A method for descending scale of plateau meteorological data by cross-terrain knowledge transfer

Abstract

The application discloses a plateau meteorological data downscaling method based on cross-terrain knowledge transfer, and belongs to the technical field of data-driven meteorological modeling and calculation. The method comprises the following steps: obtaining low-resolution meteorological data and corresponding high-resolution meteorological data of a source domain, low-resolution meteorological data and corresponding high-resolution meteorological data of a target domain, and terrain data; training a basic downscaling network model based on the low-resolution meteorological data and the corresponding high-resolution meteorological data of the source domain; performing adaptive training on the model based on the low-resolution meteorological data and the terrain data of the target domain and in combination with a composite target function containing a terrain difference regularization term; fine-tuning the model based on the meteorological data of the target domain to obtain a final downscaling model, and realizing downscaling of the low-resolution meteorological data by using the model. The application can improve the model performance, generalization ability and physical consistency.

Description

Technical Field

[0001] This invention relates to the field of data-driven meteorological modeling and computation technology, and in particular to a method for downscaling plateau meteorological data through cross-terrain knowledge transfer. Background Technology

[0002] High-resolution, gridded meteorological data, such as precipitation and temperature, are crucial inputs for fields like hydrological forecasting, disaster early warning, and climate change research. However, limited by the physical density and construction costs of observation stations, raw observational data is often sparse station data or coarse-resolution gridded data (such as reanalysis data and satellite inversion products). Therefore, "downscaling" techniques, which transform low-resolution data into high-resolution data, have become one of the core research issues in this field.

[0003] Currently, the mainstream downscaling techniques mainly include the following two categories: 1. Dynamic Downscaling Method: This method primarily relies on high-resolution regional numerical weather models (such as the WRF model). By providing initial fields and boundary conditions, the model simulates the evolution of meteorological elements at a fine scale based on atmospheric physics and hydrodynamic equations. However, this method has the following inherent limitations:

[0004] 1) High computing costs: It requires huge computing resources and time costs, making it difficult to meet the real-time business needs.

[0005] 2) Uncertainty in physical parameterization: The parameterization scheme for the physical processes of the subgrid in the model has a large deviation in plateau areas with extremely complex terrain, and the simulation results often require complex corrections before they can be used.

[0006] 2. Statistical Downscaling Methods: This method aims to establish a statistical mapping relationship between low-resolution forecast factors and high-resolution observation targets. In recent years, with the development of computational intelligence technology, data-driven statistical downscaling methods, especially those based on deep learning, have made significant progress. For example, models such as convolutional neural networks (CNNs) can automatically learn the spatial texture features of meteorological fields, achieving end-to-end downscaling. However, directly applying existing data-driven methods to special areas such as plateaus faces severe technical bottlenecks:

[0007] 1) Strong dependence on dense observations: Training such models requires a large number of paired, high-quality low-resolution-high-resolution data samples.

[0008] 2) Data issues in plateau regions: Due to the harsh natural environment, meteorological observation stations in plateau regions are extremely sparse and unevenly distributed. If these few unrepresentative samples are used directly to train the model, it is very easy to cause the model to overfit, making it unable to generalize to the vast areas without observation stations, resulting in poor accuracy and low reliability of the downscaling results.

[0009] In summary, existing data-driven downscaling methods, when applied to sparsely observed regions such as plateaus, suffer from overfitting, poor generalization ability, and low downscaling accuracy due to insufficient training samples. Therefore, how to construct a high-precision downscaling model even under conditions of severely sparse observational data in the target region (such as plateaus) is a pressing technical challenge in the field of meteorological modeling. Summary of the Invention

[0010] This invention provides a plateau meteorological data downscaling method based on cross-terrain knowledge transfer, which solves the technical problems of model overfitting, poor generalization ability and low downscaling accuracy caused by insufficient training samples when existing data-driven downscaling methods are applied to sparsely observed areas such as plateaus.

