A method, device and storage medium for interdecadal prediction of regional precipitation

Abstract

The present invention discloses a method, device and storage medium for interdecadal prediction of regional precipitation, belonging to the technical field of precipitation prediction. The method includes: dividing a prediction region into multiple sub-regions according to the main precipitation mode, and generating a regional precipitation index of the observed data; calculating the interdecadal variability based on the obtained dynamic model hindcast data, and generating a regional precipitation index of the dynamic model hindcast data; extracting the key regions of environmental variables and screening the source factors of predictability; using the screened source factors of predictability and the regional precipitation index of the dynamic model hindcast data to train and optimize a gradient boosting regression tree prediction model to obtain an optimized gradient boosting regression tree prediction model; based on the optimized gradient boosting regression tree prediction model, predicting the precipitation index of each sub-region in a future set year, and generating an interdecadal prediction result of regional precipitation; the present invention predicts the precipitation situation in a future set year in advance, saves computing resources, and improves the prediction skill of precipitation at the same time.

Description

Technical Field

[0001] The present invention relates to a method, device and storage medium for interdecadal prediction of regional precipitation, and belongs to the technical field of precipitation prediction. Background Art

[0002] The eastern region of China is significantly affected by the East Asian summer monsoon, and summer precipitation accounts for 49% of the annual total precipitation. At the same time, the summer precipitation in eastern China also shows obvious interdecadal variation characteristics. Effective interdecadal prediction is of great significance for water resource management, energy supply adjustment, etc. At present, the dynamic model is the main technical means for prediction, but affected by the initial value conditions and boundary value conditions, the prediction skills for summer precipitation in eastern China are generally very low, far from meeting the actual needs. Currently, the precipitation prediction models based on statistical or machine learning methods are basically for the weather to seasonal scale, and there are few models for interdecadal prediction. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies in the prior art, and provide a method, device and storage medium for interdecadal prediction of regional precipitation, which can predict the precipitation situation in the future set year in advance, save computing resources, and at the same time improve the prediction skills of precipitation, and help relevant departments make preparations for disaster prevention, mitigation, water resource regulation, etc. in advance.

[0004] To achieve the above purpose, the present invention is implemented by the following technical solutions:

[0005] In the first aspect, the present invention provides a method for interdecadal prediction of regional precipitation, including:

[0006] Calculating the precipitation main mode based on the observed data of the monthly average grid points in the prediction area, dividing the prediction area into multiple sub-areas according to the precipitation main mode, and generating the regional precipitation index of the observed data;

[0007] Calculating the interdecadal variability based on the obtained dynamic model hindcast data, and generating the regional precipitation index of the dynamic model hindcast data based on the divided sub-areas;

[0008] Respectively extracting the key areas of environmental variables from the regional precipitation index of the observed data and the regional precipitation index of the dynamic model hindcast data, and screening the predictability source factors according to the extracted key areas of environmental variables;

[0009] Using the screened predictability source factors and the regional precipitation index of the dynamic model hindcast data to train and optimize the gradient boosting regression tree prediction model, and obtaining the optimized gradient boosting regression tree prediction model;

[0010] Based on the optimized gradient boosting regression tree prediction model, predicting the precipitation index of each sub-area in the future set year, and generating the interdecadal prediction result of regional precipitation.

[0011] Further, calculating the precipitation principal mode based on the observed data of the monthly average grid points in the prediction area, dividing the prediction area into multiple sub-areas according to the precipitation principal mode, and generating the regional precipitation index of the observed data, including:

[0012] Obtaining the observed data of the monthly average grid points in the prediction area;

[0013] Calculating the average precipitation of each grid point in any season based on the observed data, and performing a five-year moving average process to generate the interdecadal variability;

[0014] Performing anomaly calculation on the interdecadal variability, and extracting the precipitation principal mode through empirical orthogonal decomposition;

[0015] According to the zonal tripole distribution characteristics of the precipitation principal mode, dividing the prediction area into three sub-areas, and respectively calculating the regional average precipitation index of each sub-area.

