Low-altitude atmospheric duct prediction method over the sea based on artificial intelligence ensemble error correction

Abstract

The present invention belongs to the technical field of marine meteorology and relates to a method for forecasting low-altitude atmospheric ducts at sea based on artificial intelligence ensemble error correction. This method establishes a set of forecast error correction algorithms and dynamically adjusts the weights of the ensemble members to achieve ensemble error correction of the forecast results, overcoming the high spatiotemporal uncertainty and unstable correction accuracy of a single correction algorithm. The method of the present invention incorporates numerical forecasting methods using empirical statistical post-processing techniques. Compared with purely data-driven low-altitude atmospheric duct forecasting methods, its results are more reasonable and stable, and its effective forecast time is longer.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine meteorology and relates to a method for forecasting low-altitude atmospheric duct at sea based on artificial intelligence ensemble error correction. Background Art

[0002] Atmospheric ducts are special weather phenomena caused by drastic changes in the vertical gradient of the atmospheric refractive index, which seriously affect the propagation of electromagnetic waves in the atmosphere. Based on the differences in duct height and profile shape, atmospheric ducts can generally be divided into evaporation ducts, surface ducts, and suspended ducts. Among them, surface ducts and suspended ducts with relatively high duct heights are generally referred to as low-altitude atmospheric ducts. The calculation of low-altitude atmospheric ducts is usually based on the calculation of the atmospheric corrected refractive index based on factors such as atmospheric temperature and humidity. Subsequently, the height, thickness, strength and other indicators of the low-altitude atmospheric duct are diagnosed based on the vertical gradient changes of the atmospheric corrected refractive index.

[0003] Low-altitude atmospheric duct forecasts typically use numerical forecasting methods. This involves using numerical weather models to generate numerical weather forecast products that capture the three-dimensional spatial distribution of atmospheric temperature, humidity, and pressure over the next few days. These forecasts are then used to diagnose these changes and ultimately produce a forecast of future changes in the type, height, thickness, intensity, and radio wavebands of atmospheric ducts within the target region. Generally speaking, there are three ways to improve the accuracy of low-altitude atmospheric duct forecasts: first, expanding observational data sources and developing more advanced data assimilation systems to provide more accurate physical constraints for numerical simulations. Second, improving numerical model differencing or gridding methods to refine the physical processes of numerical models. This improves local forecast simulations by coupling more comprehensive physical processes or optimizing key parameters. Third, based on regional historical observation data and feedback, various mathematical methods are used to establish a mapping relationship between observations and forecast results, and post-processing is used to eliminate systematic errors in forecast results.

[0004] However, given sparse observations at sea, relying solely on traditional methods like data assimilation and model optimization is no longer sufficient to improve forecast accuracy. In recent years, forecast post-processing methods based on artificial intelligence algorithms have provided new insights into improving low-altitude atmospheric duct forecasts. However, as a data-driven empirical method, AI-based approaches to correcting low-altitude atmospheric duct forecast errors rarely guarantee high-precision error correction under all spatiotemporal conditions. Their practical application is subject to significant uncertainty in different sea areas, seasons, and weather environments. Currently, no single artificial intelligence algorithm can guarantee superior duct correction performance over other algorithms under all spatiotemporal conditions. Summary of the Invention

[0005] Based on the problems existing in the current existing technology, the present invention provides a low-altitude atmospheric waveguide forecasting method at sea based on artificial intelligence ensemble error correction. This method establishes a forecast error correction algorithm set through multiple methods such as polynomial fitting, machine learning, and deep learning, and realizes the error set correction of the forecast results by dynamically adjusting the weights of the ensemble members, thereby overcoming the problems of high spatiotemporal uncertainty and unstable correction accuracy of a single correction algorithm.

[0006] The present invention provides a method for forecasting low-altitude atmospheric duct at sea based on artificial intelligence ensemble error correction, comprising the following steps:

[0007] (1) Construct an atmospheric corrected refractivity forecast framework based on the COAWST model to calculate the atmospheric corrected refractivity;

[0008] (2) Carry out local optimization experiments in the target area, taking the atmospheric temperature and humidity below 5 km as the main evaluation variables, screen the physical parameterization schemes in the coupled model, evaluate the impact of changes in cloud microphysics scheme, radiation scheme, cumulus convection scheme, surface layer scheme, planetary boundary layer scheme, wave base dissipation scheme, and ocean vertical turbulence mixing scheme on the atmospheric temperature and humidity simulation, and select the 4-5 schemes with the highest sensitivity to conduct further multiple groups of sensitivity experiments;

[0009] (3) Combine historical observations and reanalysis data, using the simulated observation correlation coefficient and root mean square error as evaluation indicators to evaluate the simulation capability of each option combination;

