Methods, systems, equipment and media for uncertainty assessment of meteorological and oceanographic models

Abstract

The present invention relates to a method, system, device, and medium for assessing uncertainty in a meteorological-ocean model. The method comprises: when assessing the impact of observation data uncertainty on a meteorological-ocean model, using a resampling technique to generate multiple samples through repeated sampling to construct an empirical distribution, thereby quantifying the degree of impact of observation data fluctuations on the meteorological-ocean model; when assessing the impact of model parameter uncertainty on the meteorological-ocean model, by comparing the model output results after model parameter fluctuations with the baseline results, the degree of impact of model parameter fluctuations on the meteorological-ocean model is determined; when assessing the impact of model method uncertainty on the meteorological-ocean model, by comparing the model output results obtained after the input variables are disturbed with the original output results, the degree of impact of the model method on the meteorological-ocean model output results under the condition of input variable disturbances is determined. The present invention can comprehensively reflect the impact of different factors on the model and improve the integrity of the assessment.

Description

Technical Field

[0001] The present invention relates to the technical field of uncertainty quantification, and in particular to a method, system, equipment and medium for evaluating uncertainty in a meteorological and oceanographic model. Background Art

[0002] With the deepening of meteorological and oceanographic research and the expansion of their application areas, the development and application of meteorological and oceanographic models are becoming increasingly widespread, becoming a core support for scientific decision-making in human activities. By simulating the temporal and spatial variations of the atmospheric and oceanic systems, these models provide key predictions for extreme weather warnings, climate policy formulation, and marine resource development.

[0003] Evaluating meteorological and oceanographic models under high-impact scenarios is of great significance. For example, the prediction accuracy of climate change research models can provide strong support for the scientific formulation of policies and systems.

[0004] However, due to the complexity and variability of the ocean environment, as well as differences in parameter settings, data input, core algorithms, and other aspects of its construction, the accuracy and credibility of ocean model predictions can vary significantly. Current quantitative evaluation methods for meteorological ocean models often rely on empirical judgment or single performance indicators. These methods struggle to fully reflect the model's actual performance in complex and diverse environments, exacerbating the inaccuracy and lack of credibility of predictions. This also directly impacts the model's practical application, leading to a lack of scientific basis for model optimization and limiting its widespread and in-depth application in real-world scenarios. Summary of the Invention

[0005] Based on this, it is necessary to provide a meteorological-ocean model uncertainty assessment method, system, equipment and medium that can effectively evaluate meteorological-ocean models and enhance their application value in response to the above technical problems.

[0006] A method for assessing uncertainty in a meteorological-oceanographic model, the method comprising: assessing the impact on the meteorological-oceanographic model based on uncertainty in observation data, uncertainty in model parameters, and uncertainty in model method;

[0007] When evaluating the impact of the uncertainty of the observational data on the meteorological ocean model, a resampling technique is used to generate multiple samples through repeated sampling to construct an empirical distribution to quantify the degree of impact of the fluctuation of the observational data on the meteorological ocean model;

[0008] When evaluating the impact of the model parameter uncertainty on the meteorological-ocean model, any one or more combinations of local sensitivity analysis, global sensitivity analysis, and Monte Carlo uncertainty interval estimation are used, and the model output results obtained under the input parameters are used as the baseline results. By comparing the model output results after the model parameter fluctuations with the baseline results, the degree of impact of the model parameter fluctuations on the meteorological-ocean model is determined;

[0009] When evaluating the impact of the uncertainty of the model method on the meteorological-ocean model, any one or more combinations of rank stability analysis and sensitivity analysis are used to compare the model output results obtained after the input variables are disturbed with the original output results to determine the degree of impact of the model method on the meteorological-ocean model output results when the input variables are disturbed.

[0010] On the other hand, a meteorological ocean model uncertainty assessment device is also provided, comprising:

[0011] An assessment factor determination module is used to assess the impact on the meteorological ocean model based on the uncertainty of observation data, model parameter uncertainty and model method uncertainty;

[0012] An observation data uncertainty assessment module is used to use resampling technology to generate multiple samples through repeated sampling to construct an empirical distribution when assessing the impact of the observation data uncertainty on the meteorological ocean model, so as to quantify the impact of the observation data fluctuation on the meteorological ocean model;

[0013] A model parameter uncertainty assessment module is used to use any one or more combinations of local sensitivity analysis, global sensitivity analysis, and Monte Carlo uncertainty interval estimation when assessing the impact of the model parameter uncertainty on the meteorological ocean model, using the model output results obtained under the input parameters as the baseline results, and determining the degree of impact of the model parameter fluctuation on the meteorological ocean model by comparing the model output results after the model parameter fluctuation with the baseline results;

[0014] The model method uncertainty assessment module is used to use any one or more combinations of rank stability analysis method and sensitivity analysis method when evaluating the impact of the model method uncertainty on the meteorological ocean model, and to compare the model output results obtained after the input variables are disturbed with the original output results to determine the degree of influence of the model method on the meteorological ocean model output results when the input variables are disturbed.

[0015] On the other hand, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned meteorological ocean model uncertainty assessment method when executing the computer program.

[0016] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned meteorological ocean model uncertainty assessment method are implemented.

[0017] Compared with the existing technology, the meteorological ocean model uncertainty assessment method, system, device and medium provided by the present invention have the following effects:

[0018] 1. By evaluating the various factors that affect the accuracy of meteorological and oceanographic model predictions, it comprehensively covers the sources of uncertainty, breaks through the limitations of single-dimensional evaluation, and can fully reflect the impact of different factors on the model, thereby improving the integrity of the evaluation.

[0019] 2. Design differentiated evaluation methods for different sources of uncertainty to improve the accuracy of the evaluation and enhance the accuracy of the results, providing a basis for subsequent optimization of the model.

[0020] 3. Through differential analysis, benchmark result comparison, interval estimation and other methods, abstract uncertainty is transformed into quantifiable indicators, providing clear and scientific direction guidance for model optimization, avoiding blind optimization, and enhancing the reliability and application value of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required in the embodiments. It should be understood that the following drawings only illustrate some embodiments of the present invention, and those skilled in the art can derive other relevant drawings based on these drawings without creative work.

