Numerical model and large model cooperative forecasting method based on multi-dimensional water vapor assimilation

Abstract

The application discloses a numerical model and large model cooperative prediction method based on multi-dimensional water vapor assimilation, through cooperative driving of a WRF numerical model and a Pangu large model, global large-scale characteristics trained on reanalysis data are retained, and the deficiency of a regional model in detail description is made up through local fine tuning, so that solid technical support is provided for construction of a hybrid prediction system with generalization ability and regional prediction accuracy, and the problem of insufficient capturing ability of an existing meteorological prediction large model on an extreme weather event is solved.

Description

Technical Field

[0001] This invention relates to a method for collaborative forecasting of numerical models and large models based on multidimensional water vapor assimilation. Background Technology

[0002] Current meteorological forecasting systems still face multiple severe challenges in achieving high-precision forecasts, and single numerical weather prediction models often fall short of the requirements. Given the close correlation between tropospheric parameters and meteorological elements, since the 1990s, satellite assimilation data has been applied to numerical weather prediction models to improve forecast accuracy (e.g., the forecast model published by Lorenc et al. in 2000). In 2004, Vedel and Huang published their method of assimilating tropospheric zenith delay into a high-resolution study area, improving the ability to forecast heavy precipitation. Similar experiments using numerical weather prediction models to assimilate water vapor data have been conducted both domestically and internationally. With the development of water vapor monitoring technology, its advantages in numerical weather prediction models are becoming increasingly significant. In 2012, Ma et al. conducted a water vapor assimilation experiment in a certain region, confirming that assimilating water vapor data can effectively improve the accuracy of rainfall forecasts. In 2017, He Zhixin et al. found that assimilating water vapor could improve the forecasting effect of rainfall location and intensity. In 2018, Oigawa et al. further improved the forecasting accuracy of a severe rainstorm event in a certain region by assimilating water vapor data from a higher-density observation network. In 2020, Yang et al. conducted an assimilation experiment on a heavy rainfall event using data from 66 GNSS ground observation stations in a certain region, and the study showed that GPS data assimilation has a significant effect on improving precipitation forecasts. In 2021, Risanto et al. evaluated the impact of water vapor data assimilation on short-term precipitation forecasts of the North American monsoon in northwestern a certain region, and the results showed that water vapor data assimilation significantly improved the accuracy of precipitation forecasts.

[0003] Among traditional assimilation methods, three-dimensional variational data assimilation and four-dimensional variational data assimilation techniques have gradually become hot topics in meteorological assimilation research due to their strong spatiotemporal optimization capabilities. Ding Jincai et al. (2007) found that assimilating water vapor data using a three-dimensional variational system could improve precipitation forecasting capabilities. Giannaros et al. (2020) used three-dimensional variational assimilation of tropospheric zenith delay data to forecast precipitation and water vapor in a certain area, and the results showed a positive impact. Other studies have shown that using three-dimensional / four-dimensional variational methods to assimilate external water vapor in Weather Research and Forecasting Models (WRF) can improve WRF model forecasting performance. Rohm et al. (2019) conducted a two-month WRF data assimilation experiment in a certain region, assimilating tropospheric zenith delay and water vapor data acquired from 106 GNSS stations into the model. Studies have found that compared to the WRF model with unassimilated data, assimilating water vapor data can slightly reduce the 24-hour water vapor forecast bias from 2.6 kg / m² to 2.5 kg / m². Singh et al. (2019) used the WRF model to study the impact of tropospheric zenith delay and water vapor data assimilation on a certain region, showing positive effects on precipitation forecasts, as well as forecasts of water vapor in the lower and middle troposphere, upper-level temperature, and wind fields in the upper troposphere. Leontiev et al. (2020) conducted forecasts in a certain region using assimilated GPS water vapor data and water vapor data fused from GPS and Meteosat satellite imagery, showing that assimilating water vapor data improved the accuracy of the first 3 hours of forecasts by 11% compared to the WRF model with unassimilated data. Simultaneously, a water vapor assimilation experiment was conducted based on a WRF model in the northern part of a region, finding that water vapor assimilation can improve the accuracy of precipitation prediction. Although variational assimilation methods have made some progress in local and short-term forecasts, their effectiveness under large-scale and complex weather conditions is affected by data quality, model accuracy, and boundary condition settings.

