Global lake evaporation simulation method and system

Abstract

The present invention provides a global lake evaporation simulation method and system, and relates to the technical field of climate prediction. The method comprises: downloading global latent heat flux data, extracting the longitude and latitude of the grid point where the lake subgrid is located and the latent heat flux data; extracting the corresponding meteorological variable data according to the longitude and latitude of the grid point where the lake subgrid is located; for each lake grid point, constructing a meteorological variable feature matrix using the meteorological variable data; establishing a nonlinear spatial mapping relationship between the meteorological variable feature matrix and the latent heat flux data, and adding time information as a constraint condition therein, and constructing a lake evaporation simulation model; dividing the latent heat flux data and the meteorological variable data into a training set according to a predetermined ratio to complete the training of the lake evaporation simulation model, and completing the independent verification of the evaporation simulation model with a verification set; downloading multi-mode meteorological data, importing it into the lake evaporation simulation model after training and verification, and simulating and outputting a global lake evaporation data set.

Description

Technical Field

[0001] The present invention relates to the technical field of climate prediction, and in particular to a global lake evaporation simulation method and system. Background Art

[0002] Evaporation is the main way for lakes to lose water. 30% of the precipitation on the lake surface enters the atmosphere through evaporation. This proportion even exceeds 50% in temperate and arid climate zones.

[0003] The Earth System Model is the only tool for estimating future lake evaporation on a global scale. However, compared with other land cover types, lakes cover a smaller area and are usually "invisible" at the grid resolution (~1°×1°) of the Earth System Model. As a result, only CESM2 (Community Earth System Model 2) developed by the National Center for Atmospheric Research in the United States can directly simulate lake evaporation in the IPCC's Sixth International Coupled Model Intercomparison Project (CMIP6, Coupled Model Intercomparison Project Phase 6). However, the reliability of the prediction results of a single model is low. The multi-model ensemble average can not only give the most reliable prediction results for the future, but the standard deviation between multiple models can also reflect the uncertainty of the prediction results. Therefore, it is urgent to carry out multi-model ensemble simulation of global lake evaporation.

[0004] There are two main challenges in conducting multi-model ensemble simulations of global lake evaporation. First, CMIP6 brings together 112 Earth system models from 33 countries / institutions around the world. All simulation experiments were completed in 2017, and the simulation results have been shared globally. Therefore, except for CESM, it is impossible for the other 111 models to "rework" and re-conduct simulation experiments on coupled lake processes. Secondly, Earth system model simulation experiments require high-performance computers and consume huge computing resources. Improving operating efficiency and reducing computing costs are also one of the difficulties that need to be overcome.

[0005] Based on the above problems, organically combining the Earth system model with machine learning, solving the problem of "invisibility" of lakes in the Earth system model, and developing a global lake evaporation simulator with efficient operation and accurate results are key technologies for achieving multi-model prediction of global lake evaporation. Summary of the invention

[0006] Purpose of the invention: To propose a method for simulating global lake evaporation, and further propose a system for implementing the method, so as to solve the problem that lake grids are "invisible" in other earth system models except CESM, so as to improve the credibility of future global lake evaporation estimates.

[0007] In order to solve the above technical problems, the first aspect of the present invention provides a global lake evaporation simulation method, comprising the following steps:

[0008] Download the CESM2 global latent heat flux data, extract the longitude and latitude of the grid points where the lake subgrids are located and the latent heat flux data; extract the corresponding meteorological variable data according to the longitude and latitude of the grid points where the lake subgrids are located;

[0009] For each grid point of the lake sub-grid, the meteorological variable characteristic matrix required for lake evaporation simulation is constructed using the meteorological variable data;

[0010] A nonlinear spatial mapping relationship between the characteristic matrix of meteorological variables and latent heat flux data was established, and time information was added as a constraint to construct a lake evaporation simulation model.

[0011] The latent heat flux data and the meteorological variable data are divided into training sets according to predetermined proportions to complete the training of the lake evaporation simulation model, and the lake evaporation simulation model is verified by independent samples using the verification set;

[0012] Download CMIP6 multi-model meteorological data, import it into the trained and verified lake evaporation simulation model, and simulate and output the global lake evaporation dataset.

