration ≥ 6 hours and the minimum horizontal scale ≥ 200 km are selected, and a small event library is constructed;

[0069] The contribution rate of the divergence term DIV and the relative vorticity advection RVA to the enhancement of the southwest vortex is 79.4%, among which the meridional wind plays the most important role, which can explain up to 69.1% of the southwest vortex enhancement trend;

[0070] Taking the centers of the three southwest vortex events as coordinate origin points, the average of each term of the vorticity equation is obtained based on ERA5 data, and the common features of multiple vortices are obtained: the vertical distribution of vorticity budget: the horizontal convergence and divergence term DIV at the middle and low levels (850 hPa-700 hPa) is positive (average value 1.5 × 10 -9 s -2 ), and the vorticity vertical transport term VVT at the middle layer (700 hPa-500 hPa) is positive (average value 1.3 × 10 -9 hPa -1 s -2), and the relative vorticity advection term (RVA) at the upper level (above 500 hPa) is negative (average value -0.9×10 -9 s -2 ), forming a vertical structure of low-level convergence and vorticity increase, mid-level transport maintenance, and upper-level advection weakening, which is consistent with the classical dynamic model of the southwest vortex.

[0071] Horizontal distribution characteristics: the positive contribution area of 700 hPa DIV is annularly distributed (radius 100-200 km), corresponding to the low-level convergence zone observed by radar; the positive contribution area of 500 hPa VVT is concentrated within a central radius of 50-150 km, overlapping with the low-temperature area of the FY-4A satellite cloud top, verifying the relevance of the dynamic factor and thermal characteristics.

[0072] According to the key dynamic factor, a multi-source fusion southwest vortex identification model is constructed, the multi-source fusion southwest vortex identification model is trained using the southwest vortex event library, and the identification result is output by inputting the to-be-identified data into the multi-source fusion southwest vortex identification model.

[0073] In this embodiment, the method of screening the multi-source data by using the eight-quadrant wind field determination method includes:

[0074] Randomly selecting any grid point (i, j) in the target area as a target grid point, when the zonal and meridional wind fields within a four-grid range of the target grid point satisfy v1>0, u2<0, u3<0, v4<0, v5<0, u6>0, u7>0, and v8>0, the target grid point is a cyclonic circulation; wherein u is the zonal wind speed, and v is the meridional wind speed;

[0075] When the wind directions of the upper grid point (i, j+1), the lower grid point (i, j-1), the left grid point (i-1, j), and the right grid point (i+1, j) of the cyclonic circulation satisfy u(i, j+1)<0, v(i-1, j)<0, u(i, j-1)>0, and v(i+1, j)>0, the vortex core area is a cyclonic circulation, and all grid points are traversed to output the target grid points meeting the requirements as southwest vortex candidate points.

[0076] In this embodiment, the method of determining the southwest vortex center by using the cyclonic circulation determination method combined with the maximum value of the vertical component of vorticity includes:

[0077] Based on the maximum value of the vertical component of vorticity to identify the vortex center position, when multiple maximum values appear, the vortex centers are merged, and the geometric center is taken as the vortex center; the expression of the vertical component of vorticity is:

[0078]

[0079] wherein is the longitudinal gradient of the zonal horizontal wind speed, is the lateral gradient of the zonal wind speed, and δ is the vertical component of vorticity.

[0080] In the embodiment, the method for constructing a southwest vortex event library according to the southwest vortex center comprises the following steps:

[0081] The fifth generation global reanalysis data of the European Centre for Medium-Range Weather Forecasts is used, and the time, central longitude and latitude coordinates, and central intensity of the southwest vortex event are connected in series by a spatiotemporal correlation algorithm to form an independent vortex event; the vortex event is set to have a minimum duration (more than 6 hours) and a minimum horizontal scale (diameter ≥ 200 km), and the vortex event set is output as a southwest vortex event library.