[0004] In recent years, technological innovations in artificial intelligence have injected new momentum into the field of weather forecasting. AI (Artificial Intelligence) technologies, represented by machine learning and deep learning, are reshaping the weather forecasting technology system through innovative data modeling methods. A certain company's Pangu model innovatively employs a three-dimensional neural network architecture, demonstrating computational performance surpassing traditional numerical models in tasks such as typhoon path and rainfall forecasting. Currently, the reliability and practicality of the Pangu weather model have been confirmed by numerous studies. For example, in 2022, Bi et al. proposed a deep learning-based system—the Pangoalge Weather Model—to predict global weather. The results showed that the Pangoalge Weather Model had a significant advantage in short- to medium-term forecasts. In 2024, Xu et al. successfully used the Pangoalge Weather Model to drive the WRF model, improving the accuracy of catastrophic extreme precipitation predictions in the northern part of a certain region. In 2025, Huang et al. used the Pangoalge Big Model to estimate surface air temperature from 2016 to 2019. The results showed that the deviation and root mean square error between the Pangoalge Model and the measured surface temperature at RS stations were -0.75 K and 2.54 K, respectively, which can provide a reliable and high-precision weighted average temperature estimate for real-time GNSS PWV inversion.

[0005] However, existing water vapor assimilation and weather forecasting technologies still face some challenges. On the one hand, existing assimilation methods have limitations in capturing dynamic changes in water vapor, especially in efficiently assimilating two-dimensional and three-dimensional water vapor data. On the other hand, although the Pangoal Instant Model has shown significant advantages in weather forecasting, current research focuses primarily on the prediction and analysis of meteorological data, without exploring its combined application with water vapor assimilation. Therefore, it is urgent to further improve forecast accuracy by assimilating two-dimensional and three-dimensional water vapor data and to explore how to integrate the Pangoal Instant Model with numerical weather prediction models to enhance the accuracy of weather forecasts. Summary of the Invention

[0006] To address the problem of insufficient ability of existing large-scale meteorological forecasting models to capture extreme weather events, this invention provides a collaborative forecasting method based on multidimensional water vapor assimilation of numerical models and large-scale models. Through the collaborative driving of the WRF numerical model and the Pangoal Institutional Scale (i.e., the medium-range global weather forecasting model), it not only retains the global large-scale features obtained by training the Pangoal Institutional Scale on reanalysis data, but also compensates for the lack of detailed description in regional models through local fine-tuning. This provides solid technical support for building a hybrid forecasting system with both generalization ability and regional forecasting accuracy.

[0007] According to one aspect of the present invention, a method for joint forecasting of numerical models and large models based on multidimensional water vapor assimilation is provided, comprising:

[0008] Based on the hybrid set-four-dimensional variational assimilation method, combined with an adaptive weight adjustment strategy, a two-dimensional multi-source water vapor initial field is constructed and dynamically assimilated.

[0009] Based on the conversion and assimilation strategy of tomographic three-dimensional wet refractive index data, a conversion of three-dimensional wet refractive index to temperature, humidity and air pressure is constructed and three-dimensional water vapor is assimilated.

[0010] The assimilated multidimensional water vapor data is converted to the input format of the medium-term global weather forecast model, input into the trained WRF numerical model-driven meteorological forecast model, and the forecast results are output. The WRF numerical model-driven meteorological forecast model is based on the medium-term global weather forecast model, introduces a trainable feature adapter and a weighted loss function, and is trained through a "frozen trunk + local fine-tuning" strategy.

[0011] As a further technical solution, based on the hybrid ensemble-four-dimensional variational assimilation method, a two-dimensional multi-source water vapor initial field is constructed and dynamically assimilated, including:

[0012] Based on two-dimensional water vapor data, an initial water vapor field for the WRF model is generated.

