Method, system, computer device and storage medium for generating precipitation fusion products

By extracting space-time and spatial characteristics and fusion to form a precipitation fusion model, the deviation problems existing in future multi-scenario precipitation data are solved, and the accuracy and representativeness of the data are improved.

CN118053054BActive Publication Date: 2024-11-19INST OF GEOGRAPHICAL SCI & NATURAL RESOURCE RES CAS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410021514.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-05
Publication Date
2024-11-19
Estimated Expiration
2044-01-05

AI Technical Summary

Technical Problem

The prior art has data deviations and systematic deviations in generating future multi-scenario precipitation data, which affects the trend representativeness and accuracy of the data.

Method used

By obtaining multi-source weighted set precipitation data, CMIP6 training precipitation data and auxiliary image training data, spatiotemporal and spatial characteristics are extracted, and the precipitation fusion model is fused to generate more accurate future scenario precipitation data.

Benefits of technology

The correction of data deviation is achieved, and the accuracy and trend representativeness of future multi-scenario precipitation data are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118053054B_ABST
    Figure CN118053054B_ABST
Patent Text Reader

Abstract

The present invention relates to a precipitation fusion product generation method, system, computer equipment and storage medium. The method comprises: obtaining MSEWP data, CMIP6 training precipitation data and auxiliary image training data within a first preset time, constructing corresponding sample data for different future scenarios to obtain a precipitation data fusion sample library; extracting spatiotemporal features in CMIP6 target precipitation data; extracting spatial features in auxiliary image training data; fusing spatiotemporal features and spatial features to obtain target features, fitting the nonlinear relationship between target features and MSEWP data to obtain a precipitation fusion model; training the precipitation fusion model according to the precipitation data fusion sample library, and obtaining precipitation fusion models corresponding to different future scenarios; obtaining CMIP6 target precipitation data and auxiliary image data within a second preset time, inputting the precipitation fusion model corresponding to the future scenario, and obtaining the precipitation fusion product corresponding to the scenario, which has the advantages of realizing data deviation correction and improving the precision of precipitation data in future scenarios.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of future multi-scenario precipitation, and in particular to a precipitation fusion product generation method, system, computer equipment and storage medium. Background Art

[0002] The coupled climate models of the Sixth International Coupled Model Intercomparison Project (CMIP6) have become an important reference for the United Nations Intergovernmental Panel on Climate Change (IPCC) to prepare the Sixth Assessment Report (AR6) and the United Nations Climate Change Conference and other climate change negotiations. A series of scenarios describing different future socioeconomic development models and atmospheric greenhouse gas concentrations have been constructed around CMIP6, called shared socioeconomic pathway scenarios (SSPs). Many research institutions have developed multi-scenario precipitation data for the future (2015-2100) based on SSPs and Earth System Models (ESM). These massive open and shared CMIP6 precipitation data are widely used to explore the future trends of key processes in the Earth's surface system, such as drought and flood characteristics, heat wave patterns, vegetation evolution, and land-sea interactions. In order to enhance the representativeness of data trends and reduce abnormal deviations, mainstream research usually simply calculates the ensemble average of multiple sets of future scenario data to characterize the global future spatial and temporal distribution of precipitation, but it also faces the risk of being dominated by the data with the largest deviation. At the same time, although ESM continues to undergo optimization and iteration, there are still systematic deviations and uncertainties. Summary of the invention

[0003] Based on this, the purpose of the present invention is to provide a precipitation fusion product generation method, system, computer equipment and storage medium, which have the advantages of realizing data deviation correction and improving the data accuracy of future scenario precipitation.

[0004] A method for generating precipitation fusion products comprises the following steps:

[0005] Acquire multi-source weighted ensemble precipitation data, CMIP6 training precipitation data, and auxiliary image training data within a first preset time, construct corresponding sample data for different future scenarios, and obtain a precipitation data fusion sample library, wherein the CMIP6 training precipitation data includes future scenarios and precipitation data corresponding to the future scenarios;

[0006] Extracting the spatiotemporal features in the CMIP6 training precipitation data, specifically: forming a key feature graph by spatially mapping and spatially reducing the CMIP6 training precipitation data; extracting abstract spatiotemporal features according to the key feature graph; restoring the spatial details of the abstract spatiotemporal features to obtain the spatiotemporal features;

[0007] Extracting spatial features from the auxiliary image training data, specifically: splicing a terrain information image, a geographic location information image, and each of the CMIP6 training precipitation data evaluation information images in the auxiliary image training data into a three-dimensional image block along a channel; extracting spatial features from the three-dimensional image block by stacking two-dimensional convolutional layers; wherein the auxiliary image training data includes a terrain information image, a geographic location information image, and each of the CMIP6 training precipitation data evaluation information images;

[0008] The spatiotemporal features and the spatial features are integrated to obtain target features, and a nonlinear relationship between the target features and the multi-source weighted set precipitation data is fitted to obtain a precipitation fusion model;

[0009] According to the precipitation data fusion sample library, the precipitation fusion model is trained to obtain precipitation fusion models corresponding to different future scenarios;

[0010] The CMIP6 target precipitation data and auxiliary image data within the second preset time are obtained, and input into the precipitation fusion model corresponding to the future scenario to obtain the precipitation fusion product under the corresponding scenario.

