A sea fog forecasting method, medium and system based on marine station observation data

Abstract

The application provides a sea fog forecasting method, medium and system based on marine station observation data, and belongs to the technical field of electric digital data processing. The method comprises the following steps: firstly, obtaining multi-year continuous hydro-meteorological observation data of the station, including air temperature, dew point temperature, sea water temperature, wind direction and wind speed data. Through wavelet transform, multi-scale decomposition is carried out to obtain change components of different time scales. Then, forecast factors including temperature difference and stability index are calculated, and layered Fisher discriminant analysis is carried out according to seasons and weather system types. By calculating the weight coefficient, establishing a polynomial discriminant equation set, and combining the probability density function to determine the discriminant critical value, it is determined whether there is fog weather. Finally, a three-level verification system is adopted for verification, a complete forecast product release system including the intensity grade of fog and the spatio-temporal evolution of fog is constructed, and the accurate prediction of sea fog is realized. The technical problem that the accuracy of the traditional forecasting method fluctuates greatly under different conditions is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electronic digital data processing, and in particular relates to a sea fog forecasting method, medium and system based on ocean station observation data. Background Art

[0002] As a common marine meteorological phenomenon, accurate forecasting of sea fog holds strategic significance and practical application value for maritime traffic safety, marine engineering operations, marine fishery production, and coastal military activities. In recent years, the rapid development of global maritime trade has led to a continuous increase in maritime traffic volume and a significant trend toward larger vessels, placing higher demands on the accuracy and timeliness of sea fog forecasts.

[0003] Currently, three main technical approaches are used to forecast sea fog: The first is based on numerical forecast models, which simulate the formation and evolution of sea fog by establishing numerical models that include water vapor, heat, and momentum exchange processes. Although these methods have a strong physical basis, due to the complexity of sea fog formation mechanisms and the model's high dependence on initial fields and boundary conditions, their forecast results are often less than ideal. The second is based on statistical methods, which primarily rely on historical data to establish statistical forecast models. This method is simple to operate, but it struggles to accurately describe the nonlinear characteristics of sea fog formation and has poor adaptability to abnormal weather conditions. The third is based on expert experience, relying primarily on the forecaster's judgment. While these methods can be used to make forecasts based on actual conditions, they suffer from strong subjectivity and difficulty in transferring experience.

[0004] Traditional forecasting methods often focus on a single meteorological element or a simple combination of elements, ignoring the complex interactions between multiple scales and multiple elements in the formation of sea fog, resulting in large fluctuations in forecast accuracy under different circumstances. Summary of the Invention

[0005] In view of this, the present invention provides a sea fog forecasting method, medium and system based on ocean station observation data, which can solve the technical problem that traditional forecasting methods often focus on a single meteorological element or a simple combination of elements, ignoring the complex interactions between multiple scales and multiple elements in the sea fog formation process, resulting in large fluctuations in forecast accuracy under different circumstances.

[0006] The present invention is achieved in that:

[0007] The first aspect of the application provides a sea fog forecasting method based on marine station observation data, comprising: obtaining multi-year continuous hydro-meteorological observation data of a station; performing multi-scale decomposition on the hydro-meteorological observation data by using a wavelet transform method; calculating a forecasting factor based on the multi-scale decomposition data; performing time period division and two-level classification on the forecasting factor to obtain hierarchical Fisher discriminant sample data; calculating a scatter matrix based on the Fisher two-level discriminant principle; establishing a Fisher discriminant function coefficient matrix by using a cross-validation method; constructing and solving a polynomial discriminant equation set to obtain a sea fog forecasting discriminant equation; calculating a discriminant critical value of the sea fog forecasting discriminant equation; verifying the sea fog forecasting discriminant equation by using a three-level test system; and constructing a forecasting product release system for release.

[0008] The hydro-meteorological observation data comprises station air temperature values, station dew point temperature values, upstream station dew point temperature values, station surface seawater temperature values, station wind direction and speed data, station pressure data, station relative humidity data, and near-station synchronous observation data, and the near-station synchronous observation data is sampled at a frequency of once per hour.

[0009] The multi-scale decomposition obtains interannual variation components, seasonal variation components, diurnal scale variation components, and random disturbance components.

[0010] The forecasting factor comprises a difference between upstream dew point temperature and local air temperature, a difference between station air temperature and dew point temperature, a difference between station air temperature and seawater surface temperature, a sea-air temperature difference vertical gradient, a near-surface atmospheric stability index, a water vapor flux convergence intensity, and an inversion layer stratification characteristic parameter.

[0011] The time period division is performed according to winter, spring, summer, autumn, and two transition periods of spring and autumn, the two-level classification is performed according to weather system types, and the weather system types include cold high pressure type, warm high pressure type, and frontal type.

[0012] The scatter matrix comprises a weighted inter-class scatter matrix and an intra-class scatter matrix, the weighted inter-class scatter matrix represents a difference degree between weighted means of foggy weather samples and non-foggy weather samples, and the intra-class scatter matrix represents a dispersion degree of samples within respective classes.

[0013] The cross-validation method performs multiple random grouping on the forecasting factor, calculates an intra-class dispersion cross-product sum of the forecasting factor, and establishes a Fisher discriminant function coefficient matrix containing a regularization term.

[0014] The discriminant critical value of the sea fog prediction discriminant equation is selected by analyzing the probability distribution characteristics of the fog sample and the no-fog sample, when the calculation result of the discriminant equation is greater than the discriminant critical value, it is determined that it is a foggy weather, when the calculation result of the discriminant equation is less than the discriminant critical value, it is determined that it is a no-fog weather.