[0013] A hybrid ensemble-four-dimensional variational assimilation method is used to dynamically correct the initial water vapor field of the WRF model. The hybrid ensemble-four-dimensional variational assimilation method combines the ensemble covariance with the static covariance to provide a statistical estimate of the flow-related background error, and uses an extended control variable method to incorporate the ensemble-based covariance estimate into the WRF model.

[0014] As a further technical solution, an adaptive weight adjustment strategy is incorporated, including:

[0015] Define the background error covariance of the mixed set-four-dimensional variational assimilation as: ,in, The background error covariance matrix for climate statistics. The flow-dependent error covariance matrix is ​​obtained from the ensemble prediction using the Kalman filter. This is a weighting parameter used to balance the contributions of historical statistics and real-time features;

[0016] Dynamic adjustment in different weather systems and The weight ratios between them are determined to obtain the optimal analysis field.

[0017] As a further technical solution, the method also includes optimizing the background error covariance in the following manner:

[0018] Principal component analysis was used to reduce the dimensionality of the water vapor field data and extract key temporal-spatial features.

[0019] Define the localization operation of the background error covariance matrix;

[0020] The error covariance of different weather systems is dynamically adjusted using the WRF model.

[0021] As a further technical solution, based on the conversion and assimilation strategy of tomographic three-dimensional wet refractive index data, a conversion from three-dimensional wet refractive index to temperature, humidity, and air pressure is constructed, and three-dimensional water vapor is assimilated, including:

[0022] Three-dimensional wet refractive index data are converted into relative humidity using temperature and water vapor pressure;

[0023] The relative humidity, temperature, and water vapor pressure data are assimilated into the WRF mode and used as the initial field for the WRF mode.

[0024] As a further technical solution, the training of the WRF numerical model-driven large-scale weather forecasting model includes:

[0025] Freeze the backbone network of the medium-term global weather forecast model and add a trainable feature adapter at the input, training only the weights and biases of the adapter;

[0026] Define a loss function based on weighted mean square error to directly utilize the WRF interpolation results as a monitoring signal.

[0027] As a further technical solution, the training of the large-scale weather forecasting model driven by the WRF numerical model also includes:

[0028] After the adapter converges, the selected high-level network parameters of the medium-term global weather forecast model are unfrozen, and the overall feature representation is fine-tuned.

[0029] As a further technical solution, the loss function based on weighted mean square error is:

[0030] ,

[0031] in, These are the forecast results from the medium-range global weather forecast model after adapter correction. The WRF label data is obtained through coordinate transformation and interpolation, where λ is a weighting coefficient set according to the influence of each meteorological variable on regional forecasts and the uncertainty of the WRF model. It is the residual between the predicted value and the actual value. It is the set of residuals, and P is the predicted value.

[0032] According to one aspect of the present invention, a numerical model and large model co-forecasting system based on multidimensional water vapor assimilation is provided for implementing the method, the system comprising:

[0033] The first main module is based on the hybrid set-four-dimensional variational assimilation method, combined with an adaptive weight adjustment strategy, to construct a two-dimensional multi-source water vapor initial field and perform dynamic assimilation.

[0034] The second main module, based on the conversion and assimilation strategy of tomographic three-dimensional wet refractive index data, constructs the conversion of three-dimensional wet refractive index to temperature, humidity and air pressure and performs three-dimensional water vapor assimilation.

[0035] The third main module is based on converting the assimilated multidimensional water vapor data into the input format of the medium-term global weather forecast model, inputting it into the trained WRF numerical model-driven meteorological forecast model, and outputting the forecast results. The WRF numerical model-driven meteorological forecast model is based on the medium-term global weather forecast model, introduces a trainable feature adapter and a weighted loss function, and is trained through a "frozen backbone + local fine-tuning" strategy.

[0036] According to one aspect of the present invention, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the numerical model and large model co-forecasting method based on multidimensional water vapor assimilation.