[0011] A precipitation fusion product generation system, comprising:

[0012] A sample library acquisition module is used to acquire multi-source weighted set precipitation data, CMIP6 training precipitation data and auxiliary image training data within a first preset time, construct corresponding sample data for different future scenarios, and obtain a precipitation data fusion sample library, wherein the CMIP6 training precipitation data includes future scenarios and precipitation data corresponding to the future scenarios;

[0013] The spatiotemporal feature extraction module is used to extract the spatiotemporal features in the CMIP6 training precipitation data, specifically: forming a key feature graph by performing spatial mapping and spatial dimension reduction on the CMIP6 training precipitation data; extracting abstract spatiotemporal features according to the key feature graph; restoring the spatial details of the abstract spatiotemporal features to obtain the spatiotemporal features;

[0014] The auxiliary feature extraction module is used to extract the spatial features in the auxiliary image training data, specifically: splicing the terrain information image, the geographical location information image and each of the CMIP6 training precipitation data evaluation information images in the auxiliary image training data into a three-dimensional image block along the channel; extracting the spatial features in the three-dimensional image block through stacked two-dimensional convolution layers; wherein the auxiliary image training data includes the terrain information image, the geographical location information image and each of the CMIP6 training precipitation data evaluation information images;

[0015] A relationship simulation module, used for fusing the spatiotemporal features and the spatial features, obtaining target features, fitting the nonlinear relationship between the target features and the multi-source weighted set precipitation data, and obtaining a precipitation fusion model;

[0016] A model training module, used to train the precipitation fusion model according to the precipitation data fusion sample library, and obtain precipitation fusion models corresponding to different future scenarios;

[0017] The product acquisition module is used to obtain the CMIP6 target precipitation data and auxiliary image data within the second preset time, input them into the precipitation fusion model corresponding to the future scenario, and obtain the precipitation fusion product under the corresponding scenario.

[0018] A computer device comprises: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned precipitation fusion product generation method when executing the computer program.

[0019] A readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the precipitation fusion product generation method as described above are implemented.

[0020] The precipitation fusion product generation method described in the present application obtains multi-source weighted ensemble precipitation data, CMIP6 training precipitation data and auxiliary image training data within the first preset time, constructs corresponding sample data for different future scenarios, obtains a precipitation data fusion sample library, extracts spatiotemporal features from the CMIP6 training precipitation data and spatial features from the auxiliary image training data for fusion to obtain target features, fits the nonlinear relationship between the target features and the multi-source weighted ensemble precipitation data, obtains a precipitation fusion model, and finally trains the precipitation fusion model according to the precipitation data fusion sample library to obtain precipitation fusion models corresponding to different future scenarios, obtains CMIP6 target precipitation data and auxiliary image data within the second preset time, inputs them into the precipitation fusion model corresponding to the different future scenarios, and obtains precipitation fusion products under the corresponding scenarios. The precipitation fusion product generation method described in the present application realizes data deviation correction and improves the data accuracy of future multi-scenario precipitation.

[0021] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flowchart of a method for generating a precipitation fusion product in one embodiment of the present application;

[0023] Figure 2A flowchart of the steps for obtaining the precipitation data fusion sample library in one embodiment of the present application;

[0024] Figure 3 A flowchart of the steps of obtaining the CMIP6 training precipitation data within a first preset time in one embodiment of the present application;

[0025] Figure 4 A flowchart of the steps of extracting the spatiotemporal features in the CMIP6 training precipitation data in one embodiment of the present application;

[0026] Figure 5 A flowchart of the steps of extracting spatial features from the auxiliary image training data in one embodiment of the present application;

[0027] Figure 6 This is a flowchart of the steps of preprocessing the CMIP6 target precipitation data and auxiliary image data within the second preset time in one embodiment of the present application;

[0028] Figure 7 This is a structural schematic diagram of a precipitation fusion product generation system in one embodiment of the present application;

[0029] Figure 8 Schematic diagram of computer equipment for the precipitation fusion product generation method in the embodiment of the present application. DETAILED DESCRIPTION

[0030] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0031] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0032] It should be understood that, although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" / "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determination".

[0033] See also Figure 1 , Figure 1 The following is a flowchart of a method for generating a precipitation fusion product in one embodiment of the present application. A method for generating a precipitation fusion product is characterized by comprising the following steps:

[0034] S101, obtaining multi-source weighted ensemble precipitation data, CMIP6 training precipitation data, and auxiliary image training data within a first preset time, constructing corresponding sample data for different future scenarios, and obtaining a precipitation data fusion sample library, wherein the CMIP6 training precipitation data includes future scenarios and precipitation data corresponding to the future scenarios;

[0035] S102, extracting spatiotemporal features from the CMIP6 training precipitation data;

[0036] S103, extracting spatial features from the auxiliary image training data;

[0037] S104, fusing the spatiotemporal features and the spatial features to obtain target features, fitting the nonlinear relationship between the target features and the multi-source weighted set precipitation data to obtain a precipitation fusion model;

[0038] S105, training the precipitation fusion model according to the precipitation data fusion sample library to obtain precipitation fusion models corresponding to different future scenarios;

[0039] S106, obtaining CMIP6 target precipitation data and auxiliary image data within a second preset time, inputting them into a precipitation fusion model corresponding to a future scenario, and obtaining a precipitation fusion product under the corresponding scenario.

