A climate pattern multivariate statistical downscaling method, storage medium, device

Abstract

The application discloses a climate mode multivariate statistical downscaling method, a storage medium and equipment, relates to the technical field of climate mode statistical downscaling, and comprises the following steps: acquiring mode circulation information, measured precipitation and temperature information output by a global climate mode, and dividing the information into a training set and a test set; constructing a climate mode multivariate statistical downscaling initial model, including a circulation information compression module, a precipitation processing module, a temperature processing module, a temperature and precipitation output module; training the circulation information compression module, a new circulation information compression module retaining an encoder after training, and replacing the circulation information compression module of the initial model; using the training set and the test set to train and test the climate mode multivariate statistical downscaling model, inputting mode circulation information to be processed into the model after training and testing, and obtaining down-scaled precipitation and temperature information. The application avoids multiple loading of input data and waste of running time and resources.

Description

Technical Field

[0001] This invention relates to the field of statistical downscaling technology for climate models, and more particularly to a method, storage medium, and device for multivariate statistical downscaling of climate models. Background Technology

[0002] Global climate models (GCMs) can accurately predict future global climate change and short-term climate at monthly and seasonal scales. However, their current spatial resolution (typically around 50-200 km) is low, lacking detailed regional climate information and hindering reasonable regional climate predictions. GCM simulation data also exhibits significant bias compared to meteorological data observed at the same grid scale. To address the scale mismatch between climate model output and response model input, as well as the bias in GCM output data, downscaling methods have been developed. Downscaling can overcome the limitations of global climate models in predicting regional climate change by transforming large-scale climate information from global climate models into regional-scale climate information (such as temperature and precipitation), thereby enabling regional climate predictions.

[0003] There are three main methods for downscaling global climate models: dynamic downscaling, statistical downscaling, and a combination of both. Dynamic downscaling is based on regional climate models, while statistical downscaling includes ideal forecasting, model output statistics, and stochastic weather generators, with model output statistics involving various bias correction techniques. The method of extrapolating future regional or stationary precipitation sequences by establishing the relationship between historical precipitation grid outputs and local or stationary measured precipitation can be called the model output statistics (MOS) method. Similarly, by establishing a statistical model between the GCM's temperature grid outputs and stationary or local measured temperatures, the MOS method can also be used for statistical downscaling of temperature or other meteorological elements.

[0004] Statistical downscaling is a technique that uses model circulation to correct errors in temperature and precipitation. Currently, climate model statistical downscaling is performed on individual specific meteorological elements (such as precipitation and temperature), but the input data and processing program code are largely the same. This results in multiple downscaling programs being executed separately in climate forecasting operations. This requires multiple loading of input data and multiple executions of some modules, leading to a waste of time and resources. In addition, at different climate scales, temperature and precipitation often have a certain correlation (for example, temperatures tend to be lower when there is more precipitation / snow in winter). Predicting them separately ignores the relationship between them. Summary of the Invention

[0005] The purpose of this invention is to address the problems of resource waste and neglect of inter-meteorological relationships when performing statistical downscaling on multiple meteorological elements. A multivariate statistical downscaling method for climate models is proposed, comprising the following steps:

[0006] S1. Obtain model circulation information output by global climate models, as well as local or station-measured precipitation and temperature information, and divide the model circulation information, measured precipitation and temperature information into training set and test set;

[0007] S2. Construct an initial multivariate statistical downscaling model for the climate model, including a circulation information compression module, a precipitation processing module, a temperature processing module, and temperature and precipitation output modules;

[0008] The input to the circulation information compression module is the model circulation information. The output of the circulation information compression module is input to the precipitation processing module and the temperature processing module, respectively. The outputs of the precipitation processing module and the temperature processing module are input to the temperature and precipitation output modules.

[0009] The circulating information compression module includes an encoder and a decoder;

[0010] Both the precipitation processing module and the temperature processing module include multiple stacked convolutional neural network blocks and a fully connected layer, and there is a residual connection between the precipitation processing module and the temperature processing module.

[0011] S3. Use the model circulation information to train the circulation information compression module, and retain the trained encoder as a component of the new circulation information compression module; replace the circulation information compression module of the initial model with the new circulation information compression module to obtain the climate model multivariate statistical downscaling model.

[0012] S4. Use the training set and test set to train and test the multivariate statistical downscaling model of the climate model. Input the model circulation information to be processed into the trained and tested model to obtain downscaled precipitation and temperature information.

[0013] Furthermore, the model circulation information includes: sea surface temperature, geopotential height at 500 hPa and 200 hPa, meridional and zonal winds at 200 hPa and 850 hPa, air temperature at 850 hPa, surface air temperature, and total precipitation.