[0022] Figure 1 Schematic diagram of the process of uncertainty assessment of meteorological and ocean models provided in Example 1;

[0023] Figure 2 This is a flow chart of the uncertainty assessment method for the meteorological ocean model provided in Example 1;

[0024] Figure 3 This is a structural block diagram of the meteorological ocean model uncertainty assessment device provided in Example 2;

[0025] Figure 4 This is a diagram of the internal structure of the computer device provided in Example 3.

[0026] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0028] It can be understood that the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in this field can implement it. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0029] The following describes the implementation of the present invention in detail with reference to the accompanying drawings in the embodiments of the present invention.

[0030] Example 1

[0031] This embodiment discloses a method for assessing uncertainty in meteorological and oceanographic models. By analyzing the model's construction and operational mechanisms, introducing a scientific evaluation index system and statistical analysis methods, and proposing a concise and efficient uncertainty assessment framework, this method improves the application effectiveness of meteorological and oceanographic models, promotes the standardization of evaluation methods within the field, and provides theoretical and technical support for model development and integration, user selection, and application optimization. This method enables comprehensive and accurate assessment of meteorological and oceanographic models, from data to methods. By converting abstract uncertainty into quantifiable indicators, this method provides clear and scientific guidance for model optimization, avoids blind optimization, and enhances the reliability and application value of meteorological and oceanographic models in practical applications. It is worth noting that meteorological and oceanographic models include both specialized meteorological and oceanographic models and meteorological and oceanographic impact assessment models. Specifically, specialized meteorological and oceanographic models refer to models of various meteorological and oceanographic elements within the meteorological and oceanographic field, such as wind and wave models and visibility models. Meteorological and oceanographic impact assessment models refer to models that apply meteorological and oceanographic elements to other fields and the impact and degree of impact on those fields. For example, meteorological and agricultural impact assessments assess the degree of impact of meteorological elements on agriculture, which falls within the scope of meteorological and oceanographic impact assessment models.

[0032] like Figure 1 and Figure 2 As shown, the uncertainty assessment method of the meteorological ocean model provided in this embodiment includes the following steps:

[0033] Step 201 : Evaluate the impact on the meteorological ocean model based on the uncertainty of observation data, the uncertainty of model parameters and the uncertainty of model method.

[0034] It is understandable that traditional evaluation methods often rely on empirical judgment or a single performance indicator. However, the present invention incorporates three key types of uncertainties: observation data, model parameters, and model methods, into the evaluation framework, covering the entire chain of error sources from input to output of meteorological and ocean models, thereby improving the integrity of the evaluation.

[0035] Step 202 , when evaluating the impact of observation data uncertainty on the meteorological ocean model, a resampling technique is used to generate multiple samples through repeated sampling to construct an empirical distribution to quantify the impact of observation data fluctuations on the meteorological ocean model.

[0036] It can be understood that data uncertainty assessment mainly involves selecting specific methods for different data types in combination with specific types. First, the types of meteorological and ocean observation data are classified, and the uncertainty caused by data anomalies, small samples, inherent volatility, and inherent time or spatial correlation of the data are taken into consideration. The self-service method is used to realize the uncertainty assessment of various aspects of the observation data.

[0037] Step 203, when evaluating the impact of model parameter uncertainty on the meteorological ocean model, any one or more combinations of local sensitivity analysis, global sensitivity analysis, and Monte Carlo uncertainty interval estimation are used, and the model output results obtained under the input parameters are used as the baseline results. By comparing the model output results after the model parameter fluctuations with the baseline results, the degree of impact of the model parameter fluctuations on the meteorological ocean model is determined.

[0038] It can be understood that local sensitivity analysis can identify the local impact of a single model parameter; global sensitivity analysis can capture the global impact of the interaction of multiple model parameters; Monte Carlo uncertainty interval estimation can quantify the uncertainty range through interval estimation; combining any one or more of local sensitivity analysis, global sensitivity analysis, and Monte Carlo uncertainty interval estimation according to needs can improve the accuracy of model parameter uncertainty assessment.

[0039] Step 204, when evaluating the impact of model method uncertainty on the meteorological ocean model, use any one or more combinations of rank stability analysis method and sensitivity analysis method to compare the model output results obtained after the input variables are disturbed with the original output results to determine the degree of impact of the model method on the meteorological ocean model output results when the input variables are disturbed.

[0040] It can be understood that by evaluating the uncertainty of the model method separately, confusion with observation data and model parameters is avoided, independent evaluation and judgment of uncertainty at the method level is ensured, and evaluation errors caused by cross-interference are reduced.

[0041] The present invention converts abstract uncertainty into quantifiable indicators, providing a clear direction for model optimization. If the influence of uncertainty in observation data accounts for a high proportion, the data can be improved or the observation accuracy can be improved; if the influence of uncertainty in model parameters accounts for a high proportion, the focus can be on the parameter calibration of the model; if the influence of uncertainty in model methods accounts for a high proportion, the calculation process or model structure can be improved. In this way, a scientific basis can be provided for the priority ranking of resource investment, avoiding the waste of resources caused by blind optimization. Through the method proposed in the present invention, the credible interval and error source of the model output results can be intuitively understood, so as to optimize the model in a targeted manner, reduce the decision-making errors of meteorological and ocean models in disaster warning, marine resource development, weather forecasting, etc., and improve the reliability and risk controllability of the model in practical applications.

[0042] In the specific implementation process of step 202, when evaluating the impact of observation data uncertainty on the meteorological ocean model, various Bootstrap resampling techniques are mainly used to generate a series of "possible" data set scenarios in combination with the structural characteristics of the data, targeting the volatility caused by limited data samples and random errors. Multiple samples are generated by repeated sampling to construct an empirical distribution to measure the degree of fluctuation of the statistic under different data samples, thereby achieving an assessment of the volatility of the observation data and quantifying the degree of impact of the fluctuation of the observation data on the meteorological ocean model.