[0015] Specifically, the method of the present application comprises the following steps:

[0016] S01, obtaining multi-year continuous hydro-meteorological observation data of the station, the hydro-meteorological observation data including station air temperature value, station dew point temperature value, upstream station dew point temperature value, station surface seawater temperature value, station wind direction and wind speed data, station pressure data, station relative humidity data, and near-station synchronous observation data, the near-station synchronous observation data adopting an hourly sampling frequency;

[0017] S02, performing multi-scale decomposition on the hydro-meteorological observation data by using a wavelet transform method to obtain interannual variation components, seasonal variation components, diurnal scale variation components and random disturbance components;

[0018] S03, calculating prediction factors based on the multi-scale decomposition data, the prediction factors including the difference between the upstream dew point temperature and the local air temperature, the difference between the station air temperature and the dew point temperature, the difference between the station air temperature and the seawater surface temperature, the sea-air temperature difference vertical gradient, the near-surface atmospheric stability index, the water vapor flux convergence intensity, and the inversion layer stratification characteristic parameter;

[0019] S04, dividing the prediction factors into time periods according to winter, spring, summer, autumn and the two transition periods of spring and autumn, and performing two-level classification according to the weather system type to obtain hierarchical Fisher discriminant sample data;

[0020] S05, calculating the weighted between-class scatter matrix and the within-class scatter matrix based on the Fisher two-level discriminant principle, and determining the weight coefficient through the sample representation;

[0021] S06, performing multiple random grouping on the prediction factors by using a cross-validation method, calculating the cross-product sum of the within-class scatter of the prediction factors, and establishing a Fisher discriminant function coefficient matrix containing a regularization term;

[0022] S07, constructing a polynomial discriminant equation group considering the nonlinear relationship of the prediction factors based on the Fisher discriminant function coefficient matrix and solving the polynomial discriminant equation group to obtain a sea fog prediction discriminant equation for each season and weather system type combination;

[0023] S08, calculating a discriminant threshold of the sea fog forecast discriminant equation based on a probability density function method, determining that it is a foggy day when the calculation result of the discriminant equation is greater than the discriminant threshold, and determining that it is a non-foggy day when the calculation result of the discriminant equation is less than the discriminant threshold;

[0024] S09, verifying the sea fog forecast discriminant equation by adopting a three-level verification system to obtain a forecast accuracy rate, a false alarm rate, a missed alarm rate, a threat score and a skill score;

[0025] S10, constructing a forecast product release system, wherein the forecast product includes a foggy or non-foggy discrimination result, a fog intensity level forecast, a time evolution forecast and a spatial distribution forecast.

[0026] On the basis of the above technical solution, the sea fog forecast method based on the observation data of the marine station can be further improved as follows:

[0027] The weather system type includes a cold high pressure type, a warm high pressure type and a frontal type.

[0028] The cold high pressure type usually brings clear, dry and cold weather. The cold high pressure system is a high pressure system formed by the sinking of cold air with high density, and is commonly seen in winter.

[0029] The warm high pressure type usually brings clear, warm and stable weather. The warm high pressure system is a high pressure system formed by the sinking of warm air after rising, and is commonly seen in summer.

[0030] The frontal type is formed at the junction of cold and warm air masses, and can be divided into cold front, warm front, stationary front and occluded front according to the relative position and movement direction of the cold and warm air masses. The frontal system usually brings precipitation and significant temperature change.

[0031] Further, the weighted inter-class scatter matrix represents the difference between the weighted mean values of the foggy weather and the non-foggy weather samples, and the intra-class scatter matrix represents the dispersion degree of the samples within their respective classes.

[0032] Further, the hydro-meteorological observation data is subjected to quality control to eliminate abnormal values, and objective analysis method is used to interpolate the missing data.

[0033] Further, the optimal forecast precursor period of the forecast factor is determined through lag correlation analysis.

[0034] Further, the discriminant threshold of the sea fog forecast discriminant equation is selected by analyzing the probability distribution characteristics of the foggy samples and the non-foggy samples.

[0035] Further, the three-level verification system includes a contingency table verification, a forecast skill evaluation and a forecast stability verification.

[0036] Furthermore, the forecast model is dynamically optimized through the forecast correction mechanism.

[0037] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned sea fog forecasting method based on ocean station observation data.

[0038] A third aspect of the present invention provides a sea fog forecasting system based on ocean station observation data, which includes the above-mentioned computer-readable storage medium.

[0039] Furthermore, in S01, the quality control equation for hydrological and meteorological observation data is as follows:

[0040]

[0041] Where Q i is the standardized residual; X i is the i-th observation value; is the sample mean; σ is the sample standard deviation. i When the value is >3, it is considered an abnormal value.

[0042] For the objective analysis and interpolation of missing data, the inverse distance weighting method is used:

[0043]

[0044] Where, X p is the estimated value of the point to be interpolated; X i is the observation value of the ith adjacent station; d i is the distance from the i-th station to the point to be interpolated; n is the number of adjacent stations involved in the interpolation.

[0045] In S02, continuous wavelet transform is used to decompose the time series:

[0046]

[0047] Where W f (a, b) are wavelet coefficients; f(t) is the original signal; ψ(t) is the wavelet mother function; a is the scale parameter; b is the translation parameter; ψ * represents the complex conjugate of the wavelet function.

[0048] The prediction factor calculation in S03 includes:

[0049] The difference between the upstream dew point temperature and the local air temperature:

[0050]

[0051] where ΔT ud is the temperature difference; is the upstream dew point temperature; T a is the local air temperature.

[0052] The difference between the local air temperature and the dew point temperature:

[0053] ΔT d = T a - T d ;

[0054] where ΔT d is the temperature difference; T d is the local dew point temperature.