[0013] In a further embodiment of the first aspect, the meteorological variable characteristic matrix is ​​obtained by connecting the characteristic matrices of all months within a predetermined time span forward and backward.

[0014] The characteristic matrix MET of the mth month m The surface temperature T of the month a Specific humidity q a , air pressure p, precipitation Pre, 10 m wind speed U, incident shortwave radiation K ↓ and downward longwave radiation L ↓ The numerical subscripts represent the grid points of different lake subgrids, a total of N, as follows:

[0015]

[0016] In a further embodiment of the first aspect, a nonlinear spatial mapping relationship f is established between the meteorological variable characteristic matrix MET and the latent heat flux data λE, and time information is added as a constraint condition to construct a lake evaporation simulation model, which is expressed as follows:

[0017]

[0018] Where m represents the month, lat and lon represent the latitude and longitude of the grid point where the lake is located, respectively.

[0019] In a further embodiment of the first aspect, the CESM2 global latent heat flux data includes a predetermined number of members, 70% of the meteorological variable data and latent heat flux data are randomly selected from the grid points of the lake sub-grid of each member as training sets, and the training sets of all members are merged; the training set is used to complete the training of the lake evaporation simulation model; the remaining 30% of the data is used as a validation set, and independent sample verification is performed by calculating the root mean square error and correlation coefficient.

[0020] In a further embodiment of the first aspect, before importing the CMIP6 multi-model meteorological data into the lake evaporation simulation model, the method further comprises:

[0021] The CMIP6 multi-model meteorological data are resampled to the same spatial resolution as the CESM2 global latent heat flux data. The data of each CMIP6 model are constructed into a meteorological variable characteristic matrix. The data are selected according to longitude and latitude to drive the lake evaporation simulation model at the corresponding longitude and latitude.

[0022] In a further embodiment of the first aspect, the spatial resolution is 0.94° latitude × 1.25° longitude.

[0023] A second aspect of the present invention provides a global lake evaporation simulation system, the simulation system comprising:

[0024] The data extraction module is used to download the CESM2 global latent heat flux data, extract the longitude and latitude of the grid point where the lake subgrid is located and the latent heat flux data; extract the corresponding meteorological variable data according to the longitude and latitude of the grid point where the lake subgrid is located;

[0025] The first execution module is used to construct a meteorological variable characteristic matrix required for lake evaporation simulation using the meteorological variable data for each grid point where the lake sub-grid is located;

[0026] The data construction module is used to establish a nonlinear spatial mapping relationship between the characteristic matrix of meteorological variables and the latent heat flux data, and add time information as a constraint to obtain a data matrix for constructing a lake evaporation simulation model;

[0027] A model building module, used to divide the latent heat flux data and the meteorological variable data into training sets according to predetermined proportions to complete the construction of the lake evaporation simulation model;

[0028] A model verification module, used to divide the latent heat flux data and the meteorological variable data into verification sets according to predetermined proportions to complete independent sample verification of the lake evaporation simulation model;

[0029] The second execution module is used to download the CMIP6 multi-model meteorological data, import it into the trained and verified lake evaporation simulation model, and simulate and output the global lake evaporation dataset.

[0030] According to a third aspect of the present invention, an electronic device is provided, comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the global lake evaporation simulation method as described in the first aspect is implemented.

[0031] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein at least one executable instruction is stored in the storage medium. When the executable instruction is executed on an electronic device, the electronic device executes the global lake evaporation simulation method as described in the first aspect.

[0032] Compared with the prior art, the present invention has at least the following beneficial effects:

[0033] First, the global lake evaporation projections of multiple Earth system models in CMIP6 can be realized, making the projections more reliable and providing uncertainty ranges. Second, similar methods can be used to build simulators for other variables of the lake subgrid in CESM2 (such as lake surface temperature). In addition, for land cover types with smaller areas similar to lakes (such as cities), similar schemes can be used to build simulators to achieve multi-model ensemble projections. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a technical flow chart of the present invention.