[0043] The sources of uncertainty in observation data are extensive, including: sampling errors due to errors in the measuring instruments themselves, indirect measurement errors, human recording errors and insufficient sample space coverage, errors caused by the adaptability of the observed variables and the model structure, and errors caused by the random distribution characteristics of the observed variables.

[0044] Therefore, we employ three bootstrap methods to assess uncertainty for different data types: the block bootstrap method, applicable to time series, maintains autocorrelation structure by slidingly extracting consecutive blocks, achieving reliable assessment through sampling, statistical calculation, and the generation of empirical distributions. The spatial bootstrap method, applicable to spatial or spatiotemporal data, samples groups such as sites, preserving both spatial and temporal correlations. The spatiotemporal bootstrap method, combining spatiotemporal structure with spatial location sampling and statistical analysis, assesses spatiotemporal heterogeneity. All three methods generate empirical distributions through resampling to quantify the impact of data on model uncertainty.

[0045] Among them, the first set of methods is the block bootstrap method. For time series data, the system uses the sliding block bootstrap method to divide the time series into several continuous blocks, and samples in blocks. Each time, several continuous segments of length L are randomly extracted from the original sequence and concatenated to form a new time series with the same length as the original sequence. This method ensures that the adjacent observations that are selected enter the bootstrap sample together, thereby preserving the autocorrelation and sequential structure of the original sequence. Compared with the simple independent and identically distributed Bootstrap method, the sliding block bootstrap method can effectively reproduce the correlation of the original data, making the uncertainty assessment of the data more realistic and reliable. By resampling each time series through multiple sliding block bootstraps, the distribution of model predictions or parameters under different samples is obtained, thereby calculating the variance, deviation and other metrics of the time series model results and mapping them into uncertainty scores.

[0046] The detailed steps of the block self-service method are as follows:

[0047] First determine the block length and set an appropriate block length (Usually determined by autocorrelation analysis or selected by empirical methods). Randomly extract continuous data blocks with replacement to form a new Bootstrap sequence , until the new sample length reaches the original sample length .

[0048] ;

[0049] For each generated Bootstrap sample Calculate the required statistics (mean, variance, autocorrelation coefficient, etc.).

[0050] Generate block Bootstrap empirical distribution and repeat the above steps Second, obtain the block Bootstrap empirical distribution of the statistic , further parameter estimation and confidence interval construction are performed based on this empirical distribution.

[0051] The second set of methods is spatial bootstrapping. For spatial or spatiotemporal data with spatial correlation or clustering structure (such as the joint time series of multiple stations), the system employs grouped or hierarchical bootstrap methods, using natural groups (such as stations) as sampling units, or simultaneously partitioning and sampling across both time and space. For example, for spatiotemporal data from multiple meteorological stations, one can partition the data into blocks by station and extract contiguous segments in time. This ensures that temporal correlation within the same station in the bootstrapped samples, as well as spatial correlation between different stations, is preserved in the grouped sampling (this is called an extension of spatial block bootstrapping). This multidimensional block sampling method generates a new dataset that retains the spatiotemporal correlation structure of the original data. Retraining the model based on this method can reflect the impact of the original data's uncertainty on the model, outputting key prediction metrics, error rates, and other outputs. The distribution of these outputs across all samples is then statistically analyzed to quantify the uncertainty introduced by the data into the model.

[0052] The detailed steps of the spatial self-help method are as follows:

[0053] The first is the spatial resampling process. Let the original spatial data be ,in Indicates the spatial position. From the original sequence Randomly draw spatial positions with replacement , generate a new Bootstrap sequence :

[0054] ;

[0055] Then calculate the statistics for each generated Bootstrap sample Calculate the required statistics (mean, variance, spatial uniformity, etc.).

[0056] Generate spatial Bootstrap empirical distribution and repeat the above steps Second, obtain the block Bootstrap empirical distribution of the statistic , based on this empirical distribution, parameter estimation, spatial heterogeneity assessment and confidence interval construction are further performed.

[0057] The third set of methods is the spatiotemporal bootstrap method, a nonparametric resampling method for data with both temporal and spatial structure. It combines the block sampling concept of time series with the point sampling strategy of spatial data to assess statistical uncertainty while maintaining the spatiotemporal correlation structure.

[0058] The detailed steps of the spatiotemporal bootstrap method are as follows:

[0059] The first is the spatial resampling process. Let the original spatial data be ,in Indicates spatial location, Indicates a time point. From the original sequence Randomly draw spatial positions with replacement , generate a new Bootstrap sequence :

[0060] ;

[0061] For each generated Bootstrap sample Calculate the required statistics (mean, variance, spatial uniformity, etc.).

[0062] Generate spatial Bootstrap empirical distribution and repeat the above steps Second, obtain the block Bootstrap empirical distribution of the statistic , based on this empirical distribution, parameter estimation, spatiotemporal heterogeneity assessment and confidence interval construction are further performed.

[0063] During the specific implementation of step 203, the impact assessment is mainly carried out on the uncertainty of model parameters existing in the meteorological ocean model. The uncertainty of model parameters includes the uncertainty transfer from actual input data, the uncertainty of parameter estimation method and the uncertainty of model parameters themselves, which may have a greater impact on the model output. This part is to evaluate the degree of influence of the uncertainty of model parameters on the meteorological ocean model. In order to achieve accurate quantification and impact assessment of model parameter uncertainty, the present invention provides three sets of scalable analysis methods for different model parameter types and model scenarios. In particular, considering the current situation that most meteorological ocean model output variables have no actual observations, the three sets of methods effectively solve the evaluation problem without measured data by comparing the results of benchmark parameters as benchmark values. The various methods can be used alone or in combination according to actual needs to cover the full process evaluation of parameter uncertainty.