[0055] The difference between the local air temperature and the sea surface temperature:

[0056] ΔT s = T a - T s ;

[0057] where ΔT s is the temperature difference; T s is the sea surface temperature.

[0058] The vertical gradient of the sea-air temperature difference:

[0059]

[0060] where γ t is the vertical temperature gradient; z2, z1 are two height layers.

[0061] The atmospheric stability index in the surface layer:

[0062]

[0063] where Ri is the Richardson number; g is the acceleration of gravity; is the average temperature; θ is the potential temperature; u is the wind speed.

[0064] The intensity of water vapor flux convergence:

[0065]

[0066] where Q is the water vapor flux convergence; q is the specific humidity; is the wind speed vector.

[0067] The inversion layer characteristic parameter:

[0068] I t = ΔT max · H;

[0069] where I t is the inversion intensity index; ΔTmax is the maximum temperature difference; H is the inversion layer thickness.

[0070] The Fisher discriminant analysis in S04-S07 specifically includes:

[0071] The between-class scatter matrix:

[0072]

[0073] In the formula, S B is the between-class scatter matrix; ω i is the weight of the ith class; m i is the mean vector of the ith class sample; and m is the mean vector of the overall sample.

[0074] The within-class scatter matrix:

[0075]

[0076] In the formula, S W is the within-class scatter matrix; x ij is the jth sample vector of the ith class; and n i is the number of samples of the ith class.

[0077] The regularized Fisher discriminant function:

[0078]

[0079] In the formula, J(ω) is the discriminant criterion function; ω is the discriminant vector; and λ is the regularization parameter.

[0080] The polynomial discriminant equation:

[0081]

[0082] In the formula, f(x) is the discriminant function; x i ,x j is the predictor; α ij ,β i are the coefficients; c is the constant term; and p is the number of predictors.

[0083] The determination of the discriminant threshold in S08 includes:

[0084] The probability density function based on kernel density estimation:

[0085]

[0086] In the formula, f h (x) is the probability density function; K(·) is the kernel function; h is the bandwidth parameter; and n is the sample size.

[0087] The optimal discriminant threshold:

[0088] T opt = argmin T {P miss (T)+αP false (T)};

[0089] where T opt is the optimal threshold; P miss is the false negative probability; P false is the false positive probability; and α is the weight coefficient.

[0090] The forecast verification in S09 includes:

[0091] Forecast accuracy:

[0092]

[0093] where ACC is the accuracy; H is the hit number; M is the false negative number; F is the false positive number; and CN is the correct negative number.

[0094] Threat score:

[0095]

[0096] where TS is the threat score.

[0097] Skill score:

[0098]

[0099] where SS is the skill score; and ACC r is the random forecast accuracy.

[0100] The forecast product generation in S10 includes:

[0101] Fog intensity level forecast:

[0102] V = a·exp(-b·RH) + c·ΔT d + d;

[0103] where V is the visibility; RH is the relative humidity; a, b, c, and d are fitting coefficients.

[0104] The spatial distribution forecast uses Kriging interpolation:

[0105]

[0106] where Z(x0) is the estimated value of the point to be forecasted; Z(x i ) is the observed value of the known point; and λ i is the weight coefficient.

[0107] The weight coefficient satisfies:

[0108]

[0109] Where γ(h) is the variation function; h ij is the distance between point i and point j; μ is the Lagrange multiplier.

[0110] Time evolution forecast:

[0111]

[0112] Where, f is the forecast factor; is the transport velocity; D is the diffusion coefficient; S is the source and sink term.

[0113] Characterization of forecast uncertainty:

[0114]

[0115] Where P(f|x) is the posterior probability; P(x|f) is the likelihood function; P(f) is the prior probability; and P(x) is the normalization factor.

[0116] Forecast revisions:

[0117]

[0118] Where, f c is the revised forecast value; f0 is the original forecast value; y is the observed value; is the observation operator; K is the gain matrix.

[0119] Gain matrix calculation:

[0120] K=BH T (HBH T +R) -1 ;

[0121] Where B is the background error covariance matrix; R is the observation error covariance matrix.

[0122] Compared with the prior art, the sea fog forecasting method, medium and system based on ocean station observation data provided by the present invention have the following beneficial effects:

[0123] 1. The system fully utilizes multi-source observational data from oceanographic stations. This includes not only conventional elements such as air temperature, dew point temperature, and wind speed, but also relevant indicators such as upstream dew point temperature, seawater temperature, and atmospheric stability, enabling a more comprehensive reflection of air-sea interactions. Furthermore, wavelet analysis is used to perform multi-scale decomposition of these observational data, effectively capturing patterns of change across different time scales and laying the foundation for accurate forecasting.

[0124] 2. A forecasting model based on Fisher discriminant analysis was established. By calculating weighted inter-class and intra-class scatter matrices, the optimal weights for each predictor were determined, improving the model's discriminant ability. Furthermore, by considering the nonlinear relationships between predictors, a polynomial discriminant equation system was constructed, further enhancing forecast accuracy.

[0125] 3. The forecast results are presented in a probabilistic manner. This not only identifies foggy conditions but also quantitatively predicts fog intensity, temporal evolution, and spatial distribution. The forecast uncertainty is reflected through a Bayesian probability model, enhancing the practicality of the forecast product.

[0126] 4. A comprehensive forecast verification system has been established. This includes contingency table testing, forecast skill assessment, and forecast stability testing, comprehensively evaluating the performance of the constructed model. Furthermore, a forecast correction mechanism dynamically optimizes the model, further improving the accuracy and reliability of the forecast.