[0010] (4) Based on the sensitivity assessment results, generate parameterized scheme configurations for coupled model operation and calculate the atmospheric corrected refractive index;

[0011] (6) Construct an ensemble of error correction algorithm models; conduct atmospheric refractivity forecast correction evaluation tests, and determine the final ensemble members based on the evaluation results;

[0012] (7) Using the Bayesian model averaging (BMA) algorithm, all error correction algorithm ensemble members are averaged based on the performance of each algorithm in the historical period, and initial weights are assigned. The weights are then dynamically adjusted according to the subsequent error correction performance. Based on the posterior probabilities of all ensemble members, weights are assigned to each member after normalization, and then the weighted sum is calculated to obtain the final atmospheric corrected refractivity ensemble forecast result.

[0013] (8) Based on the atmospheric corrected refractive index ensemble forecast results, low-altitude atmospheric waveguide diagnosis is performed to obtain waveguide characteristic diagnosis results.

[0014] Preferably, the error correction algorithm is selected from a combination of any two or more of polynomial fitting, machine learning algorithm, and deep learning algorithm.

[0015] Preferably, step (7) specifically includes: using the observation data of the past 120 hours in units of days, solving the maximum likelihood estimation using the expectation-maximization EM algorithm, and converting the observation data into a normal distribution; based on the forecast results after error correction in the same period, using the Markov Monte Carlo MCMC method, calculating the marginal likelihood function by constructing a Markov chain and Monte Carlo integration, and statistically calculating the posterior probability of each error correction algorithm member respectively; according to the posterior probabilities of all members, assigning weights to each member after normalization, and then performing weighted summation to obtain the final atmospheric corrected refractive index ensemble forecast result.

[0016] The method of the present invention integrates numerical forecasting methods with empirical statistical post-processing techniques. Compared with purely data-driven low-altitude atmospheric duct forecasting methods, its results are more reasonable and stable, and its effective forecast time is longer. Compared with traditional low-altitude atmospheric duct numerical forecasting methods, its forecast results are more accurate after error correction and are more adaptable to different environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic flow chart of a method for forecasting low-altitude atmospheric duct at sea based on artificial intelligence ensemble error correction provided in an embodiment of the invention;

[0018] Figure 2 This is a correction sequence diagram of the atmospheric refractive index error at an altitude of 869m calculated by different correction algorithms and the BMA method;

[0019] Figure 3 Correction sequence diagram of atmospheric refractive index error at 1046m altitude calculated by different correction algorithms and BMA method;

[0020] Figure 4 Correction sequence diagram of atmospheric refractive index error at an altitude of 1342m calculated by different correction algorithms and BMA method;

[0021] Figure 5 is the root mean square error of atmospheric refractive index correction using different correction algorithms and the BMA method. DETAILED DESCRIPTION

[0022] To facilitate understanding of the present invention, the present invention is described in more detail below with reference to the accompanying drawings and specific embodiments. Preferred embodiments of the present invention are shown in the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in this specification. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present disclosure.

[0023] The present invention provides a method for forecasting low-altitude atmospheric duct at sea based on artificial intelligence ensemble error correction, and its implementation process is as follows: Figure 1As shown, the specific steps include:

[0024] 1. First, a framework for forecasting atmospheric corrected refractivity within the forecast area is established based on the COAWST air-sea coupled model. Based on the regularly released GFS global atmosphere forecast fields, GOFS global ocean forecast fields, and WaveWatch III global wave operational forecast field data, unified processing is performed in terms of data format conversion, regional cutting, and projection transformation interpolation to provide initial and boundary field drives for each COAWST component model. In the COAWST model, the three component models of atmosphere, ocean, and wave perform numerical simulations independently, and provide the meteorological and hydrological variables calculated by each component model to the model coupler at each exchange time step, and read the variables calculated by other component models as input. Based on the simulation results, the distribution of factors such as air pressure, humidity, and temperature in the region is extracted to calculate the atmospheric corrected refractivity.

[0025] The atmospheric refractive index can be calculated by the following formula:

[0026] ;

[0027] Where N represents the refractive index (unit: N-units), T represents the temperature (unit: K), P represents the air pressure (unit: hPa), and e represents the water vapor pressure, which can be calculated using the following formula:

[0028] ;

[0029] Where q represents the specific humidity (unit: g / kg), ε is a constant (0.622);

[0030] ;

[0031] Where M represents the corrected atmospheric refractive index (unit: M-units), R represents the radius of the Earth (6371km), and h represents the altitude (unit: m).