[0064] Among them, the first set of analytical methods is local sensitivity analysis, which is applicable to continuous or discrete model parameter types, and there are no strict requirements for applicable scenarios. Local sensitivity analysis focuses on the impact of small-scale changes in input parameters near a specific point on the output. It usually measures the local impact of small changes in parameters on the output by calculating partial derivatives or finite differences. It calculates the average impact of local disturbances of input variables or model parameters on the output, thereby quickly identifying important variables and evaluating the degree of influence of each model parameter. Its output indicator is the average basic effect. The larger its value, the greater the impact of the model parameter on the model output. By calculating the basic effect of each model parameter, it can help identify which parameters are the key factors affecting the output results.

[0065] Unlike traditional methods that directly evaluate model outputs, the local sensitivity analysis provided in this embodiment focuses on measuring the error in model results. Specifically, the model output under the original parameter configuration is defined as the baseline result. By calculating the error between the model output and the baseline result under different parameter value combinations, the impact of parameter fluctuations on the model results is quantified, thereby intuitively characterizing the uncertainty level of each parameter. It is worth noting that the original parameters are the parameters under the original model.

[0066] The local sensitivity analysis process is:

[0067] Obtain input parameters; set the number of parameter levels based on the input parameters and calculate the step size to generate two or more independent tracks; on each independent track, determine a reference point, and perturb a single parameter of the reference point to obtain a perturbation point.

[0068] For each independent track, the benchmark point and disturbance point are input into the meteorological ocean model for calculation to obtain the benchmark result and disturbance result.

[0069] The mean square error between the baseline results and the perturbation results is calculated, and the basic effect of each model parameter is calculated based on the mean square error. Then, the mean absolute basic effect of each model parameter is calculated based on the basic effect.

[0070] The degree of influence of each model parameter on the output results is judged based on the mean absolute basic effect.

[0071] Specifically, we first define the problem. The parameters of the model are determined as ; The model output under the original parameter configuration is the benchmark result .

[0072] The evaluation goal is to quantify the output results and benchmark results of different input parameter value combinations The MSE error characterizes the uncertainty of the input parameter fluctuation on the model results.

[0073] Then, sample generation is performed. Parameter levels are set based on the input parameters. p , calculate the minister according to the number of parameters and parameter level, and generate independent trajectories, .

[0074] On each independent track, the reference point is . Perturb a single parameter of the reference point to obtain the perturbation point It can be seen that when generating disturbance points, only one reference point is disturbed each time, and the difference between different trajectories lies in the different reference points.

[0075] For each trajectory The sample point will Input the meteorological ocean model for calculation to obtain the baseline results and disturbance results.

[0076] Among them, the benchmark results are expressed as: ; The perturbation result is expressed as: .

[0077] Where, Represents the original parameter combination; Represents the combination after different parameter perturbations; Indicates actual data.

[0078] Calculate the mean square error between the baseline result and the perturbation result. The expression is:

[0079] ;

[0080] The basic effect of each model parameter is calculated based on the mean square error, and the expression is:

[0081] ;

[0082] The mean absolute basic effect of each model parameter is calculated based on the basic effect, and the expression is:

[0083] ;

[0084] Where, Indicates the Benchmark results for independent tracks; Indicates the Independent track The perturbation results of the model parameters; Indicates the Independent track The mean square error of the model parameters; Indicates the Independent track The basic effects of the model parameters; Indicates the the mean absolute base effects of the model parameters; Indicates the number of independent tracks.

[0085] Based on the calculated mean absolute base effect , to determine the degree of influence of each input parameter on the output result. Usually, the larger It means that the input parameter has a greater impact on the model output, and vice versa.

[0086] The second set of analytical methods is global sensitivity analysis, which is applicable to continuous or discrete parameter types and is suitable for situations where the number of parameters is less than 5 and is not a weight coefficient. Global sensitivity analysis focuses on the impact of changes in input parameters within the entire definition domain on the model output, and can quantify the interaction between model parameters. Its core is to evaluate the contribution of each model parameter and its combination to the output uncertainty by fully exploring the input space. Global sensitivity analysis is an analytical method based on variance decomposition, which can quantify the contribution of input parameters to the output results (including single variable main effects and interaction effects between variables). By decomposing the variance of the objective function into the contribution of each input parameter and the contribution of the interaction, the impact of the interaction between parameters on the uncertainty of the meteorological and ocean models is quantified.

[0087] Different from the traditional method of directly using the model output results as the evaluation object, the global sensitivity analysis provided in this embodiment focuses on the error measurement of the model results. Specifically, the model output under the original parameter configuration is defined as the benchmark result. By calculating the error between the model output and the benchmark result under different combinations of model parameter values, the degree of influence of parameter fluctuations on the model result error is quantified, and the contribution level of each model parameter and its interaction to the model uncertainty is intuitively characterized.

[0088] The global sensitivity analysis process is:

[0089] Obtain input parameters; perform calculations under the configuration of the input parameters to obtain benchmark results.

[0090] Generate two or more random sample matrices and corresponding interaction matrices based on the input parameters. Input each random sample matrix and interaction matrix into the meteorological ocean model for calculation to obtain random sample output results and interaction sample output results.

[0091] Calculate the first mean square error between the random sample output and the benchmark result, and calculate the second mean square error between the interactive sample output and the benchmark result. Calculate the variance components and sensitivity index of the model parameters based on the first and second mean square errors.

[0092] The influence of each model parameter on the output results is analyzed based on the variance components and sensitivity index.

[0093] Specifically, we first define the problem. The input parameters of the model are determined as ; The model output under the original parameter configuration is the benchmark result .

[0094] The evaluation goal is to quantify the output results and benchmark results of different parameter value combinations The MSE error is used to analyze the contribution of parameter fluctuations to the uncertainty of model results.

[0095] Then, sample generation is performed. To perform variance decomposition, it is necessary to independently generate two or more random sample matrices based on the input parameters. The random sample matrices must ensure uniform coverage of the parameter space; at the same time, the corresponding interaction matrix is ​​also generated.