[0035] Figure 2 Validation results of the lake evaporation simulator on typical lakes in five climate zones around the world. DETAILED DESCRIPTION

[0036] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features known in the art are not described.

[0037] In view of the problem that lakes are "invisible" in the Earth system model and multi-model ensemble prediction of lake evaporation cannot be achieved, this paper aims to provide a method for constructing a lake evaporation simulator using "Earth system model + machine learning". The overall technical idea is shown in Figure 1 , including three parts: (1) basic data preprocessing system; (2) construction and verification of lake evaporation simulator; (3) implementation of lake evaporation simulator. Through these three steps, multi-coupled model prediction of lake evaporation is realized to improve the reliability of global lake evaporation prediction.

[0038] First, download the CESM2 global latent heat flux data and related meteorological variable data. According to the sub-grid information given by CESM2, extract the longitude and latitude of the grid point where the lake sub-grid is located and the latent heat flux of the lake sub-grid. According to the longitude and latitude of the grid point where the lake sub-grid is located (hereinafter referred to as the lake grid point), similarly extract the meteorological data corresponding to the lake grid point. Then, for each lake grid point, use the extracted meteorological variable data to construct the characteristic matrix (composed of meteorological variables) required by the lake evaporation simulator. Taking a certain month as an example, the characteristic matrix MET1 is defined as:

[0039] (1)

[0040] In the formula, the characteristic matrix MET1 is composed of the surface temperature T a Specific humidity q a , surface pressure p, precipitation Pre, 10 m wind speed U, incident shortwave radiation K ↓ and downward longwave radiation L ↓ The numerical subscripts represent different lake grids, with a total of 4690 in the world. The characteristic matrices of all months (86 years × 12 months = 1032 months) are connected forward and backward to obtain the required characteristic matrix MET for the entire period (2015-2100).

[0041] Secondly, taking a lake grid of a member of CESM2 as an example, 70% of the meteorological and latent heat flux data of the lake were randomly selected as the training set. The training sets of the other 79 members were constructed in a similar way. Finally, the training sets of 80 members were merged, and the nonlinear spatial mapping relationship f between lake evaporation and meteorological variable matrix was established using the automatic machine learning framework FLAML, and time information was added as a constraint condition for modeling, that is:

[0042] (2)

[0043] In the formula, MET represents the characteristic matrix composed of meteorological variables, λE is the lake latent heat flux, m represents the month, and lat and lon represent the latitude and longitude of the grid point where the lake is located. For each lake, the most suitable machine learning algorithm may be different, that is, the spatial mapping relationship f may be different, which is a function of longitude and latitude. The remaining 30% of the data was used as a validation set, and independent sample validation was performed by calculating statistical indicators such as the root mean square error and correlation coefficient.

[0044] Finally, we downloaded the CMIP6 multi-model meteorological data with the same climate scenario as CESM2 LENS2 and resampled it to the spatial resolution of CESM2 (0.94° latitude × 1.25° longitude). We constructed the data of each CMIP6 model into a feature matrix consistent with the previous CESM2, selected data according to longitude and latitude, and drove the lake evaporation simulator of the corresponding longitude and latitude to generate a global lake evaporation dataset simulated by multiple models.

[0045] The embodiment of the present invention is based on an automatic machine learning algorithm to construct a lake evaporation simulator. By using conventional meteorological data and latent heat flux in CESM2, a nonlinear relationship between meteorological data and latent heat flux is constructed. In this process, because the amount of data is large, it is necessary to use a mainframe to operate. Taking the computing performance of the supercomputing center of Nanjing University of Information Science and Technology as an example, CPU: 2×Intel(R) Xeon(R) CPU E5-2680 v4 28C 2.40GHz, 20 CPU cores of a node are applied for during the construction process, and a total of 82 hours are required. Using CMIP6 model data to generate latent heat flux data in different modes, this process can be achieved with only a single machine. Taking CPU 12th Gen Intel(R) Core(TM) i7-12700F as an example, it only takes about 50 minutes to drive the simulator to generate global lake evaporation data for the entire period from 2015 to 2100 using one mode.

