Multi-stage coupled climate mode optimization and uncertainty quantification method and system

Abstract

The invention discloses a multi-stage coupled climate mode optimization and uncertainty quantification method and system, and relates to the technical field of climate prediction, and the method comprises the steps: obtaining a multi-source meteorological data set, and carrying out the preprocessing of the multi-source meteorological data set; on the basis of the preprocessed multi-source meteorological data set, a multi-dimensional evaluation index system is used for screening a GCM combination with the best adaptability to a research area; based on the screened GCM combination, establishing a segmented statistical relationship between the simulation data and the measured data in the historical period, obtaining a correction factor, and performing deviation correction on the original output of the mode by using the correction factor; and on the basis of a deviation correction result, fitting the GCM data in each scene by adopting a common least square method and a smooth fourth-order polynomial, respectively calculating the variance and proportion of each uncertainty component, and obtaining the total uncertainty through linear superposition. According to the invention, the problems of large atmospheric circulation mode selection deviation, low correction precision and insufficient uncertainty management in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of climate prediction, and particularly relates to a method and system for optimizing climate models and quantifying uncertainty through multi-stage coupling. Background Art

[0002] The General Circulation Model (GCM) is a core tool for predicting future climate change. By numerically simulating the physical processes of the atmosphere, ocean, and land surface, it provides key data support for global and regional climate research. However, the spatial resolution of climate variables output by GCMs is generally low, and there are significant biases in the simulation of elements such as precipitation and temperature in different regions, making it difficult to directly use them for assessing the impacts of climate change at local and regional scales. Although downscaling techniques (such as dynamical downscaling and statistical downscaling) can improve data quality by enhancing resolution or establishing statistical relationships, existing methods still face many bottlenecks: First, the simulation capabilities of different GCMs for regional climate elements vary significantly. Traditional evaluation methods mostly rely on a single statistical index (such as the mean square error), lacking comprehensive consideration of spatio-temporal distribution, trend changes, and probability characteristics, resulting in strong subjectivity in model selection. Second, existing bias correction methods do not fully consider the spatio-temporal heterogeneity of regional climate elements (such as precipitation frequency, intensity, and extreme event distribution). Especially in complex terrain regions, there are still systematic errors in the corrected data. In addition, the sources of uncertainty in climate prediction are complex (including internal variability, model differences, and emission scenario differences), while traditional methods mostly focus on a single source of uncertainty and lack a comprehensive quantification framework for the combined effects of the three, making it difficult to support risk classification decisions at the regional scale. These problems seriously restrict the reliability of climate change projections, and there is an urgent need for a systematic technology to optimize model selection, improve correction accuracy, and comprehensively quantify uncertainty. Summary of the Invention

[0003] Aiming at the above deficiencies in the prior art, the method and system for optimizing climate models and quantifying uncertainty through multi-stage coupling provided by the present invention solve the problems of large selection bias, low correction accuracy, and insufficient uncertainty management of the General Circulation Model in the prior art through a progressive process of "model adaptability assessment - dynamic bias correction - uncertainty quantification".

[0004] To achieve the above invention objective, the technical solution adopted by the present invention is: A method for optimizing climate models and quantifying uncertainty through multi-stage coupling, comprising the following steps: S1: Obtain multi-source meteorological data sets and perform preprocessing; S2: Based on the preprocessed multi-source meteorological data sets, use a multi-dimensional evaluation index system to screen the GCM combination with the best adaptability to the study area; S3: Based on the selected GCM combinations, establish the segmented statistical relationship between the historical simulation data and the measured data, obtain the correction factors, and use the correction factors to correct the biases in the original model outputs; S4: Based on the bias correction results, use ordinary least squares method and smooth fourth-order polynomial to fit the GCM data under each scenario, calculate the variances and proportions of each uncertainty component respectively, and obtain the total uncertainty through linear superposition, thus completing the optimization of the multi-stage coupled climate model and the quantification of uncertainty.