[0016] Further, calculating the interdecadal variability based on the obtained dynamic model reforecast data, and generating the regional precipitation index of the dynamic model reforecast data based on the divided sub-areas, including:

[0017] Obtaining the dynamic model reforecast data of the prediction area;

[0018] Based on the dynamic model reforecast data, extracting the interdecadal variability of the annual average value in the future set years and splicing them into a sequence;

[0019] Generating the regional precipitation index of the dynamic model reforecast data according to the sequence spliced by the interdecadal variability.

[0020] Further, respectively extracting the key areas of environmental variables from the regional precipitation index of the observed data and the regional precipitation index of the dynamic model reforecast data, and screening the source factors of predictability according to the extracted key areas of environmental variables, including:

[0021] Based on the regional precipitation index of the observed data and the environmental variables of the pre-generated reanalysis data, determining the key areas of environmental variables in the observation through correlation analysis;

[0022] Using the regional precipitation index of the dynamic model reforecast data to extract the key areas of environmental variables in the dynamic model;

[0023] Verifying the prediction ability of the key areas of environmental variables in the dynamic model and the consistency with the observed physical mechanism;

[0024] Screening the environmental variables that meet the consistency of prediction ability and physical mechanism as the source factors of predictability.

[0025] Further, the prediction ability of the key area of the verification environmental variable in the dynamic mode and its consistency with the observed physical mechanism include:

[0026] Verify that the key area of the environmental variable in the dynamic mode is consistent with the key area of the environmental variable in the observation, the confidence level in the correlation distribution map of the dynamic mode return data and the reanalysis data exceeds the set threshold, and the dynamic mode can reproduce the physical process of the environmental variable affecting precipitation in the observation.

[0027] Further, when training and optimizing the gradient boosting regression tree prediction model using the screened predictability source factors and the regional precipitation index of the dynamic mode return data, the hyperparameters are optimized by the five-fold cross-validation method.

[0028] Further, after predicting the regional precipitation index of each sub-region in a future set year based on the optimized gradient boosting regression tree prediction model, a decadal prediction result of the regional precipitation is generated by combining a linear regression model and a linear trend correction.

[0029] In a second aspect, the present invention provides a device for decadal prediction of regional precipitation, including:

[0030] A first generation module for calculating the main precipitation mode based on the observed data of the monthly average grid points in the prediction area, dividing the prediction area into multiple sub-regions according to the main precipitation mode, and generating a regional precipitation index of the observed data;

[0031] A second generation module for calculating the decadal variability based on the obtained dynamic mode return data, and generating a regional precipitation index of the dynamic mode return data based on the divided sub-regions;

[0032] A screening module for respectively extracting the key areas of the environmental variables from the regional precipitation index of the observed data and the regional precipitation index of the dynamic mode return data, and screening the predictability source factors according to the extracted key areas of the environmental variables;

[0033] A training and optimization module for training and optimizing a gradient boosting regression tree prediction model using the screened predictability source factors and the regional precipitation index of the dynamic mode return data to obtain an optimized gradient boosting regression tree prediction model;

[0034] A prediction module for predicting the regional precipitation index of each sub-region in a future set year based on the optimized gradient boosting regression tree prediction model, and generating a decadal prediction result of the regional precipitation.

[0035] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in any one of the foregoing are implemented.

[0036] Fourth aspect, the present invention provides a computer device, comprising:

[0037] a memory for storing computer programs / instructions;

[0038] a processor for executing the computer programs / instructions to implement the steps of the method described in any one of the foregoing.

[0039] Fifth aspect, the present invention provides a computer program product comprising computer programs / instructions which, when executed by a processor, implement the steps of the method described in any one of the foregoing.