[0040] The precipitation fusion product generation method described in the present application obtains multi-source weighted ensemble precipitation data, CMIP6 training precipitation data and auxiliary image training data within the first preset time, constructs corresponding sample data for different future scenarios, obtains a precipitation data fusion sample library, extracts spatiotemporal features in the CMIP6 training precipitation data and spatial features in the auxiliary image training data for fusion to obtain target features, fits the nonlinear relationship between the target features and the multi-source weighted ensemble precipitation data, obtains a precipitation fusion model, and finally trains the precipitation fusion model according to the precipitation data fusion sample library to obtain precipitation fusion models corresponding to different future scenarios, obtains CMIP6 precipitation data and auxiliary image training data within the second preset time, inputs them into the precipitation fusion model corresponding to the different future scenarios, and obtains precipitation fusion products under the corresponding scenarios. The precipitation fusion product generation method described in the present application realizes data deviation correction and improves the data accuracy of future multi-scenario precipitation.

[0041] For step S101, multi-source weighted set precipitation data, CMIP6 training precipitation data and auxiliary image training data within a first preset time are obtained, corresponding sample data are constructed for different future scenarios, and a precipitation data fusion sample library is obtained;

[0042] The first preset time is time information set in advance. In this embodiment, the first preset time is set to 2015-2023.

[0043] The multi-source weighted ensemble precipitation (MSEWP) is a global precipitation product. The multi-source weighted ensemble precipitation data is precipitation data corresponding to a first preset time obtained from the multi-source weighted ensemble precipitation. In this embodiment, the multi-source weighted ensemble precipitation data is precipitation data from 2015 to 2023 in the multi-source weighted ensemble precipitation.

[0044] The CMIP6 training precipitation data is precipitation data obtained from the sixth international coupled model comparison project (CMIP6) within the first preset time. The CMIP6 training precipitation data includes future scenarios and precipitation data corresponding to the future scenarios. In this embodiment, the CMIP6 training precipitation data is the future scenarios from 2015 to 2023 in the CMIP6 and the corresponding precipitation data.

[0045] The auxiliary image training data includes a terrain information image, a geographic location information image and each of the CMIP6 training precipitation data evaluation information images, wherein the terrain information image includes an elevation information image, a slope information image and a slope direction information image, and the geographic location information image includes longitude and latitude.

[0046] The sample data is data used to train the precipitation fusion model based on the multi-source weighted ensemble precipitation data, CMIP6 training precipitation data, and auxiliary image training data. The future scenarios are different scenarios determined according to the precipitation environment. In one embodiment, the future scenarios include SSP1-2.6, SSP2-4.5, SSP3-7.0, and / or SSP5-8.5. The precipitation data fusion sample data is a database obtained by collecting sample data corresponding to each future scenario.

[0047] In this embodiment, by acquiring multi-source weighted set precipitation data, CMIP6 training precipitation data and auxiliary image training data within the first preset time, sample data corresponding to different future scenarios are acquired to form the precipitation data fusion sample library, which is used to train the precipitation fusion model to obtain precipitation fusion models corresponding to different future scenarios, thereby improving the accuracy of generating future precipitation fusion products.

[0048] See also Figure 2 , Figure 2 This is a flowchart of the steps of obtaining the precipitation data fusion sample library in one embodiment of the present application. In one embodiment, the steps of obtaining multi-source weighted set precipitation data, CMIP6 training precipitation data and auxiliary image training data within a first preset time, constructing corresponding sample data for different future scenarios, and obtaining the precipitation data fusion sample library include:

[0049] S201, for each future scenario, unifying the spatial resolutions of the multi-source weighted ensemble precipitation data, the CMIP6 training precipitation data, and the auxiliary image training data to a preset spatial resolution;

[0050] S202, extracting a multi-source weighted set precipitation image block from each multi-source weighted set precipitation data in the order of first row and then column using a sliding window of preset specifications;

[0051] S203, matching the multi-source weighted set precipitation image block with the corresponding CMIP6 training precipitation data of 5 consecutive days and the auxiliary image training data according to the spatiotemporal attributes to form an input-output sample pair as the sample data;

[0052] S204, obtaining sample data of each future scenario to form the precipitation data fusion sample library.

[0053] For step S201, the spatial resolution is the size or size of the smallest unit that can be distinguished in detail on the remote sensing image, which is used to characterize the index of the image to distinguish the details of the ground target. The preset spatial resolution is a spatial resolution set in advance for unifying data information. In one embodiment, the preset spatial resolution is set to 0.5°. In other embodiments, the preset spatial resolution can be set according to the needs, for example, the preset spatial resolution is set to 0.1°, or the spatial resolution is set to 1°.

[0054] In this embodiment, by unifying the spatial resolution of the multi-source weighted set precipitation data, the CMIP6 training precipitation data, and the auxiliary image training data to the preset resolution, and further scaling them to the range of 0 to 1 through maximum and minimum value normalization, the uniformity of the precipitation fusion model data specifications is ensured, and the convenience of data processing is improved.