[0011] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, this invention provides a method for downscaling plateau meteorological data through cross-terrain knowledge transfer, the method comprising: The plateau region to which meteorological data is to be downscaled is taken as the target domain, and the region outside the plateau region is taken as the source domain; low-resolution meteorological data and corresponding high-resolution meteorological data of the source domain, low-resolution meteorological data and corresponding high-resolution meteorological data of the target domain, and topographic data are acquired. Based on low-resolution meteorological data from the source domain and corresponding high-resolution meteorological data, a basic downscaling network model is trained to learn the mapping relationship from low-resolution meteorological data to high-resolution meteorological data. A composite objective function is constructed. Based on low-resolution meteorological and topographic data of the target domain, the pre-trained basic downscaling network model is adaptively trained using the composite objective function. The composite objective function includes a topographic difference regularization term, which is used to apply a physical constraint so that the spatial gradient change of the downscaling result is structurally consistent with the gradient change of the local topography. Based on low-resolution meteorological data and corresponding high-resolution meteorological data in the target domain, the basic downscaling network model that has completed adaptive training is fine-tuned to obtain the final downscaling model. The low-resolution meteorological data to be downscaled is input into the downscaling model to obtain high-resolution downscaling results.

[0012] Furthermore, the terrain data uses high-resolution digital elevation model data.

[0013] Furthermore, the basic downscaling network model adopts an encoder-decoder architecture.

[0014] Furthermore, when training the basic downscaling network model, the objective function is a reconstruction loss function.

[0015] Furthermore, the expression for the terrain difference regularization term is: ; in, This represents the terrain difference regularization term; For spatial gradient operators; α is the L1 norm; α is a scaling factor that matches the meteorological gradient with the terrain gradient. This represents the high-resolution meteorological field output by the model in the target domain. For terrain data of the target domain; This indicates taking the average value.

[0016] Furthermore, the composite objective function also includes a domain alignment loss term; the domain alignment loss term is used to reduce the data distribution differences between the source domain and the target domain in the feature space, enabling the model to learn generalized features across regions.

[0017] Furthermore, the expression for the domain alignment loss term is: ; in, Indicates the domain alignment loss term; This represents a feature extractor for the intermediate layer of the network; Represents the first term of the source domain i A low-resolution meteorological data sample; Represents the first of the target domains j A low-resolution meteorological data sample; and These are the number of samples in the source domain and the target domain, respectively. This indicates that the square of the vector distance is calculated. This represents the feature mapping function.

[0018] Furthermore, the composite objective function is expressed as: ; in, and To balance the hyperparameters of each loss term; This represents the terrain difference regularization term; The loss is the supervised reconstruction loss over the target domain.

[0019] Furthermore, when fine-tuning the base downscaling network model that has completed adaptive training, the objective function adopts the reconstruction loss function.

[0020] Furthermore, after obtaining the final downscaling model, the method also includes: The downscaling model is validated using preset evaluation metrics; wherein the preset evaluation metrics are one or more combinations of root mean square error, mean absolute error, and correlation coefficient.

[0021] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0022] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the above method.

[0023] The beneficial effects of the technical solution provided by this invention include at least the following: 1. Reduce data dependence and solve the cold start problem: This invention greatly reduces the dependence on dense observation data of the target plateau by transferring knowledge from the data-rich source domain, and effectively solves the "cold start" problem that makes it difficult to train the model due to insufficient samples.

[0024] 2. Improved model performance and generalization ability: This invention innovatively introduces terrain features as a "bridge" and constraint for knowledge transfer, which significantly improves the accuracy and generalization ability of the downscaling model in the vast plateau areas without monitoring stations, and the prediction results are more reliable.

[0025] 3. Enhanced physical consistency: Since the model of this invention learns and transfers the physical influence of terrain on meteorological elements, its downscaling results better contain physical meaning and avoid outliers that may be generated by traditional methods and violate natural laws.

[0026] 4. Significant application value and social benefits: This invention can generate high spatiotemporal resolution meteorological data products for plateau regions, which can directly serve key areas such as early warning of flash floods and geological disasters, refined management of water resources, ecological environment protection and development of characteristic agriculture and animal husbandry, and provide reliable data support for regional disaster prevention and mitigation and sustainable economic and social development. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the execution flow of the plateau meteorological data downscaling method for cross-terrain knowledge transfer provided in an embodiment of the present invention; Figure 2 This is a system block diagram of the electronic device provided in the embodiments of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0030] First, it should be noted that in the embodiments of the present invention, the words "exemplarily," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplarily" is intended to present the concept in a specific manner. Furthermore, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either one or the other.