[0040] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0041] The present invention provides a method, device and storage medium for interdecadal prediction of regional precipitation. Based on an optimized gradient boosting regression tree prediction model, precipitation indices of each sub-region in a future set year are predicted to generate an interdecadal prediction result of regional precipitation, so as to predict the precipitation situation in a future set year in advance, saving computing resources and at the same time improving the prediction skill of precipitation, which helps relevant departments to make preparations for disaster prevention, mitigation, water resource regulation, etc. in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of a method for interdecadal prediction of regional precipitation provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be used to limit the protection scope of the present invention.

[0044] Embodiment 1. This embodiment introduces a method for interdecadal prediction of regional precipitation, comprising:

[0045] Calculating the precipitation principal mode based on the observed data of the monthly average grid points in the prediction area, dividing the prediction area into multiple sub-regions according to the precipitation principal mode, and generating the regional precipitation index of the observed data;

[0046] Calculating the interdecadal variability based on the obtained dynamic model hindcast data, and generating the regional precipitation index of the dynamic model hindcast data based on the divided sub-regions;

[0047] Extracting the key regions of environmental variables from the regional precipitation indices of the observed data and the regional precipitation indices of the dynamic model hindcast data respectively, and screening the source factors of predictability according to the extracted key regions of environmental variables;

[0048] Using the regional precipitation index of the selected predictability source factors and dynamic model return data, train and optimize the gradient boosting regression tree prediction model to obtain the optimized gradient boosting regression tree prediction model;

[0049] Based on the optimized gradient boosting regression tree prediction model, predict the precipitation index of each sub-region in the future set year to generate the interdecadal prediction results of regional precipitation.

[0050] The interdecadal prediction method of regional precipitation provided in this embodiment specifically involves the following steps in its application process:

[0051] The first step: Calculate the interdecadal variability of summer precipitation in East China: 1) Select the monthly average CN05.1 grid observation data with a horizontal resolution of 0.25°×0.25°, and calculate the summer average (June - August) of each grid point; 2) Calculate the five-year moving average of the summer average of each grid point to obtain the interdecadal variability, marked as the year in the middle of the five years. For example, the average value of 1961 - 1965 is marked as the interdecadal variability of 1963, and subtract the climatology (the average of the summer average from 1981 - 2010) from the value of that year to obtain the anomaly; 3) Conduct an empirical orthogonal function (EOF) analysis on the precipitation anomalies during the three years from 1963 to three years before the year of the advance prediction (at least after 2010) to obtain the main mode of the precipitation anomalies. For example, if the year of the advance prediction is 2013, but only the data of 2012 can be obtained, and the interdecadal variability calculated by the five-year moving average can only reach 2010, so it is three years before the year of the advance prediction; the spatial distribution of the main mode shows a zonal tripole pattern (the spatial distribution varies little with the change of years). Therefore, the East China region is divided into three sub-regions, namely South China, the Yangtze - Huaihe River Basin, and North China, and the regional average values of each sub-region are calculated as the precipitation index.

[0052] The second step: Process the dynamic model dataset and the reanalysis dataset: 1) Adopt the return data of 9 dynamic models in DCPPA of CMIP6. This data starts forecasting every year, and the average value of the future 1 - 5 years predicted each year is used as the interdecadal variability, marked as the year in the middle of the five years, and splice and combine the interdecadal variabilities predicted each year into a sequence. The variables include: precipitation (PRE), sea surface temperature (SST), sea ice (SIC), geopotential height (HGT), and zonal and meridional wind fields (UV); 2) Calculate the precipitation indices of the three sub-regions predicted by the dynamic model; 3) Select the ERA5 monthly reanalysis data with a horizontal resolution of 1°×1°, and obtain the interdecadal variability through a five-year moving average, marked as the year in the middle of the five years. The variables include: sea surface temperature (SST), sea ice (SIC), geopotential height (HGT), and zonal and meridional wind fields (UV).