[0046] The specific implementation is as follows:

[0047] 1. Data preprocessing and integration framework construction of the global lake evaporation simulator

[0048] The first step is to download the monthly latent heat flux (λE, Wm2) of 80 members of the CESM2 LargeEnsemble (CESM2 LENS) under the SSP370 scenario (2015-2100) from the Earth System Grid of the National Center for Atmospheric Research. -2 ) and related meteorological variables that drive the model to generate latent heat flux, including downward longwave radiation (L ↓ , W m -2 ), downward shortwave radiation (K ↓ , W m -2 )、Precipitation(Pre,ms -1 ), air pressure (p, Pa), specific humidity (q a , kg / kg), air temperature (T a , K), 10 m wind speed (U, ms -1 ). The members and time scales of the meteorological data downloaded are consistent with the latent heat flux.

[0049] In the second step, in the data downloaded in the first step, the data of each member is saved at intervals of 10 years (i.e., 2015~2024, 2025~2024, ..., 2095~2100), so each member has 10 files. Taking the first file of a member as an example, read the NetCDF (NC) file of the CESM2 latent heat flux, read the subgrid where the lake is located (landtype = 5) according to the subgrid information given by CESM2, and obtain 4690 lake grid points. Extract the longitude and latitude of all lake grid points and the corresponding latent heat flux data, and construct the data into a matrix containing longitude and latitude and specific time as column names. Under each column of time is the latent heat flux data corresponding to the longitude and latitude, which results in a 4690×122 (longitude and latitude + 12 months × 10 years) matrix, save the data as a csv file, and repeat this process to extract the data of all time periods in a member and save them as a csv file. According to this method, the latent heat flux data of the 80 members of the lake are saved as the same csv file to obtain the latent heat flux data of the 80 members of the CESM2 global lake grid under future scenarios.

[0050] In the third step, the downloaded meteorological data is saved in the same way as the latent heat flux data, and is also saved at intervals of 10 years. Taking the downward shortwave radiation data of a member as an example, read an NC file of CESM2 meteorological data, and according to the longitude and latitude information of the lake grid extracted in the second step, the meteorological data grid closest to the lake grid is used as the data on the lake because the longitude and latitude information has a slight deviation. Specifically, find the data that best matches the longitude and latitude of the lake grid in the NC file of downward shortwave radiation, and extract the downward shortwave radiation value of the grid. Then, organize these data in the csv format saved in the second step. The first two columns are the longitude and latitude read, and each column starting from the third column is the data of the corresponding specific time and lake grid, and also obtain a 4690×122 (longitude and latitude + 12 months × 10 years) matrix. For each member and each time period, a csv file is saved separately, so that csv files of 80 different members can be obtained, recording the downward shortwave radiation data at each time point on the global lake grid. Next, use the same method to extract data for other meteorological variables, and finally save all meteorological data and members as csv files.

[0051] The fourth step is to merge the csv files of different time periods according to longitude and latitude based on the data saved in the second and third steps, taking a member of the latent heat flux data as an example, to obtain a matrix of 4690 rows × 1034 (longitude and latitude + 12 months × 86 years) columns, that is, the monthly scale data of a member from 2015 to 2100. This matrix is ​​converted into a data matrix with longitude and latitude, time, and λE as column names, that is, a matrix of 4840080 rows (12 months × 86 years × 4690 lake grids) × 4 columns is obtained. In this way, each meteorological variable is also converted into the same data matrix, with column names of longitude and latitude, time, and the abbreviation of the corresponding meteorological variable (such as K ↓ , L ↓ The latent heat flux data and all meteorological variable data of the same member are combined according to longitude, latitude and time, and the latent heat flux data and meteorological variable data are combined to obtain a 4840080 row × 11 column (longitude and latitude + time + latent heat flux + 7 meteorological variables). The data matrix of latent heat flux and meteorological variables of 80 members is obtained by the same method.