[0055] For step S202, the sliding window of preset specifications is a sliding window of specifications set in advance. In this embodiment, the sliding window of preset specifications is a sliding window of size 32*32. The multi-source weighted set precipitation image block is extracted from the multi-source weighted set precipitation data through the sliding window of preset specifications, so as to facilitate the matching of the multi-source weighted set precipitation data with the CMIP6 training precipitation data and the auxiliary image training data.

[0056] For step S203, in this embodiment, the multi-source weighted set precipitation image block is matched with the corresponding CMIP6 training precipitation data and the auxiliary image training data for 5 consecutive days according to the spatiotemporal attributes to form an input-output sample pair, that is, the CMIP6 training precipitation data and the auxiliary image training data corresponding to the multi-source weighted set precipitation image block for 5 consecutive days at the corresponding moment are used as input, and the multi-source weighted set precipitation image block is used as output to form an input-output sample pair. Finally, the input-output sample pair is used as the sample data.

[0057] In one embodiment, the sample data includes a training set, a validation set, and a test set. After forming the input-output sample pairs, the following steps are further included:

[0058] Dividing the input-output sample pairs into a training set, a validation set, and a test set according to a preset ratio;

[0059] The training set, validation set and test set are used as the sample data.

[0060] The training set and validation set are used to train the precipitation fusion model, respectively, and the test set is used to evaluate the performance of the precipitation fusion model after training. The preset ratio is a division scheme of the input and output sample pairs set in advance. For example, if the preset ratio is set to 8:1:1, the input and output sample pairs are divided into the training set, validation set and test set according to the ratio of 8:1:1. In this embodiment, the input and output sample pairs are divided into the training set, validation set and test set according to the preset ratio to obtain the sample data, which is convenient for training, validating and testing the precipitation fusion model.

[0061] As for step S204, in this embodiment, the sample data corresponding to each future scenario is obtained by adopting the steps S201 to S203, and finally the sample database corresponding to each future scenario is assembled to obtain the precipitation data fusion sample library.

[0062] See also Figure 3 , Figure 3 This is a flowchart of the steps of obtaining the CMIP6 training precipitation data within a first preset time in one embodiment of the present application. In one embodiment, obtaining the CMIP6 training precipitation data within a first preset time includes the following steps:

[0063] S301, obtaining a number of CMIP6 precipitation data within the first preset time;

[0064] S302, performing system accuracy evaluation on each of the CMIP6 precipitation data, and obtaining evaluation information of each of the CMIP6 precipitation data;

[0065] S303: Select at least 10 sets of CMIP6 precipitation data from the plurality of CMIP6 precipitation data as the CMIP6 training precipitation data according to the evaluation information of each CMIP6 precipitation data.

[0066] For steps S301 to S302, precipitation data evaluation is performed respectively by acquiring a number of CMIP6 precipitation data within the first preset time to obtain evaluation information of each CMIP6 precipitation data, which is used to select CMIP6 precipitation data with smaller numerical error, higher trend fitting degree, and significant accuracy advantage in some regions / time periods to construct the precipitation data fusion sample library.

[0067] In one embodiment, the steps of performing system accuracy evaluation on each of the CMIP6 precipitation data and obtaining evaluation information of each of the CMIP6 precipitation data include:

[0068] Acquire multi-source weighted ensemble precipitation data and ERA5-Land precipitation data, and sample the multi-source weighted ensemble precipitation data, the plurality of CMIP6 precipitation data, and ERA5-Land precipitation data to a preset spatial resolution;

[0069] Substitute the multi-source weighted ensemble precipitation data, ERA5-Land precipitation data, and a set of CMIP6 single future scenario precipitation data into the formula:

[0070] Among them, P true is the true value of precipitation, P x , P y and P z are the multi-source weighted ensemble precipitation data, ERA5-Land precipitation data, and a set of CMIP6 precipitation data. x 、a y 、a z and b x , b y , b z are the intercept and slope of the corresponding linear equation, θ x ,θ y and θ z are the errors of the corresponding precipitation observation data sets respectively; the CMIP6 single future scenario precipitation data is the precipitation data corresponding to a future scenario in the CMIP6;

[0071]

[0072] in, and P x , P y and P z The error variance of , <·> indicates the mean calculation, They are P x , P y , P z The variance of <P x ′P y ′>、 <P x ′P z ′>、 <P y ′P z ′> are P x , P y , P z Covariance between pairs;

[0073]

[0074] Among them, CC x 、CC yand CC z P x , P y and P z The correlation coefficient between and the true value of precipitation;

[0075]

[0076] In the formula, SNR x 、SNR y and SNR z P x , P y and P z signal-to-noise ratio;

[0077] The error variance, correlation coefficient and signal-to-noise ratio are used as evaluation information of each CMIP6 precipitation data.

[0078] For step S303, at least 10 sets of CMIP6 precipitation data from the plurality of CMIP6 precipitation data are selected as the CMIP6 training precipitation data according to the evaluation information of each CMIP6 precipitation data.

[0079] In this embodiment, by calculating the evaluation information of each CMIP6 precipitation data, at least 10 sets of CMIP6 precipitation data with small error variance, high correlation coefficient and high signal-to-noise ratio are selected as the CMIP6 training precipitation data, so as to improve the accuracy of the precipitation data fusion sample library and the accuracy of the precipitation fusion product.