[0053] Step 3: Identify the sources of predictability for precipitation and clarify the physical mechanisms: 1) Use bilinear interpolation to interpolate all dynamic model data, observational, and reanalysis data into 1°×1°, which is convenient for evaluation and comparison, and remove the linear trend; 2) Identify the key sea surface temperature and sea ice regions in the same period, previous winter, and spring that are related to the precipitation index in each sub-region in the observations (regions with a confidence level exceeding 90% in the correlation distribution map of observed precipitation and reanalysis sea surface temperature or sea ice data), and use ERA5 reanalysis data to explore the physical processes affecting precipitation; 3) Identify the key sea surface temperature and sea ice regions in the same period, previous winter, and spring in each region related to the precipitation index in each sub-region in the dynamic model data (regions with a confidence level exceeding 90% in the correlation distribution map of the ensemble mean of dynamic model precipitation and the ensemble mean of dynamic model sea surface temperature or sea ice); 4) For the key sea surface temperature and sea ice identified in the previous step, evaluate the decadal prediction ability of the ensemble mean of all dynamic models (the average of all initial fields of all models), and use the temporal correlation coefficient (CC) as an indicator; 5) If the key sea surface temperature or sea ice region is consistent with the observations, the dynamic model has prediction skills for the sea surface temperature or sea ice in this region (the confidence level in the correlation distribution map of dynamic model data and reanalysis data exceeds 90%), and the dynamic model can reproduce the physical process of the factor affecting precipitation in the observations, then these key variables are considered as sources of predictability, and the regional average is calculated as the factor index; 6) Repeat the process of (2)-(5) for the three sub-regions to select all sources of predictability for each sub-region.

[0054] Step 4: Construct a machine learning model for decadal prediction of summer precipitation in eastern China for independent hindcasting: 1) For each sub-region, use the sequence of factor indices of the sources of predictability predicted by the selected models (the sequence averaged over 9 models, with 9 sequences for one factor) and the sequence of the regional average precipitation in the corresponding sub-region predicted by the models (the sequence averaged over 9 models, a total of 9 sequences), and combine them with the gradient boosting regression tree (GBRT) model to construct a prediction model. Select the three years from 1963 to the year before the prediction year as the training time period, use five-fold cross-validation to determine the optimal hyperparameters, and predict the sub-region average index of summer precipitation in the next five years to obtain the predicted precipitation indices for the 3 sub-regions. For example, in 2013, use the decadal data from 1963 to 2010 for training to predict the decadal variability from 2011 to 2015; 2) For each sub-region, use the relationship between each grid point in the sub-region and the regional average index during the three years from 1963 to the year before the prediction year to construct a linear regression model, combine it with the predicted precipitation index for the next five years in this region, obtain the predicted values for each grid point in the eastern China region, and then add the corresponding linear trend for each grid point to obtain the final predicted spatial distribution of summer precipitation in eastern China for the next five years.

[0055] Such as Figure 1The figure shows the prediction process of the entire model. First, calculate the interdecadal variability of the observed precipitation. Then, use the leading modes obtained from EOF analysis for zoning, remove the linear trend, and separately find the key sea surface temperature and sea ice factors in each region, and clarify the corresponding physical mechanism process. Also, calculate the precipitation index for each region using the data predicted by the dynamic model, find the key sea surface temperature and sea ice factors in the dynamic model data after removing the linear trend. If the factor simultaneously meets the three conditions: being consistent with the key factor region in the observation, being able to reproduce the process of the factor affecting precipitation in the observation in the dynamic model, and the dynamic model having prediction skills for this factor, then it is determined that this factor is the source factor of the predictability of precipitation in this sub-region. Combine the precipitation index predicted by the model, and use the GBRT model to predict the precipitation index. Finally, use the relationship between each point in the sub-region and the corresponding precipitation index to construct a linear model to predict the value of each point, and add the corresponding linear trend of every other point to obtain the final predicted spatial distribution of summer precipitation in eastern China.

[0056] The following combines a preferred embodiment to illustrate the content involved in the above embodiments.

[0057] Taking the prediction of the interdecadal variability of summer precipitation in eastern China from 2015 (average from 2013 to 2017) to 2019 (average from 2017 to 2021) in 2017 as an example to illustrate the specific implementation plan.