[0031] First Embodiment This embodiment provides a method for downscaling plateau meteorological data through cross-terrain knowledge transfer. This method can be implemented using a computing system equipped with a graphics processing unit (GPU), memory, and storage devices, and deployed with a deep learning framework (such as PyTorch or TensorFlow). The implementation process is as follows: Figure 1 As shown.

[0032] Specifically, the method includes the following steps: S1, taking the plateau region where meteorological data is to be downscaled as the target domain and the region outside the plateau region as the source domain; acquiring low-resolution meteorological data and corresponding high-resolution meteorological data of the source domain, as well as low-resolution meteorological data and corresponding high-resolution meteorological data of the target domain, and terrain data; It should be noted that the source domain is selected based on its climatic similarity or difference with the target domain; it can be an area with a high density of one or more observation stations. The dataset constructed in this embodiment includes:

[0033] 1. Low-resolution input data ( ): Uses global or regional reanalysis data, such as ERA5 products, with a spatial resolution typically of 0.25° × 0.25°.

[0034] 2. High-resolution tag data ( ): source domain ( ): Gridded fusion observation datasets from densely populated areas of observation stations, such as the CN05.1 dataset, are used, with a resolution of up to 0.05°×0.05°.

[0035] 3. Target domain ( ): Data were obtained from sparse meteorological stations in the plateau region.

[0036] 4. Static geographic data ( ): High-resolution digital elevation model (DEM) data, such as SRTM data, is used. This data needs to be preprocessed onto the same spatial grid as the high-resolution labeled data.

[0037] S2, based on low-resolution meteorological data from the source domain and corresponding high-resolution meteorological data, trains a basic downscaling network model to fully learn the universal physical mapping relationship between meteorological elements and local terrain. In this embodiment, the basic downscaling network model ( The algorithm employs an encoder-decoder architecture. For example, the U-Net network can be used, whose unique "skip connections" structure can effectively fuse deep semantic features with shallow spatial detail features, making it suitable for image-to-image conversion tasks such as downscaling.

[0038] The purpose of pre-training a basic downscaling network based on source domain data is to train the network in a data-rich source domain. To learn from low-resolution weather fields To high-resolution meteorological fields The general mapping relationship. In this process, the input is low-resolution data from the source domain. The output is the high-resolution prediction value of the source domain generated by the model. The objective function uses reconstruction loss, such as mean squared error (MSE). Its formula is as follows:

[0039] Training process: Minimize the above using a gradient descent algorithm (such as the Adam optimizer). Update network The weight parameters are adjusted until the model converges, thus obtaining a pre-trained base model. .

[0040] S3. Construct a composite objective function. Based on the low-resolution meteorological data and topographic data of the target domain, combine the composite objective function to adaptively train the trained basic downscaling network model. The composite objective function includes a topographic difference regularization term, which is used to apply a physical constraint so that the spatial gradient change of the downscaling result is structurally consistent with the gradient change of the local topography. The above steps are used to process the pre-trained model. The model is migrated to a sparsely observed target plateau region and trained using a composite loss function. This allows the model to adapt to the new data distribution while building an understanding of the terrain's impact. The model input consists of low-resolution meteorological data and high-resolution terrain data from the target region, as well as low-resolution meteorological data from the source region (used to calculate the domain adaptation loss). The output is the predicted value of the high-resolution meteorological data from the target region. Thus, during the migration process, the model incorporates the terrain data from the target region. Through the introduction of a terrain difference regularization term and a domain alignment loss term, the model is guided to adaptively adjust based on the unique and complex terrain features of the target region, drawing upon the knowledge learned from the source region. This stage uses the low-resolution input from the target region. DEM data of the target domain and a small number of available high-resolution target domain labels .

[0041] Composite objective function The composition is as follows: in, and To balance the hyperparameters of each loss term; Reconstruction loss in the target domain, such as mean squared error, is the fundamental guarantee for the model to achieve high performance in the target domain.

[0042] Domain alignment loss is used to reduce the data distribution differences between the source and target domains in the feature space, enabling the model to learn generalized features across regions. Its implementation involves using the Maximum Mean Discrepancy (MMD) loss, calculated as follows:

[0043] in, Representing the network Feature extractor of intermediate layer and These are the number of samples in the source domain and the target domain, respectively. The square of the Hilbert space norm is usually specifically the norm in the Reproducing Kernel Hilbert Space (RKHS). Simply put, this symbol is used to calculate the square of the vector distance in the high-dimensional feature space after mapping using the kernel method. This represents a feature mapping function that maps features extracted from a low-resolution image to the reproducing kernel Hilbert space.

