Tranformer-based airborne downscaling method, system and equipment and medium

Abstract

The invention belongs to the technical field of climate data processing, discloses a Tranformer-based climate time airborne downscaling method, system and equipment and a medium, and aims to solve the problems that the prior art lacks a combined modeling capability for time and space information and the space-time consistency of generated data is weak. The method comprises the following steps: acquiring low-resolution climate data, label data and historical high-resolution data; performing feature extraction and dimension reduction processing through a convolutional layer to generate a preliminary feature; inputting the initial features into a Transform architecture so as to output high-dimensional spatial-temporal features; the high-dimensional spatial-temporal features are recovered to be high-resolution climate data at 00 o'clock, 06 o'clock, 12 o'clock and 18 o'clock; and fusing the interpolated low-resolution climate data with the recovered high-resolution climate data by adopting a global jump connection mechanism, and outputting final high-resolution climate data.

Description

Technical Field

[0001] This invention belongs to the field of climate data processing technology, specifically relating to a Transformer-based method, system, device, and medium for climate temporal downscaling. Background Technology

[0002] With the intensification of global climate change, the demand for high-resolution climate data is growing in areas such as regional climate research, environmental monitoring, and resource management. Accurate climate data provides crucial support for refined analysis and forecasting, such as simulating extreme weather events, assessing agricultural climate adaptability, and optimizing the deployment of renewable energy. However, existing numerical climate models are limited by computational resources and typically only generate low-resolution climate data, making it difficult to meet the demands for high spatiotemporal resolution in practical applications.

[0003] Currently, climate data downscaling methods are mainly divided into two categories: physical downscaling and statistical downscaling. Physical downscaling methods are based on atmospheric dynamic equations and improve data accuracy by nesting high-resolution regional models. However, they are computationally expensive and sensitive to initial and boundary conditions, limiting their applicability. Statistical downscaling methods utilize the statistical relationships between historical observation data and low-resolution model data for modeling. While computationally efficient, they perform poorly in handling complex nonlinear relationships in the climate system and struggle to accurately capture extreme climate events and long-term spatiotemporal dependencies. In recent years, deep learning methods have been introduced into climate downscaling tasks, especially models based on convolutional neural networks (CNNs), which have made some progress in spatial feature extraction. However, CNNs have limited receptive fields, making it difficult to model global spatiotemporal dependencies in climate data. This leads to inconsistencies or distortions in the generated high-resolution data over long time series and large spatial areas.

[0004] In summary, existing climate data processing methods are insufficient in improving spatial resolution and capturing complex spatiotemporal dependencies, making it difficult to meet the needs of high-resolution climate data for refined analysis and prediction. Summary of the Invention

[0005] Based on the aforementioned shortcomings and deficiencies in the existing technology, one of the objectives of this invention is to at least solve one or more of the aforementioned problems in the existing technology. In other words, one of the objectives of this invention is to provide a Transformer-based method, system, device, and medium for climate spatiotemporal downscaling that meets one or more of the aforementioned requirements, so as to solve the problem that the existing technology lacks the ability to jointly model temporal and spatial information, and the generated data has obvious deficiencies in spatiotemporal consistency.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a climate spatiotemporal downscaling method based on Transformer, comprising the following steps: S1, acquiring low-resolution climate data, label data, and historical high-resolution data; interpolating and concatenating the low-resolution climate data and label data to form 26-channel input data; the historical high-resolution data being used directly as a high-resolution prior; S2, performing feature extraction and dimensionality reduction processing on the 26-channel input data and the historical high-resolution data through convolutional layers to generate preliminary features; S3, inputting the preliminary features into a Transformer architecture to output a high-dimensional spatiotemporal downscaling method. The Transformer architecture includes: a multi-head self-attention mechanism to capture global spatiotemporal dependencies in the initial features; a positional encoding layer to enhance the spatiotemporal dimension feature representation capability; residual connections and layer normalization layers to stabilize the training process; S4, the high-dimensional spatiotemporal features are restored to high-resolution climate data at 00:00, 06:00, 12:00 and 18:00 through a multi-temporal resolution recovery module; S5, a global skip connection mechanism is used to fuse the low-resolution climate data interpolated in step S1 with the high-resolution climate data restored in step S4 to output the final high-resolution climate data.