[0005] Further, the S1 includes the following sub-steps: S11: Obtain the measured data and model data. The measured data includes daily precipitation and temperature data, and the model data is selected from the output data of multiple GCMs in the Scenario-MIP program of CMIP6; S12: Use the bilinear interpolation method to unify the spatial resolutions of the measured data and the model data; S13: Divide the unified data into time periods of historical period, baseline period, verification period and future projection period, thus completing the acquisition and preprocessing of the multi-source meteorological data sets.

[0006] Further, the S2 includes the following sub-steps: S21: Use multiple statistical features to construct a multi-dimensional evaluation index system, and calculate the goodness-of-fit scores between each GCM simulation sequence and the measured sequence; The goodness-of-fit score is:

[0007] where is the number of statistical feature values, is the th relative difference between the model simulation sequence and the measured sequence of the statistical feature pattern, and are the maximum and minimum values of the relative differences; S22: Select the GCM combination with the best adaptability to the study area according to the goodness-of-fit scores.

[0008] Further, the S3 includes the following sub-steps: S31: Based on the quantile mapping method, sort the measured data and simulated daily data in the historical period, and divide them according to the percentiles; S32: Establish the statistical relationship between each segment of the measured data and the simulated data after division, and obtain the correction factors; S33: Add or multiply the correction factors to the future simulation data to correct the model output data. The formula is:

[0009]

[0010] Among them, is the corrected average temperature in the future period, is the corrected precipitation in the future period, is the average temperature of the original output of the model in the future period, is the precipitation of the original output of the model in the future period, is the measured average temperature during the reference period, is the average temperature of the original output of the model, is the measured precipitation during the reference period, is the precipitation of the original output of the model.

[0011] Furthermore, the S4 includes the following sub-steps: S41: Use ordinary least squares and a smoothed fourth-order polynomial to fit the GCM data in each scenario as:

[0012] Among them, is the original simulation of each model , scenario and year of the GCM data, is the fourth-order fit, is the multi-year average temperature or precipitation in the historical period, is the residual of the fitting equation; S42: Calculate the variance and proportion of each uncertainty component respectively, and obtain the total uncertainty through linear superposition. The formula is: Internal variability is:

[0013] Among them, is the number of models, is the variance of model over all scenarios and times; Model uncertainty is:

[0014] Among them, is the number of scenarios, is the variance between different models; Scenario uncertainty is:

[0015] Among them, is the variance between different scenarios; The internal variability , model uncertainty and scenario uncertainty account for the fractional uncertainty ratios of , and ; The internal variability , model uncertainty and scenario uncertainty have fractional uncertainties of as follows:

[0016]

[0017]

[0018] where is the total variance, is the average change.

[0019] The technical solution adopted by the present invention is also: a multi-stage coupled climate model optimization and uncertainty quantification system, the system comprising: A data preprocessing module: the data preprocessing module is used to obtain a multi-source meteorological data set and perform preprocessing; A model optimization module: the model optimization module uses a multi-dimensional evaluation index system to screen the GCM combination with the best adaptability to the study area; A bias correction module: the bias correction module establishes a piecewise statistical relationship between the simulated data and the measured data in the historical period, obtains a correction factor, and uses the correction factor to perform bias correction on the original output of the model; An uncertainty quantification module: the uncertainty quantification module uses ordinary least squares and a smooth fourth-order polynomial to fit the GCM data under each scenario, calculates the variances and proportions of each uncertainty component respectively, and obtains the total uncertainty through linear superposition, completing the multi-stage coupled climate model optimization and uncertainty quantification.

[0020] The beneficial effects of the present invention are: by constructing a progressive technical framework of "model optimization - dynamic correction - uncertainty quantification", the present invention significantly improves the prediction accuracy and decision-making practicality of regional climate change, and solves the core problems of strong subjectivity in model selection, insufficient adaptability of correction methods and incomplete uncertainty quantification in the prior art. Compared with traditional methods, the advantages of the present invention are reflected in: (1)Precision mode screening: Based on a dynamic optimization mechanism of multi-dimensional rank scoring, the simulation accuracy of the mode for regional climate elements is improved by 20% - 30%, providing a high-confidence data basis for subsequent analysis; (2)Efficient bias correction: A quantile mapping technique driven by spatio-temporal heterogeneity reduces the simulation errors of precipitation extreme events and temperature spatial distribution by 80% and 70% respectively; (3)Clear uncertainty management: A three-source uncertainty quantification framework reveals the dominant factors in different periods (e.g., the uncertainty ratio of the long-term scenario exceeds 70%), providing a scientific basis for formulating emission reduction and adaptation strategies in stages. Description of the Drawings