[0052] Step 5: Read the data obtained in step 4, and divide the training set and validation set into groups according to longitude and latitude from the data of one member. That is, separate each lake grid point, extract 70% of the data as the training set, and 30% of the data as the validation set, and obtain a training set and validation set for one member of one lake grid point. Read each lake grid point in turn, and divide the training set and validation set in the same way. According to this method, divide the training set and validation set of 80 members respectively, and finally merge the training set and validation set of each member respectively, and add a column indicating the source of the data (members: 001, 002, ..., 080).

[0053] 2. Construction and model verification of lake evaporation simulator

[0054] The sixth step is to store the longitude and latitude of the lake as a dat file to facilitate the subsequent loop of the lake grid. Specifically, the lake grids with the same longitude are stored as lat_ls.dat files, and the latitude and longitude of the lake grids with the same longitude are saved together as a lat_lon_dict.dat file. This is a storage method similar to that of a Python dictionary, which facilitates the subsequent loop of the simulator construction for each lake grid.

[0055] In the seventh step, add a column representing the month to the training set and validation set according to the time column, convert the original date data into a format that the machine learning model can understand, and convert each month into an independent numerical feature through one-hot encoding so that the information of each month can be used for subsequent analysis or modeling.

[0056] In the eighth step, according to the training set obtained in the seventh step and the file obtained in the sixth step, the automatic machine learning (AutoML) framework FLAML (including 5 algorithms. LGBM (lightweight gradient boosting), RF (random forest), XGBoost (extreme gradient boosting), Extra_tree (extreme random tree), XGB_limitdepth (extreme gradient boosting with limited decision tree depth)) is used to train a single lake grid. The parameters of automatic machine learning are set as: "time_budget": 50, "metric": 'r2', "task": 'regression', "seed": 2023 to build a nonlinear relationship model between meteorological elements and latent heat flux. Each lake grid will automatically select the most suitable hyperparameters and machine learning algorithms, and save the training results of each lake grid as an independent .dat file to obtain a lake evaporation simulator for each lake grid, a total of 4690. Since the final training set is 30G, a mainframe is needed in this process. Taking the computing performance of the Nanjing University of Information Science and Technology Supercomputing Center as an example, CPU: 2 * Intel(R) Xeon(R) CPU E5-2680 v4 28C 2.40GHz; memory: 128G; computing network: 56Gb infiniband. When building the lake evaporation simulator, 20 CPU cores of a node were applied, and it took a total of 82 hours.

[0057] In the ninth step, the validation set data obtained in the fifth step that was not involved in the training was used to input the meteorological element data of each lake grid point into the corresponding lake model generated in the eighth step, and the latent heat flux data was generated based on the meteorological data. This process used the same computing queue, applied for 1 CPU core of a node, and took a total of 24 hours.

[0058] In the tenth step, the training results obtained in the ninth step were compared with the latent heat flux data originally generated by CESM2 in the training set, and the spatial distribution of the root mean square error (RMSE) and Pearson correlation coefficient of each lake grid point were calculated to evaluate the effect of the simulator.

[0059] In the eleventh step, the lake grids were divided into different climate zones according to the Köppen-Geiger climate classification system, and a typical lake was selected from each climate zone (Taihu Lake in the temperate zone, Lake Tanganyika in the tropical zone, Lake Chad in the arid zone, Lake Superior in the cold zone, and Namtso Lake in the polar zone) to verify the simulator. The verification results are shown in Figure 2 . Figure 2Among them, (a) shows the simulation effect of Taihu Lake in the temperate zone; (b) shows the simulation effect of Lake Tanganyika in the tropical zone; (c) shows the simulation effect of Lake Chad in the arid zone; (d) shows the simulation effect of Lake Superior in the cold zone; (e) shows the simulation effect of Namtso Lake in the polar region.