[0058] Step 1: Select the monthly average CN05.1 grid observation data from 1961 to 2016, calculate the summer average (June - August) of each grid point, and perform a five-year moving average to obtain the interdecadal variability, marked as the middle year of the five years, and use the value of that year minus the climatology (average of the summer averages from 1981 to 2010) to obtain the anomaly. After processing, obtain the anomaly of the interdecadal variability of summer precipitation from 1963 to 2014, and perform empirical orthogonal decomposition (EOF) analysis to obtain the leading mode of the precipitation anomaly. The spatial distribution of the leading mode shows a zonal tripole pattern. Therefore, divide the eastern China region into three sub-regions, namely South China, the Yangtze - Huaihe River Basin, and North China, and calculate the regional average of each sub-region as the precipitation index.

[0059] Step 2: Processing mode and analyzing data. Select the return data of 9 dynamic models of DCPPA in CMIP6 reported and predicted annually from 1960 to 2016. The average value of the predicted future 1 - 5 years each year is used as the interdecadal variability, marked as the year in the middle of the five years, and the interdecadal variability predicted for each year is spliced and combined into a sequence, with the years becoming 1963 - 2019. The variables include: precipitation (PRE), sea surface temperature (SST), sea ice (SIC), geopotential height (HGT), and zonal and meridional wind fields (UV), and calculate the precipitation index of the three sub - regions predicted by the dynamic model from 1963 to 2019; Select the ERA5 monthly re - analysis data from 1961 to 2016, and obtain the interdecadal variability through a five - year moving average, marked as the year in the middle of the five years. The time range is 1963 - 2014, and the variables include: sea surface temperature (SST), sea ice (SIC), geopotential height (HGT), and zonal and meridional wind fields (UV).

[0060] Step 3: Find all sources of predictability for each sub - region. 1) First, use the bilinear interpolation method to interpolate all dynamic model data, observational and re - analysis data into 1°×1°, and remove the linear trend. Then, find the key sea surface temperature and sea ice regions in the same period, previous winter, and spring related to the precipitation index of each sub - region in the observations during 1963 - 2014 (regions with a confidence level exceeding 90% in the correlation distribution map of observed precipitation and re - analysis sea surface temperature or sea ice data), and use the ERA5 re - analysis data to explore the physical processes affecting precipitation; 2) Find the key sea surface temperature and sea ice regions in the same period, previous winter, and spring of each region related to the precipitation index of each sub - region in the dynamic model data during 1963 - 2014 (regions with a confidence level exceeding 90% in the correlation distribution map of the ensemble mean of dynamic model precipitation and the ensemble mean of dynamic model sea surface temperature or sea ice); 3) For the key sea surface temperature and sea ice found in the previous step, evaluate the ability of the ensemble mean of all dynamic models (the average of all initial fields of all models) to predict their interdecadal variability, and select the time - correlation coefficient (CC) as the indicator; 4) If the following three conditions are met: a. The key region of sea surface temperature or sea ice is consistent with the observation; b. The dynamic model has prediction skills for the sea surface temperature or sea ice in this region (the confidence level in the correlation distribution map of dynamic model data and re - analysis data exceeds 90%); c. The dynamic model can reproduce the physical process of the influence of this factor on precipitation in the observation; then this key variable is identified as a source of predictability, and calculate the regional average as the factor index. The time range of the index sequence is 1963 - 2019; 5) Repeat the process of (2) - (4) for the three sub - regions until all source factors of predictability for each sub - region are selected.