[0021] Figure 1 It is a flowchart of a multi-stage coupled climate model optimization and uncertainty quantification method of the present invention.

[0022] Figure 2 It is a flow block diagram of a multi-stage coupled climate model optimization and uncertainty quantification method of the present invention.

[0023] Figure 3 It is a scoring result diagram of each model in CMIP6 for precipitation and average temperature.

[0024] Figure 4 It is a spatial distribution diagram of the differences in various statistical indicators between the original simulation and the measured precipitation (GCMs-CMFD) and between the bias-corrected and the measured precipitation (DT-CMFD) during the verification period.

[0025] Figure 5 It is a spatial distribution diagram of the differences in various statistical indicators between the original simulation and the measured average temperature (GCMs-CMFD) and between the bias-corrected and the measured average temperature (DT-CMFD) during the verification period.

[0026] Figure 6 It is a comparison diagram of the climate projection uncertainty before and after bias correction.

[0027] Figure 7 It is a comparison diagram of the variance ratio of the climate projection uncertainty before and after bias correction. Detailed Embodiments

[0028] The present invention will be further described below with reference to the drawings and specific embodiments.

[0029] Example 1, as Figure 1 and Figure 2 shown, a multi-stage coupled climate model optimization and uncertainty quantification method includes the following steps: S1: Obtain multi-source meteorological data sets and perform preprocessing; S2: Based on the preprocessed multi-source meteorological dataset, use a multi-dimensional evaluation index system to screen the GCM combination with the best adaptability to the study area; S3: Based on the selected GCM combination, establish a segmented statistical relationship between historical simulation data and measured data, obtain correction factors, and use the correction factors to correct the bias of the original model output; S4: Based on the bias correction results, use ordinary least squares and a smoothing fourth-order polynomial to fit the GCM data under each scenario, calculate the variance and proportion of each uncertainty component respectively, and obtain the total uncertainty through linear superposition, completing the optimization of the multi-stage coupled climate model and the quantification of uncertainty.

[0030] The S1 includes the following sub-steps: S11: Obtain measured data and model data. The measured data includes daily-scale precipitation and temperature data, and the model data is selected from the output data of multiple GCMs in the Scenario-MIP program of CMIP6; In this embodiment, the China Regional Ground Meteorological Elements Driving Dataset (CMFD) is used as the reference data to obtain the measured daily-scale precipitation and temperature data with a spatial resolution of 0.1°×0.1°. The output data of 24 models in the Scenario-MIP program of CMIP6 are selected. Among them, the monthly-scale data is resampled to 2°×2° for model adaptability evaluation, and the daily-scale data is resampled to 1°×1° for future scenario prediction (the model information is shown in Table 1); Table 1 Specific information of CMIP6 models

[0031] S12: Use the bilinear interpolation method to unify the spatial resolution of the measured data and the model data; The bilinear interpolation method is:

[0032] where is the value of the interpolation point , x and y respectively represent the abscissa and ordinate of the meteorological element, , , and respectively represent the known coordinates before unifying the spatial resolution , , and .

[0033] S13: Divide the unified data into time periods of historical period, baseline period, verification period, and future prediction period, and complete the acquisition and preprocessing of multi-source meteorological data sets; The future climate change scenarios used are the rectangular combinations of different shared socioeconomic pathways (SSPs) and representative concentration pathways (RCPs), which can provide key data support for the research on future climate change mechanisms and climate change mitigation and adaptation. The specific information is shown in Table 2.