[0037] This invention addresses the problem of insufficient ability of existing large-scale meteorological forecasting models to capture extreme weather events by constructing a collaborative forecasting method between numerical models and large-scale models based on multidimensional water vapor assimilation. Compared with existing technologies, the advantages of this invention are:

[0038] 1. Based on the hybrid ensemble-four-dimensional variational assimilation method and combined with an adaptive weight adjustment strategy, a two-dimensional multi-source water vapor initial field optimization and dynamic assimilation method is constructed to improve the response capability of mesoscale weather forecast models to spatiotemporal variations of water vapor.

[0039] 2. Develop a conversion and assimilation strategy for tomographic three-dimensional wet refractive index data, construct a conversion method from three-dimensional wet refractive index to temperature, humidity and air pressure, and optimize the application of three-dimensional water vapor information in mesoscale weather forecasting models;

[0040] 3. Construct a cross-coordinate system data conversion method from mesoscale weather forecast model data to the input format of the Pangu model, introduce a trainable feature adapter module and a weighted loss function, and achieve high-precision weather forecasting driven by the mesoscale weather forecast model and the Pangu large model through the strategy of "freezing the backbone + local fine-tuning". Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating the collaborative forecasting method of numerical models and large models based on multidimensional water vapor assimilation provided in an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0044] This invention provides a method for collaborative forecasting of numerical models and large models based on multidimensional water vapor assimilation, such as... Figure 1 As shown, the process includes: constructing a two-dimensional multi-source water vapor initial field and performing dynamic assimilation based on a hybrid ensemble-four-dimensional variational assimilation method combined with an adaptive weight adjustment strategy; constructing a three-dimensional wet refractive index to temperature, humidity, and air pressure conversion strategy based on tomographic three-dimensional wet refractive index data and performing three-dimensional water vapor assimilation; converting the assimilated multi-dimensional water vapor data to the input format of a medium-term global weather forecast model, inputting it into a trained WRF numerical model-driven meteorological forecasting large model, and outputting forecast results; wherein, the WRF numerical model-driven meteorological forecasting large model is based on a medium-term global weather forecast model, introduces a trainable feature adapter and a weighted loss function, and is trained through a "frozen trunk + local fine-tuning" strategy.

[0045] Specifically, the steps include the following:

[0046] (1) Optimization and dynamic assimilation of the initial two-dimensional water vapor field.

[0047] The accuracy of water vapor data is a key factor affecting WRF performance. This study uses high-resolution two-dimensional water vapor data. In this embodiment of the invention, WRF is used to obtain the initial field. Furthermore, the two-dimensional water vapor data was compared with the model's initial field, and the initial field was optimized using an error function.

[0048] ,

[0049] In the formula: These are spatial grid points. Let be the error function. The initial water vapor field of the WRF model is optimized by minimizing the error function.

[0050] This invention further employs a Hybrid Ensemble-Variational Data Assimilation (Hybrid EnVAR) method to dynamically correct the two-dimensional water vapor field of the WRF model, ensuring that the model input more closely reflects actual observations. Unlike three-dimensional and four-dimensional variational assimilation methods that use static covariance models to estimate background forecast errors, the hybrid method combines ensemble covariance with static covariance to provide flow-related statistical estimates of background errors. An extended control variable method is used to incorporate ensemble-based covariance estimates into the WRF. The Hybrid4D-EnVAR method combines the advantages of ensemble methods and four-dimensional variational assimilation, using a weighted fusion of climate statistical error covariance and ensemble error covariance to make assimilation more adaptable to the changing characteristics of different weather systems. The basic variational optimization objective function of Hybrid 4D-EnVAR is defined as follows:

[0051] ,

[0052] In the formula: x is the analysis field variable, that is, the two-dimensional water vapor field to be updated; y represents the background field variable, i.e., the initial water vapor field obtained through the WRF model; y represents the observation data, i.e., the fused two-dimensional water vapor. is the observation operator; B and R are the background error covariance matrix and the observation error covariance matrix, respectively.