[0096] In this embodiment, two sets of random sample matrices A and B are generated independently, where the random sample matrix A is an N×n matrix, and each row corresponds to a set of parameters ; The random sample matrix B is an N×n matrix, and each row corresponds to a set of parameters At the same time, the corresponding interaction matrix is ​​generated ,in No. The columns are taken from matrix B and the remaining columns are taken from matrix A.

[0097] The random sample matrix A, random sample matrix B and interaction matrix Input the meteorological ocean model for calculation and obtain the random sample output results , interactive sample output results It is worth noting that the random sample output results 、 , interactive sample output results There are N output values.

[0098] Calculate the first mean square error between random sample matrix A, random sample matrix B and the benchmark results respectively 、 , expressed as:

[0099] ;

[0100] ;

[0101] And, calculate the second mean square error between the interactive sample output result and the benchmark result , expressed as:

[0102] ;

[0103] Mean square error The variance of is the core, and the contribution of the parameters to the error is decomposed. Specifically, the variance components and sensitivity index of the model parameters are calculated based on the first mean square error and the second mean square error. Among them, the variance components include the total error variance, the main effect variance, and the total effect variance; the sensitivity index includes the main effect sensitivity index and the total effect sensitivity index. The calculation expression is:

[0104] ;

[0105] ;

[0106] ;

[0107] ;

[0108] ;

[0109] Where; represents the total error variance; represents the main effect variance; represents the total effect variance; represents the main effect sensitivity index; represents the total effect sensitivity index; represents the set of all mean square errors; Indicates the model parameters.

[0110] The influence of each model parameter on the output result is analyzed based on the variance component and sensitivity index. Among them, the total error variance reflects the overall fluctuation degree of the deviation between the model result and the benchmark value; the main effect variance indicates that only the model parameter The variance of the mean squared error caused by the change, the total effect variance represents the variance of the model parameters The total error variance contribution is related to the main effect. Through these calculations, we can get the proportion of each input parameter's contribution to the result variance. In addition, the larger the main effect sensitivity index, the greater the impact of the model parameter itself on the model results.

[0111] The third set of analytical methods is Monte Carlo uncertainty interval estimation, which is applicable to continuous model parameters and is applicable to situations where the number of model parameters is less than 5. In order to further characterize the uncertainty interval of the model parameters, this embodiment uses Monte Carlo simulation. The model parameters are randomly sampled through Monte Carlo simulation, and the model output is calculated based on each set of model parameters to form an uncertainty interval. This method has obvious advantages over the parameter confidence interval of the traditional regression equation - all parameters change together, rather than independently evaluating the confidence interval of each model parameter. This enables Monte Carlo simulation to fully reflect the interaction between parameters and provide more accurate evaluation.

[0112] The Monte Carlo uncertainty interval estimation process is:

[0113] Obtain input parameters; perform calculations under the configuration of the input parameters to obtain benchmark results.

[0114] Random sampling is performed according to the interval defined by the input parameters to obtain two or more sets of parameter sample combinations; the parameter sample combinations are input into the meteorological ocean model for calculation to obtain a set of simulation output results; then the simulation output result set is screened based on the set threshold to obtain an acceptable parameter value combination.

[0115] Based on the acceptable combination of parameter values, the confidence interval of each model parameter is estimated; based on the confidence interval, the degree of influence of each model parameter on the output results is analyzed.

[0116] Specifically, we first define the problem. The input parameters of the model are determined as ; Give each input parameter an initial value , and determine its initial range; the model output under the original parameter configuration is the benchmark result .

[0117] The evaluation goal is to quantify the model output relative to the benchmark results through random sampling of parameters and multiple simulations. The uncertainty of the parameters is considered and the comprehensive impact of parameter fluctuations on the model error is analyzed.

[0118] Then, parameter sampling is performed. For each model parameter, random sampling is performed according to the interval defined by the input parameters. There are many sampling methods, such as normal distribution, truncated normal distribution, etc. to sample parameters.

[0119] Among them, the normal distribution is: For each model parameter , sampling value Draw from a normal distribution, specifically:

[0120] ;

[0121] in, is the model parameter The original estimate of is the model parameter Standard deviation; is the model parameter The mean is , the standard deviation is Normal distribution.

[0122] Truncated normal distribution: For each model parameter , sampling value Draw from a truncated normal distribution, specifically,

[0123] ;

[0124] in, is the model parameter The original estimate of is the model parameter The standard deviation, calculated from the width of the confidence interval; and are the upper and lower bounds of the parameter.

[0125] Through random sampling in the above way, we generated A set of independent parameter sample combinations , each set of parameter sample combinations contains a possible value combination of all input parameters.

[0126] Combine parameter samples Input into the meteorological ocean model for calculation and evaluation with benchmark results There are two cases here: if it is a classification model (output is discrete value), compare the classification accuracy of the output results under the original parameters with the classification accuracy of the output results of the current various simulation parameter value combinations; if it is a numerical prediction model, calculate the relative root mean square error between the output results of the current various parameter value combinations and the output results of the original parameter combination.

[0127] The model is repeatedly executed through Monte Carlo simulation to estimate its uncertainty. During each execution, the model calculations are performed using a set of parameter value combinations, and the output is recorded. In this example, 5000 simulations are performed to obtain a set of simulation output results. Then, a threshold is set based on accuracy or relative root mean square error (RMSE). The simulation output results are filtered based on the set threshold to obtain acceptable parameter value combinations.

[0128] Based on all acceptable combinations of parameter values, the confidence interval of each model parameter is estimated, the mean and standard deviation of each model parameter are calculated, and the corresponding confidence interval is calculated based on the normal distribution.

[0129] The confidence interval formula is:

[0130] ;

[0131] in, is the sample mean; is the sample standard deviation; is the critical value of the standard normal distribution.

[0132] According to the results of Monte Carlo simulation, the uncertainty interval of the parameter can be obtained. The wider the interval, the smaller the impact of the fluctuation of the parameter on the model results and the greater the uncertainty.

[0133] Monte Carlo simulations, by repeatedly sampling different combinations of model parameters, comprehensively assess the uncertainty of the model's output and provide statistical estimates for the confidence intervals of the parameters. The advantage of this method is that it simultaneously considers the interactions of all parameters, rather than evaluating the impact of each parameter independently, resulting in a more realistic reflection of the model's behavior.