[0080] For step S102, extracting the spatiotemporal features in the CMIP6 training precipitation data;

[0081] The spatiotemporal features are the temporal information and spatial information in the CMIP6 training precipitation data.

[0082] See also Figure 4 , Figure 4 This is a flowchart of the steps of extracting the spatiotemporal features in the CMIP6 training precipitation data in one embodiment of the present application. In one embodiment, the steps of extracting the spatiotemporal features in the CMIP6 training precipitation data include:

[0083] S401, forming a key feature map by performing spatial mapping and spatial dimension reduction on the CMIP6 training precipitation data;

[0084] S402, extracting abstract spatiotemporal features according to the key feature graph;

[0085] S403, restoring the spatial details of the abstract spatiotemporal feature to obtain the spatiotemporal feature.

[0086] For steps S401 to S403, by calling the spatiotemporal feature extraction module, the CMIP6 training precipitation data is spatially mapped and spatially reduced to form a key feature map, abstract spatiotemporal features are extracted according to the key feature map, and finally the spatial details in the abstract spatiotemporal features are restored to obtain the spatiotemporal features.

[0087] The spatiotemporal feature extraction module includes a three-dimensional convolution layer (Conv3d), a convolutional long short-term memory network layer (ConvLSTM) and a three-dimensional deconvolution layer (Conv3DTranspose). The three-dimensional convolution layer performs spatial mapping and spatial dimension reduction on the CMIP6 training precipitation data to form a key feature graph, the convolutional long short-term memory network layer extracts abstract spatiotemporal features according to the key feature graph, and finally the three-dimensional deconvolution layer restores the spatial details in the abstract spatiotemporal features to obtain the spatiotemporal features.

[0088] Furthermore, the spatiotemporal feature extraction module also includes a batch normalization layer (BatchNorm), which is respectively connected after the three-dimensional convolution layer, the convolutional long short-term memory network layer and the three-dimensional deconvolution layer. The batch normalization layer transforms the output features into a Gaussian distribution with a mean of 0 and a variance of 1, and adjusts the data distribution using a scale parameter and a deviation parameter to avoid internal covariance shift, so that the precipitation fusion model has the ability to autonomously adjust the data distribution, which facilitates the training of the precipitation fusion model.

[0089] In one embodiment, the spatiotemporal feature extraction module includes 3 three-dimensional convolutional layers, 3 convolutional long short-term memory network layers, 3 three-dimensional deconvolutional layers and 9 batch normalization layers, the convolution kernel size of the three-dimensional convolutional layer is 3×3×3, the convolution step size is 2, the padding size is 1, and the number of convolution kernels in each layer is 16, 64, and 128 respectively; the convolution kernel size of the convolutional long short-term memory network layer is 5×5, the convolution step size is 1, the padding size is 2, and the number of convolution kernels in each layer is 16. The first two convolutional long short-term memory network layers retain the outputs of all 5 time steps, and the third convolutional long short-term memory network layer only retains the output of the last time step; the convolution kernel size of the three-dimensional deconvolution layer is 4×4×4, the convolution step size is 2, the padding size is 1, and the number of convolution kernels in each layer is 128.

[0090] For step S103, extracting spatial features from the auxiliary image training data;

[0091] The spatial characteristics are basic characteristics of geographical phenomena expressed in different spatial dimensions according to the spatial distribution of geographical phenomena.

[0092] See also Figure 5 , Figure 5The flowchart of the steps of extracting the spatial features in the auxiliary image training data in one embodiment of the present application. In one embodiment, the auxiliary image training data includes a terrain information image, a geographic location information image and each of the CMIP6 training precipitation data evaluation information images, and the steps of extracting the spatial features in the auxiliary image training data include:

[0093] S501, splicing the terrain information image, the geographical location information image and each of the CMIP6 training precipitation data evaluation information images in the auxiliary image training data into a three-dimensional image block along a channel;

[0094] S502, extracting spatial features from the three-dimensional image block through stacked two-dimensional convolutional layers.

[0095] For steps S501 to S502, the stacked two-dimensional convolution (Conv2d) layer is used to extract spatial features. In this embodiment, the terrain information image, the geographical location information image and each of the CMIP6 training precipitation data evaluation information images in the auxiliary image training data are first spliced ​​into a three-dimensional image block along the channel, and then the spatial features are extracted from the three-dimensional image blocks through the stacked two-dimensional convolution layer.

[0096] In one embodiment, spatial features are extracted through 6 stacked two-dimensional convolutional layers, wherein the convolution kernel size of each stacked two-dimensional convolutional layer is 3*3, and the padding operation is used to ensure that the size of the feature map before and after the convolution is consistent, and the number of convolution kernels in each layer is 16, 32, 64, 64, 128, and 128 respectively.

[0097] For step S104, the spatiotemporal features and the spatial features are fused to obtain target features, and a nonlinear relationship between the target features and the multi-source weighted set precipitation data is fitted to obtain a precipitation fusion model;

[0098] The target feature is a feature obtained by fusing the spatiotemporal feature and the spatial feature. In one embodiment, the target feature is obtained by splicing the spatiotemporal feature and the spatial feature along a channel dimension.