[0053] In Hybrid EnVAR, the flow-dependent error covariance and the climate statistical background error covariance are used together to improve the accuracy of water vapor assimilation. The background error covariance B of Hybrid EnVAR is defined as:

[0054] ,

[0055] In the formula: The background error covariance matrix for climate statistics is obtained from long-term meteorological data statistics. The flow-dependent error covariance matrix is ​​obtained from the ensemble prediction of the Kalman filter set. The weighting parameters are used to balance the contributions of historical statistics and real-time features. This embodiment of the invention uses WRF to calculate the dynamic error covariance of the background field and optimizes the weight allocation of ensemble members to more accurately describe the error characteristics of the water vapor field. Furthermore, this embodiment of the invention uses an adaptive weight adjustment strategy (such as variance component estimation) to enable Hybrid EnVAR to dynamically adjust under different weather systems. and The weighting ratio between the data ensures optimal assimilation of the water vapor field, yielding the optimal analysis field x. This allows the observational data (two-dimensional water vapor data) to be integrated into the WRF model, correcting the initial water vapor field of the WRF model. This combination enables the assimilation method to consider both long-term climate statistics and the dynamic changes of instantaneous weather systems.

[0056] To enhance the sensitivity of Hybrid EnVAR to small- and medium-scale weather systems, this embodiment of the invention further optimizes the background error covariance. Specifically, principal component analysis is used to reduce the dimensionality of water vapor field data, extracting key temporal-spatial features and reducing the computational complexity of high-dimensional data. Simultaneously, to reduce unreasonable error propagation between distant grid points, localization operations are used to limit long-distance error propagation. The localization operation of the background error covariance matrix is ​​defined as follows:

[0057] ,

[0058] In the formula: The set error covariance is calculated globally. For grid points , The distance between them; The localization radius is defined as follows. This localization method effectively reduces the unreasonable propagation of error covariance between distant grid points, improving the accuracy of local water vapor field analysis. Furthermore, WRF is used to dynamically adjust the error covariance of different weather systems, enhancing the adaptability of Hybrid EnVAR to nonlinear water vapor evolution.

[0059] (2) Three-dimensional water vapor assimilation combined with chromatography technology.

[0060] In the process of assimilating two-dimensional water vapor using the WRF-DA model, only the total vertical water vapor content information is utilized, while the vertical profile information of water vapor is not, which may limit the effectiveness of data assimilation. The vertical structure of atmospheric water vapor is crucial for the formation and development of precipitation, but its high spatiotemporal resolution and complex vertical structure place higher demands on the assimilation method. Therefore, this embodiment of the invention proposes to introduce three-dimensional wet refractive index data generated using water vapor tomography into the WRF-DA model to enrich the three-dimensional structure information of water vapor and optimize the initial field of the WRF model.

[0061] It should be noted that, in this embodiment of the invention, the WRF can be considered as a forecast engine, responsible for simulating atmospheric evolution; the WRF-DA model can be considered as a preprocessing tool, responsible for optimizing the initial field of the WRF model. The output of the WRF-DA model (the assimilated analysis field) is directly used as the initial condition of the WRF model. Observational data is first input into the WRF-DA model, which generates the analysis field through variational or ensemble methods. This analysis field serves as the initial field for the WRF model, initiating the forecast simulation.

[0062] The WRF model itself does not have the ability to generate initial fields independently; it requires initial conditions provided by external data sources (such as global model outputs or reanalysis data). Common sources include GFS (USA), ECMWF (European Centre for Model Research), and NCEP FNL (reanalysis data). These data are converted into a WRF-readable format using WRF preprocessing tools, and initial fields and boundary conditions are generated using WRF's built-in functions.