[0007] As a preferred embodiment, the low-resolution climate data includes four upper-air variables and five surface variables; the labeled data includes four time channels and one topographic channel; and the historical high-resolution data includes time and data elevation.

[0008] As a preferred embodiment, the four layers correspond to 250 hPa, 500 hPa, 700 hPa, and 850 hPa. Each layer of upper-level variables includes geopotential, temperature, zonal wind, and radial wind; the five surface variables are 2-meter temperature, 10-meter zonal wind, 10-meter radial wind, sea-level air pressure, and precipitation.

[0009] As a preferred embodiment, the interpolation calculation in step S1 specifically includes: interpolating the low-resolution climate data and the label data with an original grid size of 256×512 to a uniform resolution of 720×1440 using a bilinear interpolation method.

[0010] As a preferred approach, the Transformer architecture is trained in an end-to-end manner, employing a multi-task loss function during training; the expression for the multi-task loss function is as follows: In the formula, For spatial resolution loss, is the predicted value, If it is true, For adaptive weights, It is 1e-3.

[0011] As a preferred approach, the Transformer architecture is evaluated using a variety of metrics, including spatial resolution improvement rate, temporal consistency metric, and mean absolute error.

[0012] In a second aspect, the present invention provides a Transformer-based climate temporal downscaling system for implementing the climate temporal downscaling method as described in the first aspect, comprising: Input module, feature extraction module, modeling module, and output module; The input module includes a data acquisition unit, a data preprocessing unit, and a data input unit; The data acquisition unit is used to acquire low-resolution climate data and label data to form 26-channel input data, and to acquire historical high-resolution data as high-resolution priors. The data preprocessing unit is used to perform interpolation calculations on low-resolution climate data and label data; The data input unit is used to input the 26-channel input data and the historical high-resolution data into the feature extraction module; The feature extraction module is used to receive the 26-channel input data and the historical high-resolution data, perform feature extraction and dimensionality reduction processing, and generate preliminary features; The modeling module adopts the Transformer architecture, captures the global spatiotemporal dependencies in the preliminary features through a multi-head attention mechanism, and combines positional encoding to enhance the ability to express information in the time and space dimensions. The output module includes a multi-time resolution recovery unit and a data output unit; The multi-temporal resolution recovery unit is used to restore the high-dimensional spatiotemporal features output by the Transformer architecture into high-resolution climate data at 00:00, 06:00, 12:00 and 18:00. The data output unit uses a global jump connection mechanism to fuse the low-resolution climate data acquired by the data acquisition unit with the high-resolution climate data recovered by the multi-temporal resolution recovery unit, and outputs the final high-resolution climate data.

[0013] As a preferred embodiment, the modeling module further includes a model training unit; the model training unit trains the Transformer architecture in an end-to-end manner, and the training process uses a multi-task loss function.

[0014] Thirdly, the present invention provides an electronic device, the computer device including a memory, a processor and a computer program, wherein when the computer program is executed by the processor, it implements the climate time-space downscaling method as described in the first aspect.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the climate time-space downscaling method as described in the first aspect.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. By introducing the Transformer architecture and utilizing a multi-head self-attention mechanism, this invention effectively overcomes the shortcomings of traditional methods in modeling long-term series and large-scale spatial data, accurately capturing global spatiotemporal dependencies in climate data, and significantly improving the generation quality of high-resolution data.

[0017] 2. This invention achieves effective fusion of low-resolution original data and generated high-resolution data through the synergistic effect of multi-time resolution recovery modules and global jump connection mechanism, ensuring that the model performs more consistently at different time scales, while improving the accuracy and continuity of the generated data in both spatial and temporal dimensions.

[0018] 3. This invention proposes a multi-task loss function that combines reconstruction loss, multi-resolution loss, and spatiotemporal smoothing loss. During the Transformer architecture training process, it simultaneously optimizes spatial resolution, temporal consistency, and data smoothness, making the generated high-resolution climate data superior to existing methods in both accuracy and stability, thus enhancing the model's generalization ability and practicality.

[0019] Further or more detailed beneficial effects will be described in conjunction with specific embodiments in the detailed implementation. Attached Figure Description

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 This is a schematic flowchart of the climate time-space downscaling method described in Embodiment 1 of the present invention.

[0022] Figure 2 This is a structural diagram of the electronic device described in Embodiment 3 of the present invention.