[0034] Table 2 Specific Information of Future Climate Change Scenarios

[0035] Taking into account the time spans of each data source, finally select 1979 - 2014 as the research period for simulation and comparison of the CMIP6 model in the historical period; 1979 - 2014 as the baseline period in the bias correction process of the CMIP6 model, 1995 - 2014 as the verification period of the bias correction method; select 2031 - 2050 and 2061 - 2080 as the research periods for future climate change prediction.

[0036] The S2 described above includes the following sub-steps: S21: Construct a multi-dimensional evaluation index system using multiple statistical characteristics, and calculate the goodness-of-fit score between each GCM simulation sequence and the measured sequence; The goodness-of-fit score is:

[0037] where is the number of statistical characteristic values, is the relative difference between the th statistical characteristic pattern simulation sequence and the measured sequence, and are the maximum and minimum values of the relative difference, The larger is, the closer the result of the model simulation is to the measured sequence;

[0038] In this embodiment, for precipitation and temperature, 11 statistical characteristics such as mean (Mean), coefficient of variation (Cv), spatio-temporal correlation coefficient (rspa / rtom), trend test (M-K Zc / β), probability distribution (BS / SS), etc. (see Table 3) are selected to establish a comprehensive evaluation system. By calculating the fitting degree of various statistical characteristic values between the simulated sequence and the measured sequence, the models are scored from 0 to 10 based on the quality of the fitting degree, so as to optimize the models with better adaptability to the study area.

[0039] Table 3 Information on statistical characteristic values selected in the rank scoring method

[0040] The S3 includes the following sub-steps: S31: Based on the quantile mapping method, sort the measured data and simulated daily data in the historical period and divide them according to the percentiles; S32: Establish a statistical relationship between each segment of the measured data and the simulated data after division to obtain the correction factor; S33: Add or multiply the correction factor to the future simulated data to correct the model output data. The formula is:

[0041]

[0042] Among them, is the corrected average temperature in the future period, is the corrected precipitation in the future period, is the average temperature of the original output of the model in the future period, is the precipitation of the original output of the model in the future period, is the measured average temperature during the reference period, is the average temperature of the original output of the model, is the measured precipitation during the reference period, is the precipitation of the original output of the model.

[0043] During the verification process of the bias correction method, three evaluation indicators, namely relative bias ( BIAS ), model correlation coefficient ( PCC ), and normalized root mean square error ( NRMSE ), are used for comparative analysis.

[0044]

[0045]

[0046] Among them, Represents the relative deviation of precipitation variables on a monthly scale, Represents the relative deviation of mean temperature variables on a monthly scale, and Represent the model output data and the measured data respectively.

[0047]

[0048] Among them, and Represent the model output data and the measured data at the th grid point respectively, and Represent the average values of the model output and the measured data for all grids in the study area respectively, Is the total number of grid points in the study area.

[0049]

[0050] Among them, Represents the average value of the measured data.

[0051] For precipitation and temperature, different statistical indicators were selected for method verification, and the verification period was from 1995 to 2014. For precipitation, four statistical indicators were used, including mean (MEAN), rainy day frequency (FREQ), rainy day intensity (INT), and 90th percentile (Q90); for mean temperature, the indicators used were mean (MEAN), 90th percentile (Q90), and 10th percentile (Q10). In addition, to further verify the applicability of the DT method in the Qinghai-Tibet Plateau, each verification indicator of precipitation and mean temperature was evaluated, and the evaluation indicators included BIAS, PCC, and NRMSE.

[0052] The S4 includes the following sub-steps: S41: Using ordinary least squares method and smoothing fourth-order polynomial to fit the GCM data in each scenario as:

[0053] Among them, Is the original simulation of each model , scenario and year of the GCM data, Is the fourth-order fit, Is the multi-year average temperature or precipitation in the historical period, Is the residual of the fitting equation; S42: Calculate the variance and proportion of each uncertainty component respectively, and obtain the total uncertainty through linear superposition. The formula is: Internal variability is:

[0054] wherein, is the number of patterns, is the variance of pattern across all scenarios and times; Pattern uncertainty is:

[0055] wherein, is the number of scenarios, is the variance between different patterns; Scenario uncertainty is:

[0056] wherein, is the variance between different scenarios; Said internal variability , pattern uncertainty and scenario uncertainty have fractional uncertainty proportions of , and ; Said internal variability , pattern uncertainty and scenario uncertainty have fractional uncertainties of:

[0057]

[0058]

[0059] wherein, is the total variance, is the average change.