[0063] Three-dimensional wet refractive index provides information on water vapor at different altitudes, compensating for the lack of vertical distribution information in two-dimensional water vapor data. However, the WRF-DA model cannot directly assimilate wet refractive index; it needs to be converted into assimilated basic meteorological variables such as relative humidity (RH), temperature (T), and water vapor pressure. Wet refractive index (WR) and water vapor pressure... It is closely related to temperature. The relationship with RH is as follows:

[0064] ,

[0065] In the formula, This is the saturated vapor pressure, which is temperature-dependent. and It can be calculated using the following formula:

[0066] ,

[0067] In the formula, , , The value is a constant. In the process of converting the wet refractive index into a combination of temperature, humidity, and vapor pressure, the wet refractive index is first converted into relative humidity using temperature and vapor pressure. Then, RH, temperature, and vapor pressure data are assimilated into the WRF-DA model to generate a preliminary analysis field. The Hybrid EnVAR method is employed, combined with ensemble Kalman filtering, to dynamically calculate the background error covariance. Furthermore, to optimize the adaptability of three-dimensional water vapor information to different weather systems, this embodiment of the invention further adjusts the background error covariance structure.

[0068] (3) Weather forecasting driven by the synergistic interaction between the WRF numerical model and the Pangu model.

[0069] After completing the two-dimensional and three-dimensional water vapor assimilation, this embodiment of the invention uses the output of the WRF model as a high-precision label to perform physical-guided model fine-tuning of the Pangoal Infinite Model, aiming to construct a large-scale meteorological forecasting model driven by the WRF numerical model. The Pangoal Infinite Model is a medium-term global weather forecasting model built based on 3D neural networks. This model unifies and merges meteorological information from different longitudes, latitudes, and altitudes to form a standardized atmospheric state field (such as three-dimensional grid data of isobaric surfaces or isopotential height surfaces) as model input. As a regional high-resolution forecasting tool, WRF uses terrain-following coordinates or hybrid vertical coordinates in the vertical direction, and meteorological variables are stored in the eta layer. This results in a significant difference between its output and the input format of models such as Pangoal Infinite Model. Therefore, it is necessary to convert the vertical coordinate system data of the WRF model to the standard barosphere.

[0070] For topographically following coordinates, the pressure in the eta layer can be expressed as the surface pressure. With the top pressure of the mode Linear combination:

[0071] ,

[0072] This indicates the eta layer number. For mixed coordinates, a background pressure profile is introduced. With terrain disturbance terms The superposition of these variables is achieved by calculating the actual air pressure of each eta layer in reverse and then projecting the original variables (such as potential temperature, humidity, and wind field) onto a standard isobaric surface using logarithmic linear interpolation.

[0073] The interpolation variable Q is on the isobaric surface The expression is:

[0074] ,

[0075] , These represent the meteorological variables to be interpolated. and This represents the air pressure in the adjacent eta layer.

[0076] Interpolation of the equipotential height surface requires inverting the geometric height using the potential height formula:

[0077] ,

[0078] Where PH is the perturbation potential, PHB is the ground state potential, and g is the gravitational acceleration. This represents the ground height. After interpolation, the 3D grid data needs to be resampled to a resolution matching the Pango model, and the scale difference between the WRF model and the Pango model is eliminated through bilinear interpolation.

[0079] After data transformation and preprocessing, to further fine-tune the Pangu model, this embodiment of the invention employs a strategy of freezing the model's backbone network and adding a trainable external feature adapter at the input. Specifically, while freezing all layer parameters of the Pangu model, a lightweight adapter module is added before the model input. The input feature X is transformed by the adapter and output as follows:

[0080] ,

[0081] Where W1 and W2 are weight matrices, b1 and b2 are bias terms, and σ is the Swish activation function. To ensure that the model output can fully reflect the high-precision information of WRF numerical prediction, this embodiment of the invention defines a loss function based on weighted mean square error to directly utilize the WRF interpolation results as a supervision signal.

[0082] The loss function can be expressed as:

[0083] (13)

[0084] in, The prediction results are from the Pangu model after adapter correction. The WRF label data is obtained through coordinate transformation and interpolation, where λ is a weighting coefficient set according to the influence of each meteorological variable on the regional forecast and the uncertainty of the WRF model, and P is the forecast value. It is the residual between the predicted value and the actual value. It is the set of residuals.