[0082] In the embodiment, the method for synthesizing multiple vortices according to the dynamic mechanism and taking the southwest vortex center as a reference comprises the following steps:

[0083] The multiple southwest vortices identified during the PHP event are diagnosed and analyzed by using a classical vorticity equation to obtain main factors affecting the local change and trend of vorticity; the atmospheric variables used for the diagnosis of the vorticity equation include a horizontal wind field, a vertical velocity, and specific humidity; and the expression is as follows:

[0084]

[0085] where u is a zonal wind speed, v is a meridional wind speed, ζ is vorticity, f is a Coriolis parameter, w is a vertical velocity, p is air pressure, is a vertical gradient of the vorticity, is a vertical gradient of the zonal wind speed, is a lateral gradient of the Coriolis parameter, is a longitudinal gradient of the Coriolis parameter, is a vertical gradient of the meridional wind speed, is a horizontal divergence, represents a local change rate of the vorticity, is a relative vorticity advection term RVA, is a vorticity vertical transport, is a horizontal convergence and divergence DIV, is a vorticity tilting term, is a geostrophic vorticity advection term GVA;

[0086] The relative vorticity advection term and the horizontal convergence and divergence are split in the zonal and meridional directions;

[0087]

[0088] where is a Hamiltonian operator, RVA is a relative vorticity advection term, and DIV is a horizontal convergence and divergence;

[0089] A multi-year dataset of vorticity budget is constructed by diagnosing the vorticity budget of multiple southwest vortexes. The dynamic mechanism of the interannual variability of the southwest vortex intensity is analyzed according to the multi-year dataset of vorticity budget from the perspective of atmospheric dynamics.

[0090] A synthetic analysis method based on the dynamic core of the vortex is adopted. The accurate center position of the southwest vortex is unified as the coordinate origin. The vorticity budget items are aligned and averaged to obtain the common characteristics of the vorticity budget.

[0091] In this embodiment, the method for obtaining key dynamic factors according to the common characteristics includes:

[0092] The relative vorticity advection term, horizontal convergence and divergence, vorticity tilting term, geostrophic vorticity advection term, and vorticity vertical transport are taken as dynamic factors. Triple diagnosis analysis is carried out based on the synthetic field:

[0093] (1) Quantitative contribution diagnosis: Based on the high signal-to-noise ratio three-dimensional physical field, the dynamic factors are quantitatively calculated and compared at the grid point level to obtain the local change contribution rate of the dynamic factors. The dynamic factors with a local change contribution rate greater than the contribution rate threshold are taken as the first-level diagnosis. The first-level diagnosis includes the dominant positive contribution factor, the dominant negative contribution factor, and the vertical structure difference.

[0094] (2) Spatio-temporal evolution diagnosis: The synthetic physical field is diagnosed based on the vortex center as the benchmark and divided into azimuth and life history stages. The evolution sequence of the key dynamic parameters is constructed to obtain the second-level diagnosis. The second-level diagnosis includes the causal temporal relationship between the dynamic factors and the vorticity intensity, and the spatial structure evolution of the dynamic factors.

[0095] (3) Long-term trend diagnosis: The southwest vortexes in different periods are independently synthesized and diagnosed. The differences in the contribution rates of the dynamic factors in different periods are compared to obtain the third-level diagnosis. The third-level diagnosis includes the trend dynamic root and the dominant process.

[0096] The key dynamic factors include the first-level key factor, the second-level key factor, and the modulation factor.

[0097] In actual evaluation, the first-level diagnosis (local change contribution rate): The contribution rate of each dynamic factor to the local change of vorticity is calculated (formula: contribution rate = absolute value of factor / sum of absolute values of all factors). The results show that the DIV contribution rate is 0.42, the VVT contribution rate is 0.35, the RVA contribution rate is 0.15, and the vorticity tilting term VT contribution rate is 0.08. Among them, the contribution rates of DIV and VVT are greater than the contribution rate threshold of 0.337, which are determined as the first-level key factors (dominant positive contribution factors).