[0060] Therefore, the fractional uncertainties of internal variability ( ), pattern uncertainty ( ), and scenario uncertainty ( ) (90% confidence level) are , and . The proportions of the fractional uncertainties of internal variability, pattern uncertainty, and scenario uncertainty are respectively defined as , and 。

[0061] In one embodiment of the present invention, taking the Qinghai-Tibet Plateau in the alpine source area as an example. Based on the rank scoring method, the final performance (RS) of different GCMs in simulating precipitation and temperature in the Qinghai-Tibet Plateau is as Figure 3 shown. The average score for simulated precipitation is 5.71, and the average score for simulated average temperature is 6.19. There are also significant differences in the simulation results of the same model for different elements. For example, for the CanESM5 model, the score for simulating temperature is only 3.11, while the score for simulating precipitation is 6.09. It can be found that separately selecting the general circulation models for different elements will help to give full play to the advantages of each model itself, thereby effectively improving the accuracy of simulation. According to the results of the model adaptability assessment, 10 models with the best simulation performance for precipitation and temperature in CMIP6 are respectively selected, and subsequent comparative analysis is carried out with the corresponding multi-model ensemble results.

[0062] Table 4 lists the evaluation and comparison results of various statistical indicators between the original simulated data and the measured data of precipitation (GCMs), and between the bias-corrected data and the measured data (DT). From the perspective of relative bias (BIAS), the results of DT can effectively reduce the large positive biases of MEAN, FREQ, and INT in the original simulation; from the perspective of the normalized root mean square error (NRMSE) index, the values of the original simulation results in different months are mostly greater than 1, and after correction, they are reduced to less than about 0.4, which is more consistent with the observed data; in addition, the PCC of the 4 statistical indicators after correction is mostly higher than 0.95, which is about 0.25 higher than the original simulation results.

[0063] Table 4 Evaluation results of various statistical indicators of the original simulation and bias-corrected precipitation during the verification period

[0064] Figure 4 Figure is the spatial distribution map of the deviations of various statistical indicators calculated based on the original simulation (GCMs), the bias-corrected results (DT), and the observed precipitation (CMFD). For MEAN and FREQ, the deviations of the GCMs results are mainly concentrated in the southern and eastern parts of the plateau, with high overestimations of about 240 mm; the DT results show that almost all the deviations have been basically corrected, and the annual average precipitation deviation in most parts of the plateau is controlled within 80 mm, and the deviation of FREQ is also within 0.04. For the INT index, the deviations of GCMs are significantly overestimated in most parts of the plateau, exceeding 0.9 compared with the CMFD results. After correction by the DT method, the deviation is reduced to between -0.3 and 0.3; for the Q90 index, the deviations of the GCMs simulation are mainly concentrated in the southeastern part of the plateau, with most exceeding 2.5 mm, and the DT results can reduce the deviation to within 0.5 mm.

[0065] Table 5 lists the evaluation and comparison results of various statistical indicators between the original precipitation simulation data and the measured data (GCMs), and between the bias-corrected data and the measured data (DT). Through comparative analysis from three aspects: relative bias (BIAS), pattern correlation coefficient (PCC), and normalized root mean square error (NRMSE), it can be seen that the bias of GCMs mainly concentrates during February to May and November to December, approximately -1 °C. After DT correction, the bias in most months is reduced to -0.2 °C; for NRMSE, the results of DT are all within 0.1, showing obvious improvement compared with GCMs; in addition, the PCC evaluation results further indicate that the simulation results after DT correction have improved in the ability to capture spatial distribution characteristics.