[0032] Secondly, local optimization experiments were conducted in the target area to identify the parameterization scheme configuration with the best simulation accuracy. Focusing on atmospheric temperature and humidity below 5 km as the primary evaluation variables, the physical parameterization schemes in the coupled model were screened. The impact of changes in cloud microphysics, radiation, cumulus convection, surface layer, planetary boundary layer, wave-base dissipation, and ocean vertical turbulence mixing on the atmospheric temperature and humidity simulations was evaluated. The four to five most sensitive schemes were selected for further sensitivity testing. Subsequently, the simulation performance of each scheme combination was evaluated using historical observations and reanalysis data, using simulation-observation correlation coefficients and root mean square error as evaluation metrics. Based on the sensitivity assessment results, parameterization scheme configurations were generated for use in the coupled model operation and for calculating forecast information such as atmospheric refractivity. Using the optimized parameterization scheme configurations, historical long-term return simulations were conducted to create training datasets for error correction modeling.

[0033] Subsequently, a set of error correction algorithms was constructed, using historical long-term simulated return data of the atmospheric corrected refractive index element as forecast training data and atmospheric corrected refractive index calculated from concurrent sounding, analysis field, and reanalysis data within the region as observational baseline data. In selecting error correction algorithms, algorithms with significantly different principles were selected to construct the error correction model. The algorithms are categorized into three types: the first type is traditional arithmetic methods, which use mathematical methods such as polynomial fitting (e.g., the pattern output statistics MOS method) for error correction; the second type is classical machine learning algorithms, such as BP neural networks, random forests, and support vector machines; and the third type is typical deep learning algorithms, such as convolutional neural networks (CNNs), long short-term memory (LSTMs), and gated recurrent units (GRUs).

[0034] Based on the aforementioned algorithms, we conducted an evaluation test for atmospheric refractivity forecast corrections. To ensure diversity among ensemble members, we screened out algorithms with very similar correction results and replaced them with other algorithms (such as extreme gradient boosting (XGBoost) and generative adversarial networks (GANs). After evaluation, we determined the members of the error correction ensemble for ensemble calculations.

[0035] During ensemble calculations, the Bayesian Model Averaging (BMA) algorithm is used to average all error correction algorithm ensemble members based on their historical performance. Initial weights are assigned and dynamically adjusted based on subsequent error correction performance. This ensemble averaging framework assigns different conditional probability weights to each member, skewing the ensemble results towards outstanding members.

[0036] The specific implementation steps are as follows: using the past 120 hours of observation data on a daily basis, the expectation-maximization (EM) algorithm is used to solve the maximum likelihood estimate and convert the observation data into a normal distribution. Then, based on the forecast results after error correction during the same period, the Markov Monte Carlo (MCMC) method is used to calculate the marginal likelihood function by constructing a Markov chain and Monte Carlo integration. The posterior probability of each error correction algorithm member is statistically calculated separately. Finally, based on the posterior probabilities of all members, weights are assigned to each member after normalization, and then the weighted sum is taken to obtain the final atmospheric corrected refractivity ensemble forecast result:

[0037] ;

[0038] Where yBMA is the ensemble forecast result; t is a variable, which is the atmospheric corrected refractivity; Mk is the model member numbered k; and Pr(Mk|D) is the posterior probability of Mk given the data sample D.

[0039] Finally, based on the atmospheric corrected refractivity forecast results, low-altitude atmospheric waveguide diagnosis is carried out to produce waveguide characteristic diagnosis products. The specific steps are as follows: For each grid point of the atmospheric corrected refractivity profile, diagnosis is carried out layer by layer, starting from the sea surface h0 (height 0). The specific process includes:

[0040] (1) Search upward for the first maximum peak of the corrected refractive index and record the current height h1 and the current corrected refractive index M1. If there is no h1, it is determined that no waveguide occurs and the diagnosis is complete;

[0041] (2) If h1 exists, continue searching upward to determine the first minimum value of the corrected refractive index. Then record the current height h2 and the corrected refractive index M2;

[0042] (3) If there is no increase in the corrected refractive index within the height range h0 to h1, the current waveguide type is determined to be a surface waveguide without a foundation layer. At this time, the bottom height of the waveguide is 0 meters, the top height is h2, the waveguide strength is M1–M2, and the waveguide thickness is h2–h0;

[0043] (4) If there is an increasing interval of the corrected refractive index below h1, search downward from h1 for the first point where the corrected refractive index is less than M2, and record the height h3 of this point;

[0044] (5) If no point less than M2 is found and h0 is still the initial height of 0 meters, the current waveguide is identified as a surface waveguide with a base layer. At this time, the bottom height of the waveguide is 0 meters, the top height is h2, the waveguide strength is M1–M2, and the waveguide thickness is h2–h0;

[0045] (6) If a point smaller than M2 is found, the height h3 of the point is recorded and the current waveguide type is determined to be a suspended waveguide. The bottom height of the waveguide is h3, the top height is h2, the waveguide strength is M1–M2, and the waveguide thickness is h2–h3;

[0046] (7) After completing the waveguide diagnosis process of this layer, the initial height h0 is replaced by the height h2. Then return to step (2) and continue searching upward to diagnose the waveguide of the next layer until the maximum height is reached.