[0098] Secondary diagnosis (spatiotemporal evolution relationship): analyze the timing relationship between dynamic factors and vorticity intensity in the life history stage (generation period, development period, and extinction period): the peak value of DIV (2.1 x 10 -9 s -2 ) in the development period is 2 hours ahead of the peak value of the super vorticity, and the peak value of VVT (1.6 x 10 -9 hPa -1 s -2 ) is synchronous with the peak value of the vorticity, indicating that DIV is the starting factor of the vortex enhancement, VVT is the maintenance factor, and the two constitute the secondary key factor;

[0099] Tertiary diagnosis (trend dynamic source): compare the dynamic factor differences of the three events: the contribution rate of DIV and VVT of SWV-20200717 (the strongest vortex) is 30%-50% higher than that of the other two events, and the corresponding GPM water vapor data shows that the water vapor transport flux (8 x 10 6 kg / (m·s)) in the South China Sea during the event is 1.5 times that of the other events, indicating that water vapor enhancement indirectly enhances the vortex intensity by promoting low-level convergence (increasing DIV), and the water vapor transport flux is included in the key dynamic factor set as a modulation factor;

[0100] The primary factor is DIV and VVT, the secondary factor is the timing correlation of DIV-VVT, and the modulation factor is the water vapor transport flux.

[0101] In this embodiment, the method for constructing a multi-source fusion southwest vortex recognition model according to the key dynamic factors comprises the following steps:

[0102] A multi-source fusion southwest vortex recognition model based on a three-dimensional convolutional neural network is constructed, and atmospheric reanalysis data sets, radar network data, encrypted sounding data, and satellite data are used as input data;

[0103] A hierarchical fusion strategy is used for feature fusion: a cross-modal attention gate is used to extract the spatial features of the input data in the area to be recognized, and primary fusion is performed according to the spatial features; a graph attention network is used to model the nonlinear coupling between the key dynamic factors, and secondary fusion is performed according to the nonlinear coupling; in the graph attention network, a feature vector composed of [low-level convergence intensity, middle-layer vorticity vertical transport, relative vorticity advection, and vorticity tilting] is used as a node, and the edges between the nodes are constructed by calculating the physical interaction strength between the features;

[0104] A factor feature vector is constructed according to the key dynamic factors, the low-level convergence threshold is dynamically adjusted according to the ground elevation data, and the contribution of horizontal wind shear and buoyancy effect to the transformation of vorticity is quantified based on the full-type vertical vorticity equation; wherein the factor feature vector = [low-level convergence intensity, middle-layer vorticity vertical transport, relative vorticity advection, and vorticity tilting]; the contribution of each key dynamic factor is obtained, and the fusion weight is dynamically adjusted according to the contribution;

[0105] The terrain-dependent threshold and the terrain complexity coefficient are obtained, and a dynamic threshold adaptive mechanism is used to make a decision and output a discrimination result; the dynamic threshold adaptive mechanism is as follows: when all the discrimination conditions of the southwest vortex are met, the discrimination target is output as a southwest vortex; the discrimination conditions of the southwest vortex are as follows: the low-level horizontal convergence intensity is greater than the terrain-dependent threshold, the mid-level vorticity vertical transport is greater than 1.2*10 -9 hPa -1 s -2 , the absolute value of the relative vorticity advection is greater than 0.8*10 -9 s -2 , and the vorticity tilt is greater than the terrain complexity coefficient.

[0106] In actual evaluation, a multi-source fusion southwest vortex discrimination model based on a three-dimensional convolutional neural network is constructed, and the model structure is as follows:

[0107] The input layer: four types of feature tensors, respectively, ERA5 dynamic features (u, v, w, δ, dimension: time 12 hours*space 32*32*pressure layer 3), radar reflectivity features (Z, dimension: time 12 hours*space 32*32*1), satellite Tbb features (dimension: time 12 hours*space 32*32*1), and terrain features (H, dimension: space 32*32*1);

[0108] The feature fusion layer: the primary fusion adopts a cross-modal attention gate (extracting spatial correlation features of radar and ERA5), and the secondary fusion adopts a graph attention network (modeling the nonlinear coupling of DIV and VVT, fusion weights: DIV is 0.55, VVT is 0.35, and water vapor transport is 0.1.