[0061] Step 4: Construct a machine learning model for the decadal prediction of summer precipitation in East China for independent trial forecasts. 1) Using the predictable source factor indices predicted by the models from 1963 to 2019 that have been screened (sequences averaged over 9 models, with 9 sequences for one factor) and the precipitation regional average sequences corresponding to the predicted regions by the models from 1963 to 2019 (sequences averaged over 9 models, a total of 9 sequences), a prediction model is constructed in combination with the GBRT model. The period from 1963 to 2014 is selected as the training time period, and the optimal hyperparameters are determined using five-fold cross-validation. The precipitation indices of three regions are predicted for the next five years (2015 - 2019), and the precipitation indices predicted for 3 regions are obtained; 2) For each sub-region, using the relationship between each grid point in the sub-region and the regional average index during the period from 1963 to 2014, a linear regression model is constructed. Combining with the precipitation indices predicted for the region from 2015 to 2019, the predicted values of each grid point within the East China region are obtained, and then adding the corresponding linear trend of each grid point, the spatial distribution of the predicted summer precipitation in East China from 2015 to 2019 is obtained.

[0062] Example 2. This example provides a device for decadal prediction of regional precipitation, including:

[0063] The first generation module is used to calculate the main precipitation mode based on the observed data of the monthly average grid points in the prediction region, divide the prediction region into multiple sub-regions according to the main precipitation mode, and generate the regional precipitation index of the observed data;

[0064] The second generation module is used to calculate the decadal variability based on the obtained dynamic model reforecast data, and generate the regional precipitation index of the dynamic model reforecast data based on the divided sub-regions;

[0065] The screening module is used to extract the key regions of environmental variables from the regional precipitation index of the observed data and the regional precipitation index of the dynamic model reforecast data respectively, and screen the predictable source factors according to the extracted key regions of environmental variables;

[0066] The training and optimization module is used to train and optimize the gradient boosting regression tree prediction model using the screened predictable source factors and the regional precipitation index of the dynamic model reforecast data, and obtain the optimized gradient boosting regression tree prediction model;

[0067] The prediction module is used to predict the precipitation indices of each sub-region in the future set years based on the optimized gradient boosting regression tree prediction model, and generate the decadal prediction result of regional precipitation.

[0068] For the specific function implementation of the above modules, refer to the relevant content in the method of Example 1, which will not be elaborated here.

[0069] Example 3 provides a computer-readable storage medium storing a computer program, which when executed by a processor, implements the steps of the method according to any one of Example 1.

[0070] Example 4 provides a computer device, including:

[0071] a memory for storing computer programs / instructions;

[0072] a processor for executing the computer programs / instructions to implement the steps of the method according to any one of Example 1.

[0073] Example 5 provides a computer program product including computer programs / instructions, which when executed by a processor, implement the steps of the method according to any one of Example 1.

[0074] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

[0075] Those skilled in the art should understand that the embodiments of the present disclosure can be provided as a method, a system or a computer program product. Therefore, the present disclosure can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present disclosure can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0077] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more of the processes Figure 1 one or more processes and / or blocks Figure 1 specified in the block or blocks.

[0078] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes Figure 1 one or more processes and / or blocks Figure 1 specified in the block or blocks.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure rather than to limit the scope of its protection. Although the present disclosure has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: after reading the present disclosure, those skilled in the art can still make various changes, modifications or equivalent replacements to the specific embodiments of the invention, but these changes, modifications or equivalent replacements are all within the scope of the protection of the claims pending for publication.

Claims

1. A method for regional precipitation decadal prediction, characterized in that: include: The main precipitation mode is calculated based on the observation data of the monthly average grid points in the forecast area, and the forecast area is divided into multiple sub-areas according to the main precipitation mode to generate the regional precipitation index of the observation data; Calculate the interdecadal variability based on the acquired dynamical model return data, and generate the regional precipitation index of the dynamical model return data based on the divided sub-regions; The key areas of environmental variables are extracted from the regional precipitation index of observation data and the regional precipitation index of dynamical model return data, and the predictability source factors are screened according to the extracted key areas of environmental variables; Using the selected predictability source factors and the regional precipitation index of the dynamical model return data, the gradient boosting regression tree prediction model is trained and optimized to obtain the optimized gradient boosting regression tree prediction model; Based on the optimized gradient boosting regression tree prediction model, the precipitation index of each sub-region in the future set year is predicted to generate the regional precipitation decadal prediction results; The key areas of environmental variables are extracted from the regional precipitation index of the observation data and the regional precipitation index of the dynamic model report data, and predictability source factors are screened according to the extracted key areas of environmental variables, including: Based on the regional precipitation index of the observed data and the environmental variables of the pre-generated reanalysis data, the key areas of environmental variables in the observations are determined through correlation analysis; Using the regional precipitation index of the dynamical model return data, the key areas of environmental variables in the dynamical model are extracted; Verify the predictive power of key areas of environmental variables in the dynamical model and their consistency with observed physical mechanisms; Screen environmental variables that meet the consistency between prediction ability and physical mechanism as predictability source factors; The verification of the predictive ability of key areas of environmental variables in the dynamic model and their consistency with observed physical mechanisms includes: Verify that the key areas of environmental variables in the dynamical model are consistent with the key areas of environmental variables in observations, that the confidence level in the correlation distribution map between the dynamical model report data and the reanalysis data exceeds the set threshold, and that the dynamical model can reproduce the physical process by which environmental variables in observations affect precipitation.