[0014] Furthermore, the encoder of the circulating information compression module consists of two cascaded convolutional neural network blocks;

[0015] The decoder of the circulating information compression module consists of two concatenated convolutional neural network blocks;

[0016] The encoder of the circulating current information compression module takes pattern circulating current information as input, and the encoder output of the circulating current information compression module takes the decoder as input.

[0017] Furthermore, the convolutional neural network block of the encoder in the circulating information compression module includes: convolution, ReLU activation function, and max pooling layer; the convolutional neural network block of the decoder in the circulating information compression module includes: deconvolution, ReLU activation function, and inverse max pooling layer.

[0018] Furthermore, the precipitation processing module includes three cascaded convolutional neural network blocks and one fully connected layer, while the temperature processing module includes two cascaded convolutional neural network blocks and one fully connected layer.

[0019] The output of the second convolutional neural network block in the precipitation processing module is connected as a residual between the two cascaded convolutional neural network blocks in the temperature processing module.

[0020] Furthermore, the convolutional neural network blocks of the precipitation processing module and the temperature processing module are the same, including: convolution, ReLU activation function, and max pooling layer.

[0021] The present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described multivariate statistical downscaling method for climate models.

[0022] The present invention also proposes an electronic device, including a processor and a memory, wherein the processor and the memory are interconnected, the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to execute the above-described multivariate statistical downscaling method for climate models.

[0023] The beneficial effects of the technical solution provided by this invention are:

[0024] This invention, based on statistical downscaling of climate models, proposes multivariate statistical downscaling of climate models. First, the input model circulation variables are compressed using a circulation information compression module. The compressed data is then input into precipitation and temperature processing modules respectively. Residual connections are used to realize the influence of precipitation on temperature, and downscaled temperature and precipitation are output simultaneously. This allows for the simultaneous statistical downscaling of multiple meteorological elements, such as temperature and precipitation, integrating the workflow and avoiding multiple loading of input data, as well as wasted runtime and resources. Furthermore, simultaneous downscaling of both can reveal their interrelationships. Attached Figure Description

[0025] Figure 1 This is a flowchart of the multivariate statistical downscaling method for climate models according to an embodiment of the present invention;

[0026] Figure 2This is a structural block diagram of a multivariate statistical downscaling model of a climate model according to an embodiment of the present invention;

[0027] Figure 3 This is a structural diagram of the circulating information compression module according to an embodiment of the present invention;

[0028] Figure 4 This is a structural diagram of the precipitation processing module and the temperature processing module according to an embodiment of the present invention;

[0029] Figure 5 This is a block diagram of an electronic device according to an exemplary embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0031] Example 1: The flowchart of the multivariate statistical downscaling method for climate models in this embodiment of the invention is as follows. Figure 1 Specifically, it includes the following steps:

[0032] S1. Obtain large-scale, low-resolution model circulation information output from global climate models. In a specific embodiment, the model circulation information includes: sea surface temperature, geopotential height at 500 hPa and 200 hPa, meridional and zonal winds at 200 hPa and 850 hPa, air temperature at 850 hPa, surface air temperature, and total precipitation. It also includes local or station-measured precipitation and temperature information. The model circulation information, measured precipitation, and temperature information are then divided into training and testing sets.

[0033] S2. Construct an initial multivariate statistical downscaling model of the climate model. In the preferred embodiment, refer to... Figure 2 , Figure 2 This is a block diagram of a multivariate statistical downscaling model of a climate model according to an embodiment of the present invention, including a circulation information compression module, a precipitation processing module, a temperature processing module, and a temperature and precipitation output module.

[0034] The circulation information compression module takes the model circulation information as input and compresses the circulation field data to 1 / 4 of its original value. The output of the circulation information compression module is input to the precipitation processing module and the temperature processing module, respectively. The outputs of the precipitation processing module and the temperature processing module are input to the temperature and precipitation output modules.

[0035] In a further embodiment, the structural diagram of the circulating information compression module is referenced. Figure 3 The circulating information compression module includes an encoder and a decoder. The encoder of the circulating information compression module consists of two cascaded convolutional neural network blocks, which include convolution, ReLU activation function, and max pooling layer.

[0036] The decoder of the circulating information compression module consists of two concatenated convolutional neural network blocks. The convolutional neural network blocks of the decoder of the circulating information compression module include: deconvolution, ReLU activation function, and inverse max pooling layer.

[0037] In a preferred embodiment of the present invention, the convolution kernels of the convolution and deconvolution of the circulation information compression module are both 3×3, and the pooling windows of the max pooling layer and the antimax pooling layer are both 2×2.

[0038] The encoder of the circulating current information compression module takes pattern circulating current information as input, and the encoder output of the circulating current information compression module takes the decoder as input.