[0023] Icon labels: 200. Electronic devices; 201. Processor; 202. Communication bus; 203. User interface; 204. Network interface; 205. Memory. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0025] In the following description, several embodiments of the present invention are provided. Different embodiments can be substituted or combined. Therefore, the present invention can also be considered to include all possible combinations of the same and / or different embodiments described. Thus, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present invention should also be considered to include embodiments containing one or more other possible combinations of A, B, C, and D, even if such embodiments are not explicitly described in the following text.

[0026] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of the invention. Various processes or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0027] To facilitate a better understanding of the embodiments of the present invention, its application scenarios will be explained before providing a detailed explanation of the specific implementation methods.

[0028] The climate temporal downscaling method described in the embodiments of this specification is applied to the early warning process of extreme climate events. In these scenarios, the application of the climate temporal downscaling method aims to improve the spatial and temporal accuracy of early warning data, enhance the accuracy and timeliness of early warnings, and optimize the generalization ability and practicality of early warning models.

[0029] Example 1: like Figure 1As shown, this embodiment provides a Transformer-based method for spatiotemporal downscaling of climate data, including the following steps: S1, acquiring low-resolution climate data, label data, and historical high-resolution data; interpolating and concatenating the low-resolution climate data and label data to form 26-channel input data; using the historical high-resolution data as a high-resolution prior directly; S2, performing feature extraction and dimensionality reduction on the 26-channel input data and the historical high-resolution data through convolutional layers to generate preliminary features; S3, inputting the preliminary features into a Transformer architecture to output high-dimensional spatiotemporal features. The Transformer architecture includes: a multi-head self-attention mechanism to capture global spatiotemporal dependencies in the preliminary features; a positional encoding layer to enhance the expressive power of spatiotemporal dimension features; and in each layer, a standard feed-forward network is connected after the self-attention module to further extract features and improve expressive power through nonlinear transformation. In addition, to prevent overfitting and accelerate the training process, the model uses residual connections and layer normalization layers in each layer to stabilize the training process; S4, the high-dimensional spatiotemporal features are restored to high-resolution climate data at 00:00, 06:00, 12:00 and 18:00 through the multi-temporal resolution recovery module; S5, a global skip connection mechanism is used to fuse the low-resolution climate data after interpolation in step S1 with the high-resolution climate data restored in step S4 to output the final high-resolution climate data.

[0030] More specifically, the low-resolution climate data is EC-Earth3 (70km) in CMIP6 (Global Model Coupling and Intercomparison Project), and the label data is ERA5 (25km).

[0031] Specifically, this embodiment provides a preferred implementation method, wherein the low-resolution climate data includes 4 layers of upper-air variables and 5 surface variables; the label data includes 4 time channels and 1 topographic channel; and the historical high-resolution data includes time and data elevation DEM (digital elevation model).

[0032] Specifically, this embodiment provides a preferred implementation, wherein the four layers correspond to 250hPa, 500hPa, 700hPa, and 850hPa; the upper-air variables for each layer include geopotential (Z), temperature (T), zonal wind (U), and radial wind (V); the five surface variables are 2-meter temperature (TM2), 10-meter zonal wind (U10M), 10-meter radial wind (V10M), mean sea-level pressure (MSL), and total precipitation (TP).

[0033] Specifically, this embodiment provides a preferred implementation method. The interpolation calculation in step S1 specifically includes: interpolating the low-resolution climate data and the label data with an original grid size of 256×512 to a resolution of 720×1440 using a bilinear interpolation method.

[0034] Specifically, this embodiment provides a preferred implementation method, which trains the Transformer architecture in an end-to-end manner, and the training process uses a multi-task loss function; the expression of the multi-task loss function is as follows: In the formula, For spatial resolution loss, is the predicted value, If it is true, For adaptive weights, The value is 1e-3. This embodiment uses Charbonnier loss, a smooth L1 norm approximation loss function commonly used in deep learning tasks such as image restoration, denoising, and super-resolution. Compared to directly using L1 loss, Charbonnier loss is smooth at the origin, facilitating gradient calculation and avoiding the instability of gradients near zero.

[0035] Specifically, this embodiment provides a preferred implementation method, which uses multiple metrics to evaluate the Transformer architecture; the metrics include spatial resolution improvement rate, temporal consistency index, and mean absolute error.