[0109] In this embodiment, the method for training the multi-source fusion southwest vortex discrimination model using the southwest vortex event library includes the following steps:

[0110] The multi-source fusion southwest vortex discrimination model is trained using binary cross-entropy loss to calculate the difference between the probability distribution predicted by the multi-source fusion southwest vortex discrimination model and the true label;

[0111] The multi-source fusion southwest vortex discrimination model outputs a probability value belonging to a southwest vortex for each southwest vortex event, and the loss function calculates the cross-entropy of the probability value and the true label; through a back propagation algorithm, the parameters of the multi-source fusion southwest vortex discrimination model are updated in the direction of minimizing the average cross-entropy of all training samples, so that the southwest vortex samples are classified as positive samples, and the non-southwest vortex samples are classified as negative samples;

[0112] The training is continuously performed until the difference between the probability distribution predicted by the multi-source fusion southwest vortex discrimination model and the true label converges;

[0113] In the actual evaluation, 192 samples of 3 events of the southwest vortex event library (96 samples of positive class: southwest vortex period; 96 samples of negative class: non-southwest vortex period) are taken as the training set, a binary cross-entropy loss function is used, a learning rate is 0.001, the number of iterations is 50 times, and when the loss function value converges to 0.08 (stable fluctuation ± 0.01), the training is stopped.

[0114] In a second aspect, a southwest vortex identification system based on multi-source data fusion includes:

[0115] A data acquisition module is configured to acquire multi-source data of a preset area, and pre-process the multi-source data; the multi-source data includes atmospheric reanalysis data, satellite remote sensing data, radar network observation data, encrypted sounding observation data, terrain elevation data and water vapor data.

[0116] A center determination module is configured to obtain a southwest vortex candidate point by filtering the multi-source data using an eight-quadrant wind field determination method, and determine a southwest vortex center using a cyclonic circulation determination method combined with a vertical component maximum of vorticity.

[0117] A dynamic diagnosis module is configured to construct a southwest vortex event library according to the southwest vortex center, obtain common features by synthesizing multiple vortices with the southwest vortex center as a reference according to a dynamic mechanism, and obtain key dynamic factors by dynamic diagnosis according to the common features.

[0118] A modeling output module is configured to construct a multi-source fusion southwest vortex identification model according to the key dynamic factors, train the multi-source fusion southwest vortex identification model using the southwest vortex event library, input to-be-identified data into the multi-source fusion southwest vortex identification model, and output an identification result.

[0119] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for identifying a southwest vortex based on multi-source data fusion, characterized in that, The method comprises the following steps: Collecting multi-source data of a preset area, preprocessing the multi-source data; the multi-source data comprises atmospheric reanalysis data, satellite remote sensing data, radar network observation data, encrypted sounding observation data, terrain elevation data and water vapor data; Using an eight-quadrant wind field determination method to screen the multi-source data to obtain a southwest vortex candidate point, and using a cyclonic circulation determination method combined with a vertical component maximum of vorticity to determine a southwest vortex center; Constructing a southwest vortex event library according to the southwest vortex center, synthesizing multiple vortices according to a dynamic mechanism and taking the southwest vortex center as a reference to obtain common characteristics, and performing dynamic diagnosis according to the common characteristics to obtain key dynamic factors; Constructing a multi-source fusion southwest vortex identification model according to the key dynamic factors, training the multi-source fusion southwest vortex identification model by using the southwest vortex event library, inputting to-be-identified data into the multi-source fusion southwest vortex identification model, and outputting an identification result; The method for constructing a multi-source fusion southwest vortex identification model according to the key dynamic factors comprises: Constructing a multi-source fusion southwest vortex identification model based on a three-dimensional convolutional neural network, and taking an atmospheric reanalysis data set, radar network data, encrypted sounding data and satellite data as input data; Using a hierarchical fusion strategy to perform feature fusion: using a cross-modal attention gate to extract spatial features of input data in a to-be-identified region, and performing primary fusion according to the spatial features; using a graph attention network to model nonlinear coupling between key dynamic factors, and performing secondary fusion according to the nonlinear coupling; in the graph attention network, a feature vector composed of [low-level convergence intensity, middle-layer vorticity vertical transport, relative vorticity advection and vorticity tilting] is taken as a node, and edges between nodes are constructed by calculating physical interaction intensity between features; Constructing a factor feature vector according to the key dynamic factors, dynamically adjusting a low-level convergence threshold according to terrain elevation data, and quantifying contributions of horizontal wind shear and buoyancy effect to vorticity conversion based on a full-type vertical vorticity equation; wherein the factor feature vector is [low-level convergence intensity, middle-layer vorticity vertical transport, relative vorticity advection and vorticity tilting]; the contribution degrees of the key dynamic factors are obtained, and the fusion weight is dynamically adjusted according to the contribution degrees; The terrain-dependent threshold and terrain complexity coefficient are acquired, and a dynamic threshold adaptive mechanism is used to output a discrimination result; wherein the dynamic threshold adaptive mechanism is: when all the discrimination conditions of the southwest vortex are met, the discrimination target is output as the southwest vortex; the discrimination conditions of the southwest vortex are: the low-level horizontal convergence intensity is greater than the terrain-dependent threshold, the mid-level vorticity vertical transport is greater than 1.2*10 -9 hPa -1 s -2 , the absolute value of the relative vorticity advection is greater than 0.8*10 -9 s -2 , and the vorticity tilt is greater than the terrain complexity coefficient.