[0085] During model training, only the weights and biases of the adapter are trained; all parameters of the Pangu layers are set to non-trainable. After the adapter converges, some high-level network parameters of the Pangu large model can be further unfrozen to fine-tune the overall feature representation, thereby achieving higher adaptability to regional details and prediction accuracy. In the Pangu Weather model, high-level network parameters refer to the key learnable parameters located at the top of its deep neural network architecture, responsible for processing high-level meteorological features. These parameters directly affect the model's ability to abstract complex meteorological patterns and its prediction performance. After the adapter converges, some of its corresponding parameters can also be trained.

[0086] By leveraging the synergistic effect of the WRF numerical model and the Pangoal Injection Model, not only are the large-scale global features obtained from training the Pangoal Injection Model on reanalysis data preserved, but also the shortcomings of regional models in detail description are compensated for through local fine-tuning. This provides solid technical support for building a hybrid forecasting system that combines generalization ability with regional forecast accuracy.

[0087] The implementation of the various embodiments of the present invention is based on programmed processing through a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a numerical model and large model co-forecasting system based on multidimensional water vapor assimilation. This system is used to execute the numerical model and large model co-forecasting method based on multidimensional water vapor assimilation in the above method embodiments.

[0088] The system comprises: a first main module, which constructs a two-dimensional multi-source water vapor initial field and performs dynamic assimilation based on a hybrid ensemble-four-dimensional variational assimilation method combined with an adaptive weight adjustment strategy; a second main module, which constructs a conversion from three-dimensional wet refractive index to temperature, humidity, and air pressure based on a tomographic three-dimensional wet refractive index data conversion and performs three-dimensional water vapor assimilation based on a tomographic three-dimensional wet refractive index data conversion and assimilation strategy; and a third main module, which converts the assimilated multi-dimensional water vapor data to the input format of a medium-term global weather forecast model, inputs it into a trained WRF numerical model-driven meteorological forecasting model, and outputs forecast results; wherein, the WRF numerical model-driven meteorological forecasting model is based on a medium-term global weather forecast model, introduces a trainable feature adapter and a weighted loss function, and is trained through a "frozen backbone + local fine-tuning" strategy.

[0089] The numerical model and large-scale model collaborative forecasting system based on multidimensional water vapor assimilation provided in this invention addresses the problem of insufficient ability of existing large-scale meteorological forecasting models to capture extreme weather events. By employing the aforementioned modules and through the collaborative driving of the WRF numerical model and the Pangoal Institutional Scale (i.e., the medium-range global weather forecasting model), it not only retains the global large-scale features obtained by the Pangoal Institutional Scale training on reanalysis data, but also compensates for the lack of detailed description in regional models through local fine-tuning. This provides solid technical support for building a hybrid forecasting system that combines generalization ability and regional forecasting accuracy.

[0090] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The only difference is that corresponding functional modules are set. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art can improve the modules in the above system embodiments by referring to the specific technical solutions in other method embodiments and combining technical features to obtain corresponding technical means and technical solutions composed of these technical means, on the basis of the above system embodiments, and on the premise of ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments.

[0091] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention also provides a non-transitory computer-readable storage medium that stores computer instructions that cause the computer to execute the numerical model and large model co-forecasting method based on multidimensional water vapor assimilation.

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0093] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A numerical model and large-scale model collaborative forecasting method based on multidimensional water vapor assimilation, characterized in that, include: Based on a hybrid ensemble-four-dimensional variational assimilation method combined with an adaptive weight adjustment strategy, a two-dimensional multi-source water vapor initial field is constructed and dynamically assimilated. This includes: generating a water vapor initial field for a WRF model based on two-dimensional water vapor data; dynamically correcting the water vapor initial field of the WRF model using a hybrid ensemble-four-dimensional variational assimilation method, wherein the hybrid ensemble-four-dimensional variational assimilation method combines ensemble covariance with static covariance to provide flow-related background error statistical estimates, and using an extended control variable method to incorporate the ensemble-based covariance estimates into the WRF model; further including: defining the background error covariance of the hybrid ensemble-four-dimensional variational assimilation as: ,in, The background error covariance matrix for climate statistics. The flow-dependent error covariance matrix is ​​obtained from the ensemble prediction using the Kalman filter. These are weighting parameters used to balance the contributions of historical statistics and real-time characteristics; they are dynamically adjusted across different weather systems. and The weight ratios between them are used to obtain the optimal analysis field; Based on the conversion and assimilation strategy of tomographic three-dimensional wet refractive index data, a conversion of three-dimensional wet refractive index to temperature, humidity and air pressure is constructed and three-dimensional water vapor is assimilated. The assimilated multidimensional water vapor data is converted to the input format of the medium-term global weather forecast model, input into the trained WRF numerical model-driven meteorological forecast model, and the forecast results are output. The WRF numerical model-driven meteorological forecast model is based on the medium-term global weather forecast model, introduces a trainable feature adapter and a weighted loss function, and is trained through a "frozen trunk + local fine-tuning" strategy.