[0127] In summary, the sea fog forecasting method proposed in this invention fully exploits the information potential of ocean station observation data, constructs a forecasting model based on mathematical statistics, and solves the technical problem that traditional forecasting methods often focus on a single meteorological factor or a simple combination of factors, while ignoring the complex interactions between multiple scales and multiple factors in the sea fog formation process, resulting in large fluctuations in forecast accuracy under different circumstances. BRIEF DESCRIPTION OF THE DRAWINGS

[0128] Figure 1 A flow chart of the method provided by the present invention.

[0129] Figure 2 These are four fog-causing weather situation diagrams in the embodiment, divided into four sub-diagrams: Figure a: high-pressure type entering the sea; Figure b: northward extension of the subtropical high ridge; Figure c: cyclone front type; Figure d: ridge-back trough front type; L represents the low-pressure cyclone center; H represents the high-pressure center.

[0130] Figure 3 Graph showing the wavelet decomposition results of the temperature time series in the embodiment.

[0131] Figure 4 : is the probability density distribution diagram of the foggy sample and the fog-free sample in March in the embodiment.

[0132] Figure 5 This is a comparison chart of the sea fog forecast skill scores under different weather system types in the embodiment. DETAILED DESCRIPTION

[0133] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0134] like Figure 1 FIG. 1 is a flow chart of a method for forecasting sea fog based on ocean station observation data provided by the first aspect of the present invention. The method comprises the following steps:

[0135] S01. Obtaining continuous hydrological and meteorological observation data for many years at the observation station. The hydrological and meteorological observation data include the station air temperature, station dew point temperature, upstream dew point temperature, station surface seawater temperature, station wind direction and speed data, station air pressure data, station relative humidity data, and synchronous observation data from adjacent observation stations. The synchronous observation data from adjacent observation stations are sampled once an hour.

[0136] S02. Use wavelet transform method to perform multi-scale decomposition on hydrological and meteorological observation data to obtain interannual variation component, seasonal variation component, daily scale variation component and random disturbance component;

[0137] S03. Calculate prediction factors based on multi-scale decomposition data. The prediction factors include the difference between the upstream dew point temperature and the local air temperature, the difference between the local air temperature and the dew point temperature, the difference between the local air temperature and the sea surface temperature, the vertical gradient of the sea-air temperature difference, the near-surface atmospheric stability index, the water vapor flux convergence strength, and the characteristic parameters of the inversion stratification.

[0138] S04. Divide the forecast factors into time periods according to winter, spring, summer, autumn, and the transition period between spring and autumn, and perform secondary classification according to the weather system type to obtain hierarchical Fisher discriminant sample data;

[0139] S05. Calculate the weighted inter-class scatter matrix and intra-class scatter matrix based on Fisher's second-level discriminant principle, and determine the weight coefficient based on the representativeness of the sample;

[0140] S06. Using a cross-validation method to randomly group the predictors multiple times, calculate the cross-product sum of the intra-class deviations of the predictors, and establish a Fisher discriminant function coefficient matrix including a regularization term;

[0141] S07. Based on the Fisher discriminant function coefficient matrix, a set of polynomial discriminant equations is constructed that takes into account the nonlinear relationship of the prediction factors and solved to obtain the sea fog forecast discriminant equation for each combination of season and weather system type;

[0142] S08. Calculating a critical value of a sea fog forecast discriminant equation based on a probability density function method, determining that foggy weather exists when the result of the discriminant equation is greater than the critical value, and determining that non-fog weather exists when the result of the discriminant equation is less than the critical value;

[0143] S09. Use a three-level verification system to verify the sea fog forecast discriminant equation and obtain the forecast accuracy, false alarm rate, missed alarm rate, threat score and skill score;

[0144] S10, a prediction product release system is constructed, and the prediction product includes a fog / no fog discrimination result, a fog intensity level prediction, a time evolution prediction, and a spatial distribution prediction.

[0145] The specific implementation of the above steps is described in detail as follows:

[0146] First is step S01. The purpose of this step is to obtain multi-year continuous hydro-meteorological observation data at the station, and to perform quality control and missing value interpolation on these data.

[0147] Specifically, for quality control of observation data, the standardized residual method is used to identify outliers. The calculation formula of the standardized residual is: Where Q i represents the standardized residual of the i-th observation value; X i represents the i-th observation value; represents the sample mean; and σ represents the sample standard deviation. When Q i > 3, the observation value is determined to be an outlier and is removed.

[0148] For the interpolation of missing data, the inverse distance weighting method is used for objective analysis. The calculation formula is: Where X p represents the estimated value of the point to be interpolated; X i represents the observation value of the i-th adjacent station; d i represents the distance from the i-th station to the point to be interpolated; and n represents the number of adjacent stations participating in the interpolation.

[0149] Through the above data preprocessing, complete and reliable hydro-meteorological observation data is obtained, laying a foundation for subsequent analysis.

[0150] Next is step S02, which uses continuous wavelet transform to perform multi-scale decomposition on time series data. The formula for wavelet transform is: Where W f (a, b) represents the wavelet coefficient; f(t) represents the original signal; ψ(t) represents the wavelet mother function; a represents the scale parameter; b represents the translation parameter; and ψ * represents the conjugate complex of the wavelet function.

[0151] Through wavelet transform, the original hydro-meteorological observation data can be decomposed into interannual variation components, seasonal variation components, daily scale variation components, and random disturbance components. This multi-scale decomposition helps better characterize the time variation characteristics of hydro-meteorological elements and lays a foundation for subsequent extraction of prediction factors.

[0152] In step S03, a series of predictors are calculated from the multi-scale decomposition data obtained in step S02. These predictors include:

[0153] Difference between upstream dew point temperature and local air temperature: where ΔT ud represents the temperature difference; represents the upstream dew point temperature; T a represents the local air temperature.