[0039] Both the precipitation processing module and the temperature processing module consist of multiple stacked convolutional neural network blocks and a fully connected layer, with a residual connection between them. See the structural diagrams of the precipitation processing module and the temperature processing module for reference. Figure 4 .

[0040] The precipitation processing module consists of three cascaded convolutional neural network blocks and one fully connected layer, while the temperature processing module consists of two cascaded convolutional neural network blocks and one fully connected layer.

[0041] The output of the second convolutional neural network block in the precipitation processing module is connected as a residual between the two cascaded convolutional neural network blocks in the temperature processing module.

[0042] The convolutional neural network blocks for the precipitation processing module and the temperature processing module are the same, including: convolution, ReLU activation function, and max pooling layer.

[0043] S3. Use the model circulation information as input and output to train the circulation information compression module, and retain the trained encoder as a component of the new circulation information compression module; replace the circulation information compression module of the initial model with the new circulation information compression module to obtain the climate model multivariate statistical downscaling model.

[0044] S4. Use the training and test sets to train and test the multivariate statistical downscaling model of the climate model. Input the model circulation information to be processed into the trained and tested model to obtain downscaled precipitation and temperature information of the study area.

[0045] The embodiments of the present invention achieve statistical downscaling of temperature and precipitation. Of course, the method of the present invention can also be used for statistical downscaling of other meteorological elements.

[0046] Example 2: In an exemplary embodiment, a computer-readable storage medium is included, which stores a computer program that, when executed by a processor, implements the above-described multivariate statistical downscaling method for climate models.

[0047] Example 3: Please refer to Figure 5 In one exemplary embodiment, the device further includes an electronic device including at least one processor, at least one memory, and at least one communication bus.

[0048] The memory stores a computer program, which includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the aforementioned multivariate statistical downscaling method for climate models.

[0049] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multivariate statistical downscaling method for climate models, characterized in that, Includes the following steps: S1. Obtain model circulation information output by global climate models, as well as local or station-measured precipitation and temperature information, and divide the model circulation information, measured precipitation and temperature information into training set and test set; S2. Construct an initial multivariate statistical downscaling model for the climate model, including a circulation information compression module, a precipitation processing module, a temperature processing module, and temperature and precipitation output modules; The input to the circulation information compression module is the model circulation information. The output of the circulation information compression module is input to the precipitation processing module and the temperature processing module, respectively. The outputs of the precipitation processing module and the temperature processing module are input to the temperature and precipitation output modules. The circulating information compression module includes an encoder and a decoder; Both the precipitation processing module and the temperature processing module include multiple stacked convolutional neural network blocks and a fully connected layer, and there is a residual connection between the precipitation processing module and the temperature processing module. S3. Use the model circulation information to train the circulation information compression module, and retain the trained encoder as a component of the new circulation information compression module; replace the circulation information compression module of the initial model with the new circulation information compression module to obtain the climate model multivariate statistical downscaling model. S4. Use the training set and test set to train and test the multivariate statistical downscaling model of the climate model. Input the model circulation information to be processed into the trained and tested model to obtain downscaled precipitation and temperature information.

2. The multivariate statistical downscaling method for climate models according to claim 1, characterized in that, The model circulation information includes: sea surface temperature, geopotential height at 500 hPa and 200 hPa, meridional and zonal winds at 200 hPa and 850 hPa, air temperature at 850 hPa, surface air temperature, and total precipitation.

3. The multivariate statistical downscaling method for climate models according to claim 1, characterized in that, The encoder of the circulating information compression module consists of two cascaded convolutional neural network blocks; The decoder of the circulating information compression module consists of two concatenated convolutional neural network blocks; The encoder of the circulating current information compression module takes pattern circulating current information as input, and the encoder output of the circulating current information compression module takes the decoder as input.

4. The multivariate statistical downscaling method for climate models according to claim 3, characterized in that, The convolutional neural network block of the encoder in the circulating flow information compression module includes: convolution, ReLU activation function, and max pooling layer; the convolutional neural network block of the decoder in the circulating flow information compression module includes: deconvolution, ReLU activation function, and inverse max pooling layer.

5. The multivariate statistical downscaling method for climate models according to claim 1, characterized in that, The precipitation processing module consists of three cascaded convolutional neural network blocks and one fully connected layer, while the temperature processing module consists of two cascaded convolutional neural network blocks and one fully connected layer. The output of the second convolutional neural network block in the precipitation processing module is connected as a residual between the two cascaded convolutional neural network blocks in the temperature processing module.

6. The multivariate statistical downscaling method for climate models according to claim 5, characterized in that, The convolutional neural network blocks for the precipitation processing module and the temperature processing module are the same, including: convolution, ReLU activation function, and max pooling layer.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

8. An electronic device, characterized in that, The device includes a processor and a memory, the processor being interconnected with the memory, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to perform the method as described in any one of claims 1-6.

Images