2. The southwest vortex identification method based on multi-source data fusion according to claim 1, characterized in that, The method for screening the multi-source data by using the eight-quadrant wind field determination method comprises: Randomly select any grid point (i, j) in the target area as a target grid point, when the zonal and meridional wind field in the four grid distance range of the target grid point meets v1>0, u2<0, u3<0, v4<0, v5<0, u6>0, u7>0, v8>0, then the target grid point is a cyclonic circulation; wherein is the zonal wind speed, is the meridional wind speed; When the wind directions of the target grid point (i, j+1), the lower grid point (i, j-1), the left grid point (i-1, j) and the right grid point (i+1, j) of the cyclonic circulation satisfy u(i, j+1)<0, v(i-1, j)<0, u(i, j-1)>0 and v(i+1, j)>0, the vortex core region is a cyclonic circulation, all grid points are traversed, and the target grid points meeting the requirements are output as southwest vortex candidate points.

3. The southwest vortex identification method based on multi-source data fusion according to claim 1, characterized in that, The method for determining the southwest vortex center by using the cyclonic circulation determination method combined with the vertical component maximum of vorticity comprises: Based on the vertical component maximum of vorticity, the vortex center position is identified, when multiple maximum values appear, the vortex centers are merged, and the geometric center is taken as the vortex center; the expression of the vertical component of vorticity is: wherein is the longitudinal gradient of the zonal horizontal wind speed, is the transverse gradient of the meridional horizontal wind speed, is the vertical component of the vorticity.

4. The southwest vortex identification method based on multi-source data fusion according to claim 1, characterized in that, The method for constructing a southwest vortex event library according to the southwest vortex center comprises: The fifth generation global reanalysis data of the European Centre for Medium-Range Weather Forecasts is adopted, and the time, central longitude and latitude coordinates, and central intensity of the southwest vortex event are connected by a spatiotemporal correlation algorithm to form an independent vortex event; the vortex event is set with a minimum duration and a minimum horizontal scale, and the vortex event set is output as a southwest vortex event library.

5. The southwest vortex identification method based on multi-source data fusion according to claim 1, characterized in that, According to the method for synthesizing multiple vortexes based on the center of the southwest vortex, the method comprises the following steps of: The classical vorticity equation is used to diagnose and analyze a plurality of southwest vortexes identified during the PHP event, so as to obtain main factors affecting local changes and trends of vorticity; wherein the atmospheric variables used for vorticity equation diagnosis include horizontal wind field, vertical velocity and specific humidity; and the expression is as follows: where is the zonal wind speed, is the meridional wind speed, is the vorticity, is the Coriolis parameter, w is the vertical velocity, is the air pressure, is the vertical gradient of the vorticity, is the vertical gradient of the zonal wind speed, is the horizontal gradient of the Coriolis parameter, is the vertical gradient of the Coriolis parameter, is the vertical gradient of the meridional wind speed, is the horizontal divergence, represents the local change rate of the vorticity, is the relative vorticity advection term RVA, is the vertical transport of the vorticity, is the horizontal convergence divergence DIV, is the vorticity tilting term, is the geostrophic vorticity advection term GVA; The relative vorticity advection term and the horizontal convergence and divergence are split in the zonal and meridional directions; wherein is the Hamiltonian operator, is the relative vorticity advection term, is the horizontal convergence divergence; By diagnosing and calculating the vorticity budget of a plurality of southwest vortexes, a multi-year data set of vorticity budget is constructed, which is complete in time sequence and complete in physical quantity; and the dynamic mechanism of the interannual variation of the southwest vortex intensity is analyzed from the perspective of atmospheric dynamics according to the multi-year data set of vorticity budget; A synthesis analysis method based on the dynamic core of the vortex is adopted, the accurate center position of the southwest vortex is unified as the coordinate origin, and each vorticity budget term is aligned and averaged to obtain the common characteristics of the vorticity budget.