[0099] In this embodiment, the target feature and the multi-source weighted set precipitation data are mapped through stacked fully connected layers to obtain a nonlinear relationship between the target feature and the multi-source weighted set precipitation data.

[0100] In one embodiment, the target feature and the multi-source weighted set precipitation data are mapped through three stacked fully connected layers with 64, 16 and 1 nodes respectively, so as to obtain a nonlinear relationship between the target feature and the multi-source weighted set precipitation data.

[0101] In one embodiment, a linear rectification function (ReLU) is set at the output ends of the three-dimensional convolutional layer, the convolutional long short-term memory network layer, and the stacked fully connected layer to improve the nonlinear fitting ability of the precipitation fusion model.

[0102] In step S105, the precipitation fusion model is trained according to the precipitation data fusion sample library to obtain precipitation fusion models corresponding to different future scenarios;

[0103] In this embodiment, the precipitation fusion model is trained respectively with sample data corresponding to each future scenario in the precipitation data fusion sample library. In one embodiment, the precipitation fusion model is trained with CMIP6 training precipitation data and auxiliary image training data in the sample data as input and the multi-source weighted set precipitation data as output.

[0104] In one embodiment, the precipitation fusion model is trained with a training set and a validation set in the sample data, and the performance of the precipitation fusion model is evaluated with a test set in the sample data to obtain a precipitation fusion model with good performance.

[0105] In one embodiment, during the training of the precipitation fusion model, the Xavier initialization method is used to initialize the network weights and biases, the root mean square error is used as the loss function, and the Adam optimizer is used to iteratively update the network parameters to control the training process to avoid overfitting.

[0106] The precipitation fusion model is trained respectively with sample data corresponding to each future scenario in the precipitation data fusion sample library to obtain precipitation fusion models corresponding to different future scenarios, which are used to generate precipitation fusion products corresponding to different future scenarios, thereby improving the accuracy of precipitation fusion product generation.

[0107] For step S106, the CMIP6 target precipitation data and auxiliary image data within the second preset time are obtained, and input into the precipitation fusion model corresponding to the future scenario to obtain the precipitation fusion product under the corresponding scenario.

[0108] The second preset time is the time when the precipitation fusion product needs to be obtained. In one embodiment, the second preset time is from 2015 to 2100.

[0109] See also Figure 6 , Figure 6 This is a flowchart of the steps of preprocessing the CMIP6 target precipitation data and auxiliary image data within the second preset time in one embodiment of the present application. In one embodiment, before the step of inputting into the precipitation fusion model corresponding to the future scenario, it also includes:

[0110] S601, unifying the spatial resolution of the CMIP6 target precipitation data and the auxiliary image data within the second preset time to a preset resolution;

[0111] S602, performing temporal and spatial attribute matching on the CMIP6 target precipitation data and the auxiliary image data within the second preset time;

[0112] S603, inputting the CMIP6 target precipitation data and the auxiliary image data after the temporal and spatial attribute matching into a precipitation fusion model corresponding to the future scenario.

[0113] For steps S601 to S603, the CMIP6 target precipitation data includes future scenarios and precipitation data corresponding to the future scenarios. The spatiotemporal attribute matching is to match the CMIP6 target precipitation data and auxiliary image data within the second preset time according to time information and spatial position information. The spatial resolution of the CMIP6 target precipitation data and auxiliary image data within the second preset time is unified to the preset resolution and spatiotemporal attribute matching is performed to obtain input data for the precipitation fusion model. According to the future scenario corresponding to the CMIP6 target precipitation data, the CMIP6 target precipitation data and auxiliary image data after the spatiotemporal attribute matching are input into the precipitation fusion model corresponding to the corresponding future scenario to obtain the precipitation fusion product corresponding to the future scenario.

[0114] The precipitation fusion product generation method described in the present application performs a system accuracy evaluation on the CMIP6 precipitation data within the first preset time, obtains at least 10 sets of CMIP6 precipitation data with small error variance, high correlation coefficient and high signal-to-noise ratio as the CMIP6 training precipitation data, obtains multi-source weighted ensemble precipitation data and auxiliary image training data within the first preset time, constructs corresponding sample data for different future scenarios, obtains a precipitation data fusion sample library, extracts spatiotemporal features from the CMIP6 training precipitation data and spatial features from the auxiliary image training data, fuses them to obtain target features, fits the nonlinear relationship between the target features and the multi-source weighted ensemble precipitation data, obtains a precipitation fusion model, finally trains the precipitation fusion model according to the precipitation data fusion sample library, obtains precipitation fusion models corresponding to different future scenarios, obtains CMIP6 target precipitation data and auxiliary image data within the second preset time for spatiotemporal attribute matching, inputs the CMIP6 target precipitation data and auxiliary image data after spatiotemporal attribute matching into the precipitation fusion model corresponding to the different future scenarios, and obtains precipitation fusion products under the corresponding scenarios. The precipitation fusion product generation method described in this application realizes data bias correction and improves the data accuracy of future multi-scenario precipitation.