[0154] Difference between local air temperature and dew point temperature: ΔT d = T a -T d . Where ΔT d represents the temperature difference; T d represents the local dew point temperature.

[0155] Difference between local air temperature and sea surface temperature: ΔT s = T a -T s . Where ΔT s represents the temperature difference; T s represents the sea surface temperature.

[0156] Sea-air temperature difference vertical gradient: where γ t represents the temperature vertical gradient; z2, z1 represent two height layers.

[0157] PBL atmospheric stability index: where Ri represents the Richardson number; g represents the gravitational acceleration; represents the average temperature; θ represents the potential temperature; u represents the wind speed.

[0158] Water vapor flux convergence intensity: where Q represents the water vapor flux convergence; q represents the specific humidity; represents the wind speed vector.

[0159] Inversion layer characteristics parameter: I t = ΔT max · H. Where I t represents the inversion intensity index; ΔT max represents the maximum temperature difference; H represents the inversion layer thickness.

[0160] These predictors comprehensively reflect the influence of sea-air interaction, atmospheric stability, water vapor transport, etc. on the formation of sea fog, providing rich input variables for the construction of subsequent prediction models.

[0161] In step S04, the extracted forecast factors are first divided into time periods according to season, including winter, spring, summer, autumn, and the transition period between spring and autumn. The data is then further classified into three categories based on weather system type: cold high pressure, warm high pressure, and frontal. This hierarchical processing better reflects the different characteristics of sea fog formation in different seasons and weather systems.

[0162] Next, based on the principle of Fisher discriminant analysis, the weighted inter-class scatter matrix S of foggy and non-foggy weather samples under each classification combination is calculated. B and the intra-class scatter matrix S W Among them, the calculation formula of the inter-class scatter matrix is: Among them, ω i represents the weight of the i-th category; m i Represents the mean vector of the i-th class sample; m represents the mean vector of the overall sample. The calculation formula of the intra-class scatter matrix is: Among them, x ij represents the jth sample vector of the i-th category; n i represents the number of samples in the i-th category.

[0163] These statistics lay the foundation for the subsequent construction of Fisher discriminant function.

[0164] In step S05, the S calculated in step S04 is B and S W , the weight coefficients of each prediction factor are determined by the representativeness of the sample. Specifically, the Fisher discriminant criterion function is calculated: Where J(ω) represents the discriminant function, ω represents the discriminant vector, and λ represents the regularization parameter. By optimizing J(ω), the optimal weight coefficients of each prediction factor are obtained.

[0165] The purpose of this step is to assign different weight coefficients to each predictor based on its ability to distinguish between foggy and non-fog samples. This can better reflect the relative importance of each predictor in the subsequent discriminant model.

[0166] In step S06, the extracted predictors are randomly grouped multiple times using a cross-validation method. First, the samples are randomly divided into a training set and a validation set. Then, the cross-sum of intra-class deviations is calculated based on the training set to establish a Fisher discriminant function coefficient matrix including a regularization term. Finally, the established discriminant function is tested using the validation set.

[0167] Through multiple randomized cross-validation, the discriminant ability of each predictor can be evaluated more objectively, and a robust Fisher discriminant function coefficient matrix can be obtained, which lays the foundation for the subsequent construction of the discriminant equation.

[0168] In step S07, based on the Fisher discriminant function coefficient matrix obtained in step S06, a polynomial discriminant equation group considering the nonlinear relationship between the predictor factors is constructed: Where f(x) represents the discriminant function; x i ,x j represents the predictor; α ij ,β i represents the coefficient; c represents the constant term; p represents the number of predictors.

[0169] The optimal number of polynomial terms was selected by using the stepwise regression method, which reduced the model complexity as much as possible while ensuring the discrimination accuracy. The resulting polynomial discriminant equations can be used to give different sea fog forecast discriminant equations for different seasons and weather system types.

[0170] In step S08, first, based on the probability density functions of foggy samples and fog-free samples, the optimal discriminant critical value of the sea fog forecast discriminant equation is calculated. Here, the kernel density estimation method is used to determine the probability density function: Among them, f h (x) represents the probability density function; K(·) represents the kernel function; h represents the bandwidth parameter; and n represents the number of samples.

[0171] Then, the optimal discrimination threshold is determined by minimizing the weighted sum of the probability of missed alarm and the probability of false alarm: T opt =argmin T {P miss (T)+αP false (T)}. Among them, T opt represents the optimal critical value; P miss represents the probability of missing reports; P false represents the false alarm probability; α represents the weight coefficient.

[0172] Finally, when the result of the discriminant equation is greater than the critical value, it is judged to be foggy weather; when the result is less than the critical value, it is judged to be fog-free weather. The purpose of this step is to determine the discriminant boundary between foggy and fog-free.

[0173] In step S09, a three-level verification system is used to verify the constructed sea fog forecast discriminant equation.

[0174] First, the accuracy of the forecast results is evaluated using the contingency table test method. The calculation formula for the forecast accuracy is: where H represents the number of hits; M represents the number of misses; F represents the number of false alarms; and CN represents the number of correct negatives.

[0175] Secondly, threat score and skill score are calculated to comprehensively evaluate the performance of the forecast. The formula for calculating the threat score is: The formula for calculating the skill score is: where ACC r represents the accuracy of random forecast.

[0176] Finally, the consistency of the forecast results in different time periods is evaluated through forecast stability test.

[0177] Through the above three levels of test, the performance and reliability of the constructed sea fog forecast discrimination model can be comprehensively evaluated, which provides the basis for the subsequent release of forecast products.

[0178] In step S10, based on the sea fog forecast discrimination equation constructed in the foregoing steps, comprehensive forecast products including fog / no fog discrimination results, fog intensity level forecast, time evolution forecast and spatial distribution forecast are prepared.