6. The southwest vortex identification method based on multi-source data fusion according to claim 1, characterized in that, According to the method for obtaining key dynamic factors by dynamic diagnosis based on the common characteristics, the method comprises the following steps: The relative vorticity advection term, the horizontal convergence and divergence, the vorticity tilting term, the geostrophic vorticity advection term and the vorticity vertical transport are taken as dynamic factors, and three-dimensional physical fields with high signal-to-noise ratio are used for three-dimensional diagnostic analysis based on the synthetic field: (1) Quantitative contribution diagnosis: based on the three-dimensional physical field with high signal-to-noise ratio, the dynamic factors are quantitatively calculated and compared at the grid point level to obtain the local change contribution rate of the dynamic factors, and the dynamic factors with a local change contribution rate greater than a contribution rate threshold are taken as first-level diagnosis, the first-level diagnosis includes a dominant positive contribution factor, a dominant negative contribution factor and a vertical structure difference; (2) Spatiotemporal evolution diagnosis: the vortex center is taken as the benchmark to diagnose the synthetic physical field in different directions and different life history stages, and the evolution sequence of the key dynamic parameters is constructed to obtain second-level diagnosis, the second-level diagnosis includes the causal time sequence relationship between the dynamic factors and the vorticity intensity, and the spatial structure evolution of the dynamic factors; (3) Long-term trend diagnosis: the southwest vortexes in different periods are independently synthesized and dynamically diagnosed, and the differences in the contribution rates of the dynamic factors in different periods are compared to obtain third-level diagnosis, the third-level diagnosis includes the trend dynamic root and the dominant process; The key dynamic factors include first-level key factors, second-level key factors and modulation factors.

7. The southwest vortex identification method based on multi-source data fusion according to claim 1, characterized in that, The method for training the multi-source fusion southwest vortex identification model by using the southwest vortex event library comprises the following steps: The binary cross-entropy loss is used to train the multi-source fusion southwest vortex identification model, and the difference between the probability distribution predicted by the multi-source fusion southwest vortex identification model and the real label is calculated. The multi-source fusion southwest vortex identification model outputs a probability value belonging to the southwest vortex for each southwest vortex event, and a loss function calculates the cross-entropy of the probability value and the true label; through a back propagation algorithm, parameters of the multi-source fusion southwest vortex identification model are updated in a direction of minimizing the average cross-entropy of all training samples, and southwest vortex samples are judged as positive classes, and non-southwest vortex samples are judged as negative classes; Constant training is performed until the difference between the probability distribution predicted by the multi-source fusion southwest vortex identification model and the true label converges.

8. A multi-source data fusion based Southwest Vortex identification system for performing the method of any one of claims 1-7. Comprise: A data acquisition module is configured to acquire multi-source data of a preset area, and to pre-process the multi-source data; the multi-source data comprises atmospheric reanalysis data, satellite remote sensing data, radar network observation data, encrypted sounding observation data, terrain elevation data and water vapor data; A center determination module is configured to obtain southwest vortex candidate points by screening the multi-source data using an eight-quadrant wind field determination method, and to determine a southwest vortex center using a cyclonic circulation determination method combined with a maximum value of a vertical component of vorticity; A dynamic diagnosis module is configured to construct a southwest vortex event library according to the southwest vortex center, to obtain common features by performing multi-vortex synthesis with the southwest vortex center as a reference according to a dynamic mechanism, and to obtain key dynamic factors by performing dynamic diagnosis according to the common features; A modeling output module is configured to construct a multi-source fusion southwest vortex identification model according to the key dynamic factors, to train the multi-source fusion southwest vortex identification model using the southwest vortex event library, to input to-be-identified data into the multi-source fusion southwest vortex identification model, and to output an identification result.

Images