[0066] Table 5 Evaluation results of various statistical indicators for the original simulation (GCMs) and bias-corrected average temperature (DT) during the verification period

[0067] Figure 5 It is the spatial distribution map of the bias of various statistical indicators calculated based on between the original simulation (GCMs) and the observed temperature (CMFD), as well as between the correction results (DT) and the observed temperature (CMFD). For the original simulation results, the biases of MEAN, Q90, and Q10 of the average temperature in most areas of the Qinghai-Tibet Plateau are relatively obvious. Among them, the MEAN is higher than 3 °C, and the larger positive biases where Q90 and Q10 are higher than 5 °C mainly appear at the edge of the plateau. After DT correction, most of the biases of the three indicators are reduced to -1 to 1 °C. At the same time, the three statistical indicators all show a changing pattern of gradually increasing from west to east, and the northern part of the Qaidam Basin and the southeastern part of the plateau are two regions where high values are relatively concentrated. In summary, using the DT method to correct the simulated temperature of each selected model in CMIP6 on the Qinghai-Tibet Plateau has a relatively high reliability.

[0068] Bias correction significantly reduces the total uncertainty of precipitation and temperature projections. For precipitation ( Figure 6 in a, c), the total uncertainty of the original projection (GCMs) continuously decreases from 4.57 to 1.52, and further decreases to 1.06 after DT correction. Moreover, the model uncertainty and internal variability decrease from 2.70 and 2.49 to 0.51 and 0.26 respectively; the proportion of scenario uncertainty in the long term (after 2070) increases from 26.5% to 66.5%, becoming the dominant factor. For temperature projections ( Figure 6 in b, d), the proportion of model uncertainty in the original output initially accounts for 83%, and the scenario uncertainty rises to 76% in 2080. After DT correction, the time when the minimum value of the total uncertainty appears is postponed to 2060, and the growth rate of scenario uncertainty accelerates (from 0.11 to 0.76), indicating that the emission scenario has a greater impact on the long-term temperature.

[0069] Before bias correction, the initial stage of precipitation prediction is mainly dominated by internal variability (53.9%), far exceeding 16.4% of temperature; in the long term, the contribution of temperature scenario uncertainty reaches 66.5%, significantly higher than 26.5% of precipitation ( Figure 7 in a, b). After bias correction ( Figure 7 in c, d), the uncertainty of precipitation scenario accounts for 66.5% in the long term, and the proportion of temperature scenario uncertainty rises to 57.6%, while the contribution of internal variability approaches zero. In addition, the reduction rate of precipitation model uncertainty is slower than that of temperature, indicating that the systematic bias of precipitation is more difficult to eliminate. This result highlights the key role of bias correction methods in clarifying the dominant uncertainty and optimizing long-term climate strategies.

[0070] Example 2, a multi-stage coupled climate model optimization and uncertainty quantification system, the system includes: Data preprocessing module: The data preprocessing module is used to obtain multi-source meteorological data sets and perform preprocessing; Model optimization module: The model optimization module uses a multi-dimensional evaluation index system to screen the GCM combination with the best adaptability to the study area; Bias correction module: The bias correction module establishes a segmented statistical relationship between historical simulation data and measured data, obtains correction factors, and uses the correction factors to correct the bias of the original output of the model; Uncertainty quantification module: The uncertainty quantification module uses ordinary least squares and a smoothed fourth-order polynomial to fit the GCM data under each scenario, calculates the variance and proportion of each uncertainty component respectively, and obtains the total uncertainty through linear superposition, completing the optimization of the multi-stage coupled climate model and uncertainty quantification.

[0071] Those of ordinary skill in the art will realize that the embodiments described herein are for the purpose of assisting the reader in understanding the principles of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the scope of protection of the invention.

Claims

1. A multi-stage coupled climate model optimization and uncertainty quantification method, characterized in that: The following steps are involved: S1: Obtain multi-source meteorological datasets and preprocess them; S2: Based on the preprocessed multi-source meteorological data set, a multi-dimensional evaluation index system is used to screen the GCM combination with the best adaptability to the study area; S3: Based on the selected GCM combination, the segmented statistical relationship between the simulated data and the measured data in the historical period is established to obtain the correction factor, and the correction factor is used to correct the deviation of the original output of the model; S4: Based on the bias correction results, ordinary least squares method and smoothed fourth-order polynomial are used to fit the GCM data under each scenario, the variance and proportion of each uncertainty component are calculated respectively, and the total uncertainty is obtained through linear superposition, completing the multi-stage coupled climate model optimization and uncertainty quantification.