[0044] This formula represents mapping the features of the two points to RKHS, and then calculating the square of the distance between the two mean points in the feature space.

[0045] This is a topographic difference regularization term, used to impose a physical constraint that the spatial gradient variation of the downscaling result should be structurally consistent with the gradient variation of the local topography. The aim is to minimize the high-resolution meteorological field output by the model in the target domain. Spatial gradient field and target domain DEM data The difference between the spatial gradient fields; its expression is as follows:

[0046] in, It is a spatial gradient operator (which can be approximated by the Sobel operator). Let α be the L1 norm, and α be a scaling factor that matches the meteorological gradient with the terrain gradient. This indicates taking the average value.

[0047] S4. Based on the low-resolution meteorological data and the corresponding high-resolution meteorological data of the target domain, the basic downscaling network model that has completed adaptive training is fine-tuned to obtain the final downscaling model. The above steps utilize a limited amount of labeled data from the plateau region. Fine-tuning is then performed on the adaptively trained model to further enhance its ability to capture the unique climate patterns of the plateau. The strategy in this stage is to use a relatively small learning rate to fine-tune the model. The parameters are trained using all or some of the parameters. This stage uses only the reconstruction loss, and the objective function is as follows:

[0048] in, This represents the number of available labeled samples in the plateau region.

[0049] S5 inputs the low-resolution meteorological data to be downscaled into the downscaling model to obtain high-resolution downscaling results.

[0050] Furthermore, to objectively evaluate the performance of the method of the present invention, this embodiment also uses an independent test dataset reserved from high-altitude areas that is not involved in training or fine-tuning. The downscaling results of the final model are then tested. Objective evaluation metrics such as root mean square error (RMSE), mean absolute error (MAE), and correlation coefficient (CC) are used to quantify and verify the effectiveness and accuracy of the method presented in this invention. The calculation formulas for each evaluation metric are as follows:

[0051] in, and These represent the average of the predicted value and the actual value, respectively.

[0052] Furthermore, to verify the effectiveness of the method of the present invention, it can be compared with the following baseline methods: 1. Direct Training Method: Train the network from scratch using only sparse samples from a high-altitude region. This method is used to demonstrate the performance bottleneck under sparse data conditions.

[0053] 2. Traditional transfer learning method: It only includes pre-training and fine-tuning, and does not include terrain adaptation and domain alignment.

[0054] 3. Traditional spatial interpolation methods: such as bilinear interpolation or Kriging interpolation.

[0055] The above comparative experiments can quantitatively verify the performance improvement of the method of the present invention on various evaluation indicators.

[0056] Second Embodiment This embodiment applies the plateau meteorological data downscaling method of cross-terrain knowledge transfer of the present invention to downscaling precipitation data in the Qinghai-Tibet Plateau region to verify the effectiveness of the method of the present invention.

[0057] 1. Data Description: 1) Source region: The eastern plains of China (where observation stations are densely located) were selected.

[0058] 2) Target domain: Select a region of the Qinghai-Tibet Plateau.

[0059] 3) Specific data sources: Low-resolution input uses daily precipitation data from ERA5; high-resolution labels use CN05.1 grid data in the source domain and station data from national meteorological stations in the target domain. DEM data uses SRTM90m data.

[0060] 2. Model and parameters: 1) Network structure: A U-Net network containing 5 downsampling and 5 upsampling modules is adopted.

[0061] 2) Hyperparameter settings: The learning rate is set to 1e-4. Set it to 0.5. Set it to 0.8.

[0062] 3. Execution process: Step 1, Source Domain Pre-training 1) Data pairing: The low-resolution ERA5 precipitation data from the source region (Eastern Plains of China) was spatiotemporally aligned with the corresponding high-resolution CN05.1 precipitation data, and then cropped into several... training sample pairs of pixels .

[0063] 2) Model Training: Input the above training sample pairs into the U-Net network to minimize the reconstruction loss. To achieve this, the Adam optimizer is used. The learning rate is used for iterative training.