[0179] Among them, the fog intensity level forecast adopts an exponential function model: V = a·exp(-b·RH) + c·ΔT d +d. Where V represents the visibility; RH represents the relative humidity; a, b, c, d represent the fitting coefficients.

[0180] The spatial distribution forecast adopts the method of Kriging interpolation, and its basic formula is: where Z(x0) represents the estimated value of the point to be forecast; Z(x i ) represents the observed value of the known point; λ i represents the weight coefficient. The calculation of the weight coefficient needs to satisfy the following matrix equation: where γ(h) represents the variation function; h ij represents the distance between point i and point j; μ represents the Lagrange multiplier.

[0181] The time evolution forecast adopts the form of partial differential equation: where f represents the forecast element; represents the transport speed; D represents the diffusion coefficient; S represents the source and sink term.

[0182] In addition, in order to reflect the uncertainty of the forecast results, the forecast products are also presented in the form of probability description. The specific formula is: where P(f|x) represents the posterior probability; P(x|f) represents the likelihood function; P(f) represents the prior probability; P(x) represents the normalization factor.

[0183] Finally, the forecast model is dynamically optimized through the forecast correction mechanism. The correction formula is: c =f0+K(y-Hf0). Where, f c represents the corrected forecast value; f0 represents the original forecast value; y represents the observed value; H represents the observation operator; K represents the gain matrix. The calculation formula of the gain matrix is: K = BH T (HBH T +R) -1 Where B represents the background error covariance matrix; R represents the observation error covariance matrix.

[0184] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned sea fog forecasting method based on ocean station observation data.

[0185] A third aspect of the present invention provides a sea fog forecasting system based on ocean station observation data, which includes the above-mentioned computer-readable storage medium.

[0186] Specifically, the principle of the present invention is:

[0187] 1. Data Preprocessing. First, obtain years of continuous hydrological and meteorological observation data from the observation station, including air temperature, dew point temperature, seawater temperature, wind speed and direction, air pressure, relative humidity, and other factors. Quality control and missing value supplementation are performed on these observation data to ensure data integrity and reliability, laying the foundation for subsequent analysis.

[0188] 2. Multi-scale feature extraction. Using wavelet transforms, time series data are decomposed at multiple scales to obtain components at different scales, including interannual variation, seasonal variation, daily variation, and random disturbances. This multi-scale decomposition helps better characterize the temporal variations of hydrological and meteorological elements, providing a basis for the selection of forecast factors.

[0189] 3. Calculation of Prediction Factors. Based on the multi-scale decomposition results, a series of prediction factors reflecting processes such as sea-air interaction, atmospheric stability, and water vapor transport were calculated. These factors include the difference between the upstream dew point temperature and the local air temperature, the difference between the local air temperature and the dew point temperature, the difference between the local air temperature and the seawater temperature, the vertical gradient of the seawater-air temperature difference, the near-surface atmospheric stability index, the strength of the water vapor flux convergence, and the characteristic parameters of the inversion stratification.

[0190] 4. Construction of a classification forecast model. First, samples were stratified according to season and weather system type (e.g., cold high pressure, warm high pressure, frontal, etc.) to reflect the different characteristics of sea fog formation under different environmental conditions. Then, based on the principles of Fisher discriminant analysis, weighted inter-class scatter matrices and intra-class scatter matrices were calculated to determine the optimal weight coefficients for each forecast factor. Finally, a set of discriminant equations was constructed using polynomial form to account for nonlinear relationships, and a discriminant model for sea fog forecasting was developed for different seasons and weather types.

[0191] 5. Determine the forecast critical value. By analyzing the probability density distribution of foggy and fog-free samples, the optimal critical value is determined by minimizing the probability of missed reports and false alarms. When the forecast model's calculation result is above this critical value, it is considered foggy; otherwise, it is considered fog-free.

[0192] 6. Forecast Verification and Correction. A three-tiered forecast verification system has been established, encompassing contingency table testing, forecast skill assessment, and forecast stability testing, to comprehensively evaluate the performance of the constructed model. Furthermore, through a forecast correction mechanism, the model is dynamically optimized in conjunction with real-time observational data to further improve forecast accuracy and reliability.

[0193] 7. Forecast Product Generation. Based on the above forecast model, comprehensive forecast products are generated, including fog identification results, fog intensity forecast, temporal evolution forecast, and spatial distribution forecast. The fog intensity forecast uses an exponential function model based on relative humidity and temperature differences; the temporal evolution forecast uses partial differential equations to describe the dynamic changes of fog; and the spatial distribution forecast uses Kriging interpolation to achieve regional forecasts.

[0194] The following is an example of a specific application scenario of the present invention: A coastal city oceanographic station has been facing the problem of low sea fog forecast accuracy for many years. To solve this problem, the station decided to adopt the sea fog forecast method proposed in this invention based on oceanographic station observation data.

[0195] First, the station collected continuous observation data for the past 10 years, including air temperature, dew point temperature, sea temperature, wind speed and direction, air pressure, relative humidity, and other factors. Table 1 lists some observation records for March 10, 2024.

[0196] Table 1 Observation data of marine meteorological stations on March 10, 2024

[0197] time Wind direction (°) Wind speed (m / s) Water temperature (℃) Temperature (℃) humidity(%) Dew point temperature Air pressure (hPa) 0:00 249 0.8 4.6 0.9 84 -1.4937 1025.1 1:00 217 1.1 4.6 0.8 89 -0.8039 1025 2:00 253 0.8 4.6 -0.1 92 - 1024.8 3:00 233 1.4 4.5 0.9 95 0.19077 1024.5 4:00 212 1.9 4.6 0.9 92 -0.2508 1024.2 5:00 211 1.7 4.6 0.9 92 -0.2508 1024.3

[0198] For these observation data, quality control was first performed. Taking the temperature data as an example, the standardized residual Q of each time point was calculated. i , found that the data at 02:00 was abnormal (Q i>3), so it was removed. For the missing dew point temperature data, the inverse distance weighting method was used to interpolate. After this series of preprocessing, the complete and reliable hydro-meteorological observation data were obtained.