[0036] Example 2: This embodiment provides a Transformer-based climate temporal downscaling system for implementing the climate temporal downscaling method as described in the first aspect, including: Input module, feature extraction module, modeling module, and output module; The input module includes a data acquisition unit, a data preprocessing unit, and a data input unit; The data acquisition unit is used to acquire low-resolution climate data and label data to form 26-channel input data, and to acquire historical high-resolution data as high-resolution priors. The data preprocessing unit is used to perform interpolation calculations on low-resolution climate data and label data; The data input unit is used to input the 26-channel input data and the historical high-resolution data into the feature extraction module; The feature extraction module is used to receive the 26-channel input data and the historical high-resolution data, perform feature extraction and dimensionality reduction processing, and generate preliminary features; The modeling module adopts the Transformer architecture, captures the global spatiotemporal dependencies in the preliminary features through a multi-head attention mechanism, and combines positional encoding to enhance the ability to express information in the time and space dimensions. The output module includes a multi-time resolution recovery unit and a data output unit; The multi-temporal resolution recovery unit is used to restore the high-dimensional spatiotemporal features output by the Transformer architecture into high-resolution climate data at 00:00, 06:00, 12:00 and 18:00. The data output unit uses a global jump connection mechanism to fuse the low-resolution climate data acquired by the data acquisition unit with the high-resolution climate data recovered by the multi-temporal resolution recovery unit, and outputs the final high-resolution climate data.

[0037] Specifically, this embodiment provides a preferred implementation, wherein the modeling module further includes a model training unit; the model training unit trains the Transformer architecture in an end-to-end manner, and the training process uses a multi-task loss function.

[0038] Example 3: like Figure 2 As shown, this embodiment provides an electronic device, which may include: at least one processor, at least one network interface, a user interface, a memory, and at least one communication bus.

[0039] The communication bus can be used to enable communication between the various components mentioned above.

[0040] The user interface may include buttons, and optional user interfaces may also include standard wired interfaces and wireless interfaces.

[0041] The network interface may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0042] The processor may include one or more processing cores. It connects various parts of the electronic device via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various functions and process data. Optionally, the processor can be implemented using at least one hardware form of DSP, FPGA, or PLA. The processor may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0043] The memory may include RAM or ROM. Optionally, the memory may include a non-transitory computer-readable medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. The memory, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a downscaling application. The processor can be used to call the downscaling application stored in the memory and execute the steps of the climate spatiotemporal downscaling method mentioned in the foregoing embodiments.

[0044] Example 4: This embodiment provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the above-described instructions. Figure 1 One or more steps in the illustrated embodiment. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0045] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification 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 or transmitted through a computer-readable storage medium. 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., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0046] Those skilled in the art will understand that all or part of the processes in the method of Embodiment 1 described above can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and the implementation scheme can be combined arbitrarily.

[0047] Embodiment 5: To verify the effectiveness of the Transformer-based climate temporal downscaling method described in this specification, this embodiment sets up a comparative experiment based on the actual application scenario of the climate temporal downscaling method. The control group uses the traditional method, while the verification group uses the climate temporal downscaling method described in this invention. Through this setup, this embodiment obtains the statistical results of the control and verification groups and compares them with the actual values ​​of various extreme events. The final results are as follows: annual frequency of extreme precipitation (EC-Earth: 5.72%, Pred: 51.63%, ERA5: 49.65%), annual frequency of extreme gale (EC-Earth: 3.91%, Pred: 19.2%, ERA5: 19.86%), and annual frequency of extreme high temperature (EC-Earth: 5.83%, Pred: 13.88%, ERA5: 12.37%). It should be noted that EC-Earth represents the global model, Pred represents the predicted result, ERA5 represents the true value, and the percentages following it represent the proportion of the global area where extreme events occur.

[0048] As can be seen from the above, this embodiment verifies the effectiveness of the Transformer-based climatological temporal downscaling method described in this specification.

[0049] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0050] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0051] The above description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the invention. Any equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of embodiments of the invention upon considering the specification and practicing the disclosure herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of the invention are defined by the claims.