[0064] 3) Model saving: Once the model's loss on the validation set reaches the convergence criterion, training stops, the current network weights are saved, and the pre-trained base model is obtained. This model has learned the universal downscaling rules between precipitation fields and general topography.

[0065] Step 2, Mode Adaptation 1) Data preparation: Prepare low-resolution ERA5 precipitation data for the target region (Tibetan Plateau). And the corresponding gridded DEM elevation data. .

[0066] 2) Loading the model: Load the pre-trained base model The weights are used as the initial state.

[0067] 3) Adaptive training: This involves training the target domain data... Input model. To minimize the composite objective function. Train for the target.

[0068] During this process: By continuously comparing the feature distribution of input data from the source and target domains in the intermediate layers of the network, the model is guided to learn domain-invariant features. The difference between the spatial gradient of the precipitation prediction field output by the model and the spatial gradient of the DEM elevation field is continuously calculated, which forces the model to learn and fit the unique topographic precipitation effect of the plateau region.

[0069] Step 3, Model Saving: After training convergence, the terrain adaptive model is obtained. .

[0070] Step 4, fine-tuning the target domain 1) Data preparation: Using a small amount of valuable station observation data collected in the target region (Tibetan Plateau), high-resolution labeled samples were formed after quality control and interpolation. .

[0071] 2) Loading Model: Load the terrain adaptive model The weight.

[0072] 3) Fine-tuning training: This involves adjusting the corresponding low-resolution data. Input model to minimize target domain reconstruction loss To achieve this, a significantly reduced learning rate is adopted (e.g., The model is then trained. This step aims to use real data to perform a final calibration of the model's output, making it more consistent with the actual climate characteristics of the plateau.

[0073] 4) Model finalization: After several fine-tuning cycles (epochs), the final downscaling model is obtained. .

[0074] Step 5, Effect Verification and Comparison 1) Test set evaluation: Input the reserved independent test data from the high-altitude region that has not participated in any of the above training processes. The model generates downscaling prediction results.

[0075] 2) Indicator Calculation: Compare the prediction results with the actual observations in the test set, and calculate a series of objective evaluation indicators such as RMSE, MAE, and CC.

[0076] 3) Performance comparison: On the same test set, run baseline comparison methods (such as models trained directly on only plateau data, traditional transfer learning models, and Kriging interpolation models) and calculate their evaluation metrics.

[0077] Results Analysis: By comparing various indicators, the results obtained in the embodiments of the present invention are quantitatively demonstrated. The model outperforms all baseline methods. Expected results indicate that this method not only achieves higher numerical accuracy but also reasonably reproduces the fine spatial distribution characteristics of precipitation caused by complex topography, such as the rain belt on the windward slope and the rain shadow area on the leeward slope.

[0078] Expected Results: Compared with baseline methods, the downscaled precipitation field generated in this embodiment shows significant advantages in both RMSE and CC indices. In particular, the results clearly reflect the precipitation enhancement effect along the Yarlung Tsangpo River valley and the windward slope of the Hengduan Mountains, which is in high agreement with actual physical laws, while other methods cannot generate such fine structures.

[0079] Third Embodiment This embodiment provides an electronic device, such as... Figure 2 As shown, the electronic device includes a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. Furthermore, the electronic device may also include a transceiver, the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0080] Below, in conjunction with Figure 2 A detailed introduction to each component of this electronic device is provided below: The processor is the control center of the electronic device. The electronic device may include multiple processors, each of which can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The term "processor" can refer to a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), other general-purpose processors, application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), one or more field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0081] In a specific implementation, as one example, the processor may include one or more CPUs, for example... Figure 2 CPU0 and CPU1 shown are, of course, merely illustrative examples.

[0082] The memory is used to store the software program that executes the solution of the present invention, and the processor controls its execution. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0083] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may be integrated with the processor or exist independently, and may be accessed through the interface circuit of the electronic device ( Figure 2 (Not shown in the image) is coupled to the processor; however, this embodiment of the invention does not impose specific limitations on this.

[0084] The transceiver may include a receiver and a transmitter. Figure 2 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver can be integrated with the processor or exist independently, and can be connected through the interface circuit of the electronic device (…). Figure 2 (Not shown in the image) is coupled to the processor, and this embodiment of the invention does not specifically limit this.