[0199] Next, the wavelet transform was used to decompose the observation data at multiple scales. Figure 3 The results of the multi-scale decomposition of the temperature time series are shown in five subplots from top to bottom: the original temperature series, the seasonal and trend component, the weather scale component, the daily variation component, and the random noise component. Through wavelet transform, the original complex temperature time series is decomposed into different time scale variation characteristics, revealing the multi-scale characteristics contained in the temperature variation. Among them, the seasonal and trend component reflects the long-term trend of temperature change; the weather scale component reflects the 3-7 day cycle of weather system change; the daily variation component embodies the periodic change of temperature within a day; the random noise component represents the random fluctuations that are difficult to explain with deterministic rules. This multi-scale decomposition helps to extract key information from complex meteorological element changes and provides important feature parameters for sea fog forecasting. Based on the multi-scale decomposition data, the station calculated a series of forecast factors reflecting the interaction between sea and air, as shown in Table 2.

[0200] Table 2 Forecast factor calculation results

[0201]

[0202] With these forecast factors, the station next classified the observation data. First, according to the season, it was divided into winter (December-February), spring (March-May), summer (June-August), autumn (September-November), and spring-autumn transition period. Then, according to the weather system type, it was further subdivided into cold high pressure type, warm high pressure type, and frontal type. For the case of March 10, it belongs to the spring, cold high pressure type weather.

[0203] Based on Fisher discriminant analysis, the station calculated the weighted between-class scatter matrix S B and the within-class scatter matrix S W of each forecast factor under the conditions of spring and cold high pressure type, as shown in Table 3.

[0204] Table 3 Fisher discriminant analysis statistics

[0205]

[0206] Next, the station constructed a polynomial discriminant equation considering the nonlinear relationship of the forecast factors:

[0207]

[0208] Through stepwise regression screening, the optimal number of polynomial terms was determined. Furthermore, using cross-validation, a robust discriminant function coefficient matrix was obtained. For the March 10th case, the discriminant function calculated the value to be 4.92.

[0209] In order to determine the critical value for distinguishing fog and no fog, the station calculated the probability density functions of foggy samples and no fog samples respectively based on historical observation data using the kernel density estimation method. Figure 4 The probability density distribution of the discriminant function values ​​for foggy and non-fog samples is shown. The red curve in the figure represents the probability density distribution of foggy samples, while the blue curve represents the probability density distribution of non-fog samples. The intersection of the two curves indicates that there may be misjudgments in the forecast. The vertical black dotted line in the figure marks the optimal discriminant critical value of 4.75, which is determined by minimizing the weighted sum of the probability of missed alarms and the probability of false alarms. The figure also marks the discriminant value of 4.92 for the case on March 10, which is slightly higher than the critical value and is therefore judged to be foggy weather. In addition, the figure clearly identifies the false alarm area and the missed alarm area, which correspond to the probability areas where no fog is misjudged as foggy and foggy is misjudged as non-fog, respectively. This figure intuitively demonstrates the performance of the discriminant model and the basis for its application in actual forecasting.

[0210] Since the discriminant equation calculation result on March 10 (4.92) was greater than the critical value (4.75), the station determined that there was fog on that day. To further evaluate the reliability of the forecast results, the station established a three-level verification system:

[0211] 1. Contingency Table Test. Based on the forecast results and actual observations from March 1 to March 20, a contingency table was constructed, as shown in Table 4. The resulting forecast accuracy was 85%, the threat score was 0.72, and the skill score was 0.68.

[0212] Table 4 Contingency table of forecast results (March 1-March 20)

[0213] Observation of fog Observation without fog Fog forecast 42 7 No fog forecast 5 36

[0214] 2. Forecast Skill Evaluation. In addition to the above overall evaluation indicators, the station also calculated forecast accuracy, missed alarm rate, and false alarm rate for different weather system types. The results are shown in Table 5. It can be seen that the forecast performance is good in cold high-pressure weather, but there is a certain amount of false alarm problem in frontal weather.

[0215] Table 5 Forecast performance under different weather system types

[0216] Weather system type Forecast accuracy False negative rate False positive rate Cold high pressure type 90％ 6％ 10％ Warm high pressure type 85％ 12％ 15％ Frontal 80％ 15％ 20％

[0217] 3. Forecast stability test. The station conducted a time series analysis of the forecast results from March 1 to March 20 and found that the fluctuations of various indicators within different time periods were small, indicating that the forecast results were stable.

[0218] Figure 5 This study compares the forecasting skill of sea fog under three different weather system types: cold high pressure, warm high pressure, and frontal. The figure displays five evaluation metrics in bar chart form: forecast accuracy, threat score, skill score, false negative rate, and false positive rate. The figure shows that forecast performance is best under cold high pressure weather, with a forecast accuracy of 0.90, a threat score and skill score of 0.75 and 0.70, respectively, and false negative rates of only 0.06 and 0.10. Forecast performance under frontal weather, on the other hand, is relatively poor, with a forecast accuracy of 0.80, a threat score and skill score of 0.60 and 0.58, respectively, and a false negative rate of 0.15 and 0.20. These results indicate that sea fog forecasting methods have different applicability to different weather system types, and further optimization of the forecast model is needed for frontal weather systems.