2. The numerical model and large model co-prediction method based on multidimensional water vapor assimilation according to claim 1, characterized in that, The method also includes optimizing the background error covariance in the following ways: Principal component analysis was used to reduce the dimensionality of the water vapor field data and extract key temporal-spatial features. Define the localization operation of the background error covariance matrix; The error covariance of different weather systems is dynamically adjusted using the WRF model.

3. The numerical model and large model collaborative forecasting method based on multidimensional water vapor assimilation according to claim 1, characterized in that, Based on the conversion and assimilation strategy of tomographic three-dimensional wet refractive index data, a conversion mechanism is constructed to transform the three-dimensional wet refractive index into temperature, humidity, and air pressure, and three-dimensional water vapor assimilation is performed, including: Three-dimensional wet refractive index data are converted into relative humidity using temperature and water vapor pressure. The relative humidity, temperature, and water vapor pressure data are assimilated into the WRF mode and used as the initial field for the WRF mode.

4. The numerical model and large model collaborative forecasting method based on multidimensional water vapor assimilation according to claim 1, characterized in that, The training of the large-scale weather forecast model driven by the WRF numerical model includes: Freeze the backbone network of the medium-term global weather forecast model and add a trainable feature adapter at the input, training only the weights and biases of the adapter; Define a loss function based on weighted mean square error to directly utilize the WRF interpolation results as a monitoring signal.

5. The numerical model and large model co-prediction method based on multidimensional water vapor assimilation according to claim 4, characterized in that, The training of the large-scale weather forecasting model driven by the WRF numerical model also includes: After the adapter converges, the selected high-level network parameters of the medium-term global weather forecast model are unfrozen, and the overall feature representation is fine-tuned.

6. The numerical model and large model co-prediction method based on multidimensional water vapor assimilation according to claim 4, characterized in that, The loss function based on weighted mean square error is: ， in, These are the forecast results from a medium-range global weather forecast model after adapter correction. The WRF label data is obtained through coordinate transformation and interpolation. λ is the weighting coefficient set according to the influence of each meteorological variable on the regional forecast and the uncertainty of the WRF model. v is the residual between the forecast value and the actual value. P is the forecast value.

7. A numerical model and large model collaborative forecasting system based on multidimensional water vapor assimilation, used to implement the method described in any one of claims 1-6, characterized in that, The system includes: The first main module is based on the hybrid set-four-dimensional variational assimilation method, combined with an adaptive weight adjustment strategy, to construct a two-dimensional multi-source water vapor initial field and perform dynamic assimilation. The second main module, based on the conversion and assimilation strategy of tomographic three-dimensional wet refractive index data, constructs the conversion of three-dimensional wet refractive index to temperature, humidity and air pressure and performs three-dimensional water vapor assimilation. The third main module is based on converting the assimilated multidimensional water vapor data into the input format of the medium-term global weather forecast model, inputting it into the trained WRF numerical model-driven meteorological forecast model, and outputting the forecast results. The WRF numerical model-driven meteorological forecast model is based on the medium-term global weather forecast model, introduces a trainable feature adapter and a weighted loss function, and is trained through a "frozen backbone + local fine-tuning" strategy.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the numerical model and large model co-forecasting method based on multidimensional water vapor assimilation as described in any one of claims 1 to 6.

Images