[0134] It can be seen that compared with traditional methods (such as Morris and Sobol) that directly use model output results as evaluation objects and cannot adapt to scenarios without measured data, the present invention focuses on the uncertainty assessment of meteorological and ocean model parameter fluctuations, takes the benchmark parameter output as the benchmark value, and quantifies the impact of parameter fluctuations by calculating the error between the model output and the benchmark value, effectively solving the evaluation problem without measured data.

[0135] During the specific implementation of step 204, the impact assessment is primarily focused on the uncertainty of the meteorological and oceanographic model method. Meteorological and oceanographic model calculations involve model simplification, optimization, and approximation. The omission of important factors and computational errors can impact the model's reliability and accuracy, increasing the model's uncertainty risk. The primary function of this component is to establish a system for assessing the model's inherent stability and uncertainty. The following two scalable analysis methods are provided for different input and output types and model scenarios. These methods can be used individually or in combination, depending on actual needs, to cover the entire process of assessing model method uncertainty.

[0136] The first set of analyzable methods is rank stability analysis. For meteorological and oceanographic models, rank stability analysis is a key study in model stability assessment, exploring the overall uncertainty of the model output response when considering input uncertainty.

[0137] The rank stability analysis process is:

[0138] Set the correlation coefficient threshold to determine whether the correlation coefficient between the original output variable rank and the original input variable rank meets the correlation coefficient threshold. If so, the corresponding original input variable is regarded as a key variable.

[0139] The rank stability analysis method is simulated based on key variables, and the new output variable rank is obtained by random perturbation; then the average stability index and the average rank reversal rate are calculated according to the new output variable rank; based on the average stability index and the average rank reversal rate, the degree of influence of the model method on the meteorological ocean model in terms of ranking stability is analyzed.

[0140] Specifically, meteorological ocean models often contain multiple variables, such as wind speed, air pressure, temperature, humidity and other factors, which are recorded as , calculate the specified output variable (such as rainfall) through the Spearman correlation coefficient With input variables The correlation between them is calculated as:

[0141] ;

[0142] in, Indicates the Original input variables With the specified output variable The correlation coefficient of Indicates the The rank of the original input variables, represents the rank of the original output variable, Indicates the The average rank of the original input variables, Represents the average rank of the original output variables.

[0143] According to the Spearman correlation coefficient Identify the input variables that have a greater impact on the output rank and define the correlation coefficient threshold , when the correlation coefficient If the correlation coefficient threshold requirement is met, the corresponding original input variables are regarded as key variables, which are the main risk sources for model stability.

[0144] For the identified key variables, simulation analysis is performed to change their ranks and observe the changes in the ranks of the output results. The specific perturbation method is to randomly swap the ranks of 5% of the data for each key variable, and perform 100 random perturbations to obtain 100 new output variable ranks. .

[0145] The indicator for quantitative stability is the rank stability index. The mean stability index (MRSI) is calculated based on the rank of the new output variable. The expression is:

[0146] ;

[0147] Where, represents the rank of the original output variable; Indicates the A new output variable rank; Indicates the new output variable rank number.

[0148] MRSI is used to measure the stability of model methods in different sorting scenarios. Its value range is [-1, 1]. The closer it is to 1, the more stable it is, indicating that the model method is insensitive to input fluctuations and has low uncertainty.

[0149] Considering the properties of MRSI itself, when there are multiple model methods, we consider the MRSI obtained under the same perturbation and compare the uncertainty of each model method as the relative value of MRSI. In the case of a single model method, in order to comprehensively measure the stability of the model, we consider another indicator, namely, the mean rank reversal rate (MRRR), which is expressed as:

[0150] ;

[0151] Where, represents the rank of the original output variable; Indicates the A new output variable rank; Indicates the sample size of the specified output variable.

[0152] MRRR is used to measure the proportion of samples whose output rank after disturbance is lower than the original rank. Its value range is [0, 1]. A high rank reversal rate indicates that a small disturbance of the input variable will cause a significant change in the output results of some areas.

[0153] The rank stability index evaluates the consistency of the ranking as a whole and is suitable for macro-stability analysis; the rank reversal rate focuses on drastic changes in the local ranking and is suitable for identifying sensitive areas of the model. When the rank stability index is high and the rank reversal rate is low, it indicates that the model is very robust; when the rank stability index is high and the rank reversal rate is high, it indicates that the model is stable at the macro level but unstable at the local level; when the rank stability index is low and the rank reversal rate is high, it indicates that the model is extremely unstable; when the rank stability index is low and the rank reversal rate is low, this situation is relatively rare, which means that although the overall ranking is disrupted, few samples are "degraded".

[0154] Combining the definitions and properties of MRSI and MRRR, and considering the evaluation criteria of the Spearman correlation coefficient, which is defined in the rank stability analysis method, the relationship between the absolute measurement values ​​of the two indicators and uncertainty is evaluated from the perspective of ranking stability, as shown in Table 1.

[0155] Table 1 Relationship between absolute measurement value and uncertainty

[0156]

[0157] The second set of analytical methods is sensitivity analysis. Uncertainty analysis also requires sensitivity analysis of input variables. For meteorological and oceanographic models with multiple input variables, the variables are first preliminarily screened. The impact of each perturbation on the output can be analyzed one by one, and the variables with the greatest impact on the output are selected as key variables. Key variables are the main source of risk for the model under perturbation. If such variables are uncertain, it is easy to cause drastic changes in the output and reduce the stability of the model. Subsequent perturbation of only the key variables can reduce the computational complexity of the method under conditions close to reality. For meteorological and oceanographic models with fewer input variables, a full variable perturbation method is adopted. In the evaluation, uncertainty evaluation indicators are used to quantify the uncertainty of the method.