[0219] Based on the above three-level test results, the station is relatively confident about the sea fog forecast for March 10.

[0220] As a result, the station released the following forecast products:

[0221] 1. Fog forecast: There will be fog on March 10, with an accuracy rate of 85%.

[0222] 2. Fog intensity forecast: Based on the exponential function model of relative humidity and temperature difference, visibility is expected to be between 2-3km.

[0223] 3. Temporal evolution forecast: Based on partial differential equation simulation, the fog is expected to continue until around 10:00 the next day and then gradually dissipate.

[0224] 4. Spatial distribution forecast: Using the Kriging interpolation method, it is expected that sea fog with low visibility will occur in most coastal areas.

[0225] 5. Forecast uncertainty: The posterior probability of the forecast result is estimated to be 75% through the Bayesian probability model.

[0226] Furthermore, the station also conducted further analysis based on the specific fog-causing weather conditions to improve the accuracy of the forecast, such as Figure 2 As shown, there are four types of fog-causing weather conditions:

[0227] Sea-entry high pressure type: When a high-pressure system moves from land to sea, its airflow characteristics change, creating conditions suitable for fog formation. Cold air over land flows over warmer water, forming evaporative fog or advection fog.

[0228] Northward extension of the subtropical high ridge: When the ridge line of the subtropical high pressure extends northward, the airflow characteristics and temperature and humidity conditions at its edge are conducive to the formation of fog, especially in certain geographical locations.

[0229] Cyclone front type: There is often warm and humid air flow in the front area of ​​the low-pressure system (cyclone), which meets the cold air on the ground, creating conditions suitable for fog formation.

[0230] Ridge-back trough-front type: In the area behind the high-pressure ridge and in front of the low-pressure trough, air flow convergence and changes in temperature and humidity conditions often occur, which are conducive to the formation of fog.

[0231] In subsequent monitoring, the station found that the forecast results were in good agreement with actual observations, demonstrating good forecasting skills. Through dynamic corrections, the station continuously optimized the forecast model, further improving the accuracy and reliability of sea fog forecasts.

[0232] It should be noted that the explanations of the relevant variables are shown in Table 6 below:

[0233] Table 6 Variable explanation table

[0234]

[0235]

[0236] The above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A sea fog forecasting method based on ocean station observation data, characterized in that: include: Obtaining continuous hydrological and meteorological observation data from observation stations for many years; Performing multi-scale decomposition on the hydrological and meteorological observation data using a wavelet transform method; Calculating a prediction factor based on the multi-scale decomposition data; The forecast factors are divided into time periods and classified into two levels to obtain hierarchical Fisher discriminant sample data; Calculate the scatter matrix based on Fisher's second-level discriminant principle; The Fisher discriminant function coefficient matrix was established by cross-validation method; the polynomial discriminant equation group was constructed and solved to obtain the sea fog forecast discriminant equation; Calculating a critical value of the sea fog forecast discriminant equation; A three-level verification system is used to verify the sea fog forecast discriminant equation; Build a forecast product release system for release.

2. The sea fog forecasting method based on ocean station observation data according to claim 1, characterized in that: The hydrological and meteorological observation data include the station air temperature value, the station dew point temperature value, the upstream measuring point dew point temperature value, the station surface seawater temperature value, the station wind direction and speed data, the station air pressure data, the station relative humidity data and the synchronous observation data of the adjacent measuring stations. The synchronous observation data of the adjacent measuring stations adopts a sampling frequency of once per hour.

3. The sea fog forecasting method based on ocean station observation data according to claim 2, characterized in that: The multi-scale decomposition obtains an interannual variation component, a seasonal variation component, a daily scale variation component and a random disturbance component.

4. The sea fog forecasting method based on ocean station observation data according to claim 3 is characterized in that: The prediction factors include the difference between the upstream dew point temperature and the local air temperature, the difference between the station air temperature and the dew point temperature, the difference between the station air temperature and the sea surface temperature, the vertical gradient of the sea-air temperature difference, the near-surface atmospheric stability index, the water vapor flux convergence intensity and the inversion stratification characteristic parameters.

5. The sea fog forecasting method based on ocean station observation data according to claim 4 is characterized in that: The time period division is carried out according to winter, spring, summer, autumn and the two transition periods of spring and autumn. The secondary classification is carried out according to the weather system type, and the weather system type includes cold high pressure type, warm high pressure type and front type.

6. The sea fog forecasting method based on ocean station observation data according to claim 5, characterized in that: The scatter matrix includes a weighted inter-class scatter matrix and an intra-class scatter matrix. The weighted inter-class scatter matrix represents the degree of difference between the weighted means of foggy weather samples and fog-free weather samples, and the intra-class scatter matrix represents the degree of dispersion of samples within their respective categories.

7. The sea fog forecasting method based on ocean station observation data according to claim 6, characterized in that: The cross-validation method randomly groups the predictors for multiple times, calculates the cross-product sum of the intra-class deviations of the predictors, and establishes a Fisher discriminant function coefficient matrix containing a regularization term.

8. The sea fog forecasting method based on ocean station observation data according to claim 7, characterized in that: The discriminant critical value of the sea fog forecast discriminant equation is selected by analyzing the probability distribution characteristics of foggy samples and fog-free samples to select the optimal classification threshold. When the discriminant equation calculation result is greater than the discriminant critical value, it is judged to be foggy weather, and when the discriminant equation calculation result is less than the discriminant critical value, it is judged to be fog-free weather.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the sea fog forecasting method based on ocean station observation data according to any one of claims 1 to 8.

10. A sea fog forecasting system based on ocean station observation data, characterized in that: Contains the computer-readable storage medium of claim 9.

Images