[0060] 3. Application of Lake Evaporation Simulator

[0061] Step 12: Download the CMIP6 model data from the CMIP6 dataset. Select Experiment ID: SSP370, Variant Label: ri1p1f1, Frequency: mon, Realm: atmos to download the model data with the same time length as the CESM2 LENS future scenario (2015-2100) and with specific humidity (q a , kg / kg), downward long-wave radiation (L ↓ , W m -2 ), downward shortwave radiation (K ↓ , W m -2 )、10 m wind speed(U,ms -1 ), air pressure (p, Pa), precipitation (Pre, kg m -2 s -1 ) and temperature (T a , K) variables of the CMIP6 model. Used to drive the simulator to generate multi-model lake evaporation data.

[0062] Step 13: Use the xesmf library in Python to resample the spatial resolution of all CMIP6 meteorological data to the CESM2 LENS resolution of 0.94° latitude × 1.25° longitude, and extract the data on the lake grid points according to the longitude and latitude of the lake grid points in the latent heat flux data extracted in the second step. In this process, the unit of precipitation needs to be converted. The unit of precipitation in CMIP6 is kg m -2 s -1 , while in CESM2 it is ms -1 The data of all meteorological variables were extracted and saved as csv files, and the names of meteorological variables used in the column names were changed to the names of meteorological variables in CEMS2, that is, K ↓ Radiation represents downward short wave, L ↓ represents downward long-wave radiation, etc. Finally, we get the longitude and latitude, time, meteorological variables (K ↓ , L ↓ , Pre, etc.) with a data matrix of 4,840,080 (4,690 × 12 months × 86 years) × 10 columns (latitude and longitude + time + 7 meteorological variables), and the data of each mode were saved as a csv file.

[0063] Step 14: This step can be implemented on a single machine. Taking the CPU 12th Gen Intel(R) Core(TM) i7-12700F as an example, a single mode only takes about 50 minutes through the data-driven simulator. According to the meteorological matrix data of each mode obtained in step 13, add the information indicating the month according to the processing method of step 7, and then input the meteorological data into the corresponding lake evaporation simulator according to the longitude and latitude to generate the latent heat flux data under the lake grid, and obtain the multi-mode lake evaporation data.

[0064] In addition, the present embodiment also discloses a global lake evaporation simulation system, which includes a data extraction module, a first execution module, a data construction module, a model construction module, a model verification module, and a second execution module. The data extraction module is used to download the CESM2 global latent heat flux data, and extract the longitude and latitude of the grid point where the lake sub-grid is located and the latent heat flux data; according to the longitude and latitude of the grid point where the lake sub-grid is located, the corresponding meteorological variable data is extracted. The first execution module is used to construct the meteorological variable characteristic matrix required for lake evaporation simulation for each lake grid point using the meteorological variable data. The data construction module is used to establish a nonlinear spatial mapping relationship between the meteorological variable characteristic matrix and the latent heat flux data, and add time information as a constraint condition to obtain a data matrix for constructing a lake evaporation simulation model. The model construction module is used to divide the training set from the latent heat flux data and the meteorological variable data according to a predetermined ratio to complete the construction of the lake evaporation simulation model. The model verification module is used to complete the independent sample verification of the lake evaporation simulation model for the data sets that are not involved in the training in the latent heat flux data and the meteorological variable data. The second execution module is used to download the CMIP6 multi-model meteorological data, import it into the trained lake evaporation simulation model, and simulate and output the global lake evaporation data set. The global lake evaporation simulation system can automatically execute the entire technical process of the global lake evaporation simulation disclosed in the above embodiment, which will not be repeated here.

[0065] Furthermore, the technical process of global lake evaporation simulation disclosed in the above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs.