[0158] The sensitivity analysis process is as follows:

[0159] By perturbing the original input variables one by one and analyzing their impact on the original output results, key variables are screened out; the perturbed key variables are perturbed to obtain the perturbed output results. , apply disturbances one by one (the default amplitude is 5%), and calculate the difference between the disturbance output and the original output ,according to From largest to smallest, mark ,make , No. p The contribution rate of the item is ,forward m The cumulative contribution rate is , find the smallest M such that , For a given threshold (the default value is 0.5), M key variables are finally selected.

[0160] Uncertainty measurement indicators and uncertainty evaluation indicators are calculated based on the perturbation output results and the original output results; the degree of influence of the model method on the output results of the meteorological ocean model is evaluated through uncertainty measurement indicators and uncertainty evaluation indicators.

[0161] Specifically, for the meteorological ocean model, for the specified output variable , define the original output of the model as Next, we perturb the key variables. If the uncertainty of the model is low, then when the variables are slightly perturbed, the model output will not change significantly. So we perturb the key variables in the model 100 times, and the perturbation output is .

[0162] Calculate uncertainty metrics based on the perturbed output and the original output , expressed as:

[0163] ;

[0164] In order to solve the dimensionality problem caused by the numerical size of different output variables, a more reliable uncertainty evaluation index is defined - the uncertainty evaluation index , expressed as:

[0165] ；

[0166] in, Indicates the sample size included in the specified output variable; The original output of the model The standard deviation of .

[0167] Comprehensive uncertainty evaluation index In the case of multiple model methods, consider the obtained , compare the uncertainty of each model method, as In the case of a single model approach, combined with uncertainty metrics The evaluation criteria, from the perspective of numerical accuracy, are defined as The absolute measurement values ​​are shown in Table 2.

[0168] Table 2 The relationship between the absolute measurement value and uncertainty

[0169]

[0170] Uncertainty evaluation indicators It reflects the uncertainty of the model method. The higher the index is, the more flaws the model has in theory and the higher the uncertainty is.

[0171] For meteorological and oceanographic models, uncertainty measurement can be based on the evaluation results of the model and define the level of the model's original output as , make S small perturbations to the model, and its output level is , then the uncertainty evaluation index Expressed as:

[0172] ;

[0173] Comprehensive uncertainty evaluation index In the case of multiple model methods, consider the obtained , compare the uncertainty of each model method, as In the case of a single model method, the relative value of The relationship between the absolute measurement value and uncertainty is shown in Table 3.

[0174] Table 3 The relationship between the absolute measurement value and uncertainty

[0175]

[0176] It should be understood that although this embodiment Figure 1 The steps in the diagram are shown in the order indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0177] It can be seen that the present invention addresses the stability issues of meteorological and ocean models caused by simplification and approximation, and proposes innovative indicators such as the rank stability index (MRSI), rank reversal rate (MRRR) and normalized residual error (NRMSE), which quantify the uncertainty of the model from multiple dimensions such as ranking stability, numerical accuracy and classification consistency.

[0178] Example 2

[0179] Based on the uncertainty assessment method of the meteorological ocean model in Example 1, this embodiment discloses a meteorological ocean model uncertainty assessment device, such as Figure 3 As shown, the meteorological ocean model uncertainty assessment device includes: an assessment factor determination module 401, an observation data uncertainty assessment module 402, a model parameter uncertainty assessment module 403 and a model method uncertainty assessment module 404, wherein:

[0180] The evaluation factor determination module 401 is used to evaluate the impact on the meteorological ocean model based on the uncertainty of observation data, the uncertainty of model parameters and the uncertainty of model method.

[0181] The observation data uncertainty assessment module 402 is used to use resampling technology to generate multiple samples through repeated sampling to construct an empirical distribution when assessing the impact of observation data uncertainty on the meteorological and oceanographic model, so as to quantify the impact of observation data fluctuations on the meteorological and oceanographic model;

[0182] The model parameter uncertainty assessment module 403 is used to assess the impact of model parameter uncertainty on the meteorological and oceanographic model by using any one or more combinations of local sensitivity analysis, global sensitivity analysis, and Monte Carlo uncertainty interval estimation. The model output results obtained under the input parameters are used as the baseline results. By comparing the model output results after the model parameter fluctuations with the baseline results, the degree of impact of the model parameter fluctuations on the meteorological and oceanographic model is determined.

[0183] The model method uncertainty assessment module 404 is used to evaluate the impact of model method uncertainty on the meteorological ocean model by using any one or more combinations of rank stability analysis method and sensitivity analysis method, and by comparing the model output results obtained after the input variables are disturbed with the original output results to determine the degree of influence of the model method on the meteorological ocean model output results when the input variables are disturbed.

[0184] In this embodiment, the specific working process and working principle of the evaluation factor determination module 401, the observation data uncertainty assessment module 402, the model parameter uncertainty assessment module 403 and the model method uncertainty assessment module 404 are the same as those in Example 1, and therefore will not be described in detail in this embodiment. Each unit module can be implemented in whole or in part by software, hardware or a combination thereof. Each unit module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above unit modules.

[0185] Example 3

[0186] like Figure 4 The terminal device disclosed in this embodiment includes a transmitter, a receiver, a memory, and a processor. The transmitter is used to send instructions and data, the receiver is used to receive instructions and data, the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions stored in the memory to implement the method in the above-mentioned embodiment 1.

[0187] It should be noted that the above memory can be independent or integrated with the processor. When the memory is independently provided, the terminal device further includes a bus for connecting the memory and the processor.

[0188] Example 4

[0189] This embodiment discloses a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the method in the above-mentioned embodiment 1 is implemented.

[0190] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0191] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0192] The above-described embodiments merely illustrate several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that variations and improvements are possible without departing from the spirit of the present invention, and all such variations and improvements fall within the scope of protection of the present invention.