Claims

1. A Transformer-based method for temporal downscaling of climate, characterized in that, Including the following steps: S1. Acquire low-resolution climate data, label data, and historical high-resolution data. Perform interpolation calculations on the low-resolution climate data and label data and stitch the channels together to form 26-channel input data. The historical high-resolution data is used directly as a high-resolution prior. S2. Perform feature extraction and dimensionality reduction on the 26-channel input data and the historical high-resolution data through convolutional layers to generate preliminary features; S3. Input the preliminary features into the Transformer architecture to output high-dimensional spatiotemporal features. The Transformer architecture includes: A multi-head self-attention mechanism is used to capture the global spatiotemporal dependencies in the preliminary features; Location encoding layer, used to enhance the ability to represent features in the spatiotemporal dimension; Residual connections and layer normalization layers are used to stabilize the training process; S4. The high-dimensional spatiotemporal features are restored to high-resolution climate data at 00:00, 06:00, 12:00 and 18:00 using the multi-temporal resolution recovery module; S5. Using a global jump connection mechanism, the low-resolution climate data interpolated in step S1 is fused with the high-resolution climate data recovered in step S4 to output the final high-resolution climate data.

2. The Transformer-based method for downscaling climate time as described in claim 1, characterized in that: The low-resolution climate data includes four layers of upper-air variables and five surface variables; The tag data includes four time channels and one terrain channel; The historical high-resolution data includes both time and data elevation.

3. The Transformer-based method for downscaling climate time as described in claim 2, characterized in that: The four layers correspond to 250hPa, 500hPa, 700hPa, and 850hPa; The variables at each level of altitude include geopotential, temperature, zonal wind, and radial wind; The five surface variables are 2-meter temperature, 10-meter zonal wind, 10-meter radial wind, sea level pressure, and precipitation.

4. The Transformer-based method for downscaling climate time as described in claim 1, characterized in that, The interpolation calculation in step S1 specifically includes: The low-resolution climate data and the label data, with an original grid size of 256×512, were uniformly interpolated to a resolution of 720×1440 using a bilinear interpolation method.

5. A Transformer-based method for downscaling climate time as described in claim 1, characterized in that: The Transformer architecture is trained in an end-to-end manner, and a multi-task loss function is used during the training process. The expression for the multi-task loss function is as follows: ， In the formula, For spatial resolution loss, is the predicted value, If it is true, For adaptive weights, It is 1e-3.

6. A Transformer-based method for downscaling climate time as described in claim 5, characterized in that: The Transformer architecture is evaluated using multiple metrics. The metrics include spatial resolution improvement rate, temporal consistency index, and mean absolute error.

7. A Transformer-based climatological temporal downscaling system, characterized in that, A method for implementing the climatological temporal downscaling method as described in any one of claims 1 to 6 includes: Input module, feature extraction module, modeling module, and output module; The input module includes a data acquisition unit, a data preprocessing unit, and a data input unit; The data acquisition unit is used to acquire low-resolution climate data and label data to form 26-channel input data, and to acquire historical high-resolution data as high-resolution priors. The data preprocessing unit is used to perform interpolation calculations on low-resolution climate data and label data; The data input unit is used to input the 26-channel input data and the historical high-resolution data into the feature extraction module; The feature extraction module is used to receive the 26-channel input data and the historical high-resolution data, perform feature extraction and dimensionality reduction processing, and generate preliminary features; The modeling module adopts the Transformer architecture, captures the global spatiotemporal dependencies in the preliminary features through a multi-head attention mechanism, and combines positional encoding to enhance the ability to express information in the time and space dimensions. The output module includes a multi-time resolution recovery unit and a data output unit; The multi-temporal resolution recovery unit is used to restore the high-dimensional spatiotemporal features output by the Transformer architecture into high-resolution climate data at 00:00, 06:00, 12:00 and 18:

00. The data output unit uses a global jump connection mechanism to fuse the low-resolution climate data acquired by the data acquisition unit with the high-resolution climate data recovered by the multi-temporal resolution recovery unit, and outputs the final high-resolution climate data.

8. A Transformer-based climatological temporal downscaling system according to claim 7, characterized in that: The modeling module also includes a model training unit; The model training unit trains the Transformer architecture in an end-to-end manner, and the training process uses a multi-task loss function.

9. A computer device, the computer device comprising a memory, a processor, and a computer program, characterized in that, When the computer program is executed by the processor, it implements the climatological time-space scaling method as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the climatological time-space scaling method as described in any one of claims 1 to 6.

Images