2. The method for regional precipitation decadal prediction according to claim 1, characterized in that: The method of calculating the main precipitation mode based on the observation data of the monthly average grid points in the prediction area, dividing the prediction area into multiple sub-areas according to the main precipitation mode, and generating the regional precipitation index of the observation data includes: Obtain observation data of monthly average grid points in the prediction area; The average precipitation in any season at each grid point is calculated based on the observed data, and a five-year sliding average is performed to generate decadal variability; Anomalies of decadal variability are calculated and the main modes of precipitation are extracted through empirical orthogonal decomposition; According to the latitudinal tripole distribution characteristics of the main precipitation mode, the forecast area is divided into three sub-regions, and the regional average precipitation index of each sub-region is calculated respectively.

3. The method for regional precipitation decadal prediction according to claim 1, characterized in that: The method of calculating the interdecadal variability based on the acquired dynamical model report data and generating a regional precipitation index of the dynamical model report data based on the divided sub-regions includes: Obtain dynamic model return data for the forecast area; Based on the dynamical model return data, the decadal variability of the annual average of future set years is extracted and spliced ​​into a sequence; Regional precipitation indices are generated from dynamical model return data based on the spliced ​​series of decadal variability.

4. The method for regional precipitation decadal prediction according to claim 1, characterized in that: When training and optimizing the gradient boosting regression tree prediction model using the screened predictability source factors and the regional precipitation index of the dynamical model return data, the hyperparameters are optimized by the five-fold cross validation method.

5. The method for regional precipitation decadal prediction according to claim 1, characterized in that: The optimized gradient boosting regression tree prediction model is used to predict the precipitation index of each sub-region in the future set year, and then a linear regression model and linear trend correction are combined to generate a regional precipitation decadal prediction result.

6. A regional precipitation decadal prediction device, using the regional precipitation decadal prediction method according to claim 1, characterized in that: include: The first generation module is used to calculate the main precipitation mode based on the observation data of the monthly average grid points in the prediction area, divide the prediction area into multiple sub-areas according to the main precipitation mode, and generate the regional precipitation index of the observation data; The second generation module is used to calculate the interdecadal variability based on the acquired dynamical model return data, and generate a regional precipitation index of the dynamical model return data based on the divided sub-regions; A screening module is used to extract key areas of environmental variables from the regional precipitation index of the observation data and the regional precipitation index of the dynamic model return data, and screen predictability source factors based on the extracted key areas of environmental variables; A training and optimization module is used to train and optimize the gradient boosting regression tree prediction model using the screened predictability source factors and the regional precipitation index of the dynamical model return data to obtain an optimized gradient boosting regression tree prediction model; The prediction module is used to predict the precipitation index of each sub-region in the future set year based on the optimized gradient boosting regression tree prediction model, and generate regional precipitation decadal prediction results.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 5 are implemented.

8. A computer device, characterized in that: include: Memory, for storing computer programs / instructions; A processor, configured to execute the computer program / instructions to implement the steps of the method according to any one of claims 1 to 5.

Images