2. The multi-stage coupled climate model optimization and uncertainty quantification method according to claim 1, characterized in that: The S1 includes the following sub-steps: S11: acquiring measured data and model data, wherein the measured data include daily precipitation and temperature data, and the model data are selected from multiple GCM output data in the Scenario-MIP plan of CMIP6; S12: Use bilinear interpolation to unify the spatial resolution of measured data and model data; S13: Divide the unified data into time periods of historical period, benchmark period, verification period and future forecast period, and complete the acquisition and preprocessing of multi-source meteorological data sets.

3. The multi-stage coupled climate model optimization and uncertainty quantification method according to claim 1, characterized in that: The S2 includes the following sub-steps: S21: Use multiple statistical features to construct a multi-dimensional evaluation index system and calculate the fit score between each GCM simulation series and the measured series; The fit score for: in, is the number of statistical eigenvalues, For the The relative difference between the simulated sequence and the measured sequence of the statistical characteristic pattern, and is the maximum and minimum value of the relative difference; S22: Screen the GCM combination with the best adaptability to the study area based on the goodness of fit score.

4. The multi-stage coupled climate model optimization and uncertainty quantification method according to claim 1, characterized in that: The S3 includes the following sub-steps: S31: Based on the quantile mapping method, the measured data and simulated daily data of the historical period are sorted and divided according to percentiles; S32: Establishing a statistical relationship between each segment of measured data and simulated data to obtain a correction factor; S33: Add or multiply the correction factor to the future simulation data to correct the model output data. The formula is: in, is the average temperature corrected for the future period, is the corrected precipitation for the future period, is the average temperature of the model's original output for the future period, is the original output precipitation of the model in the future period, is the measured average temperature during the reference period, is the average temperature of the model's raw output, is the measured precipitation during the reference period, is the precipitation output from the model.

5. The multi-stage coupled climate model optimization and uncertainty quantification method according to claim 1, characterized in that: The S4 comprises the following sub-steps: S41: The GCM data under each scenario are fitted using ordinary least squares method and smoothed fourth-order polynomial: in, For GCM data each mode ,scene and year The original simulation, is a fourth-order fit, is the multi-year average temperature or precipitation in the historical period, is the residual of the fitted equation; S42: Calculate the variance and proportion of each uncertainty component respectively, and obtain the total uncertainty through linear superposition. The formula is: Internal variability for: in, is the number of modes, For Mode variance across all scenarios and time; Model uncertainty for: in, is the number of scenarios, is the variance between different modes; Scenario uncertainty for: in, is the variance between different scenarios; The internal variability , Model uncertainty and scenario uncertainty The fractional uncertainty ratios are , and ; The internal variability , Model uncertainty and scenario uncertainty Uncertainty of fraction for: in, is the total variance, is the average change.

6. A system for the multi-stage coupled climate model optimization and uncertainty quantification method according to any one of claims 1 to 5, characterized in that: The system comprises: Data preprocessing module: The data preprocessing module is used to obtain multi-source meteorological data sets and perform preprocessing; Model optimization module: The model optimization module uses a multi-dimensional evaluation index system to select the GCM combination with the best adaptability to the study area; Deviation correction module: The deviation correction module establishes a segmented statistical relationship between the simulated data and the measured data in the historical period, obtains the correction factor, and uses the correction factor to perform deviation correction on the original output of the model; Uncertainty quantification module: The uncertainty quantification module uses ordinary least squares method and smooth fourth-order polynomial to fit the GCM data under each scenario, calculates the variance and proportion of each uncertainty component respectively, and obtains the total uncertainty through linear superposition, completing the multi-stage coupled climate model optimization and uncertainty quantification.

Images