Claims

1. A method for uncertainty assessment of meteorological and oceanographic models, characterized in that: The method includes: evaluating the impact on the meteorological ocean model based on observational data uncertainty, model parameter uncertainty, and model method uncertainty; When evaluating the impact of the uncertainty of the observational data on the meteorological ocean model, a resampling technique is used to generate multiple samples through repeated sampling to construct an empirical distribution to quantify the degree of impact of the fluctuation of the observational data on the meteorological ocean model; When evaluating the impact of the model parameter uncertainty on the meteorological-ocean model, any one or more combinations of local sensitivity analysis, global sensitivity analysis, and Monte Carlo uncertainty interval estimation are used, and the model output results obtained under the input parameters are used as the baseline results. By comparing the model output results after the model parameter fluctuations with the baseline results, the degree of impact of the model parameter fluctuations on the meteorological-ocean model is determined; When evaluating the impact of the uncertainty of the model method on the meteorological-ocean model, any one or more combinations of rank stability analysis and sensitivity analysis are used to compare the model output results obtained after the input variables are disturbed with the original output results to determine the degree of impact of the model method on the meteorological-ocean model output results when the input variables are disturbed.

2. The uncertainty assessment method of the meteorological ocean model according to claim 1, characterized in that: The sources of uncertainty in the observation data include: sampling errors due to errors in the measuring instruments themselves, indirect measurement errors, human recording errors and insufficient sample space coverage, errors caused by the adaptability of the observed variables and the model structure, and errors caused by the random distribution characteristics of the observed variables.

3. The uncertainty assessment method of meteorological and ocean models according to claim 1, characterized in that: When using local sensitivity analysis, it includes: Obtain input parameters; set the number of parameter levels based on the input parameters and calculate the step size to generate two or more independent tracks; On each independent track, a reference point is determined; a single parameter of the reference point is perturbed to obtain a perturbation point; For each independent track, the reference point and the disturbance point are respectively input into the meteorological ocean model for calculation to obtain a reference result and a disturbance result; Calculating a mean square error between the benchmark result and the perturbation result, calculating a basic effect of each model parameter based on the mean square error, and then calculating a mean absolute basic effect of each model parameter based on the basic effect; The degree of influence of each model parameter on the output result is judged based on the mean absolute basic effect.

4. The uncertainty assessment method of meteorological and ocean models according to claim 1, characterized in that: When using global sensitivity analysis, it includes: Obtain input parameters; perform calculations under the configuration of the input parameters to obtain benchmark results; generating two or more random sample matrices and corresponding interaction matrices based on the input parameters; Inputting each of the random sample matrices and the interaction matrix into a meteorological ocean model for calculation to obtain a random sample output result and an interaction sample output result; Calculating a first mean square error between the random sample output result and a benchmark result, and calculating a second mean square error between the interactive sample output result and the benchmark result; Calculate the variance component and sensitivity index of the model parameters according to the first mean square error and the second mean square error; The degree of influence of each model parameter on the output result is analyzed based on the variance component and the sensitivity index.

5. The uncertainty assessment method of meteorological and ocean models according to claim 1, characterized in that: When using Monte Carlo uncertainty interval estimation, including: Obtain input parameters; perform calculations under the configuration of the input parameters to obtain benchmark results; Random sampling is performed according to the interval defined by the input parameters to obtain two or more sets of parameter sample combinations; different combinations of the parameter sample combinations are input into the meteorological ocean model for calculation to obtain a set of simulation output results; and the set of simulation output results is then screened based on a set threshold to obtain an acceptable combination of parameter values; Estimate the confidence interval of each model parameter based on the acceptable parameter value combinations; The degree of influence of each model parameter on the output result is analyzed based on the confidence interval.

6. The uncertainty assessment method of meteorological and ocean models according to claim 1, characterized in that: When using the rank stability analysis method, it includes: Setting a correlation coefficient threshold to determine whether the correlation coefficient between the original output variable rank and the original input variable rank meets the correlation coefficient threshold; if so, the corresponding original input variable is considered a key variable; A rank stability analysis method is simulated based on the key variables, and a new output variable rank is obtained by random perturbation; then the average stability index and the average rank reversal rate are calculated according to the new output variable rank; based on the average stability index and the average rank reversal rate, the degree of influence of the model method on the meteorological ocean model in terms of ranking stability is analyzed.

7. The method for uncertainty assessment of meteorological and oceanic models according to claim 1, wherein: When using sensitivity analysis, it includes: By perturbing the original input variables one by one and analyzing their effects on the original output results, key variables are screened out; the key variables are perturbed to obtain perturbed output results; An uncertainty measurement index and an uncertainty evaluation index are calculated based on the disturbance output result and the original output result; and the degree of influence of the model method on the output result of the meteorological ocean model is evaluated by using the uncertainty measurement index and the uncertainty evaluation index.

8. A meteorological ocean model uncertainty assessment device, characterized in that: The device comprises: An assessment factor determination module is used to assess the impact on the meteorological ocean model based on the uncertainty of observation data, model parameter uncertainty and model method uncertainty; An observation data uncertainty assessment module is used to use resampling technology to generate multiple samples through repeated sampling to construct an empirical distribution when assessing the impact of the observation data uncertainty on the meteorological ocean model, so as to quantify the impact of the observation data fluctuation on the meteorological ocean model; A model parameter uncertainty assessment module is used to use any one or more combinations of local sensitivity analysis, global sensitivity analysis, and Monte Carlo uncertainty interval estimation when assessing the impact of the model parameter uncertainty on the meteorological ocean model, using the model output results obtained under the input parameters as the baseline results, and determining the degree of impact of the model parameter fluctuation on the meteorological ocean model by comparing the model output results after the model parameter fluctuation with the baseline results; The model method uncertainty assessment module is used to use any one or more combinations of rank stability analysis method and sensitivity analysis method when evaluating the impact of the model method uncertainty on the meteorological ocean model, and to compare the model output results obtained after the input variables are disturbed with the original output results to determine the degree of influence of the model method on the meteorological ocean model output results when the input variables are disturbed.

9. A computer device comprising a memory and a processor, characterized in that: The memory stores a computer program, and when the processor executes the computer program, the steps of the meteorological ocean model uncertainty assessment method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the uncertainty assessment method of the meteorological ocean model according to any one of claims 1 to 7 are implemented.

Images