[0066] When the computer instruction or computer program is loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instruction may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instruction may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0067] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0068] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for simulating global lake evaporation, characterized in that: The steps include: Download the CESM2 global latent heat flux data, extract the longitude and latitude of the grid points where the lake subgrids are located and the latent heat flux data; extract the corresponding meteorological variable data according to the longitude and latitude of the grid points where the lake subgrids are located; For each grid point of the lake sub-grid, the meteorological variable characteristic matrix required for lake evaporation simulation is constructed using the meteorological variable data; The meteorological variable characteristic matrix is ​​obtained by connecting the characteristic matrices of all months in a predetermined time span forward and backward; The characteristic matrix MET of the mth month m The surface temperature T of the month a Specific humidity q a , air pressure p, precipitation Pre, 10 m wind speed U, incident shortwave radiation K ↓ and downward longwave radiation L ↓ Composition, expressed as follows: ； In the formula, the numerical subscripts represent the grid points where different lake subgrids are located, with a total of N; A nonlinear spatial mapping relationship between the meteorological variable characteristic matrix and the latent heat flux data is established, and time information is added as a constraint condition to construct a lake evaporation simulation model; Dividing the latent heat flux data and the meteorological variable data into a training set according to a predetermined ratio to complete the training of the lake evaporation simulation model, and a verification set to complete the independent sample verification of the lake evaporation simulation model; Download the CMIP6 multi-model meteorological data, import it into the trained and verified lake evaporation simulation model, and simulate and output the global lake evaporation dataset.

2. The global lake evaporation simulation method according to claim 1, characterized in that: The nonlinear spatial mapping relationship f between the meteorological variable characteristic matrix MET and the latent heat flux data λE is established, and the time information is added as a constraint condition to construct the lake evaporation simulation model, which is expressed as follows: ； Where m represents the month, lat and lon represent the latitude and longitude of the grid point where the lake is located, respectively.

3. The global lake evaporation simulation method according to claim 2, characterized in that: The CESM2 global latent heat flux data contains a predetermined number of members. 70% of the meteorological variable data and latent heat flux data are randomly selected from the grid points of the lake sub-grid of each member as training sets, and the training sets of all members are merged. The training set is used to complete the training of the lake evaporation simulation model. The remaining 30% of the data is used as a validation set, and independent sample verification is performed by calculating the root mean square error and correlation coefficient.

4. The global lake evaporation simulation method according to claim 1, characterized in that: Before importing CMIP6 multi-model meteorological data into the lake evaporation simulation model, it also includes: The CMIP6 multi-model meteorological data are resampled to the same spatial resolution as the CESM2 global latent heat flux data. The data of each CMIP6 model are constructed into a meteorological variable characteristic matrix. The data are selected according to longitude and latitude to drive the lake evaporation simulation model at the corresponding longitude and latitude.

5. The global lake evaporation simulation method according to claim 4, characterized in that: The spatial resolution is 0.94° latitude × 1.25° longitude.

6. A global lake evaporation simulation system, characterized in that: include: The data extraction module is used to download the CESM2 global latent heat flux data, extract the longitude and latitude of the grid point where the lake subgrid is located and the latent heat flux data; extract the corresponding meteorological variable data according to the longitude and latitude of the grid point where the lake subgrid is located; The first execution module is used to construct a meteorological variable characteristic matrix required for lake evaporation simulation using the meteorological variable data for each grid point where the lake sub-grid is located; The meteorological variable characteristic matrix is ​​obtained by connecting the characteristic matrices of all months in a predetermined time span forward and backward; The characteristic matrix MET of the mth month m The surface temperature T of the month a Specific humidity q a , air pressure p, precipitation Pre, 10 m wind speed U, incident shortwave radiation K ↓ and downward longwave radiation L ↓ Composition, expressed as follows: ； In the formula, the numerical subscripts represent the grid points where different lake subgrids are located, with a total of N; The data construction module is used to establish the nonlinear spatial mapping relationship between the meteorological variable characteristic matrix and the latent heat flux data, and add time information as a constraint condition; A model building module, used to divide the latent heat flux data and the meteorological variable data into training sets according to predetermined proportions to complete the construction of the lake evaporation simulation model; A model verification module, used to divide the latent heat flux data and the meteorological variable data into verification sets according to predetermined proportions to complete independent sample verification of the lake evaporation simulation model; The second execution module is used to download the CMIP6 multi-model meteorological data, import it into the trained and verified lake evaporation simulation model, and simulate and output the global lake evaporation dataset.

7. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the global lake evaporation simulation method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The storage medium stores at least one executable instruction, and when the executable instruction is executed on an electronic device, the electronic device executes the global lake evaporation simulation method according to any one of claims 1 to 5.

Images