[0115] See also Figure 7 , Figure 7 The following is a schematic diagram of a precipitation fusion product generation system in one embodiment of the present application. A precipitation fusion product generation system includes:

[0116] The sample library acquisition module 11 is used to acquire multi-source weighted set precipitation data, CMIP6 training precipitation data and auxiliary image training data within a first preset time, construct corresponding sample data for different future scenarios, and obtain a precipitation data fusion sample library, wherein the CMIP6 training precipitation data includes future scenarios and precipitation data corresponding to the future scenarios;

[0117] A spatiotemporal feature extraction module 12, used to extract spatiotemporal features from the CMIP6 training precipitation data;

[0118] Auxiliary feature extraction module 13, used for extracting spatial features from the auxiliary image training data;

[0119] The relationship simulation module 14 is used to fuse the spatiotemporal features and the spatial features, obtain target features, fit the nonlinear relationship between the target features and the multi-source weighted set precipitation data, and obtain a precipitation fusion model;

[0120] A model training module 15 is used to train the precipitation fusion model according to the precipitation data fusion sample library to obtain precipitation fusion models corresponding to different future scenarios;

[0121] The product acquisition module 16 is used to obtain the CMIP6 target precipitation data and auxiliary image data within the second preset time, input them into the precipitation fusion model corresponding to the future scenario, and obtain the precipitation fusion product under the corresponding scenario.

[0122] It should be noted that the precipitation fusion product generation system provided in the above embodiment only uses the division of the above functional modules as an example when executing the precipitation fusion product generation method. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The precipitation fusion product generation system provided in the above embodiment is used to execute the precipitation fusion product generation method described in the above embodiment. Its operation method and principle are the same as the precipitation fusion product generation method recorded above, that is, the precipitation fusion product generation system provided in the above embodiment and the precipitation fusion product generation method belong to the same concept. The implementation process is detailed in the above method embodiment, which will not be repeated here.

[0123] See also Figure 8 , Figure 8Schematic diagram of a computer device for generating a precipitation fusion product in an embodiment of the present application. The computer device 21 includes: a control device 211, a memory 212, and a computer program 213 stored in the memory 212 and executable on the control device 211, such as a precipitation fusion product generating program; the control device 211 executes the computer program 213 to implement the precipitation fusion product generating method described in the above embodiment.

[0124] Among them, the control device 211 includes a processor, and the processor may include one or more processing cores. The processor uses various interfaces and lines to connect various parts in the computer device 21, and executes various functions and processes data of the computer device 21 by running or executing instructions, programs, code sets or instruction sets stored in the memory 212, and calling the data in the memory 212. Optionally, the processor can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the touch display; the modem is used to process wireless communication. It can be understood that the above-mentioned modem may not be integrated into the processor, but implemented by a single chip.

[0125] Among them, the memory 212 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read-Only Memory). Optionally, the memory 212 includes a non-transitory computer-readable storage medium. The memory 212 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 212 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 instructions, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 212 may also be optionally at least one storage device located away from the aforementioned processor.

[0126] An embodiment of the present application also provides a computer-readable storage medium, which can store multiple instructions. The instructions are suitable for loading and executing the method steps of the above embodiment by a control device. The specific execution process can be found in the specific description of the above embodiment, which will not be repeated here.

[0127] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. A method for generating precipitation fusion products, characterized in that: The following steps are involved: Acquire multi-source weighted ensemble precipitation data, CMIP6 training precipitation data, and auxiliary image training data within a first preset time, construct corresponding sample data for different future scenarios, and obtain a precipitation data fusion sample library, wherein the CMIP6 training precipitation data includes future scenarios and precipitation data corresponding to the future scenarios; Extracting the spatiotemporal features in the CMIP6 training precipitation data, specifically: forming a key feature graph by spatially mapping and spatially reducing the CMIP6 training precipitation data; extracting abstract spatiotemporal features according to the key feature graph; restoring the spatial details of the abstract spatiotemporal features to obtain the spatiotemporal features; Extracting spatial features from the auxiliary image training data, specifically: splicing the terrain information image, the geographical location information image and each of the CMIP6 training precipitation data evaluation information images in the auxiliary image training data into a three-dimensional image block along the channel; extracting spatial features from the three-dimensional image block by stacking two-dimensional convolution layers; wherein the auxiliary image training data includes the terrain information image, the geographical location information image and each of the CMIP6 training precipitation data evaluation information images; The spatiotemporal features and the spatial features are integrated to obtain target features, and a nonlinear relationship between the target features and the multi-source weighted set precipitation data is fitted to obtain a precipitation fusion model; According to the precipitation data fusion sample library, the precipitation fusion model is trained to obtain precipitation fusion models corresponding to different future scenarios; The CMIP6 target precipitation data and auxiliary image data within the second preset time are obtained, and input into the precipitation fusion model corresponding to the future scenario to obtain the precipitation fusion product under the corresponding scenario.