[0047] After the diagnosis is completed, the waveguide type, height, thickness, strength and other parameters of each grid point at each moment are output, and finally the forecast data of low-altitude atmospheric waveguides in the area are formed.

[0048] In order to evaluate the performance of the method of the present invention, a numerical simulation of the meteorological data for the whole year of 2021 was carried out based on the COAWST sea-air coupling model. According to the simulation results, the air pressure, humidity, temperature and other factors in the area were extracted, and the atmospheric corrected refractive index at different altitudes was calculated. The atmospheric corrected refractive index at different altitudes of the sounding observation data in the same period in the region was simultaneously calculated as a benchmark value. Five error correction algorithms (LSTM, GRU, CNN, BP, MOS) were used to correct the atmospheric corrected refractive index at different altitudes, and the Bayesian model averaging (BMA) method was innovatively applied to perform ensemble averaging of the multi-model correction results. The results show that compared with a single error correction algorithm, the BMA ensemble method shows better correction effects at different altitude layers, significantly improving the accuracy of the atmospheric corrected refractive index, such as Figure 2 -4 shown.

[0049] like Figure 5 As shown in the figure, the root mean square error (RMSE) between the correction results of five error correction algorithms (LSTM, GRU, CNN, BP, MOS) and the Bayesian model averaging (BMA) method and the sounding observation data of the same period was calculated respectively. It was found that the BMA method can significantly improve the calculation accuracy of the atmospheric corrected refractive index by effectively integrating the advantages of each single model.

Claims

1. A method for forecasting low-altitude atmospheric duct at sea based on artificial intelligence ensemble error correction, characterized in that: The following steps are involved: (1) Construct an atmospheric corrected refractivity forecast framework based on the COAWST model to calculate the atmospheric corrected refractivity; (2) Carry out local optimization experiments in the target area, taking the atmospheric temperature and humidity below 5 km as the main evaluation variables, screen the physical parameterization schemes in the coupled model, evaluate the impact of changes in cloud microphysics scheme, radiation scheme, cumulus convection scheme, surface layer scheme, planetary boundary layer scheme, wave base dissipation scheme, and ocean vertical turbulence mixing scheme on the atmospheric temperature and humidity simulation, and select the 4-5 schemes with the highest sensitivity to conduct further multiple groups of sensitivity experiments; (3) Combine historical observations and reanalysis data, using the simulated observation correlation coefficient and root mean square error as evaluation indicators to evaluate the simulation capability of each option combination; (4) Based on the sensitivity assessment results, generate parameterized scheme configurations for coupled model operation and calculate the atmospheric corrected refractive index; (6) Construct an ensemble of error correction algorithm models; conduct atmospheric refractivity forecast correction evaluation tests, and determine the final ensemble members based on the evaluation results; (7) Using the Bayesian model averaging (BMA) algorithm, all error correction algorithm ensemble members are averaged based on the performance of each algorithm in the historical period, and initial weights are assigned. The weights are then dynamically adjusted according to the subsequent error correction performance. Based on the posterior probabilities of all ensemble members, weights are assigned to each member after normalization, and then the weighted sum is calculated to obtain the final atmospheric corrected refractivity ensemble forecast result. (8) Based on the atmospheric corrected refractive index ensemble forecast results, low-altitude atmospheric waveguide diagnosis is performed to obtain waveguide characteristic diagnosis results.

2. The method for forecasting low-altitude atmospheric duct at sea based on artificial intelligence ensemble error correction according to claim 1, characterized in that: The error correction algorithm is selected from a combination of any two or more of polynomial fitting, machine learning algorithm, and deep learning algorithm.

3. The method for forecasting low-altitude atmospheric duct at sea based on artificial intelligence ensemble error correction according to claim 1, characterized in that: Step (7) specifically includes: using the observation data of the past 120 hours in units of days, solving the maximum likelihood estimate using the expectation-maximization EM algorithm, and converting the observation data into a normal distribution; based on the forecast results after error correction in the same period, using the Markov Monte Carlo MCMC method, calculating the marginal likelihood function by constructing a Markov chain and Monte Carlo integration, and statistically calculating the posterior probability of each error correction algorithm member; according to the posterior probabilities of all members, assigning weights to each member after normalization, and then weighted summing them to obtain the final atmospheric corrected refractive index ensemble forecast result.

Images