[0085] In addition, it should be noted that, Figure 2 The structure of the electronic device shown is not intended to limit the device. Actual devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Furthermore, the technical effects achieved by this electronic device when performing the method of the first embodiment described above can be referenced to the technical effects described in the first embodiment; therefore, they will not be repeated here.

[0086] Fourth embodiment This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0087] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely or partially hardware embodiment, a completely or partially software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented in software, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any usable medium accessible to a computer or a data storage device such as a server or data center containing one or more sets of usable media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive (SSD).

[0088] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (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, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0089] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal 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 functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal 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.

[0090] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element. Furthermore, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone, where A and B can be singular or plural. Additionally, the character " / " in this text generally indicates an "or" relationship between the preceding and following objects, but it can also indicate an "AND / OR" relationship. Please refer to the context for specific interpretations. "At least one" refers to one or more items, while "more than" refers to two or more items. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can be represented as: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0091] Furthermore, it is understood that in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0092] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0093] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of functional modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, the functional units in the various embodiments of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0094] If the method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments of the present invention have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make several improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for downscaling plateau meteorological data through cross-terrain knowledge transfer, characterized in that, include: The plateau region where meteorological data is to be downscaled is used as the target domain, and the region outside the plateau region is used as the source domain. Acquire low-resolution meteorological data and corresponding high-resolution meteorological data of the source domain, as well as low-resolution meteorological data and corresponding high-resolution meteorological data of the target domain, and terrain data; Based on low-resolution meteorological data from the source domain and corresponding high-resolution meteorological data, a basic downscaling network model is trained to learn the mapping relationship from low-resolution meteorological data to high-resolution meteorological data. A composite objective function is constructed. Based on low-resolution meteorological and topographic data of the target domain, the pre-trained basic downscaling network model is adaptively trained using the composite objective function. The composite objective function includes a topographic difference regularization term, which is used to apply a physical constraint so that the spatial gradient change of the downscaling result is structurally consistent with the gradient change of the local topography. Based on low-resolution meteorological data and corresponding high-resolution meteorological data in the target domain, the basic downscaling network model that has completed adaptive training is fine-tuned to obtain the final downscaling model. The low-resolution meteorological data to be downscaled is input into the downscaling model to obtain high-resolution downscaling results.

2. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 1, characterized in that, The terrain data used is high-resolution digital elevation model data.

3. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 1, characterized in that, The basic downscaling network model adopts an encoder-decoder architecture.

4. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 1, characterized in that, When training the basic downscaling network model, the objective function is the reconstruction loss function.

5. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 1, characterized in that, The expression for the terrain difference regularization term is: ； in, Represents the terrain difference regularization term; ∇ is the spatial gradient operator; α is the L1 norm; α is a scaling factor that matches the meteorological gradient with the terrain gradient. This represents the high-resolution meteorological field output by the model in the target domain. For terrain data of the target domain; This indicates taking the average value.

6. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 1, characterized in that, The composite objective function also includes a domain alignment loss term; the domain alignment loss term is used to reduce the data distribution differences between the source domain and the target domain in the feature space, enabling the model to learn generalized features across regions.

7. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 6, characterized in that, The expression for the domain alignment loss term is: ； in, Indicates the domain alignment loss term; This represents a feature extractor for the intermediate layer of the network; Represents the first term of the source domain i A low-resolution meteorological data sample; Represents the first of the target domains j A low-resolution meteorological data sample; and These are the number of samples in the source domain and the target domain, respectively. This indicates that the square of the vector distance is calculated. This represents the feature mapping function.

8. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 7, characterized in that, The composite objective function is expressed as: ； in, and To balance the hyperparameters of each loss term; This represents the terrain difference regularization term; The loss is the supervised reconstruction loss over the target domain.

9. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 7, characterized in that, When fine-tuning the basic downscaling network model that has completed adaptive training, the objective function is a reconstruction loss function.

10. The plateau meteorological data downscaling method for cross-terrain knowledge transfer as described in claim 1, characterized in that, After obtaining the final downscaling model, the method further includes: The downscaling model is validated using preset evaluation metrics; wherein the preset evaluation metrics are one or more combinations of root mean square error, mean absolute error, and correlation coefficient.

Images