2. The method for generating precipitation fusion products according to claim 1, characterized in that: The steps of obtaining multi-source weighted ensemble precipitation data, CMIP6 training precipitation data and auxiliary image training data within a first preset time, constructing corresponding sample data for different future scenarios, and obtaining a precipitation data fusion sample library include: For each future scenario, unifying the spatial resolutions of the multi-source weighted ensemble precipitation data, the CMIP6 training precipitation data, and the auxiliary image training data to a preset spatial resolution; Extracting a multi-source weighted set precipitation image block from each of the multi-source weighted set precipitation data in the order of first row and then column using a sliding window of preset specifications; Matching the multi-source weighted set precipitation image block with the corresponding CMIP6 training precipitation data of 5 consecutive days and the auxiliary image training data according to the spatiotemporal attributes to form an input-output sample pair as the sample data; Sample data of each future scenario is obtained to form the precipitation data fusion sample library.

3. The method for generating precipitation fusion products according to claim 2, characterized in that: Acquiring the CMIP6 training precipitation data within a first preset time includes the following steps: Obtaining some CMIP6 precipitation data within the first preset time; Conducting a system accuracy evaluation on each of the CMIP6 precipitation data, and obtaining evaluation information on each of the CMIP6 precipitation data; According to the evaluation information of each CMIP6 precipitation data, at least 10 sets of CMIP6 precipitation data from the plurality of CMIP6 precipitation data are selected as the CMIP6 training precipitation data.

4. The method for generating precipitation fusion products according to claim 3, characterized in that: The steps of performing a system accuracy evaluation on each of the CMIP6 precipitation data and obtaining evaluation information of each of the CMIP6 precipitation data include: Acquire multi-source weighted ensemble precipitation data and ERA5-Land precipitation data, and sample the multi-source weighted ensemble precipitation data, the plurality of CMIP6 precipitation data, and ERA5-Land precipitation data to the preset spatial resolution; Substitute the multi-source weighted ensemble precipitation data, ERA5-Land precipitation data, and a set of CMIP6 single future scenario precipitation data into the formula: Among them, P true is the true value of precipitation, P x , P y and P z are the multi-source weighted ensemble precipitation data, ERA5-Land precipitation data, and a set of CMIP6 precipitation data. x 、a y 、a z and b x , b y , b z are the intercept and slope of the corresponding linear equation, θ x ,θ y and θ z are the errors of the corresponding precipitation observation datasets; Arranging the formula, we get in, and P x , P y and P z The error variance of , <·> indicates the mean calculation, P ′ =- ; <P x ′2 >、 <P y ′2 >、 <P z ′2 >P x , P y , P z The variance of <P x ′ P y ′ >、 <P x ′ P z ′ >、 <P y ′ P z ′ >P x , P y , P z Covariance between pairs; Among them, CC x 、CC y and CC z P x , P y and P z The correlation coefficient between and the true value of precipitation; In the formula, SNR x 、SNR y and SNR z P x , P y and P z signal-to-noise ratio; The error variance, correlation coefficient and signal-to-noise ratio are used as evaluation information of each CMIP6 precipitation data.

5. The method for generating precipitation fusion products according to claim 1, characterized in that: Before the step of inputting into the precipitation fusion model corresponding to the future scenario, it also includes: Unifying the spatial resolution of the CMIP6 target precipitation data and the auxiliary image data within the second preset time to a preset resolution; Performing temporal and spatial attribute matching on the CMIP6 target precipitation data and the auxiliary image data within the second preset time; The CMIP6 target precipitation data and auxiliary image data after the temporal and spatial attribute matching are input into the precipitation fusion model corresponding to the future scenario.

6. A precipitation fusion product generation system, characterized in that: include: A sample library acquisition module is used to acquire multi-source weighted set precipitation data, CMIP6 training precipitation data and auxiliary image training data within a first preset time, construct corresponding sample data for different future scenarios, and obtain a precipitation data fusion sample library, wherein the CMIP6 training precipitation data includes future scenarios and precipitation data corresponding to the future scenarios; The spatiotemporal feature extraction module is used to extract the spatiotemporal features in the CMIP6 training precipitation data, specifically: forming a key feature graph by performing spatial mapping and spatial dimension reduction on the CMIP6 training precipitation data; extracting abstract spatiotemporal features according to the key feature graph; restoring the spatial details of the abstract spatiotemporal features to obtain the spatiotemporal features; The auxiliary feature extraction module is used to extract the spatial features in the auxiliary image training data, specifically: splicing the terrain information image, the geographical location information image and each of the CMIP6 training precipitation data evaluation information images in the auxiliary image training data into a three-dimensional image block along the channel; extracting the spatial features in the three-dimensional image block through stacked two-dimensional convolution layers; wherein the auxiliary image training data includes the terrain information image, the geographical location information image and each of the CMIP6 training precipitation data evaluation information images; A relationship simulation module, used for fusing the spatiotemporal features and the spatial features, obtaining target features, fitting the nonlinear relationship between the target features and the multi-source weighted set precipitation data, and obtaining a precipitation fusion model; A model training module, used to train the precipitation fusion model according to the precipitation data fusion sample library, and obtain precipitation fusion models corresponding to different future scenarios; The product acquisition module is used to obtain the CMIP6 target precipitation data and auxiliary image data within the second preset time, input them into the precipitation fusion model corresponding to the future scenario, and obtain the precipitation fusion product under the corresponding scenario.

7. A computer device comprising: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the precipitation fusion product generation method according to any one of claims 1 to 5 when executing the computer program.

8. A readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the precipitation fusion product generation method according to any one of claims 1 to 5 are implemented.